xref: /dragonfly/sys/kern/vfs_syscalls.c (revision ed183f8c)
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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
35  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/buf.h>
41 #include <sys/conf.h>
42 #include <sys/sysent.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mountctl.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/linker.h>
52 #include <sys/stat.h>
53 #include <sys/unistd.h>
54 #include <sys/vnode.h>
55 #include <sys/proc.h>
56 #include <sys/priv.h>
57 #include <sys/jail.h>
58 #include <sys/namei.h>
59 #include <sys/nlookup.h>
60 #include <sys/dirent.h>
61 #include <sys/extattr.h>
62 #include <sys/spinlock.h>
63 #include <sys/kern_syscall.h>
64 #include <sys/objcache.h>
65 #include <sys/sysctl.h>
66 
67 #include <sys/buf2.h>
68 #include <sys/file2.h>
69 #include <sys/spinlock2.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 
75 #include <machine/limits.h>
76 #include <machine/stdarg.h>
77 
78 static void mount_warning(struct mount *mp, const char *ctl, ...)
79 		__printflike(2, 3);
80 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
81 static int checkvp_chdir (struct vnode *vn, struct thread *td);
82 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
83 static int chroot_refuse_vdir_fds (thread_t td, struct filedesc *fdp);
84 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
85 static int getutimes (struct timeval *, struct timespec *);
86 static int getutimens (const struct timespec *, struct timespec *, int *);
87 static int setfown (struct mount *, struct vnode *, uid_t, gid_t);
88 static int setfmode (struct vnode *, int);
89 static int setfflags (struct vnode *, u_long);
90 static int setutimes (struct vnode *, struct vattr *,
91 			const struct timespec *, int);
92 
93 static int	usermount = 0;	/* if 1, non-root can mount fs. */
94 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
95     "Allow non-root users to mount filesystems");
96 
97 static int	debug_unmount = 0; /* if 1 loop until unmount success */
98 SYSCTL_INT(_vfs, OID_AUTO, debug_unmount, CTLFLAG_RW, &debug_unmount, 0,
99     "Stall failed unmounts in loop");
100 /*
101  * Virtual File System System Calls
102  */
103 
104 /*
105  * Mount a file system.
106  *
107  * mount_args(char *type, char *path, int flags, caddr_t data)
108  *
109  * MPALMOSTSAFE
110  */
111 int
112 sys_mount(struct mount_args *uap)
113 {
114 	struct thread *td = curthread;
115 	struct vnode *vp;
116 	struct nchandle nch;
117 	struct mount *mp, *nullmp;
118 	struct vfsconf *vfsp;
119 	int error, flag = 0, flag2 = 0;
120 	int hasmount;
121 	struct vattr va;
122 	struct nlookupdata nd;
123 	char fstypename[MFSNAMELEN];
124 	struct ucred *cred;
125 
126 	cred = td->td_ucred;
127 	if (jailed(cred)) {
128 		error = EPERM;
129 		goto done;
130 	}
131 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
132 		goto done;
133 
134 	/*
135 	 * Do not allow NFS export by non-root users.
136 	 */
137 	if (uap->flags & MNT_EXPORTED) {
138 		error = priv_check(td, PRIV_ROOT);
139 		if (error)
140 			goto done;
141 	}
142 	/*
143 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
144 	 */
145 	if (priv_check(td, PRIV_ROOT))
146 		uap->flags |= MNT_NOSUID | MNT_NODEV;
147 
148 	/*
149 	 * Lookup the requested path and extract the nch and vnode.
150 	 */
151 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
152 	if (error == 0) {
153 		if ((error = nlookup(&nd)) == 0) {
154 			if (nd.nl_nch.ncp->nc_vp == NULL)
155 				error = ENOENT;
156 		}
157 	}
158 	if (error) {
159 		nlookup_done(&nd);
160 		goto done;
161 	}
162 
163 	/*
164 	 * If the target filesystem is resolved via a nullfs mount, then
165 	 * nd.nl_nch.mount will be pointing to the nullfs mount structure
166 	 * instead of the target file system. We need it in case we are
167 	 * doing an update.
168 	 */
169 	nullmp = nd.nl_nch.mount;
170 
171 	/*
172 	 * Extract the locked+refd ncp and cleanup the nd structure
173 	 */
174 	nch = nd.nl_nch;
175 	cache_zero(&nd.nl_nch);
176 	nlookup_done(&nd);
177 
178 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
179 	    (mp = cache_findmount(&nch)) != NULL) {
180 		cache_dropmount(mp);
181 		hasmount = 1;
182 	} else {
183 		hasmount = 0;
184 	}
185 
186 
187 	/*
188 	 * now we have the locked ref'd nch and unreferenced vnode.
189 	 */
190 	vp = nch.ncp->nc_vp;
191 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
192 		cache_put(&nch);
193 		goto done;
194 	}
195 	cache_unlock(&nch);
196 
197 	/*
198 	 * Extract the file system type. We need to know this early, to take
199 	 * appropriate actions if we are dealing with a nullfs.
200 	 */
201         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
202                 cache_drop(&nch);
203                 vput(vp);
204 		goto done;
205         }
206 
207 	/*
208 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
209 	 */
210 	if (uap->flags & MNT_UPDATE) {
211 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
212 			cache_drop(&nch);
213 			vput(vp);
214 			error = EINVAL;
215 			goto done;
216 		}
217 
218 		if (strncmp(fstypename, "null", 5) == 0) {
219 			KKASSERT(nullmp);
220 			mp = nullmp;
221 		} else {
222 			mp = vp->v_mount;
223 		}
224 
225 		flag = mp->mnt_flag;
226 		flag2 = mp->mnt_kern_flag;
227 		/*
228 		 * We only allow the filesystem to be reloaded if it
229 		 * is currently mounted read-only.
230 		 */
231 		if ((uap->flags & MNT_RELOAD) &&
232 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
233 			cache_drop(&nch);
234 			vput(vp);
235 			error = EOPNOTSUPP;	/* Needs translation */
236 			goto done;
237 		}
238 		/*
239 		 * Only root, or the user that did the original mount is
240 		 * permitted to update it.
241 		 */
242 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
243 		    (error = priv_check(td, PRIV_ROOT))) {
244 			cache_drop(&nch);
245 			vput(vp);
246 			goto done;
247 		}
248 		if (vfs_busy(mp, LK_NOWAIT)) {
249 			cache_drop(&nch);
250 			vput(vp);
251 			error = EBUSY;
252 			goto done;
253 		}
254 		if (hasmount) {
255 			cache_drop(&nch);
256 			vfs_unbusy(mp);
257 			vput(vp);
258 			error = EBUSY;
259 			goto done;
260 		}
261 		mp->mnt_flag |=
262 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
263 		lwkt_gettoken(&mp->mnt_token);
264 		vn_unlock(vp);
265 		vfsp = mp->mnt_vfc;
266 		goto update;
267 	}
268 
269 	/*
270 	 * If the user is not root, ensure that they own the directory
271 	 * onto which we are attempting to mount.
272 	 */
273 	if ((error = VOP_GETATTR(vp, &va)) ||
274 	    (va.va_uid != cred->cr_uid &&
275 	     (error = priv_check(td, PRIV_ROOT)))) {
276 		cache_drop(&nch);
277 		vput(vp);
278 		goto done;
279 	}
280 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
281 		cache_drop(&nch);
282 		vput(vp);
283 		goto done;
284 	}
285 	if (vp->v_type != VDIR) {
286 		cache_drop(&nch);
287 		vput(vp);
288 		error = ENOTDIR;
289 		goto done;
290 	}
291 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
292 		cache_drop(&nch);
293 		vput(vp);
294 		error = EPERM;
295 		goto done;
296 	}
297 	vfsp = vfsconf_find_by_name(fstypename);
298 	if (vfsp == NULL) {
299 		linker_file_t lf;
300 
301 		/* Only load modules for root (very important!) */
302 		if ((error = priv_check(td, PRIV_ROOT)) != 0) {
303 			cache_drop(&nch);
304 			vput(vp);
305 			goto done;
306 		}
307 		error = linker_load_file(fstypename, &lf);
308 		if (error || lf == NULL) {
309 			cache_drop(&nch);
310 			vput(vp);
311 			if (lf == NULL)
312 				error = ENODEV;
313 			goto done;
314 		}
315 		lf->userrefs++;
316 		/* lookup again, see if the VFS was loaded */
317 		vfsp = vfsconf_find_by_name(fstypename);
318 		if (vfsp == NULL) {
319 			lf->userrefs--;
320 			linker_file_unload(lf);
321 			cache_drop(&nch);
322 			vput(vp);
323 			error = ENODEV;
324 			goto done;
325 		}
326 	}
327 	if (hasmount) {
328 		cache_drop(&nch);
329 		vput(vp);
330 		error = EBUSY;
331 		goto done;
332 	}
333 
334 	/*
335 	 * Allocate and initialize the filesystem.
336 	 */
337 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
338 	mount_init(mp);
339 	vfs_busy(mp, LK_NOWAIT);
340 	mp->mnt_op = vfsp->vfc_vfsops;
341 	mp->mnt_vfc = vfsp;
342 	mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT;
343 	vfsp->vfc_refcount++;
344 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
345 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
346 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
347 	mp->mnt_stat.f_owner = cred->cr_uid;
348 	lwkt_gettoken(&mp->mnt_token);
349 	vn_unlock(vp);
350 update:
351 	/*
352 	 * (per-mount token acquired at this point)
353 	 *
354 	 * Set the mount level flags.
355 	 */
356 	if (uap->flags & MNT_RDONLY)
357 		mp->mnt_flag |= MNT_RDONLY;
358 	else if (mp->mnt_flag & MNT_RDONLY)
359 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
360 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
361 	    MNT_SYNCHRONOUS | MNT_ASYNC | MNT_NOATIME |
362 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
363 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
364 	    MNT_AUTOMOUNTED);
365 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
366 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_ASYNC | MNT_FORCE |
367 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
368 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
369 	    MNT_AUTOMOUNTED);
370 
371 	/*
372 	 * Pre-set the mount's ALL_MPSAFE flags if specified in the vfsconf.
373 	 * This way the initial VFS_MOUNT() call will also be MPSAFE.
374 	 */
375 	if (vfsp->vfc_flags & VFCF_MPSAFE)
376 		mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
377 
378 	/*
379 	 * Mount the filesystem.
380 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
381 	 * get.
382 	 */
383 	if (mp->mnt_flag & MNT_UPDATE) {
384 		error = VFS_MOUNT(mp, uap->path, uap->data, cred);
385 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
386 			mp->mnt_flag &= ~MNT_RDONLY;
387 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
388 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
389 		if (error) {
390 			mp->mnt_flag = flag;
391 			mp->mnt_kern_flag = flag2;
392 		}
393 		lwkt_reltoken(&mp->mnt_token);
394 		vfs_unbusy(mp);
395 		vrele(vp);
396 		cache_drop(&nch);
397 		goto done;
398 	}
399 	mp->mnt_ncmounton = nch;
400 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
401 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
402 
403 	/*
404 	 * Put the new filesystem on the mount list after root.  The mount
405 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
406 	 * up) which represents the root of the mount.  The lookup code
407 	 * detects the mount point going forward and checks the root of
408 	 * the mount going backwards.
409 	 *
410 	 * It is not necessary to invalidate or purge the vnode underneath
411 	 * because elements under the mount will be given their own glue
412 	 * namecache record.
413 	 */
414 	if (!error) {
415 		if (mp->mnt_ncmountpt.ncp == NULL) {
416 			/*
417 			 * Allocate, then unlock, but leave the ref intact.
418 			 * This is the mnt_refs (1) that we will retain
419 			 * through to the unmount.
420 			 */
421 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
422 			cache_unlock(&mp->mnt_ncmountpt);
423 		}
424 		vn_unlock(vp);
425 		cache_lock(&nch);
426 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
427 		cache_unlock(&nch);
428 		cache_ismounting(mp);
429 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
430 
431 		mountlist_insert(mp, MNTINS_LAST);
432 		vn_unlock(vp);
433 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
434 		error = vfs_allocate_syncvnode(mp);
435 		lwkt_reltoken(&mp->mnt_token);
436 		vfs_unbusy(mp);
437 		error = VFS_START(mp, 0);
438 		vrele(vp);
439 		KNOTE(&fs_klist, VQ_MOUNT);
440 	} else {
441 		bzero(&mp->mnt_ncmounton, sizeof(mp->mnt_ncmounton));
442 		vn_syncer_thr_stop(mp);
443 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
444 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
445 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
446 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
447 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
448 		mp->mnt_vfc->vfc_refcount--;
449 		lwkt_reltoken(&mp->mnt_token);
450 		vfs_unbusy(mp);
451 		kfree(mp, M_MOUNT);
452 		cache_drop(&nch);
453 		vput(vp);
454 	}
455 done:
456 	return (error);
457 }
458 
459 /*
460  * Scan all active processes to see if any of them have a current
461  * or root directory onto which the new filesystem has just been
462  * mounted. If so, replace them with the new mount point.
463  *
464  * Both old_nch and new_nch are ref'd on call but not locked.
465  * new_nch must be temporarily locked so it can be associated with the
466  * vnode representing the root of the mount point.
467  */
468 struct checkdirs_info {
469 	struct nchandle old_nch;
470 	struct nchandle new_nch;
471 	struct vnode *old_vp;
472 	struct vnode *new_vp;
473 };
474 
475 static int checkdirs_callback(struct proc *p, void *data);
476 
477 static void
478 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
479 {
480 	struct checkdirs_info info;
481 	struct vnode *olddp;
482 	struct vnode *newdp;
483 	struct mount *mp;
484 
485 	/*
486 	 * If the old mount point's vnode has a usecount of 1, it is not
487 	 * being held as a descriptor anywhere.
488 	 */
489 	olddp = old_nch->ncp->nc_vp;
490 	if (olddp == NULL || VREFCNT(olddp) == 1)
491 		return;
492 
493 	/*
494 	 * Force the root vnode of the new mount point to be resolved
495 	 * so we can update any matching processes.
496 	 */
497 	mp = new_nch->mount;
498 	if (VFS_ROOT(mp, &newdp))
499 		panic("mount: lost mount");
500 	vn_unlock(newdp);
501 	cache_lock(new_nch);
502 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
503 	cache_setunresolved(new_nch);
504 	cache_setvp(new_nch, newdp);
505 	cache_unlock(new_nch);
506 
507 	/*
508 	 * Special handling of the root node
509 	 */
510 	if (rootvnode == olddp) {
511 		vref(newdp);
512 		vfs_cache_setroot(newdp, cache_hold(new_nch));
513 	}
514 
515 	/*
516 	 * Pass newdp separately so the callback does not have to access
517 	 * it via new_nch->ncp->nc_vp.
518 	 */
519 	info.old_nch = *old_nch;
520 	info.new_nch = *new_nch;
521 	info.new_vp = newdp;
522 	allproc_scan(checkdirs_callback, &info, 0);
523 	vput(newdp);
524 }
525 
526 /*
527  * NOTE: callback is not MP safe because the scanned process's filedesc
528  * structure can be ripped out from under us, amoung other things.
529  */
530 static int
531 checkdirs_callback(struct proc *p, void *data)
532 {
533 	struct checkdirs_info *info = data;
534 	struct filedesc *fdp;
535 	struct nchandle ncdrop1;
536 	struct nchandle ncdrop2;
537 	struct vnode *vprele1;
538 	struct vnode *vprele2;
539 
540 	if ((fdp = p->p_fd) != NULL) {
541 		cache_zero(&ncdrop1);
542 		cache_zero(&ncdrop2);
543 		vprele1 = NULL;
544 		vprele2 = NULL;
545 
546 		/*
547 		 * MPUNSAFE - XXX fdp can be pulled out from under a
548 		 * foreign process.
549 		 *
550 		 * A shared filedesc is ok, we don't have to copy it
551 		 * because we are making this change globally.
552 		 */
553 		spin_lock(&fdp->fd_spin);
554 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
555 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
556 			vprele1 = fdp->fd_cdir;
557 			vref(info->new_vp);
558 			fdp->fd_cdir = info->new_vp;
559 			ncdrop1 = fdp->fd_ncdir;
560 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
561 		}
562 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
563 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
564 			vprele2 = fdp->fd_rdir;
565 			vref(info->new_vp);
566 			fdp->fd_rdir = info->new_vp;
567 			ncdrop2 = fdp->fd_nrdir;
568 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
569 		}
570 		spin_unlock(&fdp->fd_spin);
571 		if (ncdrop1.ncp)
572 			cache_drop(&ncdrop1);
573 		if (ncdrop2.ncp)
574 			cache_drop(&ncdrop2);
575 		if (vprele1)
576 			vrele(vprele1);
577 		if (vprele2)
578 			vrele(vprele2);
579 	}
580 	return(0);
581 }
582 
583 /*
584  * Unmount a file system.
585  *
586  * Note: unmount takes a path to the vnode mounted on as argument,
587  * not special file (as before).
588  *
589  * umount_args(char *path, int flags)
590  *
591  * MPALMOSTSAFE
592  */
593 int
594 sys_unmount(struct unmount_args *uap)
595 {
596 	struct thread *td = curthread;
597 	struct proc *p __debugvar = td->td_proc;
598 	struct mount *mp = NULL;
599 	struct nlookupdata nd;
600 	int error;
601 
602 	KKASSERT(p);
603 	if (td->td_ucred->cr_prison != NULL) {
604 		error = EPERM;
605 		goto done;
606 	}
607 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
608 		goto done;
609 
610 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE,
611 			     NLC_FOLLOW | NLC_IGNBADDIR);
612 	if (error == 0)
613 		error = nlookup(&nd);
614 	if (error)
615 		goto out;
616 
617 	mp = nd.nl_nch.mount;
618 
619 	/*
620 	 * Only root, or the user that did the original mount is
621 	 * permitted to unmount this filesystem.
622 	 */
623 	if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
624 	    (error = priv_check(td, PRIV_ROOT)))
625 		goto out;
626 
627 	/*
628 	 * Don't allow unmounting the root file system.
629 	 */
630 	if (mp->mnt_flag & MNT_ROOTFS) {
631 		error = EINVAL;
632 		goto out;
633 	}
634 
635 	/*
636 	 * Must be the root of the filesystem
637 	 */
638 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
639 		error = EINVAL;
640 		goto out;
641 	}
642 
643 	/*
644 	 * If no error try to issue the unmount.  We lose our cache
645 	 * ref when we call nlookup_done so we must hold the mount point
646 	 * to prevent use-after-free races.
647 	 */
648 out:
649 	if (error == 0) {
650 		mount_hold(mp);
651 		nlookup_done(&nd);
652 		error = dounmount(mp, uap->flags, 0);
653 		mount_drop(mp);
654 	} else {
655 		nlookup_done(&nd);
656 	}
657 done:
658 	return (error);
659 }
660 
661 /*
662  * Do the actual file system unmount (interlocked against the mountlist
663  * token and mp->mnt_token).
664  */
665 static int
666 dounmount_interlock(struct mount *mp)
667 {
668 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
669 		return (EBUSY);
670 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
671 	return(0);
672 }
673 
674 static int
675 unmount_allproc_cb(struct proc *p, void *arg)
676 {
677 	struct mount *mp;
678 
679 	if (p->p_textnch.ncp == NULL)
680 		return 0;
681 
682 	mp = (struct mount *)arg;
683 	if (p->p_textnch.mount == mp)
684 		cache_drop(&p->p_textnch);
685 
686 	return 0;
687 }
688 
689 /*
690  * The guts of the unmount code.  The mount owns one ref and one hold
691  * count.  If we successfully interlock the unmount, those refs are ours.
692  * (The ref is from mnt_ncmountpt).
693  *
694  * When halting we shortcut certain mount types such as devfs by not actually
695  * issuing the VFS_SYNC() or VFS_UNMOUNT().  They are still disconnected
696  * from the mountlist so higher-level filesytems can unmount cleanly.
697  *
698  * The mount types that allow QUICKHALT are: devfs, tmpfs, procfs.
699  */
700 int
701 dounmount(struct mount *mp, int flags, int halting)
702 {
703 	struct namecache *ncp;
704 	struct nchandle nch;
705 	struct vnode *vp;
706 	int error;
707 	int async_flag;
708 	int lflags;
709 	int freeok = 1;
710 	int hadsyncer = 0;
711 	int retry;
712 	int quickhalt;
713 
714 	lwkt_gettoken(&mp->mnt_token);
715 
716 	/*
717 	 * When halting, certain mount points can essentially just
718 	 * be unhooked and otherwise ignored.
719 	 */
720 	if (halting && (mp->mnt_kern_flag & MNTK_QUICKHALT)) {
721 		quickhalt = 1;
722 		freeok = 0;
723 	} else {
724 		quickhalt = 0;
725 	}
726 
727 
728 	/*
729 	 * Exclusive access for unmounting purposes.
730 	 */
731 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
732 		goto out;
733 
734 	/*
735 	 * We now 'own' the last mp->mnt_refs
736 	 *
737 	 * Allow filesystems to detect that a forced unmount is in progress.
738 	 */
739 	if (flags & MNT_FORCE)
740 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
741 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK);
742 	error = lockmgr(&mp->mnt_lock, lflags);
743 	if (error) {
744 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
745 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
746 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
747 			wakeup(mp);
748 		}
749 		goto out;
750 	}
751 
752 	if (mp->mnt_flag & MNT_EXPUBLIC)
753 		vfs_setpublicfs(NULL, NULL, NULL);
754 
755 	vfs_msync(mp, MNT_WAIT);
756 	async_flag = mp->mnt_flag & MNT_ASYNC;
757 	mp->mnt_flag &=~ MNT_ASYNC;
758 
759 	/*
760 	 * Decomission our special mnt_syncer vnode.  This also stops
761 	 * the vnlru code.  If we are unable to unmount we recommission
762 	 * the vnode.
763 	 *
764 	 * Then sync the filesystem.
765 	 */
766 	if ((vp = mp->mnt_syncer) != NULL) {
767 		mp->mnt_syncer = NULL;
768 		atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
769 		vrele(vp);
770 		hadsyncer = 1;
771 	}
772 
773 	/*
774 	 * Sync normally-mounted filesystem.
775 	 */
776 	if (quickhalt == 0) {
777 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
778 			VFS_SYNC(mp, MNT_WAIT);
779 	}
780 
781 	/*
782 	 * nchandle records ref the mount structure.  Expect a count of 1
783 	 * (our mount->mnt_ncmountpt).
784 	 *
785 	 * Scans can get temporary refs on a mountpoint (thought really
786 	 * heavy duty stuff like cache_findmount() do not).
787 	 */
788 	for (retry = 0; (retry < 10 || debug_unmount); ++retry) {
789 		/*
790 		 * Invalidate the namecache topology under the mount.
791 		 * nullfs mounts alias a real mount's namecache topology
792 		 * and it should not be invalidated in that case.
793 		 */
794 		if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
795 			cache_lock(&mp->mnt_ncmountpt);
796 			cache_inval(&mp->mnt_ncmountpt,
797 				    CINV_DESTROY|CINV_CHILDREN);
798 			cache_unlock(&mp->mnt_ncmountpt);
799 		}
800 
801 		/*
802 		 * Clear pcpu caches
803 		 */
804 		cache_unmounting(mp);
805 		if (mp->mnt_refs != 1)
806 			cache_clearmntcache();
807 
808 		/*
809 		 * Break out if we are good.  Don't count ncp refs if the
810 		 * mount is aliased.
811 		 */
812 		ncp = (mp->mnt_kern_flag & MNTK_NCALIASED) ?
813 		      NULL : mp->mnt_ncmountpt.ncp;
814 		if (mp->mnt_refs == 1 &&
815 		    (ncp == NULL || (ncp->nc_refs == 1 &&
816 				     TAILQ_FIRST(&ncp->nc_list) == NULL))) {
817 			break;
818 		}
819 
820 		/*
821 		 * If forcing the unmount, clean out any p->p_textnch
822 		 * nchandles that match this mount.
823 		 */
824 		if (flags & MNT_FORCE)
825 			allproc_scan(&unmount_allproc_cb, mp, 0);
826 
827 		/*
828 		 * Sleep and retry.
829 		 */
830 		tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1);
831 		if ((retry & 15) == 15) {
832 			mount_warning(mp,
833 				      "(%p) debug - retry %d, "
834 				      "%d namecache refs, %d mount refs",
835 				      mp, retry,
836 				      (ncp ? ncp->nc_refs - 1 : 0),
837 				      mp->mnt_refs - 1);
838 		}
839 	}
840 
841 	error = 0;
842 	ncp = (mp->mnt_kern_flag & MNTK_NCALIASED) ?
843 	      NULL : mp->mnt_ncmountpt.ncp;
844 	if (mp->mnt_refs != 1 ||
845 	    (ncp != NULL && (ncp->nc_refs != 1 ||
846 			     TAILQ_FIRST(&ncp->nc_list)))) {
847 		mount_warning(mp,
848 			      "(%p): %d namecache refs, %d mount refs "
849 			      "still present",
850 			      mp,
851 			      (ncp ? ncp->nc_refs - 1 : 0),
852 			      mp->mnt_refs - 1);
853 		if (flags & MNT_FORCE) {
854 			freeok = 0;
855 			mount_warning(mp, "forcing unmount\n");
856 		} else {
857 			error = EBUSY;
858 		}
859 	}
860 
861 	/*
862 	 * So far so good, sync the filesystem once more and
863 	 * call the VFS unmount code if the sync succeeds.
864 	 */
865 	if (error == 0 && quickhalt == 0) {
866 		if (mp->mnt_flag & MNT_RDONLY) {
867 			error = VFS_UNMOUNT(mp, flags);
868 		} else {
869 			error = VFS_SYNC(mp, MNT_WAIT);
870 			if (error == 0 ||		/* no error */
871 			    error == EOPNOTSUPP ||	/* no sync avail */
872 			    (flags & MNT_FORCE)) {	/* force anyway */
873 				error = VFS_UNMOUNT(mp, flags);
874 			}
875 		}
876 		if (error) {
877 			mount_warning(mp,
878 				      "(%p) unmount: vfs refused to unmount, "
879 				      "error %d",
880 				      mp, error);
881 		}
882 	}
883 
884 	/*
885 	 * If an error occurred we can still recover, restoring the
886 	 * syncer vnode and misc flags.
887 	 */
888 	if (error) {
889 		if (mp->mnt_syncer == NULL && hadsyncer)
890 			vfs_allocate_syncvnode(mp);
891 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
892 		mp->mnt_flag |= async_flag;
893 		lockmgr(&mp->mnt_lock, LK_RELEASE);
894 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
895 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
896 			wakeup(mp);
897 		}
898 		goto out;
899 	}
900 	/*
901 	 * Clean up any journals still associated with the mount after
902 	 * filesystem activity has ceased.
903 	 */
904 	journal_remove_all_journals(mp,
905 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
906 
907 	mountlist_remove(mp);
908 
909 	/*
910 	 * Remove any installed vnode ops here so the individual VFSs don't
911 	 * have to.
912 	 *
913 	 * mnt_refs should go to zero when we scrap mnt_ncmountpt.
914 	 *
915 	 * When quickhalting we have to keep these intact because the
916 	 * underlying vnodes have not been destroyed, and some might be
917 	 * dirty.
918 	 */
919 	if (quickhalt == 0) {
920 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
921 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
922 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
923 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
924 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
925 	}
926 
927 	if (mp->mnt_ncmountpt.ncp != NULL) {
928 		nch = mp->mnt_ncmountpt;
929 		cache_zero(&mp->mnt_ncmountpt);
930 		cache_clrmountpt(&nch);
931 		cache_drop(&nch);
932 	}
933 	if (mp->mnt_ncmounton.ncp != NULL) {
934 		cache_unmounting(mp);
935 		nch = mp->mnt_ncmounton;
936 		cache_zero(&mp->mnt_ncmounton);
937 		cache_clrmountpt(&nch);
938 		cache_drop(&nch);
939 	}
940 
941 	mp->mnt_vfc->vfc_refcount--;
942 
943 	/*
944 	 * If not quickhalting the mount, we expect there to be no
945 	 * vnodes left.
946 	 */
947 	if (quickhalt == 0 && !TAILQ_EMPTY(&mp->mnt_nvnodelist))
948 		panic("unmount: dangling vnode");
949 
950 	/*
951 	 * Release the lock
952 	 */
953 	lockmgr(&mp->mnt_lock, LK_RELEASE);
954 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
955 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
956 		wakeup(mp);
957 	}
958 
959 	/*
960 	 * If we reach here and freeok != 0 we must free the mount.
961 	 * mnt_refs should already have dropped to 0, so if it is not
962 	 * zero we must cycle the caches and wait.
963 	 *
964 	 * When we are satisfied that the mount has disconnected we can
965 	 * drop the hold on the mp that represented the mount (though the
966 	 * caller might actually have another, so the caller's drop may
967 	 * do the actual free).
968 	 */
969 	if (freeok) {
970 		if (mp->mnt_refs > 0)
971 			cache_clearmntcache();
972 		while (mp->mnt_refs > 0) {
973 			cache_unmounting(mp);
974 			wakeup(mp);
975 			tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1);
976 			cache_clearmntcache();
977 		}
978 		lwkt_reltoken(&mp->mnt_token);
979 		mount_drop(mp);
980 		mp = NULL;
981 	} else {
982 		cache_clearmntcache();
983 	}
984 	error = 0;
985 	KNOTE(&fs_klist, VQ_UNMOUNT);
986 out:
987 	if (mp)
988 		lwkt_reltoken(&mp->mnt_token);
989 	return (error);
990 }
991 
992 static
993 void
994 mount_warning(struct mount *mp, const char *ctl, ...)
995 {
996 	char *ptr;
997 	char *buf;
998 	__va_list va;
999 
1000 	__va_start(va, ctl);
1001 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL,
1002 			   &ptr, &buf, 0) == 0) {
1003 		kprintf("unmount(%s): ", ptr);
1004 		kvprintf(ctl, va);
1005 		kprintf("\n");
1006 		kfree(buf, M_TEMP);
1007 	} else {
1008 		kprintf("unmount(%p", mp);
1009 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
1010 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
1011 		kprintf("): ");
1012 		kvprintf(ctl, va);
1013 		kprintf("\n");
1014 	}
1015 	__va_end(va);
1016 }
1017 
1018 /*
1019  * Shim cache_fullpath() to handle the case where a process is chrooted into
1020  * a subdirectory of a mount.  In this case if the root mount matches the
1021  * process root directory's mount we have to specify the process's root
1022  * directory instead of the mount point, because the mount point might
1023  * be above the root directory.
1024  */
1025 static
1026 int
1027 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
1028 {
1029 	struct nchandle *nch;
1030 
1031 	if (p && p->p_fd->fd_nrdir.mount == mp)
1032 		nch = &p->p_fd->fd_nrdir;
1033 	else
1034 		nch = &mp->mnt_ncmountpt;
1035 	return(cache_fullpath(p, nch, NULL, rb, fb, 0));
1036 }
1037 
1038 /*
1039  * Sync each mounted filesystem.
1040  */
1041 
1042 #ifdef DEBUG
1043 static int syncprt = 0;
1044 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
1045 #endif /* DEBUG */
1046 
1047 static int sync_callback(struct mount *mp, void *data);
1048 
1049 int
1050 sys_sync(struct sync_args *uap)
1051 {
1052 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
1053 	return (0);
1054 }
1055 
1056 static
1057 int
1058 sync_callback(struct mount *mp, void *data __unused)
1059 {
1060 	int asyncflag;
1061 
1062 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1063 		lwkt_gettoken(&mp->mnt_token);
1064 		asyncflag = mp->mnt_flag & MNT_ASYNC;
1065 		mp->mnt_flag &= ~MNT_ASYNC;
1066 		lwkt_reltoken(&mp->mnt_token);
1067 		vfs_msync(mp, MNT_NOWAIT);
1068 		VFS_SYNC(mp, MNT_NOWAIT);
1069 		lwkt_gettoken(&mp->mnt_token);
1070 		mp->mnt_flag |= asyncflag;
1071 		lwkt_reltoken(&mp->mnt_token);
1072 	}
1073 	return(0);
1074 }
1075 
1076 /* XXX PRISON: could be per prison flag */
1077 static int prison_quotas;
1078 #if 0
1079 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
1080 #endif
1081 
1082 /*
1083  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
1084  *
1085  * Change filesystem quotas.
1086  *
1087  * MPALMOSTSAFE
1088  */
1089 int
1090 sys_quotactl(struct quotactl_args *uap)
1091 {
1092 	struct nlookupdata nd;
1093 	struct thread *td;
1094 	struct mount *mp;
1095 	int error;
1096 
1097 	td = curthread;
1098 	if (td->td_ucred->cr_prison && !prison_quotas) {
1099 		error = EPERM;
1100 		goto done;
1101 	}
1102 
1103 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1104 	if (error == 0)
1105 		error = nlookup(&nd);
1106 	if (error == 0) {
1107 		mp = nd.nl_nch.mount;
1108 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
1109 				    uap->arg, nd.nl_cred);
1110 	}
1111 	nlookup_done(&nd);
1112 done:
1113 	return (error);
1114 }
1115 
1116 /*
1117  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
1118  *		void *buf, int buflen)
1119  *
1120  * This function operates on a mount point and executes the specified
1121  * operation using the specified control data, and possibly returns data.
1122  *
1123  * The actual number of bytes stored in the result buffer is returned, 0
1124  * if none, otherwise an error is returned.
1125  *
1126  * MPALMOSTSAFE
1127  */
1128 int
1129 sys_mountctl(struct mountctl_args *uap)
1130 {
1131 	struct thread *td = curthread;
1132 	struct file *fp;
1133 	void *ctl = NULL;
1134 	void *buf = NULL;
1135 	char *path = NULL;
1136 	int error;
1137 
1138 	/*
1139 	 * Sanity and permissions checks.  We must be root.
1140 	 */
1141 	if (td->td_ucred->cr_prison != NULL)
1142 		return (EPERM);
1143 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
1144 	    (error = priv_check(td, PRIV_ROOT)) != 0)
1145 		return (error);
1146 
1147 	/*
1148 	 * Argument length checks
1149 	 */
1150 	if (uap->ctllen < 0 || uap->ctllen > 1024)
1151 		return (EINVAL);
1152 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
1153 		return (EINVAL);
1154 	if (uap->path == NULL)
1155 		return (EINVAL);
1156 
1157 	/*
1158 	 * Allocate the necessary buffers and copyin data
1159 	 */
1160 	path = objcache_get(namei_oc, M_WAITOK);
1161 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1162 	if (error)
1163 		goto done;
1164 
1165 	if (uap->ctllen) {
1166 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1167 		error = copyin(uap->ctl, ctl, uap->ctllen);
1168 		if (error)
1169 			goto done;
1170 	}
1171 	if (uap->buflen)
1172 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1173 
1174 	/*
1175 	 * Validate the descriptor
1176 	 */
1177 	if (uap->fd >= 0) {
1178 		fp = holdfp(td, uap->fd, -1);
1179 		if (fp == NULL) {
1180 			error = EBADF;
1181 			goto done;
1182 		}
1183 	} else {
1184 		fp = NULL;
1185 	}
1186 
1187 	/*
1188 	 * Execute the internal kernel function and clean up.
1189 	 */
1190 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen,
1191 			      buf, uap->buflen, &uap->sysmsg_result);
1192 	if (fp)
1193 		dropfp(td, uap->fd, fp);
1194 	if (error == 0 && uap->sysmsg_result > 0)
1195 		error = copyout(buf, uap->buf, uap->sysmsg_result);
1196 done:
1197 	if (path)
1198 		objcache_put(namei_oc, path);
1199 	if (ctl)
1200 		kfree(ctl, M_TEMP);
1201 	if (buf)
1202 		kfree(buf, M_TEMP);
1203 	return (error);
1204 }
1205 
1206 /*
1207  * Execute a mount control operation by resolving the path to a mount point
1208  * and calling vop_mountctl().
1209  *
1210  * Use the mount point from the nch instead of the vnode so nullfs mounts
1211  * can properly spike the VOP.
1212  */
1213 int
1214 kern_mountctl(const char *path, int op, struct file *fp,
1215 		const void *ctl, int ctllen,
1216 		void *buf, int buflen, int *res)
1217 {
1218 	struct vnode *vp;
1219 	struct nlookupdata nd;
1220 	struct nchandle nch;
1221 	struct mount *mp;
1222 	int error;
1223 
1224 	*res = 0;
1225 	vp = NULL;
1226 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1227 	if (error)
1228 		return (error);
1229 	error = nlookup(&nd);
1230 	if (error) {
1231 		nlookup_done(&nd);
1232 		return (error);
1233 	}
1234 	error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1235 	if (error) {
1236 		nlookup_done(&nd);
1237 		return (error);
1238 	}
1239 
1240 	/*
1241 	 * Yes, all this is needed to use the nch.mount below, because
1242 	 * we must maintain a ref on the mount to avoid ripouts (e.g.
1243 	 * due to heavy mount/unmount use by synth or poudriere).
1244 	 */
1245 	nch = nd.nl_nch;
1246 	cache_zero(&nd.nl_nch);
1247 	cache_unlock(&nch);
1248 	nlookup_done(&nd);
1249 	vn_unlock(vp);
1250 
1251 	mp = nch.mount;
1252 
1253 	/*
1254 	 * Must be the root of the filesystem
1255 	 */
1256 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1257 		cache_drop(&nch);
1258 		vrele(vp);
1259 		return (EINVAL);
1260 	}
1261 	if (mp == NULL || mp->mnt_kern_flag & MNTK_UNMOUNT) {
1262 		kprintf("kern_mountctl: Warning, \"%s\" racing unmount\n",
1263 			path);
1264 		cache_drop(&nch);
1265 		vrele(vp);
1266 		return (EINVAL);
1267 	}
1268 	error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1269 			     buf, buflen, res);
1270 	vrele(vp);
1271 	cache_drop(&nch);
1272 
1273 	return (error);
1274 }
1275 
1276 int
1277 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1278 {
1279 	struct thread *td = curthread;
1280 	struct proc *p = td->td_proc;
1281 	struct mount *mp;
1282 	struct statfs *sp;
1283 	char *fullpath, *freepath;
1284 	int error;
1285 
1286 	if ((error = nlookup(nd)) != 0)
1287 		return (error);
1288 	mp = nd->nl_nch.mount;
1289 	sp = &mp->mnt_stat;
1290 
1291 	/*
1292 	 * Ignore refresh error, user should have visibility.
1293 	 * This can happen if a NFS mount goes bad (e.g. server
1294 	 * revokes perms or goes down).
1295 	 */
1296 	error = VFS_STATFS(mp, sp, nd->nl_cred);
1297 	/* ignore error */
1298 
1299 	error = mount_path(p, mp, &fullpath, &freepath);
1300 	if (error)
1301 		return(error);
1302 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1303 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1304 	kfree(freepath, M_TEMP);
1305 
1306 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1307 	bcopy(sp, buf, sizeof(*buf));
1308 	/* Only root should have access to the fsid's. */
1309 	if (priv_check(td, PRIV_ROOT))
1310 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1311 	return (0);
1312 }
1313 
1314 /*
1315  * statfs_args(char *path, struct statfs *buf)
1316  *
1317  * Get filesystem statistics.
1318  */
1319 int
1320 sys_statfs(struct statfs_args *uap)
1321 {
1322 	struct nlookupdata nd;
1323 	struct statfs buf;
1324 	int error;
1325 
1326 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1327 	if (error == 0)
1328 		error = kern_statfs(&nd, &buf);
1329 	nlookup_done(&nd);
1330 	if (error == 0)
1331 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1332 	return (error);
1333 }
1334 
1335 int
1336 kern_fstatfs(int fd, struct statfs *buf)
1337 {
1338 	struct thread *td = curthread;
1339 	struct proc *p = td->td_proc;
1340 	struct file *fp;
1341 	struct mount *mp;
1342 	struct statfs *sp;
1343 	char *fullpath, *freepath;
1344 	int error;
1345 
1346 	KKASSERT(p);
1347 	if ((error = holdvnode(td, fd, &fp)) != 0)
1348 		return (error);
1349 
1350 	/*
1351 	 * Try to use mount info from any overlays rather than the
1352 	 * mount info for the underlying vnode, otherwise we will
1353 	 * fail when operating on null-mounted paths inside a chroot.
1354 	 */
1355 	if ((mp = fp->f_nchandle.mount) == NULL)
1356 		mp = ((struct vnode *)fp->f_data)->v_mount;
1357 	if (mp == NULL) {
1358 		error = EBADF;
1359 		goto done;
1360 	}
1361 	if (fp->f_cred == NULL) {
1362 		error = EINVAL;
1363 		goto done;
1364 	}
1365 
1366 	/*
1367 	 * Ignore refresh error, user should have visibility.
1368 	 * This can happen if a NFS mount goes bad (e.g. server
1369 	 * revokes perms or goes down).
1370 	 */
1371 	sp = &mp->mnt_stat;
1372 	error = VFS_STATFS(mp, sp, fp->f_cred);
1373 
1374 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1375 		goto done;
1376 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1377 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1378 	kfree(freepath, M_TEMP);
1379 
1380 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1381 	bcopy(sp, buf, sizeof(*buf));
1382 
1383 	/* Only root should have access to the fsid's. */
1384 	if (priv_check(td, PRIV_ROOT))
1385 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1386 	error = 0;
1387 done:
1388 	fdrop(fp);
1389 	return (error);
1390 }
1391 
1392 /*
1393  * fstatfs_args(int fd, struct statfs *buf)
1394  *
1395  * Get filesystem statistics.
1396  */
1397 int
1398 sys_fstatfs(struct fstatfs_args *uap)
1399 {
1400 	struct statfs buf;
1401 	int error;
1402 
1403 	error = kern_fstatfs(uap->fd, &buf);
1404 
1405 	if (error == 0)
1406 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1407 	return (error);
1408 }
1409 
1410 int
1411 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1412 {
1413 	struct mount *mp;
1414 	struct statvfs *sp;
1415 	int error;
1416 
1417 	if ((error = nlookup(nd)) != 0)
1418 		return (error);
1419 	mp = nd->nl_nch.mount;
1420 	sp = &mp->mnt_vstat;
1421 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1422 		return (error);
1423 
1424 	sp->f_flag = 0;
1425 	if (mp->mnt_flag & MNT_RDONLY)
1426 		sp->f_flag |= ST_RDONLY;
1427 	if (mp->mnt_flag & MNT_NOSUID)
1428 		sp->f_flag |= ST_NOSUID;
1429 	bcopy(sp, buf, sizeof(*buf));
1430 	return (0);
1431 }
1432 
1433 /*
1434  * statfs_args(char *path, struct statfs *buf)
1435  *
1436  * Get filesystem statistics.
1437  */
1438 int
1439 sys_statvfs(struct statvfs_args *uap)
1440 {
1441 	struct nlookupdata nd;
1442 	struct statvfs buf;
1443 	int error;
1444 
1445 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1446 	if (error == 0)
1447 		error = kern_statvfs(&nd, &buf);
1448 	nlookup_done(&nd);
1449 	if (error == 0)
1450 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1451 	return (error);
1452 }
1453 
1454 int
1455 kern_fstatvfs(int fd, struct statvfs *buf)
1456 {
1457 	struct thread *td = curthread;
1458 	struct file *fp;
1459 	struct mount *mp;
1460 	struct statvfs *sp;
1461 	int error;
1462 
1463 	if ((error = holdvnode(td, fd, &fp)) != 0)
1464 		return (error);
1465 	if ((mp = fp->f_nchandle.mount) == NULL)
1466 		mp = ((struct vnode *)fp->f_data)->v_mount;
1467 	if (mp == NULL) {
1468 		error = EBADF;
1469 		goto done;
1470 	}
1471 	if (fp->f_cred == NULL) {
1472 		error = EINVAL;
1473 		goto done;
1474 	}
1475 	sp = &mp->mnt_vstat;
1476 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1477 		goto done;
1478 
1479 	sp->f_flag = 0;
1480 	if (mp->mnt_flag & MNT_RDONLY)
1481 		sp->f_flag |= ST_RDONLY;
1482 	if (mp->mnt_flag & MNT_NOSUID)
1483 		sp->f_flag |= ST_NOSUID;
1484 
1485 	bcopy(sp, buf, sizeof(*buf));
1486 	error = 0;
1487 done:
1488 	fdrop(fp);
1489 	return (error);
1490 }
1491 
1492 /*
1493  * fstatfs_args(int fd, struct statfs *buf)
1494  *
1495  * Get filesystem statistics.
1496  */
1497 int
1498 sys_fstatvfs(struct fstatvfs_args *uap)
1499 {
1500 	struct statvfs buf;
1501 	int error;
1502 
1503 	error = kern_fstatvfs(uap->fd, &buf);
1504 
1505 	if (error == 0)
1506 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1507 	return (error);
1508 }
1509 
1510 /*
1511  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1512  *
1513  * Get statistics on all filesystems.
1514  */
1515 
1516 struct getfsstat_info {
1517 	struct statfs *sfsp;
1518 	long count;
1519 	long maxcount;
1520 	int error;
1521 	int flags;
1522 	struct thread *td;
1523 };
1524 
1525 static int getfsstat_callback(struct mount *, void *);
1526 
1527 int
1528 sys_getfsstat(struct getfsstat_args *uap)
1529 {
1530 	struct thread *td = curthread;
1531 	struct getfsstat_info info;
1532 
1533 	bzero(&info, sizeof(info));
1534 
1535 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1536 	info.sfsp = uap->buf;
1537 	info.count = 0;
1538 	info.flags = uap->flags;
1539 	info.td = td;
1540 
1541 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1542 	if (info.sfsp && info.count > info.maxcount)
1543 		uap->sysmsg_result = info.maxcount;
1544 	else
1545 		uap->sysmsg_result = info.count;
1546 	return (info.error);
1547 }
1548 
1549 static int
1550 getfsstat_callback(struct mount *mp, void *data)
1551 {
1552 	struct getfsstat_info *info = data;
1553 	struct statfs *sp;
1554 	char *freepath;
1555 	char *fullpath;
1556 	int error;
1557 
1558 	if (info->td->td_proc && !chroot_visible_mnt(mp, info->td->td_proc))
1559 		return(0);
1560 
1561 	if (info->sfsp && info->count < info->maxcount) {
1562 		sp = &mp->mnt_stat;
1563 
1564 		/*
1565 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1566 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1567 		 * overrides MNT_WAIT.
1568 		 *
1569 		 * Ignore refresh error, user should have visibility.
1570 		 * This can happen if a NFS mount goes bad (e.g. server
1571 		 * revokes perms or goes down).
1572 		 */
1573 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1574 		    (info->flags & MNT_WAIT)) &&
1575 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1576 			/* ignore error */
1577 		}
1578 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1579 
1580 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1581 		if (error) {
1582 			info->error = error;
1583 			return(-1);
1584 		}
1585 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1586 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1587 		kfree(freepath, M_TEMP);
1588 
1589 		error = copyout(sp, info->sfsp, sizeof(*sp));
1590 		if (error) {
1591 			info->error = error;
1592 			return (-1);
1593 		}
1594 		++info->sfsp;
1595 	}
1596 	info->count++;
1597 	return(0);
1598 }
1599 
1600 /*
1601  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1602 		   long bufsize, int flags)
1603  *
1604  * Get statistics on all filesystems.
1605  */
1606 
1607 struct getvfsstat_info {
1608 	struct statfs *sfsp;
1609 	struct statvfs *vsfsp;
1610 	long count;
1611 	long maxcount;
1612 	int error;
1613 	int flags;
1614 	struct thread *td;
1615 };
1616 
1617 static int getvfsstat_callback(struct mount *, void *);
1618 
1619 int
1620 sys_getvfsstat(struct getvfsstat_args *uap)
1621 {
1622 	struct thread *td = curthread;
1623 	struct getvfsstat_info info;
1624 
1625 	bzero(&info, sizeof(info));
1626 
1627 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1628 	info.sfsp = uap->buf;
1629 	info.vsfsp = uap->vbuf;
1630 	info.count = 0;
1631 	info.flags = uap->flags;
1632 	info.td = td;
1633 
1634 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1635 	if (info.vsfsp && info.count > info.maxcount)
1636 		uap->sysmsg_result = info.maxcount;
1637 	else
1638 		uap->sysmsg_result = info.count;
1639 	return (info.error);
1640 }
1641 
1642 static int
1643 getvfsstat_callback(struct mount *mp, void *data)
1644 {
1645 	struct getvfsstat_info *info = data;
1646 	struct statfs *sp;
1647 	struct statvfs *vsp;
1648 	char *freepath;
1649 	char *fullpath;
1650 	int error;
1651 
1652 	if (info->td->td_proc && !chroot_visible_mnt(mp, info->td->td_proc))
1653 		return(0);
1654 
1655 	if (info->vsfsp && info->count < info->maxcount) {
1656 		sp = &mp->mnt_stat;
1657 		vsp = &mp->mnt_vstat;
1658 
1659 		/*
1660 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1661 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1662 		 * overrides MNT_WAIT.
1663 		 *
1664 		 * Ignore refresh error, user should have visibility.
1665 		 * This can happen if a NFS mount goes bad (e.g. server
1666 		 * revokes perms or goes down).
1667 		 */
1668 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1669 		    (info->flags & MNT_WAIT)) &&
1670 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1671 			/* ignore error */
1672 		}
1673 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1674 
1675 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1676 		    (info->flags & MNT_WAIT)) &&
1677 		    (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1678 			/* ignore error */
1679 		}
1680 		vsp->f_flag = 0;
1681 		if (mp->mnt_flag & MNT_RDONLY)
1682 			vsp->f_flag |= ST_RDONLY;
1683 		if (mp->mnt_flag & MNT_NOSUID)
1684 			vsp->f_flag |= ST_NOSUID;
1685 
1686 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1687 		if (error) {
1688 			info->error = error;
1689 			return(-1);
1690 		}
1691 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1692 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1693 		kfree(freepath, M_TEMP);
1694 
1695 		error = copyout(sp, info->sfsp, sizeof(*sp));
1696 		if (error == 0)
1697 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1698 		if (error) {
1699 			info->error = error;
1700 			return (-1);
1701 		}
1702 		++info->sfsp;
1703 		++info->vsfsp;
1704 	}
1705 	info->count++;
1706 	return(0);
1707 }
1708 
1709 
1710 /*
1711  * fchdir_args(int fd)
1712  *
1713  * Change current working directory to a given file descriptor.
1714  */
1715 int
1716 sys_fchdir(struct fchdir_args *uap)
1717 {
1718 	struct thread *td = curthread;
1719 	struct proc *p = td->td_proc;
1720 	struct filedesc *fdp = p->p_fd;
1721 	struct vnode *vp, *ovp;
1722 	struct mount *mp;
1723 	struct file *fp;
1724 	struct nchandle nch, onch, tnch;
1725 	int error;
1726 
1727 	if ((error = holdvnode(td, uap->fd, &fp)) != 0)
1728 		return (error);
1729 	lwkt_gettoken(&p->p_token);
1730 	vp = (struct vnode *)fp->f_data;
1731 	vref(vp);
1732 	vn_lock(vp, LK_SHARED | LK_RETRY);
1733 	if (fp->f_nchandle.ncp == NULL)
1734 		error = ENOTDIR;
1735 	else
1736 		error = checkvp_chdir(vp, td);
1737 	if (error) {
1738 		vput(vp);
1739 		goto done;
1740 	}
1741 	cache_copy(&fp->f_nchandle, &nch);
1742 
1743 	/*
1744 	 * If the ncp has become a mount point, traverse through
1745 	 * the mount point.
1746 	 */
1747 
1748 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1749 	       (mp = cache_findmount(&nch)) != NULL
1750 	) {
1751 		error = nlookup_mp(mp, &tnch);
1752 		if (error == 0) {
1753 			cache_unlock(&tnch);	/* leave ref intact */
1754 			vput(vp);
1755 			vp = tnch.ncp->nc_vp;
1756 			error = vget(vp, LK_SHARED);
1757 			KKASSERT(error == 0);
1758 			cache_drop(&nch);
1759 			nch = tnch;
1760 		}
1761 		cache_dropmount(mp);
1762 	}
1763 	if (error == 0) {
1764 		spin_lock(&fdp->fd_spin);
1765 		ovp = fdp->fd_cdir;
1766 		onch = fdp->fd_ncdir;
1767 		fdp->fd_cdir = vp;
1768 		fdp->fd_ncdir = nch;
1769 		spin_unlock(&fdp->fd_spin);
1770 		vn_unlock(vp);		/* leave ref intact */
1771 		cache_drop(&onch);
1772 		vrele(ovp);
1773 	} else {
1774 		cache_drop(&nch);
1775 		vput(vp);
1776 	}
1777 	fdrop(fp);
1778 done:
1779 	lwkt_reltoken(&p->p_token);
1780 	return (error);
1781 }
1782 
1783 int
1784 kern_chdir(struct nlookupdata *nd)
1785 {
1786 	struct thread *td = curthread;
1787 	struct proc *p = td->td_proc;
1788 	struct filedesc *fdp = p->p_fd;
1789 	struct vnode *vp, *ovp;
1790 	struct nchandle onch;
1791 	int error;
1792 
1793 	nd->nl_flags |= NLC_SHAREDLOCK;
1794 	if ((error = nlookup(nd)) != 0)
1795 		return (error);
1796 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1797 		return (ENOENT);
1798 	if ((error = vget(vp, LK_SHARED)) != 0)
1799 		return (error);
1800 
1801 	lwkt_gettoken(&p->p_token);
1802 	error = checkvp_chdir(vp, td);
1803 	vn_unlock(vp);
1804 	if (error == 0) {
1805 		spin_lock(&fdp->fd_spin);
1806 		ovp = fdp->fd_cdir;
1807 		onch = fdp->fd_ncdir;
1808 		fdp->fd_ncdir = nd->nl_nch;
1809 		fdp->fd_cdir = vp;
1810 		spin_unlock(&fdp->fd_spin);
1811 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1812 		cache_drop(&onch);
1813 		vrele(ovp);
1814 		cache_zero(&nd->nl_nch);
1815 	} else {
1816 		vrele(vp);
1817 	}
1818 	lwkt_reltoken(&p->p_token);
1819 	return (error);
1820 }
1821 
1822 /*
1823  * chdir_args(char *path)
1824  *
1825  * Change current working directory (``.'').
1826  */
1827 int
1828 sys_chdir(struct chdir_args *uap)
1829 {
1830 	struct nlookupdata nd;
1831 	int error;
1832 
1833 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1834 	if (error == 0)
1835 		error = kern_chdir(&nd);
1836 	nlookup_done(&nd);
1837 	return (error);
1838 }
1839 
1840 /*
1841  * Helper function for raised chroot(2) security function:  Refuse if
1842  * any filedescriptors are open directories.
1843  */
1844 static int
1845 chroot_refuse_vdir_fds(thread_t td, struct filedesc *fdp)
1846 {
1847 	struct vnode *vp;
1848 	struct file *fp;
1849 	int error;
1850 	int fd;
1851 
1852 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1853 		if ((error = holdvnode(td, fd, &fp)) != 0)
1854 			continue;
1855 		vp = (struct vnode *)fp->f_data;
1856 		if (vp->v_type != VDIR) {
1857 			fdrop(fp);
1858 			continue;
1859 		}
1860 		fdrop(fp);
1861 		return(EPERM);
1862 	}
1863 	return (0);
1864 }
1865 
1866 /*
1867  * This sysctl determines if we will allow a process to chroot(2) if it
1868  * has a directory open:
1869  *	0: disallowed for all processes.
1870  *	1: allowed for processes that were not already chroot(2)'ed.
1871  *	2: allowed for all processes.
1872  */
1873 
1874 static int chroot_allow_open_directories = 1;
1875 
1876 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1877      &chroot_allow_open_directories, 0, "");
1878 
1879 /*
1880  * chroot to the specified namecache entry.  We obtain the vp from the
1881  * namecache data.  The passed ncp must be locked and referenced and will
1882  * remain locked and referenced on return.
1883  */
1884 int
1885 kern_chroot(struct nchandle *nch)
1886 {
1887 	struct thread *td = curthread;
1888 	struct proc *p = td->td_proc;
1889 	struct filedesc *fdp = p->p_fd;
1890 	struct vnode *vp;
1891 	int error;
1892 
1893 	/*
1894 	 * Only privileged user can chroot
1895 	 */
1896 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1897 	if (error)
1898 		return (error);
1899 
1900 	/*
1901 	 * Disallow open directory descriptors (fchdir() breakouts).
1902 	 */
1903 	if (chroot_allow_open_directories == 0 ||
1904 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1905 		if ((error = chroot_refuse_vdir_fds(td, fdp)) != 0)
1906 			return (error);
1907 	}
1908 	if ((vp = nch->ncp->nc_vp) == NULL)
1909 		return (ENOENT);
1910 
1911 	if ((error = vget(vp, LK_SHARED)) != 0)
1912 		return (error);
1913 
1914 	/*
1915 	 * Check the validity of vp as a directory to change to and
1916 	 * associate it with rdir/jdir.
1917 	 */
1918 	error = checkvp_chdir(vp, td);
1919 	vn_unlock(vp);			/* leave reference intact */
1920 	if (error == 0) {
1921 		lwkt_gettoken(&p->p_token);
1922 		vrele(fdp->fd_rdir);
1923 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1924 		cache_drop(&fdp->fd_nrdir);
1925 		cache_copy(nch, &fdp->fd_nrdir);
1926 		if (fdp->fd_jdir == NULL) {
1927 			fdp->fd_jdir = vp;
1928 			vref(fdp->fd_jdir);
1929 			cache_copy(nch, &fdp->fd_njdir);
1930 		}
1931 		if ((p->p_flags & P_DIDCHROOT) == 0) {
1932 			p->p_flags |= P_DIDCHROOT;
1933 			if (p->p_depth <= 65535 - 32)
1934 				p->p_depth += 32;
1935 		}
1936 		lwkt_reltoken(&p->p_token);
1937 	} else {
1938 		vrele(vp);
1939 	}
1940 	return (error);
1941 }
1942 
1943 /*
1944  * chroot_args(char *path)
1945  *
1946  * Change notion of root (``/'') directory.
1947  */
1948 int
1949 sys_chroot(struct chroot_args *uap)
1950 {
1951 	struct thread *td __debugvar = curthread;
1952 	struct nlookupdata nd;
1953 	int error;
1954 
1955 	KKASSERT(td->td_proc);
1956 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1957 	if (error == 0) {
1958 		nd.nl_flags |= NLC_EXEC;
1959 		error = nlookup(&nd);
1960 		if (error == 0)
1961 			error = kern_chroot(&nd.nl_nch);
1962 	}
1963 	nlookup_done(&nd);
1964 	return(error);
1965 }
1966 
1967 int
1968 sys_chroot_kernel(struct chroot_kernel_args *uap)
1969 {
1970 	struct thread *td = curthread;
1971 	struct nlookupdata nd;
1972 	struct nchandle *nch;
1973 	struct vnode *vp;
1974 	int error;
1975 
1976 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1977 	if (error)
1978 		goto error_nond;
1979 
1980 	error = nlookup(&nd);
1981 	if (error)
1982 		goto error_out;
1983 
1984 	nch = &nd.nl_nch;
1985 
1986 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1987 	if (error)
1988 		goto error_out;
1989 
1990 	if ((vp = nch->ncp->nc_vp) == NULL) {
1991 		error = ENOENT;
1992 		goto error_out;
1993 	}
1994 
1995 	if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1996 		goto error_out;
1997 
1998 	vfs_cache_setroot(vp, cache_hold(nch));
1999 
2000 error_out:
2001 	nlookup_done(&nd);
2002 error_nond:
2003 	return(error);
2004 }
2005 
2006 /*
2007  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
2008  * determine whether it is legal to chdir to the vnode.  The vnode's state
2009  * is not changed by this call.
2010  */
2011 static int
2012 checkvp_chdir(struct vnode *vp, struct thread *td)
2013 {
2014 	int error;
2015 
2016 	if (vp->v_type != VDIR)
2017 		error = ENOTDIR;
2018 	else
2019 		error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
2020 	return (error);
2021 }
2022 
2023 int
2024 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
2025 {
2026 	struct thread *td = curthread;
2027 	struct proc *p = td->td_proc;
2028 	struct lwp *lp = td->td_lwp;
2029 	struct filedesc *fdp = p->p_fd;
2030 	int cmode, flags;
2031 	struct file *nfp;
2032 	struct file *fp;
2033 	struct vnode *vp;
2034 	int type, indx, error = 0;
2035 	struct flock lf;
2036 
2037 	if ((oflags & O_ACCMODE) == O_ACCMODE)
2038 		return (EINVAL);
2039 	flags = FFLAGS(oflags);
2040 	error = falloc(lp, &nfp, NULL);
2041 	if (error)
2042 		return (error);
2043 	fp = nfp;
2044 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
2045 
2046 	/*
2047 	 * XXX p_dupfd is a real mess.  It allows a device to return a
2048 	 * file descriptor to be duplicated rather then doing the open
2049 	 * itself.
2050 	 */
2051 	lp->lwp_dupfd = -1;
2052 
2053 	/*
2054 	 * Call vn_open() to do the lookup and assign the vnode to the
2055 	 * file pointer.  vn_open() does not change the ref count on fp
2056 	 * and the vnode, on success, will be inherited by the file pointer
2057 	 * and unlocked.
2058 	 *
2059 	 * Request a shared lock on the vnode if possible.
2060 	 *
2061 	 * Executable binaries can race VTEXT against O_RDWR opens, so
2062 	 * use an exclusive lock for O_RDWR opens as well.
2063 	 *
2064 	 * NOTE: We need a flag to separate terminal vnode locking from
2065 	 *	 parent locking.  O_CREAT needs parent locking, but O_TRUNC
2066 	 *	 and O_RDWR only need to lock the terminal vnode exclusively.
2067 	 */
2068 	nd->nl_flags |= NLC_LOCKVP;
2069 	if ((flags & (O_CREAT|O_TRUNC|O_RDWR)) == 0)
2070 		nd->nl_flags |= NLC_SHAREDLOCK;
2071 
2072 	error = vn_open(nd, fp, flags, cmode);
2073 	nlookup_done(nd);
2074 
2075 	if (error) {
2076 		/*
2077 		 * handle special fdopen() case.  bleh.  dupfdopen() is
2078 		 * responsible for dropping the old contents of ofiles[indx]
2079 		 * if it succeeds.
2080 		 *
2081 		 * Note that fsetfd() will add a ref to fp which represents
2082 		 * the fd_files[] assignment.  We must still drop our
2083 		 * reference.
2084 		 */
2085 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
2086 			if (fdalloc(p, 0, &indx) == 0) {
2087 				error = dupfdopen(td, indx, lp->lwp_dupfd, flags, error);
2088 				if (error == 0) {
2089 					*res = indx;
2090 					fdrop(fp);	/* our ref */
2091 					return (0);
2092 				}
2093 				fsetfd(fdp, NULL, indx);
2094 			}
2095 		}
2096 		fdrop(fp);	/* our ref */
2097 		if (error == ERESTART)
2098 			error = EINTR;
2099 		return (error);
2100 	}
2101 
2102 	/*
2103 	 * ref the vnode for ourselves so it can't be ripped out from under
2104 	 * is.  XXX need an ND flag to request that the vnode be returned
2105 	 * anyway.
2106 	 *
2107 	 * Reserve a file descriptor but do not assign it until the open
2108 	 * succeeds.
2109 	 */
2110 	vp = (struct vnode *)fp->f_data;
2111 	vref(vp);
2112 	if ((error = fdalloc(p, 0, &indx)) != 0) {
2113 		fdrop(fp);
2114 		vrele(vp);
2115 		return (error);
2116 	}
2117 
2118 	/*
2119 	 * If no error occurs the vp will have been assigned to the file
2120 	 * pointer.
2121 	 */
2122 	lp->lwp_dupfd = 0;
2123 
2124 	if (flags & (O_EXLOCK | O_SHLOCK)) {
2125 		lf.l_whence = SEEK_SET;
2126 		lf.l_start = 0;
2127 		lf.l_len = 0;
2128 		if (flags & O_EXLOCK)
2129 			lf.l_type = F_WRLCK;
2130 		else
2131 			lf.l_type = F_RDLCK;
2132 		if (flags & FNONBLOCK)
2133 			type = 0;
2134 		else
2135 			type = F_WAIT;
2136 
2137 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
2138 			/*
2139 			 * lock request failed.  Clean up the reserved
2140 			 * descriptor.
2141 			 */
2142 			vrele(vp);
2143 			fsetfd(fdp, NULL, indx);
2144 			fdrop(fp);
2145 			return (error);
2146 		}
2147 		atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
2148 	}
2149 #if 0
2150 	/*
2151 	 * Assert that all regular file vnodes were created with a object.
2152 	 */
2153 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
2154 		("open: regular file has no backing object after vn_open"));
2155 #endif
2156 
2157 	vrele(vp);
2158 
2159 	/*
2160 	 * release our private reference, leaving the one associated with the
2161 	 * descriptor table intact.
2162 	 */
2163 	if (oflags & O_CLOEXEC)
2164 		fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
2165 	fsetfd(fdp, fp, indx);
2166 	fdrop(fp);
2167 	*res = indx;
2168 
2169 	return (error);
2170 }
2171 
2172 /*
2173  * open_args(char *path, int flags, int mode)
2174  *
2175  * Check permissions, allocate an open file structure,
2176  * and call the device open routine if any.
2177  */
2178 int
2179 sys_open(struct open_args *uap)
2180 {
2181 	struct nlookupdata nd;
2182 	int error;
2183 
2184 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2185 	if (error == 0) {
2186 		error = kern_open(&nd, uap->flags,
2187 				    uap->mode, &uap->sysmsg_result);
2188 	}
2189 	nlookup_done(&nd);
2190 	return (error);
2191 }
2192 
2193 /*
2194  * openat_args(int fd, char *path, int flags, int mode)
2195  */
2196 int
2197 sys_openat(struct openat_args *uap)
2198 {
2199 	struct nlookupdata nd;
2200 	int error;
2201 	struct file *fp;
2202 
2203 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2204 	if (error == 0) {
2205 		error = kern_open(&nd, uap->flags, uap->mode,
2206 					&uap->sysmsg_result);
2207 	}
2208 	nlookup_done_at(&nd, fp);
2209 	return (error);
2210 }
2211 
2212 int
2213 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2214 {
2215 	struct thread *td = curthread;
2216 	struct proc *p = td->td_proc;
2217 	struct vnode *vp;
2218 	struct vattr vattr;
2219 	int error;
2220 	int whiteout = 0;
2221 
2222 	KKASSERT(p);
2223 
2224 	VATTR_NULL(&vattr);
2225 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2226 	vattr.va_rmajor = rmajor;
2227 	vattr.va_rminor = rminor;
2228 
2229 	switch (mode & S_IFMT) {
2230 	case S_IFMT:	/* used by badsect to flag bad sectors */
2231 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2232 		vattr.va_type = VBAD;
2233 		break;
2234 	case S_IFCHR:
2235 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2236 		vattr.va_type = VCHR;
2237 		break;
2238 	case S_IFBLK:
2239 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2240 		vattr.va_type = VBLK;
2241 		break;
2242 	case S_IFWHT:
2243 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2244 		whiteout = 1;
2245 		break;
2246 	case S_IFDIR:	/* special directories support for HAMMER */
2247 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2248 		vattr.va_type = VDIR;
2249 		break;
2250 	default:
2251 		error = EINVAL;
2252 		break;
2253 	}
2254 
2255 	if (error)
2256 		return (error);
2257 
2258 	bwillinode(1);
2259 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2260 	if ((error = nlookup(nd)) != 0)
2261 		return (error);
2262 	if (nd->nl_nch.ncp->nc_vp)
2263 		return (EEXIST);
2264 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2265 		return (error);
2266 
2267 	if (whiteout) {
2268 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2269 				      nd->nl_cred, NAMEI_CREATE);
2270 	} else {
2271 		vp = NULL;
2272 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2273 				   &vp, nd->nl_cred, &vattr);
2274 		if (error == 0)
2275 			vput(vp);
2276 	}
2277 	return (error);
2278 }
2279 
2280 /*
2281  * mknod_args(char *path, int mode, int dev)
2282  *
2283  * Create a special file.
2284  */
2285 int
2286 sys_mknod(struct mknod_args *uap)
2287 {
2288 	struct nlookupdata nd;
2289 	int error;
2290 
2291 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2292 	if (error == 0) {
2293 		error = kern_mknod(&nd, uap->mode,
2294 				   umajor(uap->dev), uminor(uap->dev));
2295 	}
2296 	nlookup_done(&nd);
2297 	return (error);
2298 }
2299 
2300 /*
2301  * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2302  *
2303  * Create a special file.  The path is relative to the directory associated
2304  * with fd.
2305  */
2306 int
2307 sys_mknodat(struct mknodat_args *uap)
2308 {
2309 	struct nlookupdata nd;
2310 	struct file *fp;
2311 	int error;
2312 
2313 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2314 	if (error == 0) {
2315 		error = kern_mknod(&nd, uap->mode,
2316 				   umajor(uap->dev), uminor(uap->dev));
2317 	}
2318 	nlookup_done_at(&nd, fp);
2319 	return (error);
2320 }
2321 
2322 int
2323 kern_mkfifo(struct nlookupdata *nd, int mode)
2324 {
2325 	struct thread *td = curthread;
2326 	struct proc *p = td->td_proc;
2327 	struct vattr vattr;
2328 	struct vnode *vp;
2329 	int error;
2330 
2331 	bwillinode(1);
2332 
2333 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2334 	if ((error = nlookup(nd)) != 0)
2335 		return (error);
2336 	if (nd->nl_nch.ncp->nc_vp)
2337 		return (EEXIST);
2338 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2339 		return (error);
2340 
2341 	VATTR_NULL(&vattr);
2342 	vattr.va_type = VFIFO;
2343 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2344 	vp = NULL;
2345 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2346 	if (error == 0)
2347 		vput(vp);
2348 	return (error);
2349 }
2350 
2351 /*
2352  * mkfifo_args(char *path, int mode)
2353  *
2354  * Create a named pipe.
2355  */
2356 int
2357 sys_mkfifo(struct mkfifo_args *uap)
2358 {
2359 	struct nlookupdata nd;
2360 	int error;
2361 
2362 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2363 	if (error == 0)
2364 		error = kern_mkfifo(&nd, uap->mode);
2365 	nlookup_done(&nd);
2366 	return (error);
2367 }
2368 
2369 /*
2370  * mkfifoat_args(int fd, char *path, mode_t mode)
2371  *
2372  * Create a named pipe.  The path is relative to the directory associated
2373  * with fd.
2374  */
2375 int
2376 sys_mkfifoat(struct mkfifoat_args *uap)
2377 {
2378 	struct nlookupdata nd;
2379 	struct file *fp;
2380 	int error;
2381 
2382 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2383 	if (error == 0)
2384 		error = kern_mkfifo(&nd, uap->mode);
2385 	nlookup_done_at(&nd, fp);
2386 	return (error);
2387 }
2388 
2389 static int hardlink_check_uid = 0;
2390 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2391     &hardlink_check_uid, 0,
2392     "Unprivileged processes cannot create hard links to files owned by other "
2393     "users");
2394 static int hardlink_check_gid = 0;
2395 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2396     &hardlink_check_gid, 0,
2397     "Unprivileged processes cannot create hard links to files owned by other "
2398     "groups");
2399 
2400 static int
2401 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2402 {
2403 	struct vattr va;
2404 	int error;
2405 
2406 	/*
2407 	 * Shortcut if disabled
2408 	 */
2409 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2410 		return (0);
2411 
2412 	/*
2413 	 * Privileged user can always hardlink
2414 	 */
2415 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2416 		return (0);
2417 
2418 	/*
2419 	 * Otherwise only if the originating file is owned by the
2420 	 * same user or group.  Note that any group is allowed if
2421 	 * the file is owned by the caller.
2422 	 */
2423 	error = VOP_GETATTR(vp, &va);
2424 	if (error != 0)
2425 		return (error);
2426 
2427 	if (hardlink_check_uid) {
2428 		if (cred->cr_uid != va.va_uid)
2429 			return (EPERM);
2430 	}
2431 
2432 	if (hardlink_check_gid) {
2433 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2434 			return (EPERM);
2435 	}
2436 
2437 	return (0);
2438 }
2439 
2440 int
2441 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2442 {
2443 	struct thread *td = curthread;
2444 	struct vnode *vp;
2445 	int error;
2446 
2447 	/*
2448 	 * Lookup the source and obtained a locked vnode.
2449 	 *
2450 	 * You may only hardlink a file which you have write permission
2451 	 * on or which you own.
2452 	 *
2453 	 * XXX relookup on vget failure / race ?
2454 	 */
2455 	bwillinode(1);
2456 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2457 	if ((error = nlookup(nd)) != 0)
2458 		return (error);
2459 	vp = nd->nl_nch.ncp->nc_vp;
2460 	KKASSERT(vp != NULL);
2461 	if (vp->v_type == VDIR)
2462 		return (EPERM);		/* POSIX */
2463 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2464 		return (error);
2465 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2466 		return (error);
2467 
2468 	/*
2469 	 * Unlock the source so we can lookup the target without deadlocking
2470 	 * (XXX vp is locked already, possible other deadlock?).  The target
2471 	 * must not exist.
2472 	 */
2473 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2474 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2475 	cache_unlock(&nd->nl_nch);
2476 	vn_unlock(vp);
2477 
2478 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2479 	if ((error = nlookup(linknd)) != 0) {
2480 		vrele(vp);
2481 		return (error);
2482 	}
2483 	if (linknd->nl_nch.ncp->nc_vp) {
2484 		vrele(vp);
2485 		return (EEXIST);
2486 	}
2487 	VFS_MODIFYING(vp->v_mount);
2488 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2489 	if (error) {
2490 		vrele(vp);
2491 		return (error);
2492 	}
2493 
2494 	/*
2495 	 * Finally run the new API VOP.
2496 	 */
2497 	error = can_hardlink(vp, td, td->td_ucred);
2498 	if (error == 0) {
2499 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2500 				  vp, linknd->nl_cred);
2501 	}
2502 	vput(vp);
2503 	return (error);
2504 }
2505 
2506 /*
2507  * link_args(char *path, char *link)
2508  *
2509  * Make a hard file link.
2510  */
2511 int
2512 sys_link(struct link_args *uap)
2513 {
2514 	struct nlookupdata nd, linknd;
2515 	int error;
2516 
2517 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2518 	if (error == 0) {
2519 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2520 		if (error == 0)
2521 			error = kern_link(&nd, &linknd);
2522 		nlookup_done(&linknd);
2523 	}
2524 	nlookup_done(&nd);
2525 	return (error);
2526 }
2527 
2528 /*
2529  * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags)
2530  *
2531  * Make a hard file link. The path1 argument is relative to the directory
2532  * associated with fd1, and similarly the path2 argument is relative to
2533  * the directory associated with fd2.
2534  */
2535 int
2536 sys_linkat(struct linkat_args *uap)
2537 {
2538 	struct nlookupdata nd, linknd;
2539 	struct file *fp1, *fp2;
2540 	int error;
2541 
2542 	error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE,
2543 	    (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0);
2544 	if (error == 0) {
2545 		error = nlookup_init_at(&linknd, &fp2, uap->fd2,
2546 		    uap->path2, UIO_USERSPACE, 0);
2547 		if (error == 0)
2548 			error = kern_link(&nd, &linknd);
2549 		nlookup_done_at(&linknd, fp2);
2550 	}
2551 	nlookup_done_at(&nd, fp1);
2552 	return (error);
2553 }
2554 
2555 int
2556 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2557 {
2558 	struct vattr vattr;
2559 	struct vnode *vp;
2560 	struct vnode *dvp;
2561 	int error;
2562 
2563 	bwillinode(1);
2564 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2565 	if ((error = nlookup(nd)) != 0)
2566 		return (error);
2567 	if (nd->nl_nch.ncp->nc_vp)
2568 		return (EEXIST);
2569 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2570 		return (error);
2571 	dvp = nd->nl_dvp;
2572 	VATTR_NULL(&vattr);
2573 	vattr.va_mode = mode;
2574 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2575 	if (error == 0)
2576 		vput(vp);
2577 	return (error);
2578 }
2579 
2580 /*
2581  * symlink(char *path, char *link)
2582  *
2583  * Make a symbolic link.
2584  */
2585 int
2586 sys_symlink(struct symlink_args *uap)
2587 {
2588 	struct thread *td = curthread;
2589 	struct nlookupdata nd;
2590 	char *path;
2591 	int error;
2592 	int mode;
2593 
2594 	path = objcache_get(namei_oc, M_WAITOK);
2595 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2596 	if (error == 0) {
2597 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2598 		if (error == 0) {
2599 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2600 			error = kern_symlink(&nd, path, mode);
2601 		}
2602 		nlookup_done(&nd);
2603 	}
2604 	objcache_put(namei_oc, path);
2605 	return (error);
2606 }
2607 
2608 /*
2609  * symlinkat_args(char *path1, int fd, char *path2)
2610  *
2611  * Make a symbolic link.  The path2 argument is relative to the directory
2612  * associated with fd.
2613  */
2614 int
2615 sys_symlinkat(struct symlinkat_args *uap)
2616 {
2617 	struct thread *td = curthread;
2618 	struct nlookupdata nd;
2619 	struct file *fp;
2620 	char *path1;
2621 	int error;
2622 	int mode;
2623 
2624 	path1 = objcache_get(namei_oc, M_WAITOK);
2625 	error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2626 	if (error == 0) {
2627 		error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2628 		    UIO_USERSPACE, 0);
2629 		if (error == 0) {
2630 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2631 			error = kern_symlink(&nd, path1, mode);
2632 		}
2633 		nlookup_done_at(&nd, fp);
2634 	}
2635 	objcache_put(namei_oc, path1);
2636 	return (error);
2637 }
2638 
2639 /*
2640  * undelete_args(char *path)
2641  *
2642  * Delete a whiteout from the filesystem.
2643  */
2644 int
2645 sys_undelete(struct undelete_args *uap)
2646 {
2647 	struct nlookupdata nd;
2648 	int error;
2649 
2650 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2651 	bwillinode(1);
2652 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2653 	if (error == 0)
2654 		error = nlookup(&nd);
2655 	if (error == 0)
2656 		error = ncp_writechk(&nd.nl_nch);
2657 	if (error == 0) {
2658 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2659 				      NAMEI_DELETE);
2660 	}
2661 	nlookup_done(&nd);
2662 	return (error);
2663 }
2664 
2665 int
2666 kern_unlink(struct nlookupdata *nd)
2667 {
2668 	int error;
2669 
2670 	bwillinode(1);
2671 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2672 	if ((error = nlookup(nd)) != 0)
2673 		return (error);
2674 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2675 		return (error);
2676 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2677 	return (error);
2678 }
2679 
2680 /*
2681  * unlink_args(char *path)
2682  *
2683  * Delete a name from the filesystem.
2684  */
2685 int
2686 sys_unlink(struct unlink_args *uap)
2687 {
2688 	struct nlookupdata nd;
2689 	int error;
2690 
2691 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2692 	if (error == 0)
2693 		error = kern_unlink(&nd);
2694 	nlookup_done(&nd);
2695 	return (error);
2696 }
2697 
2698 
2699 /*
2700  * unlinkat_args(int fd, char *path, int flags)
2701  *
2702  * Delete the file or directory entry pointed to by fd/path.
2703  */
2704 int
2705 sys_unlinkat(struct unlinkat_args *uap)
2706 {
2707 	struct nlookupdata nd;
2708 	struct file *fp;
2709 	int error;
2710 
2711 	if (uap->flags & ~AT_REMOVEDIR)
2712 		return (EINVAL);
2713 
2714 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2715 	if (error == 0) {
2716 		if (uap->flags & AT_REMOVEDIR)
2717 			error = kern_rmdir(&nd);
2718 		else
2719 			error = kern_unlink(&nd);
2720 	}
2721 	nlookup_done_at(&nd, fp);
2722 	return (error);
2723 }
2724 
2725 int
2726 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2727 {
2728 	struct thread *td = curthread;
2729 	struct file *fp;
2730 	struct vnode *vp;
2731 	struct vattr vattr;
2732 	off_t new_offset;
2733 	int error;
2734 
2735 	fp = holdfp(td, fd, -1);
2736 	if (fp == NULL)
2737 		return (EBADF);
2738 	if (fp->f_type != DTYPE_VNODE) {
2739 		error = ESPIPE;
2740 		goto done;
2741 	}
2742 	vp = (struct vnode *)fp->f_data;
2743 
2744 	switch (whence) {
2745 	case L_INCR:
2746 		spin_lock(&fp->f_spin);
2747 		new_offset = fp->f_offset + offset;
2748 		error = 0;
2749 		break;
2750 	case L_XTND:
2751 		error = VOP_GETATTR_FP(vp, &vattr, fp);
2752 		spin_lock(&fp->f_spin);
2753 		new_offset = offset + vattr.va_size;
2754 		break;
2755 	case L_SET:
2756 		new_offset = offset;
2757 		error = 0;
2758 		spin_lock(&fp->f_spin);
2759 		break;
2760 	default:
2761 		new_offset = 0;
2762 		error = EINVAL;
2763 		spin_lock(&fp->f_spin);
2764 		break;
2765 	}
2766 
2767 	/*
2768 	 * Validate the seek position.  Negative offsets are not allowed
2769 	 * for regular files or directories.
2770 	 *
2771 	 * Normally we would also not want to allow negative offsets for
2772 	 * character and block-special devices.  However kvm addresses
2773 	 * on 64 bit architectures might appear to be negative and must
2774 	 * be allowed.
2775 	 */
2776 	if (error == 0) {
2777 		if (new_offset < 0 &&
2778 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2779 			error = EINVAL;
2780 		} else {
2781 			fp->f_offset = new_offset;
2782 		}
2783 	}
2784 	*res = fp->f_offset;
2785 	spin_unlock(&fp->f_spin);
2786 done:
2787 	dropfp(td, fd, fp);
2788 
2789 	return (error);
2790 }
2791 
2792 /*
2793  * lseek_args(int fd, int pad, off_t offset, int whence)
2794  *
2795  * Reposition read/write file offset.
2796  */
2797 int
2798 sys_lseek(struct lseek_args *uap)
2799 {
2800 	int error;
2801 
2802 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2803 			   &uap->sysmsg_offset);
2804 
2805 	return (error);
2806 }
2807 
2808 /*
2809  * Check if current process can access given file.  amode is a bitmask of *_OK
2810  * access bits.  flags is a bitmask of AT_* flags.
2811  */
2812 int
2813 kern_access(struct nlookupdata *nd, int amode, int flags)
2814 {
2815 	struct vnode *vp;
2816 	int error, mode;
2817 
2818 	if (flags & ~AT_EACCESS)
2819 		return (EINVAL);
2820 	nd->nl_flags |= NLC_SHAREDLOCK;
2821 	if ((error = nlookup(nd)) != 0)
2822 		return (error);
2823 retry:
2824 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
2825 	if (error)
2826 		return (error);
2827 
2828 	/* Flags == 0 means only check for existence. */
2829 	if (amode) {
2830 		mode = 0;
2831 		if (amode & R_OK)
2832 			mode |= VREAD;
2833 		if (amode & W_OK)
2834 			mode |= VWRITE;
2835 		if (amode & X_OK)
2836 			mode |= VEXEC;
2837 		if ((mode & VWRITE) == 0 ||
2838 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2839 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2840 
2841 		/*
2842 		 * If the file handle is stale we have to re-resolve the
2843 		 * entry with the ncp held exclusively.  This is a hack
2844 		 * at the moment.
2845 		 */
2846 		if (error == ESTALE) {
2847 			vput(vp);
2848 			cache_unlock(&nd->nl_nch);
2849 			cache_lock(&nd->nl_nch);
2850 			cache_setunresolved(&nd->nl_nch);
2851 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2852 			if (error == 0) {
2853 				vp = NULL;
2854 				goto retry;
2855 			}
2856 			return(error);
2857 		}
2858 	}
2859 	vput(vp);
2860 	return (error);
2861 }
2862 
2863 /*
2864  * access_args(char *path, int flags)
2865  *
2866  * Check access permissions.
2867  */
2868 int
2869 sys_access(struct access_args *uap)
2870 {
2871 	struct nlookupdata nd;
2872 	int error;
2873 
2874 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2875 	if (error == 0)
2876 		error = kern_access(&nd, uap->flags, 0);
2877 	nlookup_done(&nd);
2878 	return (error);
2879 }
2880 
2881 
2882 /*
2883  * eaccess_args(char *path, int flags)
2884  *
2885  * Check access permissions.
2886  */
2887 int
2888 sys_eaccess(struct eaccess_args *uap)
2889 {
2890 	struct nlookupdata nd;
2891 	int error;
2892 
2893 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2894 	if (error == 0)
2895 		error = kern_access(&nd, uap->flags, AT_EACCESS);
2896 	nlookup_done(&nd);
2897 	return (error);
2898 }
2899 
2900 
2901 /*
2902  * faccessat_args(int fd, char *path, int amode, int flags)
2903  *
2904  * Check access permissions.
2905  */
2906 int
2907 sys_faccessat(struct faccessat_args *uap)
2908 {
2909 	struct nlookupdata nd;
2910 	struct file *fp;
2911 	int error;
2912 
2913 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2914 				NLC_FOLLOW);
2915 	if (error == 0)
2916 		error = kern_access(&nd, uap->amode, uap->flags);
2917 	nlookup_done_at(&nd, fp);
2918 	return (error);
2919 }
2920 
2921 int
2922 kern_stat(struct nlookupdata *nd, struct stat *st)
2923 {
2924 	int error;
2925 	struct vnode *vp;
2926 
2927 	nd->nl_flags |= NLC_SHAREDLOCK;
2928 	if ((error = nlookup(nd)) != 0)
2929 		return (error);
2930 again:
2931 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2932 		return (ENOENT);
2933 
2934 	if ((error = vget(vp, LK_SHARED)) != 0)
2935 		return (error);
2936 	error = vn_stat(vp, st, nd->nl_cred);
2937 
2938 	/*
2939 	 * If the file handle is stale we have to re-resolve the
2940 	 * entry with the ncp held exclusively.  This is a hack
2941 	 * at the moment.
2942 	 */
2943 	if (error == ESTALE) {
2944 		vput(vp);
2945 		cache_unlock(&nd->nl_nch);
2946 		cache_lock(&nd->nl_nch);
2947 		cache_setunresolved(&nd->nl_nch);
2948 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2949 		if (error == 0)
2950 			goto again;
2951 	} else {
2952 		vput(vp);
2953 	}
2954 	return (error);
2955 }
2956 
2957 /*
2958  * stat_args(char *path, struct stat *ub)
2959  *
2960  * Get file status; this version follows links.
2961  */
2962 int
2963 sys_stat(struct stat_args *uap)
2964 {
2965 	struct nlookupdata nd;
2966 	struct stat st;
2967 	int error;
2968 
2969 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2970 	if (error == 0) {
2971 		error = kern_stat(&nd, &st);
2972 		if (error == 0)
2973 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2974 	}
2975 	nlookup_done(&nd);
2976 	return (error);
2977 }
2978 
2979 /*
2980  * lstat_args(char *path, struct stat *ub)
2981  *
2982  * Get file status; this version does not follow links.
2983  */
2984 int
2985 sys_lstat(struct lstat_args *uap)
2986 {
2987 	struct nlookupdata nd;
2988 	struct stat st;
2989 	int error;
2990 
2991 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2992 	if (error == 0) {
2993 		error = kern_stat(&nd, &st);
2994 		if (error == 0)
2995 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2996 	}
2997 	nlookup_done(&nd);
2998 	return (error);
2999 }
3000 
3001 /*
3002  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
3003  *
3004  * Get status of file pointed to by fd/path.
3005  */
3006 int
3007 sys_fstatat(struct fstatat_args *uap)
3008 {
3009 	struct nlookupdata nd;
3010 	struct stat st;
3011 	int error;
3012 	int flags;
3013 	struct file *fp;
3014 
3015 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3016 		return (EINVAL);
3017 
3018 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3019 
3020 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3021 				UIO_USERSPACE, flags);
3022 	if (error == 0) {
3023 		error = kern_stat(&nd, &st);
3024 		if (error == 0)
3025 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
3026 	}
3027 	nlookup_done_at(&nd, fp);
3028 	return (error);
3029 }
3030 
3031 static int
3032 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp)
3033 {
3034 	struct nlookupdata nd;
3035 	struct vnode *vp;
3036 	int error;
3037 
3038 	vp = NULL;
3039 	error = nlookup_init(&nd, path, UIO_USERSPACE, flags);
3040 	if (error == 0)
3041 		error = nlookup(&nd);
3042 	if (error == 0)
3043 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3044 	nlookup_done(&nd);
3045 	if (error == 0) {
3046 		error = VOP_PATHCONF(vp, name, sysmsg_regp);
3047 		vput(vp);
3048 	}
3049 	return (error);
3050 }
3051 
3052 /*
3053  * pathconf_Args(char *path, int name)
3054  *
3055  * Get configurable pathname variables.
3056  */
3057 int
3058 sys_pathconf(struct pathconf_args *uap)
3059 {
3060 	return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW,
3061 		&uap->sysmsg_reg));
3062 }
3063 
3064 /*
3065  * lpathconf_Args(char *path, int name)
3066  *
3067  * Get configurable pathname variables, but don't follow symlinks.
3068  */
3069 int
3070 sys_lpathconf(struct lpathconf_args *uap)
3071 {
3072 	return (kern_pathconf(uap->path, uap->name, 0, &uap->sysmsg_reg));
3073 }
3074 
3075 /*
3076  * XXX: daver
3077  * kern_readlink isn't properly split yet.  There is a copyin burried
3078  * in VOP_READLINK().
3079  */
3080 int
3081 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
3082 {
3083 	struct thread *td = curthread;
3084 	struct vnode *vp;
3085 	struct iovec aiov;
3086 	struct uio auio;
3087 	int error;
3088 
3089 	nd->nl_flags |= NLC_SHAREDLOCK;
3090 	if ((error = nlookup(nd)) != 0)
3091 		return (error);
3092 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
3093 	if (error)
3094 		return (error);
3095 	if (vp->v_type != VLNK) {
3096 		error = EINVAL;
3097 	} else {
3098 		aiov.iov_base = buf;
3099 		aiov.iov_len = count;
3100 		auio.uio_iov = &aiov;
3101 		auio.uio_iovcnt = 1;
3102 		auio.uio_offset = 0;
3103 		auio.uio_rw = UIO_READ;
3104 		auio.uio_segflg = UIO_USERSPACE;
3105 		auio.uio_td = td;
3106 		auio.uio_resid = count;
3107 		error = VOP_READLINK(vp, &auio, td->td_ucred);
3108 	}
3109 	vput(vp);
3110 	*res = count - auio.uio_resid;
3111 	return (error);
3112 }
3113 
3114 /*
3115  * readlink_args(char *path, char *buf, int count)
3116  *
3117  * Return target name of a symbolic link.
3118  */
3119 int
3120 sys_readlink(struct readlink_args *uap)
3121 {
3122 	struct nlookupdata nd;
3123 	int error;
3124 
3125 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3126 	if (error == 0) {
3127 		error = kern_readlink(&nd, uap->buf, uap->count,
3128 					&uap->sysmsg_result);
3129 	}
3130 	nlookup_done(&nd);
3131 	return (error);
3132 }
3133 
3134 /*
3135  * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
3136  *
3137  * Return target name of a symbolic link.  The path is relative to the
3138  * directory associated with fd.
3139  */
3140 int
3141 sys_readlinkat(struct readlinkat_args *uap)
3142 {
3143 	struct nlookupdata nd;
3144 	struct file *fp;
3145 	int error;
3146 
3147 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
3148 	if (error == 0) {
3149 		error = kern_readlink(&nd, uap->buf, uap->bufsize,
3150 					&uap->sysmsg_result);
3151 	}
3152 	nlookup_done_at(&nd, fp);
3153 	return (error);
3154 }
3155 
3156 static int
3157 setfflags(struct vnode *vp, u_long flags)
3158 {
3159 	struct thread *td = curthread;
3160 	int error;
3161 	struct vattr vattr;
3162 
3163 	/*
3164 	 * Prevent non-root users from setting flags on devices.  When
3165 	 * a device is reused, users can retain ownership of the device
3166 	 * if they are allowed to set flags and programs assume that
3167 	 * chown can't fail when done as root.
3168 	 */
3169 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
3170 	    ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
3171 		return (error);
3172 
3173 	/*
3174 	 * note: vget is required for any operation that might mod the vnode
3175 	 * so VINACTIVE is properly cleared.
3176 	 */
3177 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3178 		VATTR_NULL(&vattr);
3179 		vattr.va_flags = flags;
3180 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3181 		vput(vp);
3182 	}
3183 	return (error);
3184 }
3185 
3186 /*
3187  * chflags(const char *path, u_long flags)
3188  *
3189  * Change flags of a file given a path name.
3190  */
3191 int
3192 sys_chflags(struct chflags_args *uap)
3193 {
3194 	struct nlookupdata nd;
3195 	struct vnode *vp;
3196 	int error;
3197 
3198 	vp = NULL;
3199 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3200 	if (error == 0)
3201 		error = nlookup(&nd);
3202 	if (error == 0)
3203 		error = ncp_writechk(&nd.nl_nch);
3204 	if (error == 0)
3205 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3206 	nlookup_done(&nd);
3207 	if (error == 0) {
3208 		error = setfflags(vp, uap->flags);
3209 		vrele(vp);
3210 	}
3211 	return (error);
3212 }
3213 
3214 /*
3215  * lchflags(const char *path, u_long flags)
3216  *
3217  * Change flags of a file given a path name, but don't follow symlinks.
3218  */
3219 int
3220 sys_lchflags(struct lchflags_args *uap)
3221 {
3222 	struct nlookupdata nd;
3223 	struct vnode *vp;
3224 	int error;
3225 
3226 	vp = NULL;
3227 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3228 	if (error == 0)
3229 		error = nlookup(&nd);
3230 	if (error == 0)
3231 		error = ncp_writechk(&nd.nl_nch);
3232 	if (error == 0)
3233 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3234 	nlookup_done(&nd);
3235 	if (error == 0) {
3236 		error = setfflags(vp, uap->flags);
3237 		vrele(vp);
3238 	}
3239 	return (error);
3240 }
3241 
3242 /*
3243  * fchflags_args(int fd, u_flags flags)
3244  *
3245  * Change flags of a file given a file descriptor.
3246  */
3247 int
3248 sys_fchflags(struct fchflags_args *uap)
3249 {
3250 	struct thread *td = curthread;
3251 	struct file *fp;
3252 	int error;
3253 
3254 	if ((error = holdvnode(td, uap->fd, &fp)) != 0)
3255 		return (error);
3256 	if (fp->f_nchandle.ncp)
3257 		error = ncp_writechk(&fp->f_nchandle);
3258 	if (error == 0)
3259 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
3260 	fdrop(fp);
3261 	return (error);
3262 }
3263 
3264 /*
3265  * chflagsat_args(int fd, const char *path, u_long flags, int atflags)
3266  * change flags given a pathname relative to a filedescriptor
3267  */
3268 int sys_chflagsat(struct chflagsat_args *uap)
3269 {
3270 	struct nlookupdata nd;
3271 	struct vnode *vp;
3272 	struct file *fp;
3273 	int error;
3274 	int lookupflags;
3275 
3276 	if (uap->atflags & ~AT_SYMLINK_NOFOLLOW)
3277 		return (EINVAL);
3278 
3279 	lookupflags = (uap->atflags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3280 
3281 	vp = NULL;
3282 	error = nlookup_init_at(&nd, &fp, uap->fd,  uap->path, UIO_USERSPACE, lookupflags);
3283 	if (error == 0)
3284 		error = nlookup(&nd);
3285 	if (error == 0)
3286 		error = ncp_writechk(&nd.nl_nch);
3287 	if (error == 0)
3288 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3289 	nlookup_done_at(&nd, fp);
3290 	if (error == 0) {
3291 		error = setfflags(vp, uap->flags);
3292 		vrele(vp);
3293 	}
3294 	return (error);
3295 }
3296 
3297 
3298 static int
3299 setfmode(struct vnode *vp, int mode)
3300 {
3301 	struct thread *td = curthread;
3302 	int error;
3303 	struct vattr vattr;
3304 
3305 	/*
3306 	 * note: vget is required for any operation that might mod the vnode
3307 	 * so VINACTIVE is properly cleared.
3308 	 */
3309 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3310 		VATTR_NULL(&vattr);
3311 		vattr.va_mode = mode & ALLPERMS;
3312 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3313 		cache_inval_wxok(vp);
3314 		vput(vp);
3315 	}
3316 	return error;
3317 }
3318 
3319 int
3320 kern_chmod(struct nlookupdata *nd, int mode)
3321 {
3322 	struct vnode *vp;
3323 	int error;
3324 
3325 	if ((error = nlookup(nd)) != 0)
3326 		return (error);
3327 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3328 		return (error);
3329 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3330 		error = setfmode(vp, mode);
3331 	vrele(vp);
3332 	return (error);
3333 }
3334 
3335 /*
3336  * chmod_args(char *path, int mode)
3337  *
3338  * Change mode of a file given path name.
3339  */
3340 int
3341 sys_chmod(struct chmod_args *uap)
3342 {
3343 	struct nlookupdata nd;
3344 	int error;
3345 
3346 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3347 	if (error == 0)
3348 		error = kern_chmod(&nd, uap->mode);
3349 	nlookup_done(&nd);
3350 	return (error);
3351 }
3352 
3353 /*
3354  * lchmod_args(char *path, int mode)
3355  *
3356  * Change mode of a file given path name (don't follow links.)
3357  */
3358 int
3359 sys_lchmod(struct lchmod_args *uap)
3360 {
3361 	struct nlookupdata nd;
3362 	int error;
3363 
3364 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3365 	if (error == 0)
3366 		error = kern_chmod(&nd, uap->mode);
3367 	nlookup_done(&nd);
3368 	return (error);
3369 }
3370 
3371 /*
3372  * fchmod_args(int fd, int mode)
3373  *
3374  * Change mode of a file given a file descriptor.
3375  */
3376 int
3377 sys_fchmod(struct fchmod_args *uap)
3378 {
3379 	struct thread *td = curthread;
3380 	struct file *fp;
3381 	int error;
3382 
3383 	if ((error = holdvnode(td, uap->fd, &fp)) != 0)
3384 		return (error);
3385 	if (fp->f_nchandle.ncp)
3386 		error = ncp_writechk(&fp->f_nchandle);
3387 	if (error == 0)
3388 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
3389 	fdrop(fp);
3390 	return (error);
3391 }
3392 
3393 /*
3394  * fchmodat_args(char *path, int mode)
3395  *
3396  * Change mode of a file pointed to by fd/path.
3397  */
3398 int
3399 sys_fchmodat(struct fchmodat_args *uap)
3400 {
3401 	struct nlookupdata nd;
3402 	struct file *fp;
3403 	int error;
3404 	int flags;
3405 
3406 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3407 		return (EINVAL);
3408 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3409 
3410 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3411 				UIO_USERSPACE, flags);
3412 	if (error == 0)
3413 		error = kern_chmod(&nd, uap->mode);
3414 	nlookup_done_at(&nd, fp);
3415 	return (error);
3416 }
3417 
3418 static int
3419 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid)
3420 {
3421 	struct thread *td = curthread;
3422 	int error;
3423 	struct vattr vattr;
3424 	uid_t o_uid;
3425 	gid_t o_gid;
3426 	uint64_t size;
3427 
3428 	/*
3429 	 * note: vget is required for any operation that might mod the vnode
3430 	 * so VINACTIVE is properly cleared.
3431 	 */
3432 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3433 		if ((error = VOP_GETATTR(vp, &vattr)) != 0)
3434 			return error;
3435 		o_uid = vattr.va_uid;
3436 		o_gid = vattr.va_gid;
3437 		size = vattr.va_size;
3438 
3439 		VATTR_NULL(&vattr);
3440 		vattr.va_uid = uid;
3441 		vattr.va_gid = gid;
3442 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3443 		vput(vp);
3444 	}
3445 
3446 	if (error == 0) {
3447 		if (uid == -1)
3448 			uid = o_uid;
3449 		if (gid == -1)
3450 			gid = o_gid;
3451 		VFS_ACCOUNT(mp, o_uid, o_gid, -size);
3452 		VFS_ACCOUNT(mp,   uid,   gid,  size);
3453 	}
3454 
3455 	return error;
3456 }
3457 
3458 int
3459 kern_chown(struct nlookupdata *nd, int uid, int gid)
3460 {
3461 	struct vnode *vp;
3462 	int error;
3463 
3464 	if ((error = nlookup(nd)) != 0)
3465 		return (error);
3466 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3467 		return (error);
3468 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3469 		error = setfown(nd->nl_nch.mount, vp, uid, gid);
3470 	vrele(vp);
3471 	return (error);
3472 }
3473 
3474 /*
3475  * chown(char *path, int uid, int gid)
3476  *
3477  * Set ownership given a path name.
3478  */
3479 int
3480 sys_chown(struct chown_args *uap)
3481 {
3482 	struct nlookupdata nd;
3483 	int error;
3484 
3485 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3486 	if (error == 0)
3487 		error = kern_chown(&nd, uap->uid, uap->gid);
3488 	nlookup_done(&nd);
3489 	return (error);
3490 }
3491 
3492 /*
3493  * lchown_args(char *path, int uid, int gid)
3494  *
3495  * Set ownership given a path name, do not cross symlinks.
3496  */
3497 int
3498 sys_lchown(struct lchown_args *uap)
3499 {
3500 	struct nlookupdata nd;
3501 	int error;
3502 
3503 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3504 	if (error == 0)
3505 		error = kern_chown(&nd, uap->uid, uap->gid);
3506 	nlookup_done(&nd);
3507 	return (error);
3508 }
3509 
3510 /*
3511  * fchown_args(int fd, int uid, int gid)
3512  *
3513  * Set ownership given a file descriptor.
3514  */
3515 int
3516 sys_fchown(struct fchown_args *uap)
3517 {
3518 	struct thread *td = curthread;
3519 	struct proc *p = td->td_proc;
3520 	struct file *fp;
3521 	int error;
3522 
3523 	if ((error = holdvnode(td, uap->fd, &fp)) != 0)
3524 		return (error);
3525 	if (fp->f_nchandle.ncp)
3526 		error = ncp_writechk(&fp->f_nchandle);
3527 	if (error == 0)
3528 		error = setfown(p->p_fd->fd_ncdir.mount,
3529 			(struct vnode *)fp->f_data, uap->uid, uap->gid);
3530 	fdrop(fp);
3531 	return (error);
3532 }
3533 
3534 /*
3535  * fchownat(int fd, char *path, int uid, int gid, int flags)
3536  *
3537  * Set ownership of file pointed to by fd/path.
3538  */
3539 int
3540 sys_fchownat(struct fchownat_args *uap)
3541 {
3542 	struct nlookupdata nd;
3543 	struct file *fp;
3544 	int error;
3545 	int flags;
3546 
3547 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3548 		return (EINVAL);
3549 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3550 
3551 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3552 				UIO_USERSPACE, flags);
3553 	if (error == 0)
3554 		error = kern_chown(&nd, uap->uid, uap->gid);
3555 	nlookup_done_at(&nd, fp);
3556 	return (error);
3557 }
3558 
3559 
3560 static int
3561 getutimes(struct timeval *tvp, struct timespec *tsp)
3562 {
3563 	struct timeval tv[2];
3564 	int error;
3565 
3566 	if (tvp == NULL) {
3567 		microtime(&tv[0]);
3568 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3569 		tsp[1] = tsp[0];
3570 	} else {
3571 		if ((error = itimerfix(tvp)) != 0)
3572 			return (error);
3573 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3574 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3575 	}
3576 	return 0;
3577 }
3578 
3579 static int
3580 getutimens(const struct timespec *ts, struct timespec *newts, int *nullflag)
3581 {
3582 	struct timespec tsnow;
3583 	int error;
3584 
3585 	*nullflag = 0;
3586 	nanotime(&tsnow);
3587 	if (ts == NULL) {
3588 		newts[0] = tsnow;
3589 		newts[1] = tsnow;
3590 		*nullflag = 1;
3591 		return (0);
3592 	}
3593 
3594 	newts[0] = ts[0];
3595 	newts[1] = ts[1];
3596 	if (newts[0].tv_nsec == UTIME_OMIT && newts[1].tv_nsec == UTIME_OMIT)
3597 		return (0);
3598 	if (newts[0].tv_nsec == UTIME_NOW && newts[1].tv_nsec == UTIME_NOW)
3599 		*nullflag = 1;
3600 
3601 	if (newts[0].tv_nsec == UTIME_OMIT)
3602 		newts[0].tv_sec = VNOVAL;
3603 	else if (newts[0].tv_nsec == UTIME_NOW)
3604 		newts[0] = tsnow;
3605 	else if ((error = itimespecfix(&newts[0])) != 0)
3606 		return (error);
3607 
3608 	if (newts[1].tv_nsec == UTIME_OMIT)
3609 		newts[1].tv_sec = VNOVAL;
3610 	else if (newts[1].tv_nsec == UTIME_NOW)
3611 		newts[1] = tsnow;
3612 	else if ((error = itimespecfix(&newts[1])) != 0)
3613 		return (error);
3614 
3615 	return (0);
3616 }
3617 
3618 static int
3619 setutimes(struct vnode *vp, struct vattr *vattr,
3620 	  const struct timespec *ts, int nullflag)
3621 {
3622 	struct thread *td = curthread;
3623 	int error;
3624 
3625 	VATTR_NULL(vattr);
3626 	vattr->va_atime = ts[0];
3627 	vattr->va_mtime = ts[1];
3628 	if (nullflag)
3629 		vattr->va_vaflags |= VA_UTIMES_NULL;
3630 	error = VOP_SETATTR(vp, vattr, td->td_ucred);
3631 
3632 	return error;
3633 }
3634 
3635 int
3636 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3637 {
3638 	struct timespec ts[2];
3639 	int error;
3640 
3641 	if (tptr) {
3642 		if ((error = getutimes(tptr, ts)) != 0)
3643 			return (error);
3644 	}
3645 	error = kern_utimensat(nd, tptr ? ts : NULL, 0);
3646 	return (error);
3647 }
3648 
3649 /*
3650  * utimes_args(char *path, struct timeval *tptr)
3651  *
3652  * Set the access and modification times of a file.
3653  */
3654 int
3655 sys_utimes(struct utimes_args *uap)
3656 {
3657 	struct timeval tv[2];
3658 	struct nlookupdata nd;
3659 	int error;
3660 
3661 	if (uap->tptr) {
3662  		error = copyin(uap->tptr, tv, sizeof(tv));
3663 		if (error)
3664 			return (error);
3665 	}
3666 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3667 	if (error == 0)
3668 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3669 	nlookup_done(&nd);
3670 	return (error);
3671 }
3672 
3673 /*
3674  * lutimes_args(char *path, struct timeval *tptr)
3675  *
3676  * Set the access and modification times of a file.
3677  */
3678 int
3679 sys_lutimes(struct lutimes_args *uap)
3680 {
3681 	struct timeval tv[2];
3682 	struct nlookupdata nd;
3683 	int error;
3684 
3685 	if (uap->tptr) {
3686 		error = copyin(uap->tptr, tv, sizeof(tv));
3687 		if (error)
3688 			return (error);
3689 	}
3690 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3691 	if (error == 0)
3692 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3693 	nlookup_done(&nd);
3694 	return (error);
3695 }
3696 
3697 /*
3698  * Set utimes on a file descriptor.  The creds used to open the
3699  * file are used to determine whether the operation is allowed
3700  * or not.
3701  */
3702 int
3703 kern_futimens(int fd, struct timespec *ts)
3704 {
3705 	struct thread *td = curthread;
3706 	struct timespec newts[2];
3707 	struct file *fp;
3708 	struct vnode *vp;
3709 	struct vattr vattr;
3710 	int nullflag;
3711 	int error;
3712 
3713 	error = getutimens(ts, newts, &nullflag);
3714 	if (error)
3715 		return (error);
3716 	if ((error = holdvnode(td, fd, &fp)) != 0)
3717 		return (error);
3718 	if (fp->f_nchandle.ncp)
3719 		error = ncp_writechk(&fp->f_nchandle);
3720 	if (error == 0) {
3721 		vp = fp->f_data;
3722 		error = vget(vp, LK_EXCLUSIVE);
3723 		if (error == 0) {
3724 			error = VOP_GETATTR_FP(vp, &vattr, fp);
3725 			if (error == 0) {
3726 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3727 						   fp->f_cred);
3728 			}
3729 			if (error == 0) {
3730 				error = setutimes(vp, &vattr, newts, nullflag);
3731 			}
3732 			vput(vp);
3733 		}
3734 	}
3735 	fdrop(fp);
3736 	return (error);
3737 }
3738 
3739 /*
3740  * futimens_args(int fd, struct timespec *ts)
3741  *
3742  * Set the access and modification times of a file.
3743  */
3744 int
3745 sys_futimens(struct futimens_args *uap)
3746 {
3747 	struct timespec ts[2];
3748 	int error;
3749 
3750 	if (uap->ts) {
3751 		error = copyin(uap->ts, ts, sizeof(ts));
3752 		if (error)
3753 			return (error);
3754 	}
3755 	error = kern_futimens(uap->fd, uap->ts ? ts : NULL);
3756 	return (error);
3757 }
3758 
3759 int
3760 kern_futimes(int fd, struct timeval *tptr)
3761 {
3762 	struct timespec ts[2];
3763 	int error;
3764 
3765 	if (tptr) {
3766 		if ((error = getutimes(tptr, ts)) != 0)
3767 			return (error);
3768 	}
3769 	error = kern_futimens(fd, tptr ? ts : NULL);
3770 	return (error);
3771 }
3772 
3773 /*
3774  * futimes_args(int fd, struct timeval *tptr)
3775  *
3776  * Set the access and modification times of a file.
3777  */
3778 int
3779 sys_futimes(struct futimes_args *uap)
3780 {
3781 	struct timeval tv[2];
3782 	int error;
3783 
3784 	if (uap->tptr) {
3785 		error = copyin(uap->tptr, tv, sizeof(tv));
3786 		if (error)
3787 			return (error);
3788 	}
3789 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3790 	return (error);
3791 }
3792 
3793 int
3794 kern_utimensat(struct nlookupdata *nd, const struct timespec *ts, int flags)
3795 {
3796 	struct timespec newts[2];
3797 	struct vnode *vp;
3798 	struct vattr vattr;
3799 	int nullflag;
3800 	int error;
3801 
3802 	if (flags & ~AT_SYMLINK_NOFOLLOW)
3803 		return (EINVAL);
3804 
3805 	error = getutimens(ts, newts, &nullflag);
3806 	if (error)
3807 		return (error);
3808 
3809 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3810 	if ((error = nlookup(nd)) != 0)
3811 		return (error);
3812 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3813 		return (error);
3814 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3815 		return (error);
3816 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3817 		error = vget(vp, LK_EXCLUSIVE);
3818 		if (error == 0) {
3819 			error = setutimes(vp, &vattr, newts, nullflag);
3820 			vput(vp);
3821 		}
3822 	}
3823 	vrele(vp);
3824 	return (error);
3825 }
3826 
3827 /*
3828  * utimensat_args(int fd, const char *path, const struct timespec *ts, int flags);
3829  *
3830  * Set file access and modification times of a file.
3831  */
3832 int
3833 sys_utimensat(struct utimensat_args *uap)
3834 {
3835 	struct timespec ts[2];
3836 	struct nlookupdata nd;
3837 	struct file *fp;
3838 	int error;
3839 	int flags;
3840 
3841 	if (uap->ts) {
3842 		error = copyin(uap->ts, ts, sizeof(ts));
3843 		if (error)
3844 			return (error);
3845 	}
3846 
3847 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3848 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3849 	                        UIO_USERSPACE, flags);
3850 	if (error == 0)
3851 		error = kern_utimensat(&nd, uap->ts ? ts : NULL, uap->flags);
3852 	nlookup_done_at(&nd, fp);
3853 	return (error);
3854 }
3855 
3856 int
3857 kern_truncate(struct nlookupdata *nd, off_t length)
3858 {
3859 	struct vnode *vp;
3860 	struct vattr vattr;
3861 	int error;
3862 	uid_t uid = 0;
3863 	gid_t gid = 0;
3864 	uint64_t old_size = 0;
3865 
3866 	if (length < 0)
3867 		return(EINVAL);
3868 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3869 	if ((error = nlookup(nd)) != 0)
3870 		return (error);
3871 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3872 		return (error);
3873 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3874 		return (error);
3875 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
3876 	if (error) {
3877 		vrele(vp);
3878 		return (error);
3879 	}
3880 	if (vp->v_type == VDIR) {
3881 		error = EISDIR;
3882 		goto done;
3883 	}
3884 	if (vfs_quota_enabled) {
3885 		error = VOP_GETATTR(vp, &vattr);
3886 		KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0"));
3887 		uid = vattr.va_uid;
3888 		gid = vattr.va_gid;
3889 		old_size = vattr.va_size;
3890 	}
3891 
3892 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3893 		VATTR_NULL(&vattr);
3894 		vattr.va_size = length;
3895 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3896 		VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size);
3897 	}
3898 done:
3899 	vput(vp);
3900 	return (error);
3901 }
3902 
3903 /*
3904  * truncate(char *path, int pad, off_t length)
3905  *
3906  * Truncate a file given its path name.
3907  */
3908 int
3909 sys_truncate(struct truncate_args *uap)
3910 {
3911 	struct nlookupdata nd;
3912 	int error;
3913 
3914 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3915 	if (error == 0)
3916 		error = kern_truncate(&nd, uap->length);
3917 	nlookup_done(&nd);
3918 	return error;
3919 }
3920 
3921 int
3922 kern_ftruncate(int fd, off_t length)
3923 {
3924 	struct thread *td = curthread;
3925 	struct vattr vattr;
3926 	struct vnode *vp;
3927 	struct file *fp;
3928 	int error;
3929 	uid_t uid = 0;
3930 	gid_t gid = 0;
3931 	uint64_t old_size = 0;
3932 	struct mount *mp;
3933 
3934 	if (length < 0)
3935 		return(EINVAL);
3936 	if ((error = holdvnode(td, fd, &fp)) != 0)
3937 		return (error);
3938 	if (fp->f_nchandle.ncp) {
3939 		error = ncp_writechk(&fp->f_nchandle);
3940 		if (error)
3941 			goto done;
3942 	}
3943 	if ((fp->f_flag & FWRITE) == 0) {
3944 		error = EINVAL;
3945 		goto done;
3946 	}
3947 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3948 		error = EINVAL;
3949 		goto done;
3950 	}
3951 	vp = (struct vnode *)fp->f_data;
3952 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3953 	if (vp->v_type == VDIR) {
3954 		error = EISDIR;
3955 		vn_unlock(vp);
3956 		goto done;
3957 	}
3958 
3959 	if (vfs_quota_enabled) {
3960 		error = VOP_GETATTR_FP(vp, &vattr, fp);
3961 		KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0"));
3962 		uid = vattr.va_uid;
3963 		gid = vattr.va_gid;
3964 		old_size = vattr.va_size;
3965 	}
3966 
3967 	if ((error = vn_writechk(vp, NULL)) == 0) {
3968 		VATTR_NULL(&vattr);
3969 		vattr.va_size = length;
3970 		error = VOP_SETATTR_FP(vp, &vattr, fp->f_cred, fp);
3971 		mp = vq_vptomp(vp);
3972 		VFS_ACCOUNT(mp, uid, gid, length - old_size);
3973 	}
3974 	vn_unlock(vp);
3975 done:
3976 	fdrop(fp);
3977 	return (error);
3978 }
3979 
3980 /*
3981  * ftruncate_args(int fd, int pad, off_t length)
3982  *
3983  * Truncate a file given a file descriptor.
3984  */
3985 int
3986 sys_ftruncate(struct ftruncate_args *uap)
3987 {
3988 	int error;
3989 
3990 	error = kern_ftruncate(uap->fd, uap->length);
3991 
3992 	return (error);
3993 }
3994 
3995 /*
3996  * fsync(int fd)
3997  *
3998  * Sync an open file.
3999  */
4000 int
4001 sys_fsync(struct fsync_args *uap)
4002 {
4003 	struct thread *td = curthread;
4004 	struct vnode *vp;
4005 	struct file *fp;
4006 	vm_object_t obj;
4007 	int error;
4008 
4009 	if ((error = holdvnode(td, uap->fd, &fp)) != 0)
4010 		return (error);
4011 	vp = (struct vnode *)fp->f_data;
4012 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4013 	if ((obj = vp->v_object) != NULL) {
4014 		if (vp->v_mount == NULL ||
4015 		    (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) {
4016 			vm_object_page_clean(obj, 0, 0, 0);
4017 		}
4018 	}
4019 	error = VOP_FSYNC_FP(vp, MNT_WAIT, VOP_FSYNC_SYSCALL, fp);
4020 	if (error == 0 && vp->v_mount)
4021 		error = buf_fsync(vp);
4022 	vn_unlock(vp);
4023 	fdrop(fp);
4024 
4025 	return (error);
4026 }
4027 
4028 int
4029 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
4030 {
4031 	struct nchandle fnchd;
4032 	struct nchandle tnchd;
4033 	struct namecache *ncp;
4034 	struct vnode *fdvp;
4035 	struct vnode *tdvp;
4036 	struct mount *mp;
4037 	int error;
4038 	u_int fncp_gen;
4039 	u_int tncp_gen;
4040 
4041 	bwillinode(1);
4042 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
4043 	if ((error = nlookup(fromnd)) != 0)
4044 		return (error);
4045 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
4046 		return (ENOENT);
4047 	fnchd.mount = fromnd->nl_nch.mount;
4048 	cache_hold(&fnchd);
4049 
4050 	/*
4051 	 * unlock the source nch so we can lookup the target nch without
4052 	 * deadlocking.  The target may or may not exist so we do not check
4053 	 * for a target vp like kern_mkdir() and other creation functions do.
4054 	 *
4055 	 * The source and target directories are ref'd and rechecked after
4056 	 * everything is relocked to determine if the source or target file
4057 	 * has been renamed.
4058 	 */
4059 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
4060 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
4061 
4062 	fncp_gen = fromnd->nl_nch.ncp->nc_generation;
4063 
4064 	cache_unlock(&fromnd->nl_nch);
4065 
4066 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
4067 	if ((error = nlookup(tond)) != 0) {
4068 		cache_drop(&fnchd);
4069 		return (error);
4070 	}
4071 	tncp_gen = tond->nl_nch.ncp->nc_generation;
4072 
4073 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
4074 		cache_drop(&fnchd);
4075 		return (ENOENT);
4076 	}
4077 	tnchd.mount = tond->nl_nch.mount;
4078 	cache_hold(&tnchd);
4079 
4080 	/*
4081 	 * If the source and target are the same there is nothing to do
4082 	 */
4083 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
4084 		cache_drop(&fnchd);
4085 		cache_drop(&tnchd);
4086 		return (0);
4087 	}
4088 
4089 	/*
4090 	 * Mount points cannot be renamed or overwritten
4091 	 */
4092 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
4093 	    NCF_ISMOUNTPT
4094 	) {
4095 		cache_drop(&fnchd);
4096 		cache_drop(&tnchd);
4097 		return (EINVAL);
4098 	}
4099 
4100 	/*
4101 	 * Relock the source ncp.  cache_relock() will deal with any
4102 	 * deadlocks against the already-locked tond and will also
4103 	 * make sure both are resolved.
4104 	 *
4105 	 * NOTE AFTER RELOCKING: The source or target ncp may have become
4106 	 * invalid while they were unlocked, nc_vp and nc_mount could
4107 	 * be NULL.
4108 	 */
4109 	cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
4110 		     &tond->nl_nch, tond->nl_cred);
4111 	fromnd->nl_flags |= NLC_NCPISLOCKED;
4112 
4113 	/*
4114 	 * If the namecache generation changed for either fromnd or tond,
4115 	 * we must retry.
4116 	 */
4117 	if (fromnd->nl_nch.ncp->nc_generation != fncp_gen ||
4118 	    tond->nl_nch.ncp->nc_generation != tncp_gen) {
4119 		kprintf("kern_rename: retry due to gen on: "
4120 			"\"%s\" -> \"%s\"\n",
4121 			fromnd->nl_nch.ncp->nc_name,
4122 			tond->nl_nch.ncp->nc_name);
4123 		cache_drop(&fnchd);
4124 		cache_drop(&tnchd);
4125 		return (EAGAIN);
4126 	}
4127 
4128 	/*
4129 	 * If either fromnd or tond are marked destroyed a ripout occured
4130 	 * out from under us and we must retry.
4131 	 */
4132 	if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) ||
4133 	    fromnd->nl_nch.ncp->nc_vp == NULL ||
4134 	    (tond->nl_nch.ncp->nc_flag & NCF_DESTROYED)) {
4135 		kprintf("kern_rename: retry due to ripout on: "
4136 			"\"%s\" -> \"%s\"\n",
4137 			fromnd->nl_nch.ncp->nc_name,
4138 			tond->nl_nch.ncp->nc_name);
4139 		cache_drop(&fnchd);
4140 		cache_drop(&tnchd);
4141 		return (EAGAIN);
4142 	}
4143 
4144 	/*
4145 	 * Make sure the parent directories linkages are the same.
4146 	 * XXX shouldn't be needed any more w/ generation check above.
4147 	 */
4148 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
4149 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
4150 		cache_drop(&fnchd);
4151 		cache_drop(&tnchd);
4152 		return (ENOENT);
4153 	}
4154 
4155 	/*
4156 	 * Both the source and target must be within the same filesystem and
4157 	 * in the same filesystem as their parent directories within the
4158 	 * namecache topology.
4159 	 *
4160 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
4161 	 */
4162 	mp = fnchd.mount;
4163 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
4164 	    mp != tond->nl_nch.mount) {
4165 		cache_drop(&fnchd);
4166 		cache_drop(&tnchd);
4167 		return (EXDEV);
4168 	}
4169 
4170 	/*
4171 	 * Make sure the mount point is writable
4172 	 */
4173 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
4174 		cache_drop(&fnchd);
4175 		cache_drop(&tnchd);
4176 		return (error);
4177 	}
4178 
4179 	/*
4180 	 * If the target exists and either the source or target is a directory,
4181 	 * then both must be directories.
4182 	 *
4183 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
4184 	 * have become NULL.
4185 	 */
4186 	if (tond->nl_nch.ncp->nc_vp) {
4187 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
4188 			error = ENOENT;
4189 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
4190 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
4191 				error = ENOTDIR;
4192 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
4193 			error = EISDIR;
4194 		}
4195 	}
4196 
4197 	/*
4198 	 * You cannot rename a source into itself or a subdirectory of itself.
4199 	 * We check this by travsersing the target directory upwards looking
4200 	 * for a match against the source.
4201 	 *
4202 	 * XXX MPSAFE
4203 	 */
4204 	if (error == 0) {
4205 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
4206 			if (fromnd->nl_nch.ncp == ncp) {
4207 				error = EINVAL;
4208 				break;
4209 			}
4210 		}
4211 	}
4212 
4213 	cache_drop(&fnchd);
4214 	cache_drop(&tnchd);
4215 
4216 	/*
4217 	 * Even though the namespaces are different, they may still represent
4218 	 * hardlinks to the same file.  The filesystem might have a hard time
4219 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
4220 	 * when we detect the situation.
4221 	 */
4222 	if (error == 0) {
4223 		fdvp = fromnd->nl_dvp;
4224 		tdvp = tond->nl_dvp;
4225 		if (fdvp == NULL || tdvp == NULL) {
4226 			error = EPERM;
4227 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
4228 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
4229 					    fromnd->nl_cred);
4230 		} else {
4231 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
4232 					    fdvp, tdvp, tond->nl_cred);
4233 		}
4234 	}
4235 	return (error);
4236 }
4237 
4238 /*
4239  * rename_args(char *from, char *to)
4240  *
4241  * Rename files.  Source and destination must either both be directories,
4242  * or both not be directories.  If target is a directory, it must be empty.
4243  */
4244 int
4245 sys_rename(struct rename_args *uap)
4246 {
4247 	struct nlookupdata fromnd, tond;
4248 	int error;
4249 
4250 	do {
4251 		error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
4252 		if (error == 0) {
4253 			error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
4254 			if (error == 0)
4255 				error = kern_rename(&fromnd, &tond);
4256 			nlookup_done(&tond);
4257 		}
4258 		nlookup_done(&fromnd);
4259 	} while (error == EAGAIN);
4260 	return (error);
4261 }
4262 
4263 /*
4264  * renameat_args(int oldfd, char *old, int newfd, char *new)
4265  *
4266  * Rename files using paths relative to the directories associated with
4267  * oldfd and newfd.  Source and destination must either both be directories,
4268  * or both not be directories.  If target is a directory, it must be empty.
4269  */
4270 int
4271 sys_renameat(struct renameat_args *uap)
4272 {
4273 	struct nlookupdata oldnd, newnd;
4274 	struct file *oldfp, *newfp;
4275 	int error;
4276 
4277 	do {
4278 		error = nlookup_init_at(&oldnd, &oldfp,
4279 					uap->oldfd, uap->old,
4280 					UIO_USERSPACE, 0);
4281 		if (error == 0) {
4282 			error = nlookup_init_at(&newnd, &newfp,
4283 						uap->newfd, uap->new,
4284 						UIO_USERSPACE, 0);
4285 			if (error == 0)
4286 				error = kern_rename(&oldnd, &newnd);
4287 			nlookup_done_at(&newnd, newfp);
4288 		}
4289 		nlookup_done_at(&oldnd, oldfp);
4290 	} while (error == EAGAIN);
4291 	return (error);
4292 }
4293 
4294 int
4295 kern_mkdir(struct nlookupdata *nd, int mode)
4296 {
4297 	struct thread *td = curthread;
4298 	struct proc *p = td->td_proc;
4299 	struct vnode *vp;
4300 	struct vattr vattr;
4301 	int error;
4302 
4303 	bwillinode(1);
4304 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
4305 	if ((error = nlookup(nd)) != 0)
4306 		return (error);
4307 
4308 	if (nd->nl_nch.ncp->nc_vp)
4309 		return (EEXIST);
4310 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4311 		return (error);
4312 	VATTR_NULL(&vattr);
4313 	vattr.va_type = VDIR;
4314 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
4315 
4316 	vp = NULL;
4317 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
4318 	if (error == 0)
4319 		vput(vp);
4320 	return (error);
4321 }
4322 
4323 /*
4324  * mkdir_args(char *path, int mode)
4325  *
4326  * Make a directory file.
4327  */
4328 int
4329 sys_mkdir(struct mkdir_args *uap)
4330 {
4331 	struct nlookupdata nd;
4332 	int error;
4333 
4334 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4335 	if (error == 0)
4336 		error = kern_mkdir(&nd, uap->mode);
4337 	nlookup_done(&nd);
4338 	return (error);
4339 }
4340 
4341 /*
4342  * mkdirat_args(int fd, char *path, mode_t mode)
4343  *
4344  * Make a directory file.  The path is relative to the directory associated
4345  * with fd.
4346  */
4347 int
4348 sys_mkdirat(struct mkdirat_args *uap)
4349 {
4350 	struct nlookupdata nd;
4351 	struct file *fp;
4352 	int error;
4353 
4354 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
4355 	if (error == 0)
4356 		error = kern_mkdir(&nd, uap->mode);
4357 	nlookup_done_at(&nd, fp);
4358 	return (error);
4359 }
4360 
4361 int
4362 kern_rmdir(struct nlookupdata *nd)
4363 {
4364 	int error;
4365 
4366 	bwillinode(1);
4367 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
4368 	if ((error = nlookup(nd)) != 0)
4369 		return (error);
4370 
4371 	/*
4372 	 * Do not allow directories representing mount points to be
4373 	 * deleted, even if empty.  Check write perms on mount point
4374 	 * in case the vnode is aliased (aka nullfs).
4375 	 */
4376 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
4377 		return (EBUSY);
4378 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4379 		return (error);
4380 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
4381 	return (error);
4382 }
4383 
4384 /*
4385  * rmdir_args(char *path)
4386  *
4387  * Remove a directory file.
4388  */
4389 int
4390 sys_rmdir(struct rmdir_args *uap)
4391 {
4392 	struct nlookupdata nd;
4393 	int error;
4394 
4395 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4396 	if (error == 0)
4397 		error = kern_rmdir(&nd);
4398 	nlookup_done(&nd);
4399 	return (error);
4400 }
4401 
4402 int
4403 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
4404 		   enum uio_seg direction)
4405 {
4406 	struct thread *td = curthread;
4407 	struct vnode *vp;
4408 	struct file *fp;
4409 	struct uio auio;
4410 	struct iovec aiov;
4411 	off_t loff;
4412 	int error, eofflag;
4413 
4414 	if ((error = holdvnode(td, fd, &fp)) != 0)
4415 		return (error);
4416 	if ((fp->f_flag & FREAD) == 0) {
4417 		error = EBADF;
4418 		goto done;
4419 	}
4420 	vp = (struct vnode *)fp->f_data;
4421 	if (vp->v_type != VDIR) {
4422 		error = EINVAL;
4423 		goto done;
4424 	}
4425 	aiov.iov_base = buf;
4426 	aiov.iov_len = count;
4427 	auio.uio_iov = &aiov;
4428 	auio.uio_iovcnt = 1;
4429 	auio.uio_rw = UIO_READ;
4430 	auio.uio_segflg = direction;
4431 	auio.uio_td = td;
4432 	auio.uio_resid = count;
4433 	loff = auio.uio_offset = fp->f_offset;
4434 	error = VOP_READDIR_FP(vp, &auio, fp->f_cred, &eofflag, NULL, NULL, fp);
4435 	fp->f_offset = auio.uio_offset;
4436 	if (error)
4437 		goto done;
4438 
4439 	/*
4440 	 * WARNING!  *basep may not be wide enough to accomodate the
4441 	 * seek offset.   XXX should we hack this to return the upper 32 bits
4442 	 * for offsets greater then 4G?
4443 	 */
4444 	if (basep) {
4445 		*basep = (long)loff;
4446 	}
4447 	*res = count - auio.uio_resid;
4448 done:
4449 	fdrop(fp);
4450 	return (error);
4451 }
4452 
4453 /*
4454  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4455  *
4456  * Read a block of directory entries in a file system independent format.
4457  */
4458 int
4459 sys_getdirentries(struct getdirentries_args *uap)
4460 {
4461 	long base;
4462 	int error;
4463 
4464 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4465 				   &uap->sysmsg_result, UIO_USERSPACE);
4466 
4467 	if (error == 0 && uap->basep)
4468 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
4469 	return (error);
4470 }
4471 
4472 /*
4473  * getdents_args(int fd, char *buf, size_t count)
4474  */
4475 int
4476 sys_getdents(struct getdents_args *uap)
4477 {
4478 	int error;
4479 
4480 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4481 				   &uap->sysmsg_result, UIO_USERSPACE);
4482 
4483 	return (error);
4484 }
4485 
4486 /*
4487  * Set the mode mask for creation of filesystem nodes.
4488  *
4489  * umask(int newmask)
4490  */
4491 int
4492 sys_umask(struct umask_args *uap)
4493 {
4494 	struct thread *td = curthread;
4495 	struct proc *p = td->td_proc;
4496 	struct filedesc *fdp;
4497 
4498 	fdp = p->p_fd;
4499 	uap->sysmsg_result = fdp->fd_cmask;
4500 	fdp->fd_cmask = uap->newmask & ALLPERMS;
4501 	return (0);
4502 }
4503 
4504 /*
4505  * revoke(char *path)
4506  *
4507  * Void all references to file by ripping underlying filesystem
4508  * away from vnode.
4509  */
4510 int
4511 sys_revoke(struct revoke_args *uap)
4512 {
4513 	struct nlookupdata nd;
4514 	struct vattr vattr;
4515 	struct vnode *vp;
4516 	struct ucred *cred;
4517 	int error;
4518 
4519 	vp = NULL;
4520 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4521 	if (error == 0)
4522 		error = nlookup(&nd);
4523 	if (error == 0)
4524 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4525 	cred = crhold(nd.nl_cred);
4526 	nlookup_done(&nd);
4527 	if (error == 0) {
4528 		if (error == 0)
4529 			error = VOP_GETATTR(vp, &vattr);
4530 		if (error == 0 && cred->cr_uid != vattr.va_uid)
4531 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4532 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4533 			if (vcount(vp) > 0)
4534 				error = vrevoke(vp, cred);
4535 		} else if (error == 0) {
4536 			error = vrevoke(vp, cred);
4537 		}
4538 		vrele(vp);
4539 	}
4540 	if (cred)
4541 		crfree(cred);
4542 	return (error);
4543 }
4544 
4545 /*
4546  * getfh_args(char *fname, fhandle_t *fhp)
4547  *
4548  * Get (NFS) file handle
4549  *
4550  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4551  * mount.  This allows nullfs mounts to be explicitly exported.
4552  *
4553  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4554  *
4555  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
4556  *	    work, but you do not really have protection against access to
4557  *	    the related parent directories.
4558  */
4559 int
4560 sys_getfh(struct getfh_args *uap)
4561 {
4562 	struct thread *td = curthread;
4563 	struct nlookupdata nd;
4564 	fhandle_t fh;
4565 	struct vnode *vp;
4566 	struct mount *mp;
4567 	int error;
4568 
4569 	/*
4570 	 * Must be super user
4571 	 */
4572 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
4573 		return (error);
4574 
4575 	vp = NULL;
4576 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4577 	if (error == 0)
4578 		error = nlookup(&nd);
4579 	if (error == 0)
4580 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4581 	mp = nd.nl_nch.mount;
4582 	nlookup_done(&nd);
4583 	if (error == 0) {
4584 		bzero(&fh, sizeof(fh));
4585 		fh.fh_fsid = mp->mnt_stat.f_fsid;
4586 		error = VFS_VPTOFH(vp, &fh.fh_fid);
4587 		vput(vp);
4588 		if (error == 0)
4589 			error = copyout(&fh, uap->fhp, sizeof(fh));
4590 	}
4591 	return (error);
4592 }
4593 
4594 /*
4595  * fhopen_args(const struct fhandle *u_fhp, int flags)
4596  *
4597  * syscall for the rpc.lockd to use to translate a NFS file handle into
4598  * an open descriptor.
4599  *
4600  * warning: do not remove the priv_check() call or this becomes one giant
4601  * security hole.
4602  */
4603 int
4604 sys_fhopen(struct fhopen_args *uap)
4605 {
4606 	struct thread *td = curthread;
4607 	struct filedesc *fdp = td->td_proc->p_fd;
4608 	struct mount *mp;
4609 	struct vnode *vp;
4610 	struct fhandle fhp;
4611 	struct vattr vat;
4612 	struct vattr *vap = &vat;
4613 	struct flock lf;
4614 	int fmode, mode, error = 0, type;
4615 	struct file *nfp;
4616 	struct file *fp;
4617 	int indx;
4618 
4619 	/*
4620 	 * Must be super user
4621 	 */
4622 	error = priv_check(td, PRIV_ROOT);
4623 	if (error)
4624 		return (error);
4625 
4626 	fmode = FFLAGS(uap->flags);
4627 
4628 	/*
4629 	 * Why not allow a non-read/write open for our lockd?
4630 	 */
4631 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4632 		return (EINVAL);
4633 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4634 	if (error)
4635 		return(error);
4636 
4637 	/*
4638 	 * Find the mount point
4639 	 */
4640 	mp = vfs_getvfs(&fhp.fh_fsid);
4641 	if (mp == NULL) {
4642 		error = ESTALE;
4643 		goto done2;
4644 	}
4645 	/* now give me my vnode, it gets returned to me locked */
4646 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4647 	if (error)
4648 		goto done;
4649  	/*
4650 	 * from now on we have to make sure not
4651 	 * to forget about the vnode
4652 	 * any error that causes an abort must vput(vp)
4653 	 * just set error = err and 'goto bad;'.
4654 	 */
4655 
4656 	/*
4657 	 * from vn_open
4658 	 */
4659 	if (vp->v_type == VLNK) {
4660 		error = EMLINK;
4661 		goto bad;
4662 	}
4663 	if (vp->v_type == VSOCK) {
4664 		error = EOPNOTSUPP;
4665 		goto bad;
4666 	}
4667 	mode = 0;
4668 	if (fmode & (FWRITE | O_TRUNC)) {
4669 		if (vp->v_type == VDIR) {
4670 			error = EISDIR;
4671 			goto bad;
4672 		}
4673 		error = vn_writechk(vp, NULL);
4674 		if (error)
4675 			goto bad;
4676 		mode |= VWRITE;
4677 	}
4678 	if (fmode & FREAD)
4679 		mode |= VREAD;
4680 	if (mode) {
4681 		error = VOP_ACCESS(vp, mode, td->td_ucred);
4682 		if (error)
4683 			goto bad;
4684 	}
4685 	if (fmode & O_TRUNC) {
4686 		vn_unlock(vp);				/* XXX */
4687 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
4688 		VATTR_NULL(vap);
4689 		vap->va_size = 0;
4690 		error = VOP_SETATTR(vp, vap, td->td_ucred);
4691 		if (error)
4692 			goto bad;
4693 	}
4694 
4695 	/*
4696 	 * VOP_OPEN needs the file pointer so it can potentially override
4697 	 * it.
4698 	 *
4699 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
4700 	 * directory.  XXX
4701 	 */
4702 	if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4703 		goto bad;
4704 	fp = nfp;
4705 
4706 	error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4707 	if (error) {
4708 		/*
4709 		 * setting f_ops this way prevents VOP_CLOSE from being
4710 		 * called or fdrop() releasing the vp from v_data.   Since
4711 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4712 		 */
4713 		fp->f_ops = &badfileops;
4714 		fp->f_data = NULL;
4715 		goto bad_drop;
4716 	}
4717 
4718 	/*
4719 	 * The fp is given its own reference, we still have our ref and lock.
4720 	 *
4721 	 * Assert that all regular files must be created with a VM object.
4722 	 */
4723 	if (vp->v_type == VREG && vp->v_object == NULL) {
4724 		kprintf("fhopen: regular file did not "
4725 			"have VM object: %p\n",
4726 			vp);
4727 		goto bad_drop;
4728 	}
4729 
4730 	/*
4731 	 * The open was successful.  Handle any locking requirements.
4732 	 */
4733 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
4734 		lf.l_whence = SEEK_SET;
4735 		lf.l_start = 0;
4736 		lf.l_len = 0;
4737 		if (fmode & O_EXLOCK)
4738 			lf.l_type = F_WRLCK;
4739 		else
4740 			lf.l_type = F_RDLCK;
4741 		if (fmode & FNONBLOCK)
4742 			type = 0;
4743 		else
4744 			type = F_WAIT;
4745 		vn_unlock(vp);
4746 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK,
4747 					 &lf, type)) != 0) {
4748 			/*
4749 			 * release our private reference.
4750 			 */
4751 			fsetfd(fdp, NULL, indx);
4752 			fdrop(fp);
4753 			vrele(vp);
4754 			goto done;
4755 		}
4756 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4757 		atomic_set_int(&fp->f_flag, FHASLOCK);	/* race ok */
4758 	}
4759 
4760 	/*
4761 	 * Clean up.  Associate the file pointer with the previously
4762 	 * reserved descriptor and return it.
4763 	 */
4764 	vput(vp);
4765 	if (uap->flags & O_CLOEXEC)
4766 		fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
4767 	fsetfd(fdp, fp, indx);
4768 	fdrop(fp);
4769 	uap->sysmsg_result = indx;
4770 	mount_drop(mp);
4771 
4772 	return (error);
4773 
4774 bad_drop:
4775 	fsetfd(fdp, NULL, indx);
4776 	fdrop(fp);
4777 bad:
4778 	vput(vp);
4779 done:
4780 	mount_drop(mp);
4781 done2:
4782 	return (error);
4783 }
4784 
4785 /*
4786  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4787  */
4788 int
4789 sys_fhstat(struct fhstat_args *uap)
4790 {
4791 	struct thread *td = curthread;
4792 	struct stat sb;
4793 	fhandle_t fh;
4794 	struct mount *mp;
4795 	struct vnode *vp;
4796 	int error;
4797 
4798 	/*
4799 	 * Must be super user
4800 	 */
4801 	error = priv_check(td, PRIV_ROOT);
4802 	if (error)
4803 		return (error);
4804 
4805 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4806 	if (error)
4807 		return (error);
4808 
4809 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4810 		error = ESTALE;
4811 	if (error == 0) {
4812 		if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4813 			error = vn_stat(vp, &sb, td->td_ucred);
4814 			vput(vp);
4815 		}
4816 	}
4817 	if (error == 0)
4818 		error = copyout(&sb, uap->sb, sizeof(sb));
4819 	if (mp)
4820 		mount_drop(mp);
4821 
4822 	return (error);
4823 }
4824 
4825 /*
4826  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4827  */
4828 int
4829 sys_fhstatfs(struct fhstatfs_args *uap)
4830 {
4831 	struct thread *td = curthread;
4832 	struct proc *p = td->td_proc;
4833 	struct statfs *sp;
4834 	struct mount *mp;
4835 	struct vnode *vp;
4836 	struct statfs sb;
4837 	char *fullpath, *freepath;
4838 	fhandle_t fh;
4839 	int error;
4840 
4841 	/*
4842 	 * Must be super user
4843 	 */
4844 	if ((error = priv_check(td, PRIV_ROOT)))
4845 		return (error);
4846 
4847 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4848 		return (error);
4849 
4850 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4851 		error = ESTALE;
4852 		goto done;
4853 	}
4854 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4855 		error = ESTALE;
4856 		goto done;
4857 	}
4858 
4859 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4860 		goto done;
4861 	mp = vp->v_mount;
4862 	sp = &mp->mnt_stat;
4863 	vput(vp);
4864 	if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4865 		goto done;
4866 
4867 	error = mount_path(p, mp, &fullpath, &freepath);
4868 	if (error)
4869 		goto done;
4870 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4871 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4872 	kfree(freepath, M_TEMP);
4873 
4874 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4875 	if (priv_check(td, PRIV_ROOT)) {
4876 		bcopy(sp, &sb, sizeof(sb));
4877 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4878 		sp = &sb;
4879 	}
4880 	error = copyout(sp, uap->buf, sizeof(*sp));
4881 done:
4882 	if (mp)
4883 		mount_drop(mp);
4884 
4885 	return (error);
4886 }
4887 
4888 /*
4889  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4890  */
4891 int
4892 sys_fhstatvfs(struct fhstatvfs_args *uap)
4893 {
4894 	struct thread *td = curthread;
4895 	struct proc *p = td->td_proc;
4896 	struct statvfs *sp;
4897 	struct mount *mp;
4898 	struct vnode *vp;
4899 	fhandle_t fh;
4900 	int error;
4901 
4902 	/*
4903 	 * Must be super user
4904 	 */
4905 	if ((error = priv_check(td, PRIV_ROOT)))
4906 		return (error);
4907 
4908 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4909 		return (error);
4910 
4911 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4912 		error = ESTALE;
4913 		goto done;
4914 	}
4915 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4916 		error = ESTALE;
4917 		goto done;
4918 	}
4919 
4920 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4921 		goto done;
4922 	mp = vp->v_mount;
4923 	sp = &mp->mnt_vstat;
4924 	vput(vp);
4925 	if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4926 		goto done;
4927 
4928 	sp->f_flag = 0;
4929 	if (mp->mnt_flag & MNT_RDONLY)
4930 		sp->f_flag |= ST_RDONLY;
4931 	if (mp->mnt_flag & MNT_NOSUID)
4932 		sp->f_flag |= ST_NOSUID;
4933 	error = copyout(sp, uap->buf, sizeof(*sp));
4934 done:
4935 	if (mp)
4936 		mount_drop(mp);
4937 	return (error);
4938 }
4939 
4940 
4941 /*
4942  * Syscall to push extended attribute configuration information into the
4943  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4944  * a command (int cmd), and attribute name and misc data.  For now, the
4945  * attribute name is left in userspace for consumption by the VFS_op.
4946  * It will probably be changed to be copied into sysspace by the
4947  * syscall in the future, once issues with various consumers of the
4948  * attribute code have raised their hands.
4949  *
4950  * Currently this is used only by UFS Extended Attributes.
4951  */
4952 int
4953 sys_extattrctl(struct extattrctl_args *uap)
4954 {
4955 	struct nlookupdata nd;
4956 	struct vnode *vp;
4957 	char attrname[EXTATTR_MAXNAMELEN];
4958 	int error;
4959 	size_t size;
4960 
4961 	attrname[0] = 0;
4962 	vp = NULL;
4963 	error = 0;
4964 
4965 	if (error == 0 && uap->filename) {
4966 		error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4967 				     NLC_FOLLOW);
4968 		if (error == 0)
4969 			error = nlookup(&nd);
4970 		if (error == 0)
4971 			error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4972 		nlookup_done(&nd);
4973 	}
4974 
4975 	if (error == 0 && uap->attrname) {
4976 		error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4977 				  &size);
4978 	}
4979 
4980 	if (error == 0) {
4981 		error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4982 		if (error == 0)
4983 			error = nlookup(&nd);
4984 		if (error == 0)
4985 			error = ncp_writechk(&nd.nl_nch);
4986 		if (error == 0) {
4987 			error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4988 					       uap->attrnamespace,
4989 					       uap->attrname, nd.nl_cred);
4990 		}
4991 		nlookup_done(&nd);
4992 	}
4993 
4994 	return (error);
4995 }
4996 
4997 /*
4998  * Syscall to get a named extended attribute on a file or directory.
4999  */
5000 int
5001 sys_extattr_set_file(struct extattr_set_file_args *uap)
5002 {
5003 	char attrname[EXTATTR_MAXNAMELEN];
5004 	struct nlookupdata nd;
5005 	struct vnode *vp;
5006 	struct uio auio;
5007 	struct iovec aiov;
5008 	int error;
5009 
5010 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
5011 	if (error)
5012 		return (error);
5013 
5014 	vp = NULL;
5015 
5016 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5017 	if (error == 0)
5018 		error = nlookup(&nd);
5019 	if (error == 0)
5020 		error = ncp_writechk(&nd.nl_nch);
5021 	if (error == 0)
5022 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
5023 	if (error) {
5024 		nlookup_done(&nd);
5025 		return (error);
5026 	}
5027 
5028 	bzero(&auio, sizeof(auio));
5029 	aiov.iov_base = uap->data;
5030 	aiov.iov_len = uap->nbytes;
5031 	auio.uio_iov = &aiov;
5032 	auio.uio_iovcnt = 1;
5033 	auio.uio_offset = 0;
5034 	auio.uio_resid = uap->nbytes;
5035 	auio.uio_rw = UIO_WRITE;
5036 	auio.uio_td = curthread;
5037 
5038 	error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
5039 			       &auio, nd.nl_cred);
5040 
5041 	vput(vp);
5042 	nlookup_done(&nd);
5043 	return (error);
5044 }
5045 
5046 /*
5047  * Syscall to get a named extended attribute on a file or directory.
5048  */
5049 int
5050 sys_extattr_get_file(struct extattr_get_file_args *uap)
5051 {
5052 	char attrname[EXTATTR_MAXNAMELEN];
5053 	struct nlookupdata nd;
5054 	struct uio auio;
5055 	struct iovec aiov;
5056 	struct vnode *vp;
5057 	int error;
5058 
5059 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
5060 	if (error)
5061 		return (error);
5062 
5063 	vp = NULL;
5064 
5065 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5066 	if (error == 0)
5067 		error = nlookup(&nd);
5068 	if (error == 0)
5069 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_SHARED, &vp);
5070 	if (error) {
5071 		nlookup_done(&nd);
5072 		return (error);
5073 	}
5074 
5075 	bzero(&auio, sizeof(auio));
5076 	aiov.iov_base = uap->data;
5077 	aiov.iov_len = uap->nbytes;
5078 	auio.uio_iov = &aiov;
5079 	auio.uio_iovcnt = 1;
5080 	auio.uio_offset = 0;
5081 	auio.uio_resid = uap->nbytes;
5082 	auio.uio_rw = UIO_READ;
5083 	auio.uio_td = curthread;
5084 
5085 	error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
5086 				&auio, nd.nl_cred);
5087 	uap->sysmsg_result = uap->nbytes - auio.uio_resid;
5088 
5089 	vput(vp);
5090 	nlookup_done(&nd);
5091 	return(error);
5092 }
5093 
5094 /*
5095  * Syscall to delete a named extended attribute from a file or directory.
5096  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
5097  */
5098 int
5099 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
5100 {
5101 	char attrname[EXTATTR_MAXNAMELEN];
5102 	struct nlookupdata nd;
5103 	struct vnode *vp;
5104 	int error;
5105 
5106 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
5107 	if (error)
5108 		return(error);
5109 
5110 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5111 	if (error == 0)
5112 		error = nlookup(&nd);
5113 	if (error == 0)
5114 		error = ncp_writechk(&nd.nl_nch);
5115 	if (error == 0) {
5116 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
5117 		if (error == 0) {
5118 			error = VOP_SETEXTATTR(vp, uap->attrnamespace,
5119 					       attrname, NULL, nd.nl_cred);
5120 			vput(vp);
5121 		}
5122 	}
5123 	nlookup_done(&nd);
5124 	return(error);
5125 }
5126 
5127 /*
5128  * Determine if the mount is visible to the process.
5129  */
5130 static int
5131 chroot_visible_mnt(struct mount *mp, struct proc *p)
5132 {
5133 	struct nchandle nch;
5134 
5135 	/*
5136 	 * Traverse from the mount point upwards.  If we hit the process
5137 	 * root then the mount point is visible to the process.
5138 	 */
5139 	nch = mp->mnt_ncmountpt;
5140 	while (nch.ncp) {
5141 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
5142 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
5143 			return(1);
5144 		}
5145 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
5146 			nch = nch.mount->mnt_ncmounton;
5147 		} else {
5148 			nch.ncp = nch.ncp->nc_parent;
5149 		}
5150 	}
5151 
5152 	/*
5153 	 * If the mount point is not visible to the process, but the
5154 	 * process root is in a subdirectory of the mount, return
5155 	 * TRUE anyway.
5156 	 */
5157 	if (p->p_fd->fd_nrdir.mount == mp)
5158 		return(1);
5159 
5160 	return(0);
5161 }
5162 
5163