xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 2e7bf158f373428dba2c765c927f14d9e94f00a4)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.135 2008/11/11 00:55:49 pavalos Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/linker.h>
57 #include <sys/stat.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/proc.h>
61 #include <sys/priv.h>
62 #include <sys/jail.h>
63 #include <sys/namei.h>
64 #include <sys/nlookup.h>
65 #include <sys/dirent.h>
66 #include <sys/extattr.h>
67 #include <sys/spinlock.h>
68 #include <sys/kern_syscall.h>
69 #include <sys/objcache.h>
70 #include <sys/sysctl.h>
71 
72 #include <sys/buf2.h>
73 #include <sys/file2.h>
74 #include <sys/spinlock2.h>
75 #include <sys/mplock2.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
80 
81 #include <machine/limits.h>
82 #include <machine/stdarg.h>
83 
84 #include <vfs/union/union.h>
85 
86 static void mount_warning(struct mount *mp, const char *ctl, ...);
87 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
88 static int checkvp_chdir (struct vnode *vn, struct thread *td);
89 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
90 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
91 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
92 static int getutimes (const struct timeval *, struct timespec *);
93 static int setfown (struct vnode *, uid_t, gid_t);
94 static int setfmode (struct vnode *, int);
95 static int setfflags (struct vnode *, int);
96 static int setutimes (struct vnode *, struct vattr *,
97 			const struct timespec *, int);
98 static int	usermount = 0;	/* if 1, non-root can mount fs. */
99 
100 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
101 
102 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
103 
104 /*
105  * Virtual File System System Calls
106  */
107 
108 /*
109  * Mount a file system.
110  *
111  * mount_args(char *type, char *path, int flags, caddr_t data)
112  *
113  * MPALMOSTSAFE
114  */
115 int
116 sys_mount(struct mount_args *uap)
117 {
118 	struct thread *td = curthread;
119 	struct vnode *vp;
120 	struct nchandle nch;
121 	struct mount *mp, *nullmp;
122 	struct vfsconf *vfsp;
123 	int error, flag = 0, flag2 = 0;
124 	int hasmount;
125 	struct vattr va;
126 	struct nlookupdata nd;
127 	char fstypename[MFSNAMELEN];
128 	struct ucred *cred;
129 
130 	get_mplock();
131 	cred = td->td_ucred;
132 	if (jailed(cred)) {
133 		error = EPERM;
134 		goto done;
135 	}
136 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
137 		goto done;
138 
139 	/*
140 	 * Do not allow NFS export by non-root users.
141 	 */
142 	if (uap->flags & MNT_EXPORTED) {
143 		error = priv_check(td, PRIV_ROOT);
144 		if (error)
145 			goto done;
146 	}
147 	/*
148 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
149 	 */
150 	if (priv_check(td, PRIV_ROOT))
151 		uap->flags |= MNT_NOSUID | MNT_NODEV;
152 
153 	/*
154 	 * Lookup the requested path and extract the nch and vnode.
155 	 */
156 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
157 	if (error == 0) {
158 		if ((error = nlookup(&nd)) == 0) {
159 			if (nd.nl_nch.ncp->nc_vp == NULL)
160 				error = ENOENT;
161 		}
162 	}
163 	if (error) {
164 		nlookup_done(&nd);
165 		goto done;
166 	}
167 
168 	/*
169 	 * If the target filesystem is resolved via a nullfs mount, then
170 	 * nd.nl_nch.mount will be pointing to the nullfs mount structure
171 	 * instead of the target file system. We need it in case we are
172 	 * doing an update.
173 	 */
174 	nullmp = nd.nl_nch.mount;
175 
176 	/*
177 	 * Extract the locked+refd ncp and cleanup the nd structure
178 	 */
179 	nch = nd.nl_nch;
180 	cache_zero(&nd.nl_nch);
181 	nlookup_done(&nd);
182 
183 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
184 		hasmount = 1;
185 	else
186 		hasmount = 0;
187 
188 
189 	/*
190 	 * now we have the locked ref'd nch and unreferenced vnode.
191 	 */
192 	vp = nch.ncp->nc_vp;
193 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
194 		cache_put(&nch);
195 		goto done;
196 	}
197 	cache_unlock(&nch);
198 
199 	/*
200 	 * Extract the file system type. We need to know this early, to take
201 	 * appropriate actions if we are dealing with a nullfs.
202 	 */
203         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
204                 cache_drop(&nch);
205                 vput(vp);
206 		goto done;
207         }
208 
209 	/*
210 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
211 	 */
212 	if (uap->flags & MNT_UPDATE) {
213 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
214 			cache_drop(&nch);
215 			vput(vp);
216 			error = EINVAL;
217 			goto done;
218 		}
219 
220 		if (strncmp(fstypename, "null", 5) == 0) {
221 			KKASSERT(nullmp);
222 			mp = nullmp;
223 		} else {
224 			mp = vp->v_mount;
225 		}
226 
227 		flag = mp->mnt_flag;
228 		flag2 = mp->mnt_kern_flag;
229 		/*
230 		 * We only allow the filesystem to be reloaded if it
231 		 * is currently mounted read-only.
232 		 */
233 		if ((uap->flags & MNT_RELOAD) &&
234 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
235 			cache_drop(&nch);
236 			vput(vp);
237 			error = EOPNOTSUPP;	/* Needs translation */
238 			goto done;
239 		}
240 		/*
241 		 * Only root, or the user that did the original mount is
242 		 * permitted to update it.
243 		 */
244 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
245 		    (error = priv_check(td, PRIV_ROOT))) {
246 			cache_drop(&nch);
247 			vput(vp);
248 			goto done;
249 		}
250 		if (vfs_busy(mp, LK_NOWAIT)) {
251 			cache_drop(&nch);
252 			vput(vp);
253 			error = EBUSY;
254 			goto done;
255 		}
256 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
257 			cache_drop(&nch);
258 			vfs_unbusy(mp);
259 			vput(vp);
260 			error = EBUSY;
261 			goto done;
262 		}
263 		vsetflags(vp, VMOUNT);
264 		mp->mnt_flag |=
265 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
266 		vn_unlock(vp);
267 		goto update;
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 && (error = priv_check(td, PRIV_ROOT)))) {
275 		cache_drop(&nch);
276 		vput(vp);
277 		goto done;
278 	}
279 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
280 		cache_drop(&nch);
281 		vput(vp);
282 		goto done;
283 	}
284 	if (vp->v_type != VDIR) {
285 		cache_drop(&nch);
286 		vput(vp);
287 		error = ENOTDIR;
288 		goto done;
289 	}
290 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
291 		cache_drop(&nch);
292 		vput(vp);
293 		error = EPERM;
294 		goto done;
295 	}
296 	vfsp = vfsconf_find_by_name(fstypename);
297 	if (vfsp == NULL) {
298 		linker_file_t lf;
299 
300 		/* Only load modules for root (very important!) */
301 		if ((error = priv_check(td, PRIV_ROOT)) != 0) {
302 			cache_drop(&nch);
303 			vput(vp);
304 			goto done;
305 		}
306 		error = linker_load_file(fstypename, &lf);
307 		if (error || lf == NULL) {
308 			cache_drop(&nch);
309 			vput(vp);
310 			if (lf == NULL)
311 				error = ENODEV;
312 			goto done;
313 		}
314 		lf->userrefs++;
315 		/* lookup again, see if the VFS was loaded */
316 		vfsp = vfsconf_find_by_name(fstypename);
317 		if (vfsp == NULL) {
318 			lf->userrefs--;
319 			linker_file_unload(lf);
320 			cache_drop(&nch);
321 			vput(vp);
322 			error = ENODEV;
323 			goto done;
324 		}
325 	}
326 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
327 		cache_drop(&nch);
328 		vput(vp);
329 		error = EBUSY;
330 		goto done;
331 	}
332 	vsetflags(vp, VMOUNT);
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 	vfsp->vfc_refcount++;
343 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
344 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
345 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
346 	mp->mnt_stat.f_owner = cred->cr_uid;
347 	vn_unlock(vp);
348 update:
349 	/*
350 	 * Set the mount level flags.
351 	 */
352 	if (uap->flags & MNT_RDONLY)
353 		mp->mnt_flag |= MNT_RDONLY;
354 	else if (mp->mnt_flag & MNT_RDONLY)
355 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
356 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
357 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
358 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
359 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
360 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
361 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
362 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
363 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
364 	/*
365 	 * Mount the filesystem.
366 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
367 	 * get.
368 	 */
369 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
370 	if (mp->mnt_flag & MNT_UPDATE) {
371 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
372 			mp->mnt_flag &= ~MNT_RDONLY;
373 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
374 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
375 		if (error) {
376 			mp->mnt_flag = flag;
377 			mp->mnt_kern_flag = flag2;
378 		}
379 		vfs_unbusy(mp);
380 		vclrflags(vp, VMOUNT);
381 		vrele(vp);
382 		cache_drop(&nch);
383 		goto done;
384 	}
385 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
386 	/*
387 	 * Put the new filesystem on the mount list after root.  The mount
388 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
389 	 * up) which represents the root of the mount.  The lookup code
390 	 * detects the mount point going forward and checks the root of
391 	 * the mount going backwards.
392 	 *
393 	 * It is not necessary to invalidate or purge the vnode underneath
394 	 * because elements under the mount will be given their own glue
395 	 * namecache record.
396 	 */
397 	if (!error) {
398 		if (mp->mnt_ncmountpt.ncp == NULL) {
399 			/*
400 			 * allocate, then unlock, but leave the ref intact
401 			 */
402 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
403 			cache_unlock(&mp->mnt_ncmountpt);
404 		}
405 		mp->mnt_ncmounton = nch;		/* inherits ref */
406 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
407 
408 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
409 		vclrflags(vp, VMOUNT);
410 		mountlist_insert(mp, MNTINS_LAST);
411 		vn_unlock(vp);
412 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
413 		error = vfs_allocate_syncvnode(mp);
414 		vfs_unbusy(mp);
415 		error = VFS_START(mp, 0);
416 		vrele(vp);
417 	} else {
418 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
419 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
420 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
421 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
422 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
423 		vclrflags(vp, VMOUNT);
424 		mp->mnt_vfc->vfc_refcount--;
425 		vfs_unbusy(mp);
426 		kfree(mp, M_MOUNT);
427 		cache_drop(&nch);
428 		vput(vp);
429 	}
430 done:
431 	rel_mplock();
432 	return (error);
433 }
434 
435 /*
436  * Scan all active processes to see if any of them have a current
437  * or root directory onto which the new filesystem has just been
438  * mounted. If so, replace them with the new mount point.
439  *
440  * The passed ncp is ref'd and locked (from the mount code) and
441  * must be associated with the vnode representing the root of the
442  * mount point.
443  */
444 struct checkdirs_info {
445 	struct nchandle old_nch;
446 	struct nchandle new_nch;
447 	struct vnode *old_vp;
448 	struct vnode *new_vp;
449 };
450 
451 static int checkdirs_callback(struct proc *p, void *data);
452 
453 static void
454 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
455 {
456 	struct checkdirs_info info;
457 	struct vnode *olddp;
458 	struct vnode *newdp;
459 	struct mount *mp;
460 
461 	/*
462 	 * If the old mount point's vnode has a usecount of 1, it is not
463 	 * being held as a descriptor anywhere.
464 	 */
465 	olddp = old_nch->ncp->nc_vp;
466 	if (olddp == NULL || olddp->v_sysref.refcnt == 1)
467 		return;
468 
469 	/*
470 	 * Force the root vnode of the new mount point to be resolved
471 	 * so we can update any matching processes.
472 	 */
473 	mp = new_nch->mount;
474 	if (VFS_ROOT(mp, &newdp))
475 		panic("mount: lost mount");
476 	cache_setunresolved(new_nch);
477 	cache_setvp(new_nch, newdp);
478 
479 	/*
480 	 * Special handling of the root node
481 	 */
482 	if (rootvnode == olddp) {
483 		vref(newdp);
484 		vfs_cache_setroot(newdp, cache_hold(new_nch));
485 	}
486 
487 	/*
488 	 * Pass newdp separately so the callback does not have to access
489 	 * it via new_nch->ncp->nc_vp.
490 	 */
491 	info.old_nch = *old_nch;
492 	info.new_nch = *new_nch;
493 	info.new_vp = newdp;
494 	allproc_scan(checkdirs_callback, &info);
495 	vput(newdp);
496 }
497 
498 /*
499  * NOTE: callback is not MP safe because the scanned process's filedesc
500  * structure can be ripped out from under us, amoung other things.
501  */
502 static int
503 checkdirs_callback(struct proc *p, void *data)
504 {
505 	struct checkdirs_info *info = data;
506 	struct filedesc *fdp;
507 	struct nchandle ncdrop1;
508 	struct nchandle ncdrop2;
509 	struct vnode *vprele1;
510 	struct vnode *vprele2;
511 
512 	if ((fdp = p->p_fd) != NULL) {
513 		cache_zero(&ncdrop1);
514 		cache_zero(&ncdrop2);
515 		vprele1 = NULL;
516 		vprele2 = NULL;
517 
518 		/*
519 		 * MPUNSAFE - XXX fdp can be pulled out from under a
520 		 * foreign process.
521 		 *
522 		 * A shared filedesc is ok, we don't have to copy it
523 		 * because we are making this change globally.
524 		 */
525 		spin_lock_wr(&fdp->fd_spin);
526 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
527 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
528 			vprele1 = fdp->fd_cdir;
529 			vref(info->new_vp);
530 			fdp->fd_cdir = info->new_vp;
531 			ncdrop1 = fdp->fd_ncdir;
532 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
533 		}
534 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
535 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
536 			vprele2 = fdp->fd_rdir;
537 			vref(info->new_vp);
538 			fdp->fd_rdir = info->new_vp;
539 			ncdrop2 = fdp->fd_nrdir;
540 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
541 		}
542 		spin_unlock_wr(&fdp->fd_spin);
543 		if (ncdrop1.ncp)
544 			cache_drop(&ncdrop1);
545 		if (ncdrop2.ncp)
546 			cache_drop(&ncdrop2);
547 		if (vprele1)
548 			vrele(vprele1);
549 		if (vprele2)
550 			vrele(vprele2);
551 	}
552 	return(0);
553 }
554 
555 /*
556  * Unmount a file system.
557  *
558  * Note: unmount takes a path to the vnode mounted on as argument,
559  * not special file (as before).
560  *
561  * umount_args(char *path, int flags)
562  *
563  * MPALMOSTSAFE
564  */
565 int
566 sys_unmount(struct unmount_args *uap)
567 {
568 	struct thread *td = curthread;
569 	struct proc *p __debugvar = td->td_proc;
570 	struct mount *mp = NULL;
571 	struct nlookupdata nd;
572 	int error;
573 
574 	KKASSERT(p);
575 	get_mplock();
576 	if (td->td_ucred->cr_prison != NULL) {
577 		error = EPERM;
578 		goto done;
579 	}
580 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
581 		goto done;
582 
583 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
584 	if (error == 0)
585 		error = nlookup(&nd);
586 	if (error)
587 		goto out;
588 
589 	mp = nd.nl_nch.mount;
590 
591 	/*
592 	 * Only root, or the user that did the original mount is
593 	 * permitted to unmount this filesystem.
594 	 */
595 	if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
596 	    (error = priv_check(td, PRIV_ROOT)))
597 		goto out;
598 
599 	/*
600 	 * Don't allow unmounting the root file system.
601 	 */
602 	if (mp->mnt_flag & MNT_ROOTFS) {
603 		error = EINVAL;
604 		goto out;
605 	}
606 
607 	/*
608 	 * Must be the root of the filesystem
609 	 */
610 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
611 		error = EINVAL;
612 		goto out;
613 	}
614 
615 out:
616 	nlookup_done(&nd);
617 	if (error == 0)
618 		error = dounmount(mp, uap->flags);
619 done:
620 	rel_mplock();
621 	return (error);
622 }
623 
624 /*
625  * Do the actual file system unmount.
626  */
627 static int
628 dounmount_interlock(struct mount *mp)
629 {
630 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
631 		return (EBUSY);
632 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
633 	return(0);
634 }
635 
636 static int
637 unmount_allproc_cb(struct proc *p, void *arg)
638 {
639 	struct mount *mp;
640 
641 	if (p->p_textnch.ncp == NULL)
642 		return 0;
643 
644 	mp = (struct mount *)arg;
645 	if (p->p_textnch.mount == mp)
646 		cache_drop(&p->p_textnch);
647 
648 	return 0;
649 }
650 
651 int
652 dounmount(struct mount *mp, int flags)
653 {
654 	struct namecache *ncp;
655 	struct nchandle nch;
656 	struct vnode *vp;
657 	int error;
658 	int async_flag;
659 	int lflags;
660 	int freeok = 1;
661 
662 	/*
663 	 * Exclusive access for unmounting purposes
664 	 */
665 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
666 		return (error);
667 
668 	/*
669 	 * Allow filesystems to detect that a forced unmount is in progress.
670 	 */
671 	if (flags & MNT_FORCE)
672 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
673 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
674 	error = lockmgr(&mp->mnt_lock, lflags);
675 	if (error) {
676 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
677 		if (mp->mnt_kern_flag & MNTK_MWAIT)
678 			wakeup(mp);
679 		return (error);
680 	}
681 
682 	if (mp->mnt_flag & MNT_EXPUBLIC)
683 		vfs_setpublicfs(NULL, NULL, NULL);
684 
685 	vfs_msync(mp, MNT_WAIT);
686 	async_flag = mp->mnt_flag & MNT_ASYNC;
687 	mp->mnt_flag &=~ MNT_ASYNC;
688 
689 	/*
690 	 * If this filesystem isn't aliasing other filesystems,
691 	 * try to invalidate any remaining namecache entries and
692 	 * check the count afterwords.
693 	 */
694 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
695 		cache_lock(&mp->mnt_ncmountpt);
696 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
697 		cache_unlock(&mp->mnt_ncmountpt);
698 
699 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
700 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
701 			allproc_scan(&unmount_allproc_cb, mp);
702 		}
703 
704 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
705 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
706 
707 			if ((flags & MNT_FORCE) == 0) {
708 				error = EBUSY;
709 				mount_warning(mp, "Cannot unmount: "
710 						  "%d namecache "
711 						  "references still "
712 						  "present",
713 						  ncp->nc_refs - 1);
714 			} else {
715 				mount_warning(mp, "Forced unmount: "
716 						  "%d namecache "
717 						  "references still "
718 						  "present",
719 						  ncp->nc_refs - 1);
720 				freeok = 0;
721 			}
722 		}
723 	}
724 
725 	/*
726 	 * nchandle records ref the mount structure.  Expect a count of 1
727 	 * (our mount->mnt_ncmountpt).
728 	 */
729 	if (mp->mnt_refs != 1) {
730 		if ((flags & MNT_FORCE) == 0) {
731 			mount_warning(mp, "Cannot unmount: "
732 					  "%d process references still "
733 					  "present", mp->mnt_refs);
734 			error = EBUSY;
735 		} else {
736 			mount_warning(mp, "Forced unmount: "
737 					  "%d process references still "
738 					  "present", mp->mnt_refs);
739 			freeok = 0;
740 		}
741 	}
742 
743 	/*
744 	 * Decomission our special mnt_syncer vnode.  This also stops
745 	 * the vnlru code.  If we are unable to unmount we recommission
746 	 * the vnode.
747 	 */
748 	if (error == 0) {
749 		if ((vp = mp->mnt_syncer) != NULL) {
750 			mp->mnt_syncer = NULL;
751 			vrele(vp);
752 		}
753 		if (((mp->mnt_flag & MNT_RDONLY) ||
754 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
755 		    (flags & MNT_FORCE)) {
756 			error = VFS_UNMOUNT(mp, flags);
757 		}
758 	}
759 	if (error) {
760 		if (mp->mnt_syncer == NULL)
761 			vfs_allocate_syncvnode(mp);
762 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
763 		mp->mnt_flag |= async_flag;
764 		lockmgr(&mp->mnt_lock, LK_RELEASE);
765 		if (mp->mnt_kern_flag & MNTK_MWAIT)
766 			wakeup(mp);
767 		return (error);
768 	}
769 	/*
770 	 * Clean up any journals still associated with the mount after
771 	 * filesystem activity has ceased.
772 	 */
773 	journal_remove_all_journals(mp,
774 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
775 
776 	mountlist_remove(mp);
777 
778 	/*
779 	 * Remove any installed vnode ops here so the individual VFSs don't
780 	 * have to.
781 	 */
782 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
783 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
784 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
785 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
786 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
787 
788 	if (mp->mnt_ncmountpt.ncp != NULL) {
789 		nch = mp->mnt_ncmountpt;
790 		cache_zero(&mp->mnt_ncmountpt);
791 		cache_clrmountpt(&nch);
792 		cache_drop(&nch);
793 	}
794 	if (mp->mnt_ncmounton.ncp != NULL) {
795 		nch = mp->mnt_ncmounton;
796 		cache_zero(&mp->mnt_ncmounton);
797 		cache_clrmountpt(&nch);
798 		cache_drop(&nch);
799 	}
800 
801 	mp->mnt_vfc->vfc_refcount--;
802 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
803 		panic("unmount: dangling vnode");
804 	lockmgr(&mp->mnt_lock, LK_RELEASE);
805 	if (mp->mnt_kern_flag & MNTK_MWAIT)
806 		wakeup(mp);
807 	if (freeok)
808 		kfree(mp, M_MOUNT);
809 	return (0);
810 }
811 
812 static
813 void
814 mount_warning(struct mount *mp, const char *ctl, ...)
815 {
816 	char *ptr;
817 	char *buf;
818 	__va_list va;
819 
820 	__va_start(va, ctl);
821 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf, 0) == 0) {
822 		kprintf("unmount(%s): ", ptr);
823 		kvprintf(ctl, va);
824 		kprintf("\n");
825 		kfree(buf, M_TEMP);
826 	} else {
827 		kprintf("unmount(%p", mp);
828 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
829 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
830 		kprintf("): ");
831 		kvprintf(ctl, va);
832 		kprintf("\n");
833 	}
834 	__va_end(va);
835 }
836 
837 /*
838  * Shim cache_fullpath() to handle the case where a process is chrooted into
839  * a subdirectory of a mount.  In this case if the root mount matches the
840  * process root directory's mount we have to specify the process's root
841  * directory instead of the mount point, because the mount point might
842  * be above the root directory.
843  */
844 static
845 int
846 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
847 {
848 	struct nchandle *nch;
849 
850 	if (p && p->p_fd->fd_nrdir.mount == mp)
851 		nch = &p->p_fd->fd_nrdir;
852 	else
853 		nch = &mp->mnt_ncmountpt;
854 	return(cache_fullpath(p, nch, rb, fb, 0));
855 }
856 
857 /*
858  * Sync each mounted filesystem.
859  */
860 
861 #ifdef DEBUG
862 static int syncprt = 0;
863 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
864 #endif /* DEBUG */
865 
866 static int sync_callback(struct mount *mp, void *data);
867 
868 /*
869  * MPALMOSTSAFE
870  */
871 int
872 sys_sync(struct sync_args *uap)
873 {
874 	get_mplock();
875 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
876 #ifdef DEBUG
877 	/*
878 	 * print out buffer pool stat information on each sync() call.
879 	 */
880 	if (syncprt)
881 		vfs_bufstats();
882 #endif /* DEBUG */
883 	rel_mplock();
884 	return (0);
885 }
886 
887 static
888 int
889 sync_callback(struct mount *mp, void *data __unused)
890 {
891 	int asyncflag;
892 
893 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
894 		asyncflag = mp->mnt_flag & MNT_ASYNC;
895 		mp->mnt_flag &= ~MNT_ASYNC;
896 		vfs_msync(mp, MNT_NOWAIT);
897 		VFS_SYNC(mp, MNT_NOWAIT);
898 		mp->mnt_flag |= asyncflag;
899 	}
900 	return(0);
901 }
902 
903 /* XXX PRISON: could be per prison flag */
904 static int prison_quotas;
905 #if 0
906 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
907 #endif
908 
909 /*
910  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
911  *
912  * Change filesystem quotas.
913  *
914  * MPALMOSTSAFE
915  */
916 int
917 sys_quotactl(struct quotactl_args *uap)
918 {
919 	struct nlookupdata nd;
920 	struct thread *td;
921 	struct proc *p;
922 	struct mount *mp;
923 	int error;
924 
925 	get_mplock();
926 	td = curthread;
927 	p = td->td_proc;
928 	if (td->td_ucred->cr_prison && !prison_quotas) {
929 		error = EPERM;
930 		goto done;
931 	}
932 
933 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
934 	if (error == 0)
935 		error = nlookup(&nd);
936 	if (error == 0) {
937 		mp = nd.nl_nch.mount;
938 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
939 				    uap->arg, nd.nl_cred);
940 	}
941 	nlookup_done(&nd);
942 done:
943 	rel_mplock();
944 	return (error);
945 }
946 
947 /*
948  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
949  *		void *buf, int buflen)
950  *
951  * This function operates on a mount point and executes the specified
952  * operation using the specified control data, and possibly returns data.
953  *
954  * The actual number of bytes stored in the result buffer is returned, 0
955  * if none, otherwise an error is returned.
956  *
957  * MPALMOSTSAFE
958  */
959 int
960 sys_mountctl(struct mountctl_args *uap)
961 {
962 	struct thread *td = curthread;
963 	struct proc *p = td->td_proc;
964 	struct file *fp;
965 	void *ctl = NULL;
966 	void *buf = NULL;
967 	char *path = NULL;
968 	int error;
969 
970 	/*
971 	 * Sanity and permissions checks.  We must be root.
972 	 */
973 	KKASSERT(p);
974 	if (td->td_ucred->cr_prison != NULL)
975 		return (EPERM);
976 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
977 	    (error = priv_check(td, PRIV_ROOT)) != 0)
978 		return (error);
979 
980 	/*
981 	 * Argument length checks
982 	 */
983 	if (uap->ctllen < 0 || uap->ctllen > 1024)
984 		return (EINVAL);
985 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
986 		return (EINVAL);
987 	if (uap->path == NULL)
988 		return (EINVAL);
989 
990 	/*
991 	 * Allocate the necessary buffers and copyin data
992 	 */
993 	path = objcache_get(namei_oc, M_WAITOK);
994 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
995 	if (error)
996 		goto done;
997 
998 	if (uap->ctllen) {
999 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1000 		error = copyin(uap->ctl, ctl, uap->ctllen);
1001 		if (error)
1002 			goto done;
1003 	}
1004 	if (uap->buflen)
1005 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1006 
1007 	/*
1008 	 * Validate the descriptor
1009 	 */
1010 	if (uap->fd >= 0) {
1011 		fp = holdfp(p->p_fd, uap->fd, -1);
1012 		if (fp == NULL) {
1013 			error = EBADF;
1014 			goto done;
1015 		}
1016 	} else {
1017 		fp = NULL;
1018 	}
1019 
1020 	/*
1021 	 * Execute the internal kernel function and clean up.
1022 	 */
1023 	get_mplock();
1024 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1025 	rel_mplock();
1026 	if (fp)
1027 		fdrop(fp);
1028 	if (error == 0 && uap->sysmsg_result > 0)
1029 		error = copyout(buf, uap->buf, uap->sysmsg_result);
1030 done:
1031 	if (path)
1032 		objcache_put(namei_oc, path);
1033 	if (ctl)
1034 		kfree(ctl, M_TEMP);
1035 	if (buf)
1036 		kfree(buf, M_TEMP);
1037 	return (error);
1038 }
1039 
1040 /*
1041  * Execute a mount control operation by resolving the path to a mount point
1042  * and calling vop_mountctl().
1043  *
1044  * Use the mount point from the nch instead of the vnode so nullfs mounts
1045  * can properly spike the VOP.
1046  */
1047 int
1048 kern_mountctl(const char *path, int op, struct file *fp,
1049 		const void *ctl, int ctllen,
1050 		void *buf, int buflen, int *res)
1051 {
1052 	struct vnode *vp;
1053 	struct mount *mp;
1054 	struct nlookupdata nd;
1055 	int error;
1056 
1057 	*res = 0;
1058 	vp = NULL;
1059 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1060 	if (error == 0)
1061 		error = nlookup(&nd);
1062 	if (error == 0)
1063 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1064 	mp = nd.nl_nch.mount;
1065 	nlookup_done(&nd);
1066 	if (error)
1067 		return (error);
1068 	vn_unlock(vp);
1069 
1070 	/*
1071 	 * Must be the root of the filesystem
1072 	 */
1073 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1074 		vrele(vp);
1075 		return (EINVAL);
1076 	}
1077 	error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1078 			     buf, buflen, res);
1079 	vrele(vp);
1080 	return (error);
1081 }
1082 
1083 int
1084 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1085 {
1086 	struct thread *td = curthread;
1087 	struct proc *p = td->td_proc;
1088 	struct mount *mp;
1089 	struct statfs *sp;
1090 	char *fullpath, *freepath;
1091 	int error;
1092 
1093 	if ((error = nlookup(nd)) != 0)
1094 		return (error);
1095 	mp = nd->nl_nch.mount;
1096 	sp = &mp->mnt_stat;
1097 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1098 		return (error);
1099 
1100 	error = mount_path(p, mp, &fullpath, &freepath);
1101 	if (error)
1102 		return(error);
1103 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1104 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1105 	kfree(freepath, M_TEMP);
1106 
1107 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1108 	bcopy(sp, buf, sizeof(*buf));
1109 	/* Only root should have access to the fsid's. */
1110 	if (priv_check(td, PRIV_ROOT))
1111 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1112 	return (0);
1113 }
1114 
1115 /*
1116  * statfs_args(char *path, struct statfs *buf)
1117  *
1118  * Get filesystem statistics.
1119  *
1120  * MPALMOSTSAFE
1121  */
1122 int
1123 sys_statfs(struct statfs_args *uap)
1124 {
1125 	struct nlookupdata nd;
1126 	struct statfs buf;
1127 	int error;
1128 
1129 	get_mplock();
1130 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1131 	if (error == 0)
1132 		error = kern_statfs(&nd, &buf);
1133 	nlookup_done(&nd);
1134 	if (error == 0)
1135 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1136 	rel_mplock();
1137 	return (error);
1138 }
1139 
1140 /*
1141  * MPALMOSTSAFE
1142  */
1143 int
1144 kern_fstatfs(int fd, struct statfs *buf)
1145 {
1146 	struct thread *td = curthread;
1147 	struct proc *p = td->td_proc;
1148 	struct file *fp;
1149 	struct mount *mp;
1150 	struct statfs *sp;
1151 	char *fullpath, *freepath;
1152 	int error;
1153 
1154 	KKASSERT(p);
1155 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1156 		return (error);
1157 	get_mplock();
1158 	mp = ((struct vnode *)fp->f_data)->v_mount;
1159 	if (mp == NULL) {
1160 		error = EBADF;
1161 		goto done;
1162 	}
1163 	if (fp->f_cred == NULL) {
1164 		error = EINVAL;
1165 		goto done;
1166 	}
1167 	sp = &mp->mnt_stat;
1168 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1169 		goto done;
1170 
1171 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1172 		goto done;
1173 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1174 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1175 	kfree(freepath, M_TEMP);
1176 
1177 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1178 	bcopy(sp, buf, sizeof(*buf));
1179 
1180 	/* Only root should have access to the fsid's. */
1181 	if (priv_check(td, PRIV_ROOT))
1182 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1183 	error = 0;
1184 done:
1185 	rel_mplock();
1186 	fdrop(fp);
1187 	return (error);
1188 }
1189 
1190 /*
1191  * fstatfs_args(int fd, struct statfs *buf)
1192  *
1193  * Get filesystem statistics.
1194  *
1195  * MPSAFE
1196  */
1197 int
1198 sys_fstatfs(struct fstatfs_args *uap)
1199 {
1200 	struct statfs buf;
1201 	int error;
1202 
1203 	error = kern_fstatfs(uap->fd, &buf);
1204 
1205 	if (error == 0)
1206 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1207 	return (error);
1208 }
1209 
1210 int
1211 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1212 {
1213 	struct mount *mp;
1214 	struct statvfs *sp;
1215 	int error;
1216 
1217 	if ((error = nlookup(nd)) != 0)
1218 		return (error);
1219 	mp = nd->nl_nch.mount;
1220 	sp = &mp->mnt_vstat;
1221 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1222 		return (error);
1223 
1224 	sp->f_flag = 0;
1225 	if (mp->mnt_flag & MNT_RDONLY)
1226 		sp->f_flag |= ST_RDONLY;
1227 	if (mp->mnt_flag & MNT_NOSUID)
1228 		sp->f_flag |= ST_NOSUID;
1229 	bcopy(sp, buf, sizeof(*buf));
1230 	return (0);
1231 }
1232 
1233 /*
1234  * statfs_args(char *path, struct statfs *buf)
1235  *
1236  * Get filesystem statistics.
1237  *
1238  * MPALMOSTSAFE
1239  */
1240 int
1241 sys_statvfs(struct statvfs_args *uap)
1242 {
1243 	struct nlookupdata nd;
1244 	struct statvfs buf;
1245 	int error;
1246 
1247 	get_mplock();
1248 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1249 	if (error == 0)
1250 		error = kern_statvfs(&nd, &buf);
1251 	nlookup_done(&nd);
1252 	if (error == 0)
1253 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1254 	rel_mplock();
1255 	return (error);
1256 }
1257 
1258 int
1259 kern_fstatvfs(int fd, struct statvfs *buf)
1260 {
1261 	struct thread *td = curthread;
1262 	struct proc *p = td->td_proc;
1263 	struct file *fp;
1264 	struct mount *mp;
1265 	struct statvfs *sp;
1266 	int error;
1267 
1268 	KKASSERT(p);
1269 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1270 		return (error);
1271 	mp = ((struct vnode *)fp->f_data)->v_mount;
1272 	if (mp == NULL) {
1273 		error = EBADF;
1274 		goto done;
1275 	}
1276 	if (fp->f_cred == NULL) {
1277 		error = EINVAL;
1278 		goto done;
1279 	}
1280 	sp = &mp->mnt_vstat;
1281 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1282 		goto done;
1283 
1284 	sp->f_flag = 0;
1285 	if (mp->mnt_flag & MNT_RDONLY)
1286 		sp->f_flag |= ST_RDONLY;
1287 	if (mp->mnt_flag & MNT_NOSUID)
1288 		sp->f_flag |= ST_NOSUID;
1289 
1290 	bcopy(sp, buf, sizeof(*buf));
1291 	error = 0;
1292 done:
1293 	fdrop(fp);
1294 	return (error);
1295 }
1296 
1297 /*
1298  * fstatfs_args(int fd, struct statfs *buf)
1299  *
1300  * Get filesystem statistics.
1301  *
1302  * MPALMOSTSAFE
1303  */
1304 int
1305 sys_fstatvfs(struct fstatvfs_args *uap)
1306 {
1307 	struct statvfs buf;
1308 	int error;
1309 
1310 	get_mplock();
1311 	error = kern_fstatvfs(uap->fd, &buf);
1312 	rel_mplock();
1313 
1314 	if (error == 0)
1315 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1316 	return (error);
1317 }
1318 
1319 /*
1320  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1321  *
1322  * Get statistics on all filesystems.
1323  */
1324 
1325 struct getfsstat_info {
1326 	struct statfs *sfsp;
1327 	long count;
1328 	long maxcount;
1329 	int error;
1330 	int flags;
1331 	struct thread *td;
1332 };
1333 
1334 static int getfsstat_callback(struct mount *, void *);
1335 
1336 /*
1337  * MPALMOSTSAFE
1338  */
1339 int
1340 sys_getfsstat(struct getfsstat_args *uap)
1341 {
1342 	struct thread *td = curthread;
1343 	struct getfsstat_info info;
1344 
1345 	bzero(&info, sizeof(info));
1346 
1347 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1348 	info.sfsp = uap->buf;
1349 	info.count = 0;
1350 	info.flags = uap->flags;
1351 	info.td = td;
1352 
1353 	get_mplock();
1354 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1355 	rel_mplock();
1356 	if (info.sfsp && info.count > info.maxcount)
1357 		uap->sysmsg_result = info.maxcount;
1358 	else
1359 		uap->sysmsg_result = info.count;
1360 	return (info.error);
1361 }
1362 
1363 static int
1364 getfsstat_callback(struct mount *mp, void *data)
1365 {
1366 	struct getfsstat_info *info = data;
1367 	struct statfs *sp;
1368 	char *freepath;
1369 	char *fullpath;
1370 	int error;
1371 
1372 	if (info->sfsp && info->count < info->maxcount) {
1373 		if (info->td->td_proc &&
1374 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1375 			return(0);
1376 		}
1377 		sp = &mp->mnt_stat;
1378 
1379 		/*
1380 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1381 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1382 		 * overrides MNT_WAIT.
1383 		 */
1384 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1385 		    (info->flags & MNT_WAIT)) &&
1386 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1387 			return(0);
1388 		}
1389 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1390 
1391 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1392 		if (error) {
1393 			info->error = error;
1394 			return(-1);
1395 		}
1396 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1397 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1398 		kfree(freepath, M_TEMP);
1399 
1400 		error = copyout(sp, info->sfsp, sizeof(*sp));
1401 		if (error) {
1402 			info->error = error;
1403 			return (-1);
1404 		}
1405 		++info->sfsp;
1406 	}
1407 	info->count++;
1408 	return(0);
1409 }
1410 
1411 /*
1412  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1413 		   long bufsize, int flags)
1414  *
1415  * Get statistics on all filesystems.
1416  */
1417 
1418 struct getvfsstat_info {
1419 	struct statfs *sfsp;
1420 	struct statvfs *vsfsp;
1421 	long count;
1422 	long maxcount;
1423 	int error;
1424 	int flags;
1425 	struct thread *td;
1426 };
1427 
1428 static int getvfsstat_callback(struct mount *, void *);
1429 
1430 /*
1431  * MPALMOSTSAFE
1432  */
1433 int
1434 sys_getvfsstat(struct getvfsstat_args *uap)
1435 {
1436 	struct thread *td = curthread;
1437 	struct getvfsstat_info info;
1438 
1439 	bzero(&info, sizeof(info));
1440 
1441 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1442 	info.sfsp = uap->buf;
1443 	info.vsfsp = uap->vbuf;
1444 	info.count = 0;
1445 	info.flags = uap->flags;
1446 	info.td = td;
1447 
1448 	get_mplock();
1449 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1450 	if (info.vsfsp && info.count > info.maxcount)
1451 		uap->sysmsg_result = info.maxcount;
1452 	else
1453 		uap->sysmsg_result = info.count;
1454 	rel_mplock();
1455 	return (info.error);
1456 }
1457 
1458 static int
1459 getvfsstat_callback(struct mount *mp, void *data)
1460 {
1461 	struct getvfsstat_info *info = data;
1462 	struct statfs *sp;
1463 	struct statvfs *vsp;
1464 	char *freepath;
1465 	char *fullpath;
1466 	int error;
1467 
1468 	if (info->vsfsp && info->count < info->maxcount) {
1469 		if (info->td->td_proc &&
1470 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1471 			return(0);
1472 		}
1473 		sp = &mp->mnt_stat;
1474 		vsp = &mp->mnt_vstat;
1475 
1476 		/*
1477 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1478 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1479 		 * overrides MNT_WAIT.
1480 		 */
1481 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1482 		    (info->flags & MNT_WAIT)) &&
1483 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1484 			return(0);
1485 		}
1486 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1487 
1488 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1489 		    (info->flags & MNT_WAIT)) &&
1490 		    (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1491 			return(0);
1492 		}
1493 		vsp->f_flag = 0;
1494 		if (mp->mnt_flag & MNT_RDONLY)
1495 			vsp->f_flag |= ST_RDONLY;
1496 		if (mp->mnt_flag & MNT_NOSUID)
1497 			vsp->f_flag |= ST_NOSUID;
1498 
1499 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1500 		if (error) {
1501 			info->error = error;
1502 			return(-1);
1503 		}
1504 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1505 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1506 		kfree(freepath, M_TEMP);
1507 
1508 		error = copyout(sp, info->sfsp, sizeof(*sp));
1509 		if (error == 0)
1510 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1511 		if (error) {
1512 			info->error = error;
1513 			return (-1);
1514 		}
1515 		++info->sfsp;
1516 		++info->vsfsp;
1517 	}
1518 	info->count++;
1519 	return(0);
1520 }
1521 
1522 
1523 /*
1524  * fchdir_args(int fd)
1525  *
1526  * Change current working directory to a given file descriptor.
1527  *
1528  * MPALMOSTSAFE
1529  */
1530 int
1531 sys_fchdir(struct fchdir_args *uap)
1532 {
1533 	struct thread *td = curthread;
1534 	struct proc *p = td->td_proc;
1535 	struct filedesc *fdp = p->p_fd;
1536 	struct vnode *vp, *ovp;
1537 	struct mount *mp;
1538 	struct file *fp;
1539 	struct nchandle nch, onch, tnch;
1540 	int error;
1541 
1542 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1543 		return (error);
1544 	get_mplock();
1545 	vp = (struct vnode *)fp->f_data;
1546 	vref(vp);
1547 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1548 	if (fp->f_nchandle.ncp == NULL)
1549 		error = ENOTDIR;
1550 	else
1551 		error = checkvp_chdir(vp, td);
1552 	if (error) {
1553 		vput(vp);
1554 		goto done;
1555 	}
1556 	cache_copy(&fp->f_nchandle, &nch);
1557 
1558 	/*
1559 	 * If the ncp has become a mount point, traverse through
1560 	 * the mount point.
1561 	 */
1562 
1563 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1564 	       (mp = cache_findmount(&nch)) != NULL
1565 	) {
1566 		error = nlookup_mp(mp, &tnch);
1567 		if (error == 0) {
1568 			cache_unlock(&tnch);	/* leave ref intact */
1569 			vput(vp);
1570 			vp = tnch.ncp->nc_vp;
1571 			error = vget(vp, LK_SHARED);
1572 			KKASSERT(error == 0);
1573 			cache_drop(&nch);
1574 			nch = tnch;
1575 		}
1576 	}
1577 	if (error == 0) {
1578 		ovp = fdp->fd_cdir;
1579 		onch = fdp->fd_ncdir;
1580 		vn_unlock(vp);		/* leave ref intact */
1581 		fdp->fd_cdir = vp;
1582 		fdp->fd_ncdir = nch;
1583 		cache_drop(&onch);
1584 		vrele(ovp);
1585 	} else {
1586 		cache_drop(&nch);
1587 		vput(vp);
1588 	}
1589 	fdrop(fp);
1590 done:
1591 	rel_mplock();
1592 	return (error);
1593 }
1594 
1595 int
1596 kern_chdir(struct nlookupdata *nd)
1597 {
1598 	struct thread *td = curthread;
1599 	struct proc *p = td->td_proc;
1600 	struct filedesc *fdp = p->p_fd;
1601 	struct vnode *vp, *ovp;
1602 	struct nchandle onch;
1603 	int error;
1604 
1605 	if ((error = nlookup(nd)) != 0)
1606 		return (error);
1607 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1608 		return (ENOENT);
1609 	if ((error = vget(vp, LK_SHARED)) != 0)
1610 		return (error);
1611 
1612 	error = checkvp_chdir(vp, td);
1613 	vn_unlock(vp);
1614 	if (error == 0) {
1615 		ovp = fdp->fd_cdir;
1616 		onch = fdp->fd_ncdir;
1617 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1618 		fdp->fd_ncdir = nd->nl_nch;
1619 		fdp->fd_cdir = vp;
1620 		cache_drop(&onch);
1621 		vrele(ovp);
1622 		cache_zero(&nd->nl_nch);
1623 	} else {
1624 		vrele(vp);
1625 	}
1626 	return (error);
1627 }
1628 
1629 /*
1630  * chdir_args(char *path)
1631  *
1632  * Change current working directory (``.'').
1633  *
1634  * MPALMOSTSAFE
1635  */
1636 int
1637 sys_chdir(struct chdir_args *uap)
1638 {
1639 	struct nlookupdata nd;
1640 	int error;
1641 
1642 	get_mplock();
1643 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1644 	if (error == 0)
1645 		error = kern_chdir(&nd);
1646 	nlookup_done(&nd);
1647 	rel_mplock();
1648 	return (error);
1649 }
1650 
1651 /*
1652  * Helper function for raised chroot(2) security function:  Refuse if
1653  * any filedescriptors are open directories.
1654  */
1655 static int
1656 chroot_refuse_vdir_fds(struct filedesc *fdp)
1657 {
1658 	struct vnode *vp;
1659 	struct file *fp;
1660 	int error;
1661 	int fd;
1662 
1663 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1664 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1665 			continue;
1666 		vp = (struct vnode *)fp->f_data;
1667 		if (vp->v_type != VDIR) {
1668 			fdrop(fp);
1669 			continue;
1670 		}
1671 		fdrop(fp);
1672 		return(EPERM);
1673 	}
1674 	return (0);
1675 }
1676 
1677 /*
1678  * This sysctl determines if we will allow a process to chroot(2) if it
1679  * has a directory open:
1680  *	0: disallowed for all processes.
1681  *	1: allowed for processes that were not already chroot(2)'ed.
1682  *	2: allowed for all processes.
1683  */
1684 
1685 static int chroot_allow_open_directories = 1;
1686 
1687 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1688      &chroot_allow_open_directories, 0, "");
1689 
1690 /*
1691  * chroot to the specified namecache entry.  We obtain the vp from the
1692  * namecache data.  The passed ncp must be locked and referenced and will
1693  * remain locked and referenced on return.
1694  */
1695 int
1696 kern_chroot(struct nchandle *nch)
1697 {
1698 	struct thread *td = curthread;
1699 	struct proc *p = td->td_proc;
1700 	struct filedesc *fdp = p->p_fd;
1701 	struct vnode *vp;
1702 	int error;
1703 
1704 	/*
1705 	 * Only privileged user can chroot
1706 	 */
1707 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1708 	if (error)
1709 		return (error);
1710 
1711 	/*
1712 	 * Disallow open directory descriptors (fchdir() breakouts).
1713 	 */
1714 	if (chroot_allow_open_directories == 0 ||
1715 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1716 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1717 			return (error);
1718 	}
1719 	if ((vp = nch->ncp->nc_vp) == NULL)
1720 		return (ENOENT);
1721 
1722 	if ((error = vget(vp, LK_SHARED)) != 0)
1723 		return (error);
1724 
1725 	/*
1726 	 * Check the validity of vp as a directory to change to and
1727 	 * associate it with rdir/jdir.
1728 	 */
1729 	error = checkvp_chdir(vp, td);
1730 	vn_unlock(vp);			/* leave reference intact */
1731 	if (error == 0) {
1732 		vrele(fdp->fd_rdir);
1733 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1734 		cache_drop(&fdp->fd_nrdir);
1735 		cache_copy(nch, &fdp->fd_nrdir);
1736 		if (fdp->fd_jdir == NULL) {
1737 			fdp->fd_jdir = vp;
1738 			vref(fdp->fd_jdir);
1739 			cache_copy(nch, &fdp->fd_njdir);
1740 		}
1741 	} else {
1742 		vrele(vp);
1743 	}
1744 	return (error);
1745 }
1746 
1747 /*
1748  * chroot_args(char *path)
1749  *
1750  * Change notion of root (``/'') directory.
1751  *
1752  * MPALMOSTSAFE
1753  */
1754 int
1755 sys_chroot(struct chroot_args *uap)
1756 {
1757 	struct thread *td __debugvar = curthread;
1758 	struct nlookupdata nd;
1759 	int error;
1760 
1761 	KKASSERT(td->td_proc);
1762 	get_mplock();
1763 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1764 	if (error == 0) {
1765 		nd.nl_flags |= NLC_EXEC;
1766 		error = nlookup(&nd);
1767 		if (error == 0)
1768 			error = kern_chroot(&nd.nl_nch);
1769 	}
1770 	nlookup_done(&nd);
1771 	rel_mplock();
1772 	return(error);
1773 }
1774 
1775 int
1776 sys_chroot_kernel(struct chroot_kernel_args *uap)
1777 {
1778 	struct thread *td = curthread;
1779 	struct nlookupdata nd;
1780 	struct nchandle *nch;
1781 	struct vnode *vp;
1782 	int error;
1783 
1784 	get_mplock();
1785 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1786 	if (error)
1787 		goto error_nond;
1788 
1789 	error = nlookup(&nd);
1790 	if (error)
1791 		goto error_out;
1792 
1793 	nch = &nd.nl_nch;
1794 
1795 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1796 	if (error)
1797 		goto error_out;
1798 
1799 	if ((vp = nch->ncp->nc_vp) == NULL) {
1800 		error = ENOENT;
1801 		goto error_out;
1802 	}
1803 
1804 	if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1805 		goto error_out;
1806 
1807 	kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1808 	vfs_cache_setroot(vp, cache_hold(nch));
1809 
1810 error_out:
1811 	nlookup_done(&nd);
1812 error_nond:
1813 	rel_mplock();
1814 	return(error);
1815 }
1816 
1817 /*
1818  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1819  * determine whether it is legal to chdir to the vnode.  The vnode's state
1820  * is not changed by this call.
1821  */
1822 int
1823 checkvp_chdir(struct vnode *vp, struct thread *td)
1824 {
1825 	int error;
1826 
1827 	if (vp->v_type != VDIR)
1828 		error = ENOTDIR;
1829 	else
1830 		error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1831 	return (error);
1832 }
1833 
1834 /*
1835  * MPSAFE
1836  */
1837 int
1838 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1839 {
1840 	struct thread *td = curthread;
1841 	struct proc *p = td->td_proc;
1842 	struct lwp *lp = td->td_lwp;
1843 	struct filedesc *fdp = p->p_fd;
1844 	int cmode, flags;
1845 	struct file *nfp;
1846 	struct file *fp;
1847 	struct vnode *vp;
1848 	int type, indx, error;
1849 	struct flock lf;
1850 
1851 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1852 		return (EINVAL);
1853 	flags = FFLAGS(oflags);
1854 	error = falloc(lp, &nfp, NULL);
1855 	if (error)
1856 		return (error);
1857 	fp = nfp;
1858 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1859 
1860 	/*
1861 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1862 	 * file descriptor to be duplicated rather then doing the open
1863 	 * itself.
1864 	 */
1865 	lp->lwp_dupfd = -1;
1866 
1867 	/*
1868 	 * Call vn_open() to do the lookup and assign the vnode to the
1869 	 * file pointer.  vn_open() does not change the ref count on fp
1870 	 * and the vnode, on success, will be inherited by the file pointer
1871 	 * and unlocked.
1872 	 */
1873 	nd->nl_flags |= NLC_LOCKVP;
1874 	error = vn_open(nd, fp, flags, cmode);
1875 	nlookup_done(nd);
1876 	if (error) {
1877 		/*
1878 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1879 		 * responsible for dropping the old contents of ofiles[indx]
1880 		 * if it succeeds.
1881 		 *
1882 		 * Note that fsetfd() will add a ref to fp which represents
1883 		 * the fd_files[] assignment.  We must still drop our
1884 		 * reference.
1885 		 */
1886 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1887 			if (fdalloc(p, 0, &indx) == 0) {
1888 				error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1889 				if (error == 0) {
1890 					*res = indx;
1891 					fdrop(fp);	/* our ref */
1892 					return (0);
1893 				}
1894 				fsetfd(fdp, NULL, indx);
1895 			}
1896 		}
1897 		fdrop(fp);	/* our ref */
1898 		if (error == ERESTART)
1899 			error = EINTR;
1900 		return (error);
1901 	}
1902 
1903 	/*
1904 	 * ref the vnode for ourselves so it can't be ripped out from under
1905 	 * is.  XXX need an ND flag to request that the vnode be returned
1906 	 * anyway.
1907 	 *
1908 	 * Reserve a file descriptor but do not assign it until the open
1909 	 * succeeds.
1910 	 */
1911 	vp = (struct vnode *)fp->f_data;
1912 	vref(vp);
1913 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1914 		fdrop(fp);
1915 		vrele(vp);
1916 		return (error);
1917 	}
1918 
1919 	/*
1920 	 * If no error occurs the vp will have been assigned to the file
1921 	 * pointer.
1922 	 */
1923 	lp->lwp_dupfd = 0;
1924 
1925 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1926 		lf.l_whence = SEEK_SET;
1927 		lf.l_start = 0;
1928 		lf.l_len = 0;
1929 		if (flags & O_EXLOCK)
1930 			lf.l_type = F_WRLCK;
1931 		else
1932 			lf.l_type = F_RDLCK;
1933 		if (flags & FNONBLOCK)
1934 			type = 0;
1935 		else
1936 			type = F_WAIT;
1937 
1938 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1939 			/*
1940 			 * lock request failed.  Clean up the reserved
1941 			 * descriptor.
1942 			 */
1943 			vrele(vp);
1944 			fsetfd(fdp, NULL, indx);
1945 			fdrop(fp);
1946 			return (error);
1947 		}
1948 		fp->f_flag |= FHASLOCK;
1949 	}
1950 #if 0
1951 	/*
1952 	 * Assert that all regular file vnodes were created with a object.
1953 	 */
1954 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1955 		("open: regular file has no backing object after vn_open"));
1956 #endif
1957 
1958 	vrele(vp);
1959 
1960 	/*
1961 	 * release our private reference, leaving the one associated with the
1962 	 * descriptor table intact.
1963 	 */
1964 	fsetfd(fdp, fp, indx);
1965 	fdrop(fp);
1966 	*res = indx;
1967 	return (0);
1968 }
1969 
1970 /*
1971  * open_args(char *path, int flags, int mode)
1972  *
1973  * Check permissions, allocate an open file structure,
1974  * and call the device open routine if any.
1975  *
1976  * MPALMOSTSAFE
1977  */
1978 int
1979 sys_open(struct open_args *uap)
1980 {
1981 	CACHE_MPLOCK_DECLARE;
1982 	struct nlookupdata nd;
1983 	int error;
1984 
1985 	CACHE_GETMPLOCK1();
1986 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1987 	if (error == 0) {
1988 		error = kern_open(&nd, uap->flags,
1989 				    uap->mode, &uap->sysmsg_result);
1990 	}
1991 	nlookup_done(&nd);
1992 	CACHE_RELMPLOCK();
1993 	return (error);
1994 }
1995 
1996 /*
1997  * openat_args(int fd, char *path, int flags, int mode)
1998  *
1999  * MPALMOSTSAFE
2000  */
2001 int
2002 sys_openat(struct openat_args *uap)
2003 {
2004 	CACHE_MPLOCK_DECLARE;
2005 	struct nlookupdata nd;
2006 	int error;
2007 	struct file *fp;
2008 
2009 	CACHE_GETMPLOCK1();
2010 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2011 	if (error == 0) {
2012 		error = kern_open(&nd, uap->flags, uap->mode,
2013 					&uap->sysmsg_result);
2014 	}
2015 	nlookup_done_at(&nd, fp);
2016 	CACHE_RELMPLOCK();
2017 	return (error);
2018 }
2019 
2020 int
2021 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2022 {
2023 	struct thread *td = curthread;
2024 	struct proc *p = td->td_proc;
2025 	struct vnode *vp;
2026 	struct vattr vattr;
2027 	int error;
2028 	int whiteout = 0;
2029 
2030 	KKASSERT(p);
2031 
2032 	VATTR_NULL(&vattr);
2033 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2034 	vattr.va_rmajor = rmajor;
2035 	vattr.va_rminor = rminor;
2036 
2037 	switch (mode & S_IFMT) {
2038 	case S_IFMT:	/* used by badsect to flag bad sectors */
2039 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2040 		vattr.va_type = VBAD;
2041 		break;
2042 	case S_IFCHR:
2043 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2044 		vattr.va_type = VCHR;
2045 		break;
2046 	case S_IFBLK:
2047 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2048 		vattr.va_type = VBLK;
2049 		break;
2050 	case S_IFWHT:
2051 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2052 		whiteout = 1;
2053 		break;
2054 	case S_IFDIR:	/* special directories support for HAMMER */
2055 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2056 		vattr.va_type = VDIR;
2057 		break;
2058 	default:
2059 		error = EINVAL;
2060 		break;
2061 	}
2062 
2063 	if (error)
2064 		return (error);
2065 
2066 	bwillinode(1);
2067 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2068 	if ((error = nlookup(nd)) != 0)
2069 		return (error);
2070 	if (nd->nl_nch.ncp->nc_vp)
2071 		return (EEXIST);
2072 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2073 		return (error);
2074 
2075 	if (whiteout) {
2076 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2077 				      nd->nl_cred, NAMEI_CREATE);
2078 	} else {
2079 		vp = NULL;
2080 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2081 				   &vp, nd->nl_cred, &vattr);
2082 		if (error == 0)
2083 			vput(vp);
2084 	}
2085 	return (error);
2086 }
2087 
2088 /*
2089  * mknod_args(char *path, int mode, int dev)
2090  *
2091  * Create a special file.
2092  *
2093  * MPALMOSTSAFE
2094  */
2095 int
2096 sys_mknod(struct mknod_args *uap)
2097 {
2098 	struct nlookupdata nd;
2099 	int error;
2100 
2101 	get_mplock();
2102 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2103 	if (error == 0) {
2104 		error = kern_mknod(&nd, uap->mode,
2105 				   umajor(uap->dev), uminor(uap->dev));
2106 	}
2107 	nlookup_done(&nd);
2108 	rel_mplock();
2109 	return (error);
2110 }
2111 
2112 int
2113 kern_mkfifo(struct nlookupdata *nd, int mode)
2114 {
2115 	struct thread *td = curthread;
2116 	struct proc *p = td->td_proc;
2117 	struct vattr vattr;
2118 	struct vnode *vp;
2119 	int error;
2120 
2121 	bwillinode(1);
2122 
2123 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2124 	if ((error = nlookup(nd)) != 0)
2125 		return (error);
2126 	if (nd->nl_nch.ncp->nc_vp)
2127 		return (EEXIST);
2128 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2129 		return (error);
2130 
2131 	VATTR_NULL(&vattr);
2132 	vattr.va_type = VFIFO;
2133 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2134 	vp = NULL;
2135 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2136 	if (error == 0)
2137 		vput(vp);
2138 	return (error);
2139 }
2140 
2141 /*
2142  * mkfifo_args(char *path, int mode)
2143  *
2144  * Create a named pipe.
2145  *
2146  * MPALMOSTSAFE
2147  */
2148 int
2149 sys_mkfifo(struct mkfifo_args *uap)
2150 {
2151 	struct nlookupdata nd;
2152 	int error;
2153 
2154 	get_mplock();
2155 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2156 	if (error == 0)
2157 		error = kern_mkfifo(&nd, uap->mode);
2158 	nlookup_done(&nd);
2159 	rel_mplock();
2160 	return (error);
2161 }
2162 
2163 static int hardlink_check_uid = 0;
2164 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2165     &hardlink_check_uid, 0,
2166     "Unprivileged processes cannot create hard links to files owned by other "
2167     "users");
2168 static int hardlink_check_gid = 0;
2169 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2170     &hardlink_check_gid, 0,
2171     "Unprivileged processes cannot create hard links to files owned by other "
2172     "groups");
2173 
2174 static int
2175 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2176 {
2177 	struct vattr va;
2178 	int error;
2179 
2180 	/*
2181 	 * Shortcut if disabled
2182 	 */
2183 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2184 		return (0);
2185 
2186 	/*
2187 	 * Privileged user can always hardlink
2188 	 */
2189 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2190 		return (0);
2191 
2192 	/*
2193 	 * Otherwise only if the originating file is owned by the
2194 	 * same user or group.  Note that any group is allowed if
2195 	 * the file is owned by the caller.
2196 	 */
2197 	error = VOP_GETATTR(vp, &va);
2198 	if (error != 0)
2199 		return (error);
2200 
2201 	if (hardlink_check_uid) {
2202 		if (cred->cr_uid != va.va_uid)
2203 			return (EPERM);
2204 	}
2205 
2206 	if (hardlink_check_gid) {
2207 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2208 			return (EPERM);
2209 	}
2210 
2211 	return (0);
2212 }
2213 
2214 int
2215 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2216 {
2217 	struct thread *td = curthread;
2218 	struct vnode *vp;
2219 	int error;
2220 
2221 	/*
2222 	 * Lookup the source and obtained a locked vnode.
2223 	 *
2224 	 * You may only hardlink a file which you have write permission
2225 	 * on or which you own.
2226 	 *
2227 	 * XXX relookup on vget failure / race ?
2228 	 */
2229 	bwillinode(1);
2230 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2231 	if ((error = nlookup(nd)) != 0)
2232 		return (error);
2233 	vp = nd->nl_nch.ncp->nc_vp;
2234 	KKASSERT(vp != NULL);
2235 	if (vp->v_type == VDIR)
2236 		return (EPERM);		/* POSIX */
2237 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2238 		return (error);
2239 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2240 		return (error);
2241 
2242 	/*
2243 	 * Unlock the source so we can lookup the target without deadlocking
2244 	 * (XXX vp is locked already, possible other deadlock?).  The target
2245 	 * must not exist.
2246 	 */
2247 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2248 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2249 	cache_unlock(&nd->nl_nch);
2250 	vn_unlock(vp);
2251 
2252 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2253 	if ((error = nlookup(linknd)) != 0) {
2254 		vrele(vp);
2255 		return (error);
2256 	}
2257 	if (linknd->nl_nch.ncp->nc_vp) {
2258 		vrele(vp);
2259 		return (EEXIST);
2260 	}
2261 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2262 		vrele(vp);
2263 		return (error);
2264 	}
2265 
2266 	/*
2267 	 * Finally run the new API VOP.
2268 	 */
2269 	error = can_hardlink(vp, td, td->td_ucred);
2270 	if (error == 0) {
2271 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2272 				  vp, linknd->nl_cred);
2273 	}
2274 	vput(vp);
2275 	return (error);
2276 }
2277 
2278 /*
2279  * link_args(char *path, char *link)
2280  *
2281  * Make a hard file link.
2282  *
2283  * MPALMOSTSAFE
2284  */
2285 int
2286 sys_link(struct link_args *uap)
2287 {
2288 	struct nlookupdata nd, linknd;
2289 	int error;
2290 
2291 	get_mplock();
2292 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2293 	if (error == 0) {
2294 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2295 		if (error == 0)
2296 			error = kern_link(&nd, &linknd);
2297 		nlookup_done(&linknd);
2298 	}
2299 	nlookup_done(&nd);
2300 	rel_mplock();
2301 	return (error);
2302 }
2303 
2304 int
2305 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2306 {
2307 	struct vattr vattr;
2308 	struct vnode *vp;
2309 	struct vnode *dvp;
2310 	int error;
2311 
2312 	bwillinode(1);
2313 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2314 	if ((error = nlookup(nd)) != 0)
2315 		return (error);
2316 	if (nd->nl_nch.ncp->nc_vp)
2317 		return (EEXIST);
2318 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2319 		return (error);
2320 	dvp = nd->nl_dvp;
2321 	VATTR_NULL(&vattr);
2322 	vattr.va_mode = mode;
2323 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2324 	if (error == 0)
2325 		vput(vp);
2326 	return (error);
2327 }
2328 
2329 /*
2330  * symlink(char *path, char *link)
2331  *
2332  * Make a symbolic link.
2333  *
2334  * MPALMOSTSAFE
2335  */
2336 int
2337 sys_symlink(struct symlink_args *uap)
2338 {
2339 	struct thread *td = curthread;
2340 	struct nlookupdata nd;
2341 	char *path;
2342 	int error;
2343 	int mode;
2344 
2345 	path = objcache_get(namei_oc, M_WAITOK);
2346 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2347 	if (error == 0) {
2348 		get_mplock();
2349 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2350 		if (error == 0) {
2351 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2352 			error = kern_symlink(&nd, path, mode);
2353 		}
2354 		nlookup_done(&nd);
2355 		rel_mplock();
2356 	}
2357 	objcache_put(namei_oc, path);
2358 	return (error);
2359 }
2360 
2361 /*
2362  * undelete_args(char *path)
2363  *
2364  * Delete a whiteout from the filesystem.
2365  *
2366  * MPALMOSTSAFE
2367  */
2368 int
2369 sys_undelete(struct undelete_args *uap)
2370 {
2371 	struct nlookupdata nd;
2372 	int error;
2373 
2374 	get_mplock();
2375 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2376 	bwillinode(1);
2377 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2378 	if (error == 0)
2379 		error = nlookup(&nd);
2380 	if (error == 0)
2381 		error = ncp_writechk(&nd.nl_nch);
2382 	if (error == 0) {
2383 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2384 				      NAMEI_DELETE);
2385 	}
2386 	nlookup_done(&nd);
2387 	rel_mplock();
2388 	return (error);
2389 }
2390 
2391 int
2392 kern_unlink(struct nlookupdata *nd)
2393 {
2394 	int error;
2395 
2396 	bwillinode(1);
2397 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2398 	if ((error = nlookup(nd)) != 0)
2399 		return (error);
2400 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2401 		return (error);
2402 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2403 	return (error);
2404 }
2405 
2406 /*
2407  * unlink_args(char *path)
2408  *
2409  * Delete a name from the filesystem.
2410  *
2411  * MPALMOSTSAFE
2412  */
2413 int
2414 sys_unlink(struct unlink_args *uap)
2415 {
2416 	struct nlookupdata nd;
2417 	int error;
2418 
2419 	get_mplock();
2420 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2421 	if (error == 0)
2422 		error = kern_unlink(&nd);
2423 	nlookup_done(&nd);
2424 	rel_mplock();
2425 	return (error);
2426 }
2427 
2428 
2429 /*
2430  * unlinkat_args(int fd, char *path, int flags)
2431  *
2432  * Delete the file or directory entry pointed to by fd/path.
2433  *
2434  * MPALMOSTSAFE
2435  */
2436 int
2437 sys_unlinkat(struct unlinkat_args *uap)
2438 {
2439 	struct nlookupdata nd;
2440 	struct file *fp;
2441 	int error;
2442 
2443 	if (uap->flags & ~AT_REMOVEDIR)
2444 		return (EINVAL);
2445 
2446 	get_mplock();
2447 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2448 	if (error == 0) {
2449 		if (uap->flags & AT_REMOVEDIR)
2450 			error = kern_rmdir(&nd);
2451 		else
2452 			error = kern_unlink(&nd);
2453 	}
2454 	nlookup_done_at(&nd, fp);
2455 	rel_mplock();
2456 	return (error);
2457 }
2458 
2459 /*
2460  * MPALMOSTSAFE
2461  */
2462 int
2463 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2464 {
2465 	struct thread *td = curthread;
2466 	struct proc *p = td->td_proc;
2467 	struct file *fp;
2468 	struct vnode *vp;
2469 	struct vattr vattr;
2470 	off_t new_offset;
2471 	int error;
2472 
2473 	fp = holdfp(p->p_fd, fd, -1);
2474 	if (fp == NULL)
2475 		return (EBADF);
2476 	if (fp->f_type != DTYPE_VNODE) {
2477 		error = ESPIPE;
2478 		goto done;
2479 	}
2480 	vp = (struct vnode *)fp->f_data;
2481 
2482 	switch (whence) {
2483 	case L_INCR:
2484 		spin_lock_wr(&fp->f_spin);
2485 		new_offset = fp->f_offset + offset;
2486 		error = 0;
2487 		break;
2488 	case L_XTND:
2489 		get_mplock();
2490 		error = VOP_GETATTR(vp, &vattr);
2491 		rel_mplock();
2492 		spin_lock_wr(&fp->f_spin);
2493 		new_offset = offset + vattr.va_size;
2494 		break;
2495 	case L_SET:
2496 		new_offset = offset;
2497 		error = 0;
2498 		spin_lock_wr(&fp->f_spin);
2499 		break;
2500 	default:
2501 		new_offset = 0;
2502 		error = EINVAL;
2503 		spin_lock_wr(&fp->f_spin);
2504 		break;
2505 	}
2506 
2507 	/*
2508 	 * Validate the seek position.  Negative offsets are not allowed
2509 	 * for regular files or directories.
2510 	 *
2511 	 * Normally we would also not want to allow negative offsets for
2512 	 * character and block-special devices.  However kvm addresses
2513 	 * on 64 bit architectures might appear to be negative and must
2514 	 * be allowed.
2515 	 */
2516 	if (error == 0) {
2517 		if (new_offset < 0 &&
2518 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2519 			error = EINVAL;
2520 		} else {
2521 			fp->f_offset = new_offset;
2522 		}
2523 	}
2524 	*res = fp->f_offset;
2525 	spin_unlock_wr(&fp->f_spin);
2526 done:
2527 	fdrop(fp);
2528 	return (error);
2529 }
2530 
2531 /*
2532  * lseek_args(int fd, int pad, off_t offset, int whence)
2533  *
2534  * Reposition read/write file offset.
2535  *
2536  * MPSAFE
2537  */
2538 int
2539 sys_lseek(struct lseek_args *uap)
2540 {
2541 	int error;
2542 
2543 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2544 			   &uap->sysmsg_offset);
2545 
2546 	return (error);
2547 }
2548 
2549 /*
2550  * Check if current process can access given file.  amode is a bitmask of *_OK
2551  * access bits.  flags is a bitmask of AT_* flags.
2552  */
2553 int
2554 kern_access(struct nlookupdata *nd, int amode, int flags)
2555 {
2556 	struct vnode *vp;
2557 	int error, mode;
2558 
2559 	if (flags & ~AT_EACCESS)
2560 		return (EINVAL);
2561 	if ((error = nlookup(nd)) != 0)
2562 		return (error);
2563 retry:
2564 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2565 	if (error)
2566 		return (error);
2567 
2568 	/* Flags == 0 means only check for existence. */
2569 	if (amode) {
2570 		mode = 0;
2571 		if (amode & R_OK)
2572 			mode |= VREAD;
2573 		if (amode & W_OK)
2574 			mode |= VWRITE;
2575 		if (amode & X_OK)
2576 			mode |= VEXEC;
2577 		if ((mode & VWRITE) == 0 ||
2578 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2579 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2580 
2581 		/*
2582 		 * If the file handle is stale we have to re-resolve the
2583 		 * entry.  This is a hack at the moment.
2584 		 */
2585 		if (error == ESTALE) {
2586 			vput(vp);
2587 			cache_setunresolved(&nd->nl_nch);
2588 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2589 			if (error == 0) {
2590 				vp = NULL;
2591 				goto retry;
2592 			}
2593 			return(error);
2594 		}
2595 	}
2596 	vput(vp);
2597 	return (error);
2598 }
2599 
2600 /*
2601  * access_args(char *path, int flags)
2602  *
2603  * Check access permissions.
2604  *
2605  * MPALMOSTSAFE
2606  */
2607 int
2608 sys_access(struct access_args *uap)
2609 {
2610 	struct nlookupdata nd;
2611 	int error;
2612 
2613 	get_mplock();
2614 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2615 	if (error == 0)
2616 		error = kern_access(&nd, uap->flags, 0);
2617 	nlookup_done(&nd);
2618 	rel_mplock();
2619 	return (error);
2620 }
2621 
2622 
2623 /*
2624  * faccessat_args(int fd, char *path, int amode, int flags)
2625  *
2626  * Check access permissions.
2627  *
2628  * MPALMOSTSAFE
2629  */
2630 int
2631 sys_faccessat(struct faccessat_args *uap)
2632 {
2633 	struct nlookupdata nd;
2634 	struct file *fp;
2635 	int error;
2636 
2637 	get_mplock();
2638 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2639 				NLC_FOLLOW);
2640 	if (error == 0)
2641 		error = kern_access(&nd, uap->amode, uap->flags);
2642 	nlookup_done_at(&nd, fp);
2643 	rel_mplock();
2644 	return (error);
2645 }
2646 
2647 
2648 /*
2649  * MPSAFE
2650  */
2651 int
2652 kern_stat(struct nlookupdata *nd, struct stat *st)
2653 {
2654 	int error;
2655 	struct vnode *vp;
2656 	thread_t td;
2657 
2658 	if ((error = nlookup(nd)) != 0)
2659 		return (error);
2660 again:
2661 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2662 		return (ENOENT);
2663 
2664 	td = curthread;
2665 	if ((error = vget(vp, LK_SHARED)) != 0)
2666 		return (error);
2667 	error = vn_stat(vp, st, nd->nl_cred);
2668 
2669 	/*
2670 	 * If the file handle is stale we have to re-resolve the entry.  This
2671 	 * is a hack at the moment.
2672 	 */
2673 	if (error == ESTALE) {
2674 		vput(vp);
2675 		cache_setunresolved(&nd->nl_nch);
2676 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2677 		if (error == 0)
2678 			goto again;
2679 	} else {
2680 		vput(vp);
2681 	}
2682 	return (error);
2683 }
2684 
2685 /*
2686  * stat_args(char *path, struct stat *ub)
2687  *
2688  * Get file status; this version follows links.
2689  *
2690  * MPSAFE
2691  */
2692 int
2693 sys_stat(struct stat_args *uap)
2694 {
2695 	CACHE_MPLOCK_DECLARE;
2696 	struct nlookupdata nd;
2697 	struct stat st;
2698 	int error;
2699 
2700 	CACHE_GETMPLOCK1();
2701 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2702 	if (error == 0) {
2703 		error = kern_stat(&nd, &st);
2704 		if (error == 0)
2705 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2706 	}
2707 	nlookup_done(&nd);
2708 	CACHE_RELMPLOCK();
2709 	return (error);
2710 }
2711 
2712 /*
2713  * lstat_args(char *path, struct stat *ub)
2714  *
2715  * Get file status; this version does not follow links.
2716  *
2717  * MPALMOSTSAFE
2718  */
2719 int
2720 sys_lstat(struct lstat_args *uap)
2721 {
2722 	CACHE_MPLOCK_DECLARE;
2723 	struct nlookupdata nd;
2724 	struct stat st;
2725 	int error;
2726 
2727 	CACHE_GETMPLOCK1();
2728 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2729 	if (error == 0) {
2730 		error = kern_stat(&nd, &st);
2731 		if (error == 0)
2732 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2733 	}
2734 	nlookup_done(&nd);
2735 	CACHE_RELMPLOCK();
2736 	return (error);
2737 }
2738 
2739 /*
2740  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2741  *
2742  * Get status of file pointed to by fd/path.
2743  *
2744  * MPALMOSTSAFE
2745  */
2746 int
2747 sys_fstatat(struct fstatat_args *uap)
2748 {
2749 	CACHE_MPLOCK_DECLARE;
2750 	struct nlookupdata nd;
2751 	struct stat st;
2752 	int error;
2753 	int flags;
2754 	struct file *fp;
2755 
2756 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2757 		return (EINVAL);
2758 
2759 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2760 
2761 	CACHE_GETMPLOCK1();
2762 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2763 				UIO_USERSPACE, flags);
2764 	if (error == 0) {
2765 		error = kern_stat(&nd, &st);
2766 		if (error == 0)
2767 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
2768 	}
2769 	nlookup_done_at(&nd, fp);
2770 	CACHE_RELMPLOCK();
2771 	return (error);
2772 }
2773 
2774 /*
2775  * pathconf_Args(char *path, int name)
2776  *
2777  * Get configurable pathname variables.
2778  *
2779  * MPALMOSTSAFE
2780  */
2781 int
2782 sys_pathconf(struct pathconf_args *uap)
2783 {
2784 	struct nlookupdata nd;
2785 	struct vnode *vp;
2786 	int error;
2787 
2788 	vp = NULL;
2789 	get_mplock();
2790 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2791 	if (error == 0)
2792 		error = nlookup(&nd);
2793 	if (error == 0)
2794 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2795 	nlookup_done(&nd);
2796 	if (error == 0) {
2797 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2798 		vput(vp);
2799 	}
2800 	rel_mplock();
2801 	return (error);
2802 }
2803 
2804 /*
2805  * XXX: daver
2806  * kern_readlink isn't properly split yet.  There is a copyin burried
2807  * in VOP_READLINK().
2808  */
2809 int
2810 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2811 {
2812 	struct thread *td = curthread;
2813 	struct vnode *vp;
2814 	struct iovec aiov;
2815 	struct uio auio;
2816 	int error;
2817 
2818 	if ((error = nlookup(nd)) != 0)
2819 		return (error);
2820 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2821 	if (error)
2822 		return (error);
2823 	if (vp->v_type != VLNK) {
2824 		error = EINVAL;
2825 	} else {
2826 		aiov.iov_base = buf;
2827 		aiov.iov_len = count;
2828 		auio.uio_iov = &aiov;
2829 		auio.uio_iovcnt = 1;
2830 		auio.uio_offset = 0;
2831 		auio.uio_rw = UIO_READ;
2832 		auio.uio_segflg = UIO_USERSPACE;
2833 		auio.uio_td = td;
2834 		auio.uio_resid = count;
2835 		error = VOP_READLINK(vp, &auio, td->td_ucred);
2836 	}
2837 	vput(vp);
2838 	*res = count - auio.uio_resid;
2839 	return (error);
2840 }
2841 
2842 /*
2843  * readlink_args(char *path, char *buf, int count)
2844  *
2845  * Return target name of a symbolic link.
2846  *
2847  * MPALMOSTSAFE
2848  */
2849 int
2850 sys_readlink(struct readlink_args *uap)
2851 {
2852 	struct nlookupdata nd;
2853 	int error;
2854 
2855 	get_mplock();
2856 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2857 	if (error == 0) {
2858 		error = kern_readlink(&nd, uap->buf, uap->count,
2859 					&uap->sysmsg_result);
2860 	}
2861 	nlookup_done(&nd);
2862 	rel_mplock();
2863 	return (error);
2864 }
2865 
2866 static int
2867 setfflags(struct vnode *vp, int flags)
2868 {
2869 	struct thread *td = curthread;
2870 	int error;
2871 	struct vattr vattr;
2872 
2873 	/*
2874 	 * Prevent non-root users from setting flags on devices.  When
2875 	 * a device is reused, users can retain ownership of the device
2876 	 * if they are allowed to set flags and programs assume that
2877 	 * chown can't fail when done as root.
2878 	 */
2879 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2880 	    ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2881 		return (error);
2882 
2883 	/*
2884 	 * note: vget is required for any operation that might mod the vnode
2885 	 * so VINACTIVE is properly cleared.
2886 	 */
2887 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2888 		VATTR_NULL(&vattr);
2889 		vattr.va_flags = flags;
2890 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2891 		vput(vp);
2892 	}
2893 	return (error);
2894 }
2895 
2896 /*
2897  * chflags(char *path, int flags)
2898  *
2899  * Change flags of a file given a path name.
2900  *
2901  * MPALMOSTSAFE
2902  */
2903 int
2904 sys_chflags(struct chflags_args *uap)
2905 {
2906 	struct nlookupdata nd;
2907 	struct vnode *vp;
2908 	int error;
2909 
2910 	vp = NULL;
2911 	get_mplock();
2912 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2913 	if (error == 0)
2914 		error = nlookup(&nd);
2915 	if (error == 0)
2916 		error = ncp_writechk(&nd.nl_nch);
2917 	if (error == 0)
2918 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2919 	nlookup_done(&nd);
2920 	if (error == 0) {
2921 		error = setfflags(vp, uap->flags);
2922 		vrele(vp);
2923 	}
2924 	rel_mplock();
2925 	return (error);
2926 }
2927 
2928 /*
2929  * lchflags(char *path, int flags)
2930  *
2931  * Change flags of a file given a path name, but don't follow symlinks.
2932  *
2933  * MPALMOSTSAFE
2934  */
2935 int
2936 sys_lchflags(struct lchflags_args *uap)
2937 {
2938 	struct nlookupdata nd;
2939 	struct vnode *vp;
2940 	int error;
2941 
2942 	vp = NULL;
2943 	get_mplock();
2944 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2945 	if (error == 0)
2946 		error = nlookup(&nd);
2947 	if (error == 0)
2948 		error = ncp_writechk(&nd.nl_nch);
2949 	if (error == 0)
2950 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2951 	nlookup_done(&nd);
2952 	if (error == 0) {
2953 		error = setfflags(vp, uap->flags);
2954 		vrele(vp);
2955 	}
2956 	rel_mplock();
2957 	return (error);
2958 }
2959 
2960 /*
2961  * fchflags_args(int fd, int flags)
2962  *
2963  * Change flags of a file given a file descriptor.
2964  *
2965  * MPALMOSTSAFE
2966  */
2967 int
2968 sys_fchflags(struct fchflags_args *uap)
2969 {
2970 	struct thread *td = curthread;
2971 	struct proc *p = td->td_proc;
2972 	struct file *fp;
2973 	int error;
2974 
2975 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2976 		return (error);
2977 	get_mplock();
2978 	if (fp->f_nchandle.ncp)
2979 		error = ncp_writechk(&fp->f_nchandle);
2980 	if (error == 0)
2981 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2982 	rel_mplock();
2983 	fdrop(fp);
2984 	return (error);
2985 }
2986 
2987 static int
2988 setfmode(struct vnode *vp, int mode)
2989 {
2990 	struct thread *td = curthread;
2991 	int error;
2992 	struct vattr vattr;
2993 
2994 	/*
2995 	 * note: vget is required for any operation that might mod the vnode
2996 	 * so VINACTIVE is properly cleared.
2997 	 */
2998 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2999 		VATTR_NULL(&vattr);
3000 		vattr.va_mode = mode & ALLPERMS;
3001 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3002 		vput(vp);
3003 	}
3004 	return error;
3005 }
3006 
3007 int
3008 kern_chmod(struct nlookupdata *nd, int mode)
3009 {
3010 	struct vnode *vp;
3011 	int error;
3012 
3013 	if ((error = nlookup(nd)) != 0)
3014 		return (error);
3015 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3016 		return (error);
3017 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3018 		error = setfmode(vp, mode);
3019 	vrele(vp);
3020 	return (error);
3021 }
3022 
3023 /*
3024  * chmod_args(char *path, int mode)
3025  *
3026  * Change mode of a file given path name.
3027  *
3028  * MPALMOSTSAFE
3029  */
3030 int
3031 sys_chmod(struct chmod_args *uap)
3032 {
3033 	struct nlookupdata nd;
3034 	int error;
3035 
3036 	get_mplock();
3037 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3038 	if (error == 0)
3039 		error = kern_chmod(&nd, uap->mode);
3040 	nlookup_done(&nd);
3041 	rel_mplock();
3042 	return (error);
3043 }
3044 
3045 /*
3046  * lchmod_args(char *path, int mode)
3047  *
3048  * Change mode of a file given path name (don't follow links.)
3049  *
3050  * MPALMOSTSAFE
3051  */
3052 int
3053 sys_lchmod(struct lchmod_args *uap)
3054 {
3055 	struct nlookupdata nd;
3056 	int error;
3057 
3058 	get_mplock();
3059 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3060 	if (error == 0)
3061 		error = kern_chmod(&nd, uap->mode);
3062 	nlookup_done(&nd);
3063 	rel_mplock();
3064 	return (error);
3065 }
3066 
3067 /*
3068  * fchmod_args(int fd, int mode)
3069  *
3070  * Change mode of a file given a file descriptor.
3071  *
3072  * MPALMOSTSAFE
3073  */
3074 int
3075 sys_fchmod(struct fchmod_args *uap)
3076 {
3077 	struct thread *td = curthread;
3078 	struct proc *p = td->td_proc;
3079 	struct file *fp;
3080 	int error;
3081 
3082 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3083 		return (error);
3084 	get_mplock();
3085 	if (fp->f_nchandle.ncp)
3086 		error = ncp_writechk(&fp->f_nchandle);
3087 	if (error == 0)
3088 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
3089 	rel_mplock();
3090 	fdrop(fp);
3091 	return (error);
3092 }
3093 
3094 /*
3095  * fchmodat_args(char *path, int mode)
3096  *
3097  * Change mode of a file pointed to by fd/path.
3098  *
3099  * MPALMOSTSAFE
3100  */
3101 int
3102 sys_fchmodat(struct fchmodat_args *uap)
3103 {
3104 	struct nlookupdata nd;
3105 	struct file *fp;
3106 	int error;
3107 	int flags;
3108 
3109 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3110 		return (EINVAL);
3111 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3112 
3113 	get_mplock();
3114 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3115 				UIO_USERSPACE, flags);
3116 	if (error == 0)
3117 		error = kern_chmod(&nd, uap->mode);
3118 	nlookup_done_at(&nd, fp);
3119 	rel_mplock();
3120 	return (error);
3121 }
3122 
3123 static int
3124 setfown(struct vnode *vp, uid_t uid, gid_t gid)
3125 {
3126 	struct thread *td = curthread;
3127 	int error;
3128 	struct vattr vattr;
3129 
3130 	/*
3131 	 * note: vget is required for any operation that might mod the vnode
3132 	 * so VINACTIVE is properly cleared.
3133 	 */
3134 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3135 		VATTR_NULL(&vattr);
3136 		vattr.va_uid = uid;
3137 		vattr.va_gid = gid;
3138 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3139 		vput(vp);
3140 	}
3141 	return error;
3142 }
3143 
3144 int
3145 kern_chown(struct nlookupdata *nd, int uid, int gid)
3146 {
3147 	struct vnode *vp;
3148 	int error;
3149 
3150 	if ((error = nlookup(nd)) != 0)
3151 		return (error);
3152 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3153 		return (error);
3154 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3155 		error = setfown(vp, uid, gid);
3156 	vrele(vp);
3157 	return (error);
3158 }
3159 
3160 /*
3161  * chown(char *path, int uid, int gid)
3162  *
3163  * Set ownership given a path name.
3164  *
3165  * MPALMOSTSAFE
3166  */
3167 int
3168 sys_chown(struct chown_args *uap)
3169 {
3170 	struct nlookupdata nd;
3171 	int error;
3172 
3173 	get_mplock();
3174 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3175 	if (error == 0)
3176 		error = kern_chown(&nd, uap->uid, uap->gid);
3177 	nlookup_done(&nd);
3178 	rel_mplock();
3179 	return (error);
3180 }
3181 
3182 /*
3183  * lchown_args(char *path, int uid, int gid)
3184  *
3185  * Set ownership given a path name, do not cross symlinks.
3186  *
3187  * MPALMOSTSAFE
3188  */
3189 int
3190 sys_lchown(struct lchown_args *uap)
3191 {
3192 	struct nlookupdata nd;
3193 	int error;
3194 
3195 	get_mplock();
3196 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3197 	if (error == 0)
3198 		error = kern_chown(&nd, uap->uid, uap->gid);
3199 	nlookup_done(&nd);
3200 	rel_mplock();
3201 	return (error);
3202 }
3203 
3204 /*
3205  * fchown_args(int fd, int uid, int gid)
3206  *
3207  * Set ownership given a file descriptor.
3208  *
3209  * MPALMOSTSAFE
3210  */
3211 int
3212 sys_fchown(struct fchown_args *uap)
3213 {
3214 	struct thread *td = curthread;
3215 	struct proc *p = td->td_proc;
3216 	struct file *fp;
3217 	int error;
3218 
3219 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3220 		return (error);
3221 	get_mplock();
3222 	if (fp->f_nchandle.ncp)
3223 		error = ncp_writechk(&fp->f_nchandle);
3224 	if (error == 0)
3225 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
3226 	rel_mplock();
3227 	fdrop(fp);
3228 	return (error);
3229 }
3230 
3231 /*
3232  * fchownat(int fd, char *path, int uid, int gid, int flags)
3233  *
3234  * Set ownership of file pointed to by fd/path.
3235  *
3236  * MPALMOSTSAFE
3237  */
3238 int
3239 sys_fchownat(struct fchownat_args *uap)
3240 {
3241 	struct nlookupdata nd;
3242 	struct file *fp;
3243 	int error;
3244 	int flags;
3245 
3246 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3247 		return (EINVAL);
3248 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3249 
3250 	get_mplock();
3251 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3252 				UIO_USERSPACE, flags);
3253 	if (error == 0)
3254 		error = kern_chown(&nd, uap->uid, uap->gid);
3255 	nlookup_done_at(&nd, fp);
3256 	rel_mplock();
3257 	return (error);
3258 }
3259 
3260 
3261 static int
3262 getutimes(const struct timeval *tvp, struct timespec *tsp)
3263 {
3264 	struct timeval tv[2];
3265 
3266 	if (tvp == NULL) {
3267 		microtime(&tv[0]);
3268 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3269 		tsp[1] = tsp[0];
3270 	} else {
3271 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3272 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3273 	}
3274 	return 0;
3275 }
3276 
3277 static int
3278 setutimes(struct vnode *vp, struct vattr *vattr,
3279 	  const struct timespec *ts, int nullflag)
3280 {
3281 	struct thread *td = curthread;
3282 	int error;
3283 
3284 	VATTR_NULL(vattr);
3285 	vattr->va_atime = ts[0];
3286 	vattr->va_mtime = ts[1];
3287 	if (nullflag)
3288 		vattr->va_vaflags |= VA_UTIMES_NULL;
3289 	error = VOP_SETATTR(vp, vattr, td->td_ucred);
3290 
3291 	return error;
3292 }
3293 
3294 int
3295 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3296 {
3297 	struct timespec ts[2];
3298 	struct vnode *vp;
3299 	struct vattr vattr;
3300 	int error;
3301 
3302 	if ((error = getutimes(tptr, ts)) != 0)
3303 		return (error);
3304 
3305 	/*
3306 	 * NOTE: utimes() succeeds for the owner even if the file
3307 	 * is not user-writable.
3308 	 */
3309 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3310 
3311 	if ((error = nlookup(nd)) != 0)
3312 		return (error);
3313 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3314 		return (error);
3315 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3316 		return (error);
3317 
3318 	/*
3319 	 * note: vget is required for any operation that might mod the vnode
3320 	 * so VINACTIVE is properly cleared.
3321 	 */
3322 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3323 		error = vget(vp, LK_EXCLUSIVE);
3324 		if (error == 0) {
3325 			error = setutimes(vp, &vattr, ts, (tptr == NULL));
3326 			vput(vp);
3327 		}
3328 	}
3329 	vrele(vp);
3330 	return (error);
3331 }
3332 
3333 /*
3334  * utimes_args(char *path, struct timeval *tptr)
3335  *
3336  * Set the access and modification times of a file.
3337  *
3338  * MPALMOSTSAFE
3339  */
3340 int
3341 sys_utimes(struct utimes_args *uap)
3342 {
3343 	struct timeval tv[2];
3344 	struct nlookupdata nd;
3345 	int error;
3346 
3347 	if (uap->tptr) {
3348  		error = copyin(uap->tptr, tv, sizeof(tv));
3349 		if (error)
3350 			return (error);
3351 	}
3352 	get_mplock();
3353 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3354 	if (error == 0)
3355 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3356 	nlookup_done(&nd);
3357 	rel_mplock();
3358 	return (error);
3359 }
3360 
3361 /*
3362  * lutimes_args(char *path, struct timeval *tptr)
3363  *
3364  * Set the access and modification times of a file.
3365  *
3366  * MPALMOSTSAFE
3367  */
3368 int
3369 sys_lutimes(struct lutimes_args *uap)
3370 {
3371 	struct timeval tv[2];
3372 	struct nlookupdata nd;
3373 	int error;
3374 
3375 	if (uap->tptr) {
3376 		error = copyin(uap->tptr, tv, sizeof(tv));
3377 		if (error)
3378 			return (error);
3379 	}
3380 	get_mplock();
3381 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3382 	if (error == 0)
3383 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3384 	nlookup_done(&nd);
3385 	rel_mplock();
3386 	return (error);
3387 }
3388 
3389 /*
3390  * Set utimes on a file descriptor.  The creds used to open the
3391  * file are used to determine whether the operation is allowed
3392  * or not.
3393  */
3394 int
3395 kern_futimes(int fd, struct timeval *tptr)
3396 {
3397 	struct thread *td = curthread;
3398 	struct proc *p = td->td_proc;
3399 	struct timespec ts[2];
3400 	struct file *fp;
3401 	struct vnode *vp;
3402 	struct vattr vattr;
3403 	int error;
3404 
3405 	error = getutimes(tptr, ts);
3406 	if (error)
3407 		return (error);
3408 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3409 		return (error);
3410 	if (fp->f_nchandle.ncp)
3411 		error = ncp_writechk(&fp->f_nchandle);
3412 	if (error == 0) {
3413 		vp = fp->f_data;
3414 		error = vget(vp, LK_EXCLUSIVE);
3415 		if (error == 0) {
3416 			error = VOP_GETATTR(vp, &vattr);
3417 			if (error == 0) {
3418 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3419 						   fp->f_cred);
3420 			}
3421 			if (error == 0) {
3422 				error = setutimes(vp, &vattr, ts,
3423 						  (tptr == NULL));
3424 			}
3425 			vput(vp);
3426 		}
3427 	}
3428 	fdrop(fp);
3429 	return (error);
3430 }
3431 
3432 /*
3433  * futimes_args(int fd, struct timeval *tptr)
3434  *
3435  * Set the access and modification times of a file.
3436  *
3437  * MPALMOSTSAFE
3438  */
3439 int
3440 sys_futimes(struct futimes_args *uap)
3441 {
3442 	struct timeval tv[2];
3443 	int error;
3444 
3445 	if (uap->tptr) {
3446 		error = copyin(uap->tptr, tv, sizeof(tv));
3447 		if (error)
3448 			return (error);
3449 	}
3450 	get_mplock();
3451 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3452 	rel_mplock();
3453 
3454 	return (error);
3455 }
3456 
3457 int
3458 kern_truncate(struct nlookupdata *nd, off_t length)
3459 {
3460 	struct vnode *vp;
3461 	struct vattr vattr;
3462 	int error;
3463 
3464 	if (length < 0)
3465 		return(EINVAL);
3466 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3467 	if ((error = nlookup(nd)) != 0)
3468 		return (error);
3469 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3470 		return (error);
3471 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3472 		return (error);
3473 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3474 		vrele(vp);
3475 		return (error);
3476 	}
3477 	if (vp->v_type == VDIR) {
3478 		error = EISDIR;
3479 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3480 		VATTR_NULL(&vattr);
3481 		vattr.va_size = length;
3482 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3483 	}
3484 	vput(vp);
3485 	return (error);
3486 }
3487 
3488 /*
3489  * truncate(char *path, int pad, off_t length)
3490  *
3491  * Truncate a file given its path name.
3492  *
3493  * MPALMOSTSAFE
3494  */
3495 int
3496 sys_truncate(struct truncate_args *uap)
3497 {
3498 	struct nlookupdata nd;
3499 	int error;
3500 
3501 	get_mplock();
3502 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3503 	if (error == 0)
3504 		error = kern_truncate(&nd, uap->length);
3505 	nlookup_done(&nd);
3506 	rel_mplock();
3507 	return error;
3508 }
3509 
3510 int
3511 kern_ftruncate(int fd, off_t length)
3512 {
3513 	struct thread *td = curthread;
3514 	struct proc *p = td->td_proc;
3515 	struct vattr vattr;
3516 	struct vnode *vp;
3517 	struct file *fp;
3518 	int error;
3519 
3520 	if (length < 0)
3521 		return(EINVAL);
3522 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3523 		return (error);
3524 	if (fp->f_nchandle.ncp) {
3525 		error = ncp_writechk(&fp->f_nchandle);
3526 		if (error)
3527 			goto done;
3528 	}
3529 	if ((fp->f_flag & FWRITE) == 0) {
3530 		error = EINVAL;
3531 		goto done;
3532 	}
3533 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3534 		error = EINVAL;
3535 		goto done;
3536 	}
3537 	vp = (struct vnode *)fp->f_data;
3538 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3539 	if (vp->v_type == VDIR) {
3540 		error = EISDIR;
3541 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
3542 		VATTR_NULL(&vattr);
3543 		vattr.va_size = length;
3544 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3545 	}
3546 	vn_unlock(vp);
3547 done:
3548 	fdrop(fp);
3549 	return (error);
3550 }
3551 
3552 /*
3553  * ftruncate_args(int fd, int pad, off_t length)
3554  *
3555  * Truncate a file given a file descriptor.
3556  *
3557  * MPALMOSTSAFE
3558  */
3559 int
3560 sys_ftruncate(struct ftruncate_args *uap)
3561 {
3562 	int error;
3563 
3564 	get_mplock();
3565 	error = kern_ftruncate(uap->fd, uap->length);
3566 	rel_mplock();
3567 
3568 	return (error);
3569 }
3570 
3571 /*
3572  * fsync(int fd)
3573  *
3574  * Sync an open file.
3575  *
3576  * MPALMOSTSAFE
3577  */
3578 int
3579 sys_fsync(struct fsync_args *uap)
3580 {
3581 	struct thread *td = curthread;
3582 	struct proc *p = td->td_proc;
3583 	struct vnode *vp;
3584 	struct file *fp;
3585 	vm_object_t obj;
3586 	int error;
3587 
3588 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3589 		return (error);
3590 	get_mplock();
3591 	vp = (struct vnode *)fp->f_data;
3592 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3593 	if ((obj = vp->v_object) != NULL)
3594 		vm_object_page_clean(obj, 0, 0, 0);
3595 	error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3596 	if (error == 0 && vp->v_mount)
3597 		error = buf_fsync(vp);
3598 	vn_unlock(vp);
3599 	rel_mplock();
3600 	fdrop(fp);
3601 
3602 	return (error);
3603 }
3604 
3605 int
3606 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3607 {
3608 	struct nchandle fnchd;
3609 	struct nchandle tnchd;
3610 	struct namecache *ncp;
3611 	struct vnode *fdvp;
3612 	struct vnode *tdvp;
3613 	struct mount *mp;
3614 	int error;
3615 
3616 	bwillinode(1);
3617 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3618 	if ((error = nlookup(fromnd)) != 0)
3619 		return (error);
3620 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3621 		return (ENOENT);
3622 	fnchd.mount = fromnd->nl_nch.mount;
3623 	cache_hold(&fnchd);
3624 
3625 	/*
3626 	 * unlock the source nch so we can lookup the target nch without
3627 	 * deadlocking.  The target may or may not exist so we do not check
3628 	 * for a target vp like kern_mkdir() and other creation functions do.
3629 	 *
3630 	 * The source and target directories are ref'd and rechecked after
3631 	 * everything is relocked to determine if the source or target file
3632 	 * has been renamed.
3633 	 */
3634 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3635 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3636 	cache_unlock(&fromnd->nl_nch);
3637 
3638 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3639 	if ((error = nlookup(tond)) != 0) {
3640 		cache_drop(&fnchd);
3641 		return (error);
3642 	}
3643 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3644 		cache_drop(&fnchd);
3645 		return (ENOENT);
3646 	}
3647 	tnchd.mount = tond->nl_nch.mount;
3648 	cache_hold(&tnchd);
3649 
3650 	/*
3651 	 * If the source and target are the same there is nothing to do
3652 	 */
3653 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3654 		cache_drop(&fnchd);
3655 		cache_drop(&tnchd);
3656 		return (0);
3657 	}
3658 
3659 	/*
3660 	 * Mount points cannot be renamed or overwritten
3661 	 */
3662 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3663 	    NCF_ISMOUNTPT
3664 	) {
3665 		cache_drop(&fnchd);
3666 		cache_drop(&tnchd);
3667 		return (EINVAL);
3668 	}
3669 
3670 	/*
3671 	 * Relock the source ncp.  cache_relock() will deal with any
3672 	 * deadlocks against the already-locked tond and will also
3673 	 * make sure both are resolved.
3674 	 *
3675 	 * NOTE AFTER RELOCKING: The source or target ncp may have become
3676 	 * invalid while they were unlocked, nc_vp and nc_mount could
3677 	 * be NULL.
3678 	 */
3679 	cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
3680 		     &tond->nl_nch, tond->nl_cred);
3681 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3682 
3683 	/*
3684 	 * make sure the parent directories linkages are the same
3685 	 */
3686 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3687 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3688 		cache_drop(&fnchd);
3689 		cache_drop(&tnchd);
3690 		return (ENOENT);
3691 	}
3692 
3693 	/*
3694 	 * Both the source and target must be within the same filesystem and
3695 	 * in the same filesystem as their parent directories within the
3696 	 * namecache topology.
3697 	 *
3698 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3699 	 */
3700 	mp = fnchd.mount;
3701 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3702 	    mp != tond->nl_nch.mount) {
3703 		cache_drop(&fnchd);
3704 		cache_drop(&tnchd);
3705 		return (EXDEV);
3706 	}
3707 
3708 	/*
3709 	 * Make sure the mount point is writable
3710 	 */
3711 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3712 		cache_drop(&fnchd);
3713 		cache_drop(&tnchd);
3714 		return (error);
3715 	}
3716 
3717 	/*
3718 	 * If the target exists and either the source or target is a directory,
3719 	 * then both must be directories.
3720 	 *
3721 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3722 	 * have become NULL.
3723 	 */
3724 	if (tond->nl_nch.ncp->nc_vp) {
3725 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3726 			error = ENOENT;
3727 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3728 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3729 				error = ENOTDIR;
3730 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3731 			error = EISDIR;
3732 		}
3733 	}
3734 
3735 	/*
3736 	 * You cannot rename a source into itself or a subdirectory of itself.
3737 	 * We check this by travsersing the target directory upwards looking
3738 	 * for a match against the source.
3739 	 *
3740 	 * XXX MPSAFE
3741 	 */
3742 	if (error == 0) {
3743 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3744 			if (fromnd->nl_nch.ncp == ncp) {
3745 				error = EINVAL;
3746 				break;
3747 			}
3748 		}
3749 	}
3750 
3751 	cache_drop(&fnchd);
3752 	cache_drop(&tnchd);
3753 
3754 	/*
3755 	 * Even though the namespaces are different, they may still represent
3756 	 * hardlinks to the same file.  The filesystem might have a hard time
3757 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3758 	 * when we detect the situation.
3759 	 */
3760 	if (error == 0) {
3761 		fdvp = fromnd->nl_dvp;
3762 		tdvp = tond->nl_dvp;
3763 		if (fdvp == NULL || tdvp == NULL) {
3764 			error = EPERM;
3765 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3766 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3767 					    fromnd->nl_cred);
3768 		} else {
3769 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3770 					    fdvp, tdvp, tond->nl_cred);
3771 		}
3772 	}
3773 	return (error);
3774 }
3775 
3776 /*
3777  * rename_args(char *from, char *to)
3778  *
3779  * Rename files.  Source and destination must either both be directories,
3780  * or both not be directories.  If target is a directory, it must be empty.
3781  *
3782  * MPALMOSTSAFE
3783  */
3784 int
3785 sys_rename(struct rename_args *uap)
3786 {
3787 	struct nlookupdata fromnd, tond;
3788 	int error;
3789 
3790 	get_mplock();
3791 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3792 	if (error == 0) {
3793 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3794 		if (error == 0)
3795 			error = kern_rename(&fromnd, &tond);
3796 		nlookup_done(&tond);
3797 	}
3798 	nlookup_done(&fromnd);
3799 	rel_mplock();
3800 	return (error);
3801 }
3802 
3803 int
3804 kern_mkdir(struct nlookupdata *nd, int mode)
3805 {
3806 	struct thread *td = curthread;
3807 	struct proc *p = td->td_proc;
3808 	struct vnode *vp;
3809 	struct vattr vattr;
3810 	int error;
3811 
3812 	bwillinode(1);
3813 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3814 	if ((error = nlookup(nd)) != 0)
3815 		return (error);
3816 
3817 	if (nd->nl_nch.ncp->nc_vp)
3818 		return (EEXIST);
3819 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3820 		return (error);
3821 	VATTR_NULL(&vattr);
3822 	vattr.va_type = VDIR;
3823 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3824 
3825 	vp = NULL;
3826 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
3827 	if (error == 0)
3828 		vput(vp);
3829 	return (error);
3830 }
3831 
3832 /*
3833  * mkdir_args(char *path, int mode)
3834  *
3835  * Make a directory file.
3836  *
3837  * MPALMOSTSAFE
3838  */
3839 int
3840 sys_mkdir(struct mkdir_args *uap)
3841 {
3842 	struct nlookupdata nd;
3843 	int error;
3844 
3845 	get_mplock();
3846 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3847 	if (error == 0)
3848 		error = kern_mkdir(&nd, uap->mode);
3849 	nlookup_done(&nd);
3850 	rel_mplock();
3851 	return (error);
3852 }
3853 
3854 int
3855 kern_rmdir(struct nlookupdata *nd)
3856 {
3857 	int error;
3858 
3859 	bwillinode(1);
3860 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3861 	if ((error = nlookup(nd)) != 0)
3862 		return (error);
3863 
3864 	/*
3865 	 * Do not allow directories representing mount points to be
3866 	 * deleted, even if empty.  Check write perms on mount point
3867 	 * in case the vnode is aliased (aka nullfs).
3868 	 */
3869 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3870 		return (EINVAL);
3871 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3872 		return (error);
3873 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3874 	return (error);
3875 }
3876 
3877 /*
3878  * rmdir_args(char *path)
3879  *
3880  * Remove a directory file.
3881  *
3882  * MPALMOSTSAFE
3883  */
3884 int
3885 sys_rmdir(struct rmdir_args *uap)
3886 {
3887 	struct nlookupdata nd;
3888 	int error;
3889 
3890 	get_mplock();
3891 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3892 	if (error == 0)
3893 		error = kern_rmdir(&nd);
3894 	nlookup_done(&nd);
3895 	rel_mplock();
3896 	return (error);
3897 }
3898 
3899 int
3900 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3901     enum uio_seg direction)
3902 {
3903 	struct thread *td = curthread;
3904 	struct proc *p = td->td_proc;
3905 	struct vnode *vp;
3906 	struct file *fp;
3907 	struct uio auio;
3908 	struct iovec aiov;
3909 	off_t loff;
3910 	int error, eofflag;
3911 
3912 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3913 		return (error);
3914 	if ((fp->f_flag & FREAD) == 0) {
3915 		error = EBADF;
3916 		goto done;
3917 	}
3918 	vp = (struct vnode *)fp->f_data;
3919 unionread:
3920 	if (vp->v_type != VDIR) {
3921 		error = EINVAL;
3922 		goto done;
3923 	}
3924 	aiov.iov_base = buf;
3925 	aiov.iov_len = count;
3926 	auio.uio_iov = &aiov;
3927 	auio.uio_iovcnt = 1;
3928 	auio.uio_rw = UIO_READ;
3929 	auio.uio_segflg = direction;
3930 	auio.uio_td = td;
3931 	auio.uio_resid = count;
3932 	loff = auio.uio_offset = fp->f_offset;
3933 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3934 	fp->f_offset = auio.uio_offset;
3935 	if (error)
3936 		goto done;
3937 	if (count == auio.uio_resid) {
3938 		if (union_dircheckp) {
3939 			error = union_dircheckp(td, &vp, fp);
3940 			if (error == -1)
3941 				goto unionread;
3942 			if (error)
3943 				goto done;
3944 		}
3945 #if 0
3946 		if ((vp->v_flag & VROOT) &&
3947 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3948 			struct vnode *tvp = vp;
3949 			vp = vp->v_mount->mnt_vnodecovered;
3950 			vref(vp);
3951 			fp->f_data = vp;
3952 			fp->f_offset = 0;
3953 			vrele(tvp);
3954 			goto unionread;
3955 		}
3956 #endif
3957 	}
3958 
3959 	/*
3960 	 * WARNING!  *basep may not be wide enough to accomodate the
3961 	 * seek offset.   XXX should we hack this to return the upper 32 bits
3962 	 * for offsets greater then 4G?
3963 	 */
3964 	if (basep) {
3965 		*basep = (long)loff;
3966 	}
3967 	*res = count - auio.uio_resid;
3968 done:
3969 	fdrop(fp);
3970 	return (error);
3971 }
3972 
3973 /*
3974  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3975  *
3976  * Read a block of directory entries in a file system independent format.
3977  *
3978  * MPALMOSTSAFE
3979  */
3980 int
3981 sys_getdirentries(struct getdirentries_args *uap)
3982 {
3983 	long base;
3984 	int error;
3985 
3986 	get_mplock();
3987 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3988 				   &uap->sysmsg_result, UIO_USERSPACE);
3989 	rel_mplock();
3990 
3991 	if (error == 0 && uap->basep)
3992 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3993 	return (error);
3994 }
3995 
3996 /*
3997  * getdents_args(int fd, char *buf, size_t count)
3998  *
3999  * MPALMOSTSAFE
4000  */
4001 int
4002 sys_getdents(struct getdents_args *uap)
4003 {
4004 	int error;
4005 
4006 	get_mplock();
4007 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4008 				   &uap->sysmsg_result, UIO_USERSPACE);
4009 	rel_mplock();
4010 
4011 	return (error);
4012 }
4013 
4014 /*
4015  * Set the mode mask for creation of filesystem nodes.
4016  *
4017  * umask(int newmask)
4018  *
4019  * MPSAFE
4020  */
4021 int
4022 sys_umask(struct umask_args *uap)
4023 {
4024 	struct thread *td = curthread;
4025 	struct proc *p = td->td_proc;
4026 	struct filedesc *fdp;
4027 
4028 	fdp = p->p_fd;
4029 	uap->sysmsg_result = fdp->fd_cmask;
4030 	fdp->fd_cmask = uap->newmask & ALLPERMS;
4031 	return (0);
4032 }
4033 
4034 /*
4035  * revoke(char *path)
4036  *
4037  * Void all references to file by ripping underlying filesystem
4038  * away from vnode.
4039  *
4040  * MPALMOSTSAFE
4041  */
4042 int
4043 sys_revoke(struct revoke_args *uap)
4044 {
4045 	struct nlookupdata nd;
4046 	struct vattr vattr;
4047 	struct vnode *vp;
4048 	struct ucred *cred;
4049 	int error;
4050 
4051 	vp = NULL;
4052 	get_mplock();
4053 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4054 	if (error == 0)
4055 		error = nlookup(&nd);
4056 	if (error == 0)
4057 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4058 	cred = crhold(nd.nl_cred);
4059 	nlookup_done(&nd);
4060 	if (error == 0) {
4061 		if (error == 0)
4062 			error = VOP_GETATTR(vp, &vattr);
4063 		if (error == 0 && cred->cr_uid != vattr.va_uid)
4064 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4065 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4066 			if (vcount(vp) > 0)
4067 				error = vrevoke(vp, cred);
4068 		} else if (error == 0) {
4069 			error = vrevoke(vp, cred);
4070 		}
4071 		vrele(vp);
4072 	}
4073 	if (cred)
4074 		crfree(cred);
4075 	rel_mplock();
4076 	return (error);
4077 }
4078 
4079 /*
4080  * getfh_args(char *fname, fhandle_t *fhp)
4081  *
4082  * Get (NFS) file handle
4083  *
4084  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4085  * mount.  This allows nullfs mounts to be explicitly exported.
4086  *
4087  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4088  *
4089  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
4090  *	    work, but you do not really have protection against access to
4091  *	    the related parent directories.
4092  *
4093  * MPALMOSTSAFE
4094  */
4095 int
4096 sys_getfh(struct getfh_args *uap)
4097 {
4098 	struct thread *td = curthread;
4099 	struct nlookupdata nd;
4100 	fhandle_t fh;
4101 	struct vnode *vp;
4102 	struct mount *mp;
4103 	int error;
4104 
4105 	/*
4106 	 * Must be super user
4107 	 */
4108 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
4109 		return (error);
4110 
4111 	vp = NULL;
4112 	get_mplock();
4113 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4114 	if (error == 0)
4115 		error = nlookup(&nd);
4116 	if (error == 0)
4117 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4118 	mp = nd.nl_nch.mount;
4119 	nlookup_done(&nd);
4120 	if (error == 0) {
4121 		bzero(&fh, sizeof(fh));
4122 		fh.fh_fsid = mp->mnt_stat.f_fsid;
4123 		error = VFS_VPTOFH(vp, &fh.fh_fid);
4124 		vput(vp);
4125 		if (error == 0)
4126 			error = copyout(&fh, uap->fhp, sizeof(fh));
4127 	}
4128 	rel_mplock();
4129 	return (error);
4130 }
4131 
4132 /*
4133  * fhopen_args(const struct fhandle *u_fhp, int flags)
4134  *
4135  * syscall for the rpc.lockd to use to translate a NFS file handle into
4136  * an open descriptor.
4137  *
4138  * warning: do not remove the priv_check() call or this becomes one giant
4139  * security hole.
4140  *
4141  * MPALMOSTSAFE
4142  */
4143 int
4144 sys_fhopen(struct fhopen_args *uap)
4145 {
4146 	struct thread *td = curthread;
4147 	struct filedesc *fdp = td->td_proc->p_fd;
4148 	struct mount *mp;
4149 	struct vnode *vp;
4150 	struct fhandle fhp;
4151 	struct vattr vat;
4152 	struct vattr *vap = &vat;
4153 	struct flock lf;
4154 	int fmode, mode, error, type;
4155 	struct file *nfp;
4156 	struct file *fp;
4157 	int indx;
4158 
4159 	/*
4160 	 * Must be super user
4161 	 */
4162 	error = priv_check(td, PRIV_ROOT);
4163 	if (error)
4164 		return (error);
4165 
4166 	fmode = FFLAGS(uap->flags);
4167 
4168 	/*
4169 	 * Why not allow a non-read/write open for our lockd?
4170 	 */
4171 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4172 		return (EINVAL);
4173 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4174 	if (error)
4175 		return(error);
4176 
4177 	/*
4178 	 * Find the mount point
4179 	 */
4180 	get_mplock();
4181 	mp = vfs_getvfs(&fhp.fh_fsid);
4182 	if (mp == NULL) {
4183 		error = ESTALE;
4184 		goto  done;
4185 	}
4186 	/* now give me my vnode, it gets returned to me locked */
4187 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4188 	if (error)
4189 		goto done;
4190  	/*
4191 	 * from now on we have to make sure not
4192 	 * to forget about the vnode
4193 	 * any error that causes an abort must vput(vp)
4194 	 * just set error = err and 'goto bad;'.
4195 	 */
4196 
4197 	/*
4198 	 * from vn_open
4199 	 */
4200 	if (vp->v_type == VLNK) {
4201 		error = EMLINK;
4202 		goto bad;
4203 	}
4204 	if (vp->v_type == VSOCK) {
4205 		error = EOPNOTSUPP;
4206 		goto bad;
4207 	}
4208 	mode = 0;
4209 	if (fmode & (FWRITE | O_TRUNC)) {
4210 		if (vp->v_type == VDIR) {
4211 			error = EISDIR;
4212 			goto bad;
4213 		}
4214 		error = vn_writechk(vp, NULL);
4215 		if (error)
4216 			goto bad;
4217 		mode |= VWRITE;
4218 	}
4219 	if (fmode & FREAD)
4220 		mode |= VREAD;
4221 	if (mode) {
4222 		error = VOP_ACCESS(vp, mode, td->td_ucred);
4223 		if (error)
4224 			goto bad;
4225 	}
4226 	if (fmode & O_TRUNC) {
4227 		vn_unlock(vp);				/* XXX */
4228 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
4229 		VATTR_NULL(vap);
4230 		vap->va_size = 0;
4231 		error = VOP_SETATTR(vp, vap, td->td_ucred);
4232 		if (error)
4233 			goto bad;
4234 	}
4235 
4236 	/*
4237 	 * VOP_OPEN needs the file pointer so it can potentially override
4238 	 * it.
4239 	 *
4240 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
4241 	 * directory.  XXX
4242 	 */
4243 	if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4244 		goto bad;
4245 	fp = nfp;
4246 
4247 	error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4248 	if (error) {
4249 		/*
4250 		 * setting f_ops this way prevents VOP_CLOSE from being
4251 		 * called or fdrop() releasing the vp from v_data.   Since
4252 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4253 		 */
4254 		fp->f_ops = &badfileops;
4255 		fp->f_data = NULL;
4256 		goto bad_drop;
4257 	}
4258 
4259 	/*
4260 	 * The fp is given its own reference, we still have our ref and lock.
4261 	 *
4262 	 * Assert that all regular files must be created with a VM object.
4263 	 */
4264 	if (vp->v_type == VREG && vp->v_object == NULL) {
4265 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4266 		goto bad_drop;
4267 	}
4268 
4269 	/*
4270 	 * The open was successful.  Handle any locking requirements.
4271 	 */
4272 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
4273 		lf.l_whence = SEEK_SET;
4274 		lf.l_start = 0;
4275 		lf.l_len = 0;
4276 		if (fmode & O_EXLOCK)
4277 			lf.l_type = F_WRLCK;
4278 		else
4279 			lf.l_type = F_RDLCK;
4280 		if (fmode & FNONBLOCK)
4281 			type = 0;
4282 		else
4283 			type = F_WAIT;
4284 		vn_unlock(vp);
4285 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4286 			/*
4287 			 * release our private reference.
4288 			 */
4289 			fsetfd(fdp, NULL, indx);
4290 			fdrop(fp);
4291 			vrele(vp);
4292 			goto done;
4293 		}
4294 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4295 		fp->f_flag |= FHASLOCK;
4296 	}
4297 
4298 	/*
4299 	 * Clean up.  Associate the file pointer with the previously
4300 	 * reserved descriptor and return it.
4301 	 */
4302 	vput(vp);
4303 	rel_mplock();
4304 	fsetfd(fdp, fp, indx);
4305 	fdrop(fp);
4306 	uap->sysmsg_result = indx;
4307 	return (0);
4308 
4309 bad_drop:
4310 	fsetfd(fdp, NULL, indx);
4311 	fdrop(fp);
4312 bad:
4313 	vput(vp);
4314 done:
4315 	rel_mplock();
4316 	return (error);
4317 }
4318 
4319 /*
4320  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4321  *
4322  * MPALMOSTSAFE
4323  */
4324 int
4325 sys_fhstat(struct fhstat_args *uap)
4326 {
4327 	struct thread *td = curthread;
4328 	struct stat sb;
4329 	fhandle_t fh;
4330 	struct mount *mp;
4331 	struct vnode *vp;
4332 	int error;
4333 
4334 	/*
4335 	 * Must be super user
4336 	 */
4337 	error = priv_check(td, PRIV_ROOT);
4338 	if (error)
4339 		return (error);
4340 
4341 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4342 	if (error)
4343 		return (error);
4344 
4345 	get_mplock();
4346 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4347 		error = ESTALE;
4348 	if (error == 0) {
4349 		if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4350 			error = vn_stat(vp, &sb, td->td_ucred);
4351 			vput(vp);
4352 		}
4353 	}
4354 	rel_mplock();
4355 	if (error == 0)
4356 		error = copyout(&sb, uap->sb, sizeof(sb));
4357 	return (error);
4358 }
4359 
4360 /*
4361  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4362  *
4363  * MPALMOSTSAFE
4364  */
4365 int
4366 sys_fhstatfs(struct fhstatfs_args *uap)
4367 {
4368 	struct thread *td = curthread;
4369 	struct proc *p = td->td_proc;
4370 	struct statfs *sp;
4371 	struct mount *mp;
4372 	struct vnode *vp;
4373 	struct statfs sb;
4374 	char *fullpath, *freepath;
4375 	fhandle_t fh;
4376 	int error;
4377 
4378 	/*
4379 	 * Must be super user
4380 	 */
4381 	if ((error = priv_check(td, PRIV_ROOT)))
4382 		return (error);
4383 
4384 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4385 		return (error);
4386 
4387 	get_mplock();
4388 
4389 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4390 		error = ESTALE;
4391 		goto done;
4392 	}
4393 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4394 		error = ESTALE;
4395 		goto done;
4396 	}
4397 
4398 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4399 		goto done;
4400 	mp = vp->v_mount;
4401 	sp = &mp->mnt_stat;
4402 	vput(vp);
4403 	if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4404 		goto done;
4405 
4406 	error = mount_path(p, mp, &fullpath, &freepath);
4407 	if (error)
4408 		goto done;
4409 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4410 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4411 	kfree(freepath, M_TEMP);
4412 
4413 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4414 	if (priv_check(td, PRIV_ROOT)) {
4415 		bcopy(sp, &sb, sizeof(sb));
4416 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4417 		sp = &sb;
4418 	}
4419 	error = copyout(sp, uap->buf, sizeof(*sp));
4420 done:
4421 	rel_mplock();
4422 	return (error);
4423 }
4424 
4425 /*
4426  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4427  *
4428  * MPALMOSTSAFE
4429  */
4430 int
4431 sys_fhstatvfs(struct fhstatvfs_args *uap)
4432 {
4433 	struct thread *td = curthread;
4434 	struct proc *p = td->td_proc;
4435 	struct statvfs *sp;
4436 	struct mount *mp;
4437 	struct vnode *vp;
4438 	fhandle_t fh;
4439 	int error;
4440 
4441 	/*
4442 	 * Must be super user
4443 	 */
4444 	if ((error = priv_check(td, PRIV_ROOT)))
4445 		return (error);
4446 
4447 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4448 		return (error);
4449 
4450 	get_mplock();
4451 
4452 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4453 		error = ESTALE;
4454 		goto done;
4455 	}
4456 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4457 		error = ESTALE;
4458 		goto done;
4459 	}
4460 
4461 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4462 		goto done;
4463 	mp = vp->v_mount;
4464 	sp = &mp->mnt_vstat;
4465 	vput(vp);
4466 	if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4467 		goto done;
4468 
4469 	sp->f_flag = 0;
4470 	if (mp->mnt_flag & MNT_RDONLY)
4471 		sp->f_flag |= ST_RDONLY;
4472 	if (mp->mnt_flag & MNT_NOSUID)
4473 		sp->f_flag |= ST_NOSUID;
4474 	error = copyout(sp, uap->buf, sizeof(*sp));
4475 done:
4476 	rel_mplock();
4477 	return (error);
4478 }
4479 
4480 
4481 /*
4482  * Syscall to push extended attribute configuration information into the
4483  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4484  * a command (int cmd), and attribute name and misc data.  For now, the
4485  * attribute name is left in userspace for consumption by the VFS_op.
4486  * It will probably be changed to be copied into sysspace by the
4487  * syscall in the future, once issues with various consumers of the
4488  * attribute code have raised their hands.
4489  *
4490  * Currently this is used only by UFS Extended Attributes.
4491  *
4492  * MPALMOSTSAFE
4493  */
4494 int
4495 sys_extattrctl(struct extattrctl_args *uap)
4496 {
4497 	struct nlookupdata nd;
4498 	struct vnode *vp;
4499 	char attrname[EXTATTR_MAXNAMELEN];
4500 	int error;
4501 	size_t size;
4502 
4503 	get_mplock();
4504 
4505 	attrname[0] = 0;
4506 	vp = NULL;
4507 	error = 0;
4508 
4509 	if (error == 0 && uap->filename) {
4510 		error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4511 				     NLC_FOLLOW);
4512 		if (error == 0)
4513 			error = nlookup(&nd);
4514 		if (error == 0)
4515 			error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4516 		nlookup_done(&nd);
4517 	}
4518 
4519 	if (error == 0 && uap->attrname) {
4520 		error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4521 				  &size);
4522 	}
4523 
4524 	if (error == 0) {
4525 		error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4526 		if (error == 0)
4527 			error = nlookup(&nd);
4528 		if (error == 0)
4529 			error = ncp_writechk(&nd.nl_nch);
4530 		if (error == 0) {
4531 			error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4532 					       uap->attrnamespace,
4533 					       uap->attrname, nd.nl_cred);
4534 		}
4535 		nlookup_done(&nd);
4536 	}
4537 
4538 	rel_mplock();
4539 
4540 	return (error);
4541 }
4542 
4543 /*
4544  * Syscall to get a named extended attribute on a file or directory.
4545  *
4546  * MPALMOSTSAFE
4547  */
4548 int
4549 sys_extattr_set_file(struct extattr_set_file_args *uap)
4550 {
4551 	char attrname[EXTATTR_MAXNAMELEN];
4552 	struct nlookupdata nd;
4553 	struct vnode *vp;
4554 	struct uio auio;
4555 	struct iovec aiov;
4556 	int error;
4557 
4558 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4559 	if (error)
4560 		return (error);
4561 
4562 	vp = NULL;
4563 	get_mplock();
4564 
4565 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4566 	if (error == 0)
4567 		error = nlookup(&nd);
4568 	if (error == 0)
4569 		error = ncp_writechk(&nd.nl_nch);
4570 	if (error == 0)
4571 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4572 	if (error) {
4573 		nlookup_done(&nd);
4574 		rel_mplock();
4575 		return (error);
4576 	}
4577 
4578 	bzero(&auio, sizeof(auio));
4579 	aiov.iov_base = uap->data;
4580 	aiov.iov_len = uap->nbytes;
4581 	auio.uio_iov = &aiov;
4582 	auio.uio_iovcnt = 1;
4583 	auio.uio_offset = 0;
4584 	auio.uio_resid = uap->nbytes;
4585 	auio.uio_rw = UIO_WRITE;
4586 	auio.uio_td = curthread;
4587 
4588 	error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4589 			       &auio, nd.nl_cred);
4590 
4591 	vput(vp);
4592 	nlookup_done(&nd);
4593 	rel_mplock();
4594 	return (error);
4595 }
4596 
4597 /*
4598  * Syscall to get a named extended attribute on a file or directory.
4599  *
4600  * MPALMOSTSAFE
4601  */
4602 int
4603 sys_extattr_get_file(struct extattr_get_file_args *uap)
4604 {
4605 	char attrname[EXTATTR_MAXNAMELEN];
4606 	struct nlookupdata nd;
4607 	struct uio auio;
4608 	struct iovec aiov;
4609 	struct vnode *vp;
4610 	int error;
4611 
4612 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4613 	if (error)
4614 		return (error);
4615 
4616 	vp = NULL;
4617 	get_mplock();
4618 
4619 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4620 	if (error == 0)
4621 		error = nlookup(&nd);
4622 	if (error == 0)
4623 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4624 	if (error) {
4625 		nlookup_done(&nd);
4626 		rel_mplock();
4627 		return (error);
4628 	}
4629 
4630 	bzero(&auio, sizeof(auio));
4631 	aiov.iov_base = uap->data;
4632 	aiov.iov_len = uap->nbytes;
4633 	auio.uio_iov = &aiov;
4634 	auio.uio_iovcnt = 1;
4635 	auio.uio_offset = 0;
4636 	auio.uio_resid = uap->nbytes;
4637 	auio.uio_rw = UIO_READ;
4638 	auio.uio_td = curthread;
4639 
4640 	error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4641 				&auio, nd.nl_cred);
4642 	uap->sysmsg_result = uap->nbytes - auio.uio_resid;
4643 
4644 	vput(vp);
4645 	nlookup_done(&nd);
4646 	rel_mplock();
4647 	return(error);
4648 }
4649 
4650 /*
4651  * Syscall to delete a named extended attribute from a file or directory.
4652  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4653  *
4654  * MPALMOSTSAFE
4655  */
4656 int
4657 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4658 {
4659 	char attrname[EXTATTR_MAXNAMELEN];
4660 	struct nlookupdata nd;
4661 	struct vnode *vp;
4662 	int error;
4663 
4664 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4665 	if (error)
4666 		return(error);
4667 
4668 	get_mplock();
4669 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4670 	if (error == 0)
4671 		error = nlookup(&nd);
4672 	if (error == 0)
4673 		error = ncp_writechk(&nd.nl_nch);
4674 	if (error == 0) {
4675 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4676 		if (error == 0) {
4677 			error = VOP_SETEXTATTR(vp, uap->attrnamespace,
4678 					       attrname, NULL, nd.nl_cred);
4679 			vput(vp);
4680 		}
4681 	}
4682 	nlookup_done(&nd);
4683 	rel_mplock();
4684 	return(error);
4685 }
4686 
4687 /*
4688  * Determine if the mount is visible to the process.
4689  */
4690 static int
4691 chroot_visible_mnt(struct mount *mp, struct proc *p)
4692 {
4693 	struct nchandle nch;
4694 
4695 	/*
4696 	 * Traverse from the mount point upwards.  If we hit the process
4697 	 * root then the mount point is visible to the process.
4698 	 */
4699 	nch = mp->mnt_ncmountpt;
4700 	while (nch.ncp) {
4701 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4702 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4703 			return(1);
4704 		}
4705 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4706 			nch = nch.mount->mnt_ncmounton;
4707 		} else {
4708 			nch.ncp = nch.ncp->nc_parent;
4709 		}
4710 	}
4711 
4712 	/*
4713 	 * If the mount point is not visible to the process, but the
4714 	 * process root is in a subdirectory of the mount, return
4715 	 * TRUE anyway.
4716 	 */
4717 	if (p->p_fd->fd_nrdir.mount == mp)
4718 		return(1);
4719 
4720 	return(0);
4721 }
4722 
4723