xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 70e491c09b1cb18f96facdbc91052080f2a455c3)
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 int
637 dounmount(struct mount *mp, int flags)
638 {
639 	struct namecache *ncp;
640 	struct nchandle nch;
641 	struct vnode *vp;
642 	int error;
643 	int async_flag;
644 	int lflags;
645 	int freeok = 1;
646 
647 	/*
648 	 * Exclusive access for unmounting purposes
649 	 */
650 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
651 		return (error);
652 
653 	/*
654 	 * Allow filesystems to detect that a forced unmount is in progress.
655 	 */
656 	if (flags & MNT_FORCE)
657 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
658 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
659 	error = lockmgr(&mp->mnt_lock, lflags);
660 	if (error) {
661 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
662 		if (mp->mnt_kern_flag & MNTK_MWAIT)
663 			wakeup(mp);
664 		return (error);
665 	}
666 
667 	if (mp->mnt_flag & MNT_EXPUBLIC)
668 		vfs_setpublicfs(NULL, NULL, NULL);
669 
670 	vfs_msync(mp, MNT_WAIT);
671 	async_flag = mp->mnt_flag & MNT_ASYNC;
672 	mp->mnt_flag &=~ MNT_ASYNC;
673 
674 	/*
675 	 * If this filesystem isn't aliasing other filesystems,
676 	 * try to invalidate any remaining namecache entries and
677 	 * check the count afterwords.
678 	 */
679 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
680 		cache_lock(&mp->mnt_ncmountpt);
681 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
682 		cache_unlock(&mp->mnt_ncmountpt);
683 
684 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
685 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
686 
687 			if ((flags & MNT_FORCE) == 0) {
688 				error = EBUSY;
689 				mount_warning(mp, "Cannot unmount: "
690 						  "%d namecache "
691 						  "references still "
692 						  "present",
693 						  ncp->nc_refs - 1);
694 			} else {
695 				mount_warning(mp, "Forced unmount: "
696 						  "%d namecache "
697 						  "references still "
698 						  "present",
699 						  ncp->nc_refs - 1);
700 				freeok = 0;
701 			}
702 		}
703 	}
704 
705 	/*
706 	 * nchandle records ref the mount structure.  Expect a count of 1
707 	 * (our mount->mnt_ncmountpt).
708 	 */
709 	if (mp->mnt_refs != 1) {
710 		if ((flags & MNT_FORCE) == 0) {
711 			mount_warning(mp, "Cannot unmount: "
712 					  "%d process references still "
713 					  "present", mp->mnt_refs);
714 			error = EBUSY;
715 		} else {
716 			mount_warning(mp, "Forced unmount: "
717 					  "%d process references still "
718 					  "present", mp->mnt_refs);
719 			freeok = 0;
720 		}
721 	}
722 
723 	/*
724 	 * Decomission our special mnt_syncer vnode.  This also stops
725 	 * the vnlru code.  If we are unable to unmount we recommission
726 	 * the vnode.
727 	 */
728 	if (error == 0) {
729 		if ((vp = mp->mnt_syncer) != NULL) {
730 			mp->mnt_syncer = NULL;
731 			vrele(vp);
732 		}
733 		if (((mp->mnt_flag & MNT_RDONLY) ||
734 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
735 		    (flags & MNT_FORCE)) {
736 			error = VFS_UNMOUNT(mp, flags);
737 		}
738 	}
739 	if (error) {
740 		if (mp->mnt_syncer == NULL)
741 			vfs_allocate_syncvnode(mp);
742 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
743 		mp->mnt_flag |= async_flag;
744 		lockmgr(&mp->mnt_lock, LK_RELEASE);
745 		if (mp->mnt_kern_flag & MNTK_MWAIT)
746 			wakeup(mp);
747 		return (error);
748 	}
749 	/*
750 	 * Clean up any journals still associated with the mount after
751 	 * filesystem activity has ceased.
752 	 */
753 	journal_remove_all_journals(mp,
754 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
755 
756 	mountlist_remove(mp);
757 
758 	/*
759 	 * Remove any installed vnode ops here so the individual VFSs don't
760 	 * have to.
761 	 */
762 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
763 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
764 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
765 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
766 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
767 
768 	if (mp->mnt_ncmountpt.ncp != NULL) {
769 		nch = mp->mnt_ncmountpt;
770 		cache_zero(&mp->mnt_ncmountpt);
771 		cache_clrmountpt(&nch);
772 		cache_drop(&nch);
773 	}
774 	if (mp->mnt_ncmounton.ncp != NULL) {
775 		nch = mp->mnt_ncmounton;
776 		cache_zero(&mp->mnt_ncmounton);
777 		cache_clrmountpt(&nch);
778 		cache_drop(&nch);
779 	}
780 
781 	mp->mnt_vfc->vfc_refcount--;
782 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
783 		panic("unmount: dangling vnode");
784 	lockmgr(&mp->mnt_lock, LK_RELEASE);
785 	if (mp->mnt_kern_flag & MNTK_MWAIT)
786 		wakeup(mp);
787 	if (freeok)
788 		kfree(mp, M_MOUNT);
789 	return (0);
790 }
791 
792 static
793 void
794 mount_warning(struct mount *mp, const char *ctl, ...)
795 {
796 	char *ptr;
797 	char *buf;
798 	__va_list va;
799 
800 	__va_start(va, ctl);
801 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf, 0) == 0) {
802 		kprintf("unmount(%s): ", ptr);
803 		kvprintf(ctl, va);
804 		kprintf("\n");
805 		kfree(buf, M_TEMP);
806 	} else {
807 		kprintf("unmount(%p", mp);
808 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
809 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
810 		kprintf("): ");
811 		kvprintf(ctl, va);
812 		kprintf("\n");
813 	}
814 	__va_end(va);
815 }
816 
817 /*
818  * Shim cache_fullpath() to handle the case where a process is chrooted into
819  * a subdirectory of a mount.  In this case if the root mount matches the
820  * process root directory's mount we have to specify the process's root
821  * directory instead of the mount point, because the mount point might
822  * be above the root directory.
823  */
824 static
825 int
826 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
827 {
828 	struct nchandle *nch;
829 
830 	if (p && p->p_fd->fd_nrdir.mount == mp)
831 		nch = &p->p_fd->fd_nrdir;
832 	else
833 		nch = &mp->mnt_ncmountpt;
834 	return(cache_fullpath(p, nch, rb, fb, 0));
835 }
836 
837 /*
838  * Sync each mounted filesystem.
839  */
840 
841 #ifdef DEBUG
842 static int syncprt = 0;
843 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
844 #endif /* DEBUG */
845 
846 static int sync_callback(struct mount *mp, void *data);
847 
848 /*
849  * MPALMOSTSAFE
850  */
851 int
852 sys_sync(struct sync_args *uap)
853 {
854 	get_mplock();
855 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
856 #ifdef DEBUG
857 	/*
858 	 * print out buffer pool stat information on each sync() call.
859 	 */
860 	if (syncprt)
861 		vfs_bufstats();
862 #endif /* DEBUG */
863 	rel_mplock();
864 	return (0);
865 }
866 
867 static
868 int
869 sync_callback(struct mount *mp, void *data __unused)
870 {
871 	int asyncflag;
872 
873 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
874 		asyncflag = mp->mnt_flag & MNT_ASYNC;
875 		mp->mnt_flag &= ~MNT_ASYNC;
876 		vfs_msync(mp, MNT_NOWAIT);
877 		VFS_SYNC(mp, MNT_NOWAIT);
878 		mp->mnt_flag |= asyncflag;
879 	}
880 	return(0);
881 }
882 
883 /* XXX PRISON: could be per prison flag */
884 static int prison_quotas;
885 #if 0
886 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
887 #endif
888 
889 /*
890  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
891  *
892  * Change filesystem quotas.
893  *
894  * MPALMOSTSAFE
895  */
896 int
897 sys_quotactl(struct quotactl_args *uap)
898 {
899 	struct nlookupdata nd;
900 	struct thread *td;
901 	struct proc *p;
902 	struct mount *mp;
903 	int error;
904 
905 	get_mplock();
906 	td = curthread;
907 	p = td->td_proc;
908 	if (td->td_ucred->cr_prison && !prison_quotas) {
909 		error = EPERM;
910 		goto done;
911 	}
912 
913 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
914 	if (error == 0)
915 		error = nlookup(&nd);
916 	if (error == 0) {
917 		mp = nd.nl_nch.mount;
918 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
919 				    uap->arg, nd.nl_cred);
920 	}
921 	nlookup_done(&nd);
922 done:
923 	rel_mplock();
924 	return (error);
925 }
926 
927 /*
928  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
929  *		void *buf, int buflen)
930  *
931  * This function operates on a mount point and executes the specified
932  * operation using the specified control data, and possibly returns data.
933  *
934  * The actual number of bytes stored in the result buffer is returned, 0
935  * if none, otherwise an error is returned.
936  *
937  * MPALMOSTSAFE
938  */
939 int
940 sys_mountctl(struct mountctl_args *uap)
941 {
942 	struct thread *td = curthread;
943 	struct proc *p = td->td_proc;
944 	struct file *fp;
945 	void *ctl = NULL;
946 	void *buf = NULL;
947 	char *path = NULL;
948 	int error;
949 
950 	/*
951 	 * Sanity and permissions checks.  We must be root.
952 	 */
953 	KKASSERT(p);
954 	if (td->td_ucred->cr_prison != NULL)
955 		return (EPERM);
956 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
957 	    (error = priv_check(td, PRIV_ROOT)) != 0)
958 		return (error);
959 
960 	/*
961 	 * Argument length checks
962 	 */
963 	if (uap->ctllen < 0 || uap->ctllen > 1024)
964 		return (EINVAL);
965 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
966 		return (EINVAL);
967 	if (uap->path == NULL)
968 		return (EINVAL);
969 
970 	/*
971 	 * Allocate the necessary buffers and copyin data
972 	 */
973 	path = objcache_get(namei_oc, M_WAITOK);
974 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
975 	if (error)
976 		goto done;
977 
978 	if (uap->ctllen) {
979 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
980 		error = copyin(uap->ctl, ctl, uap->ctllen);
981 		if (error)
982 			goto done;
983 	}
984 	if (uap->buflen)
985 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
986 
987 	/*
988 	 * Validate the descriptor
989 	 */
990 	if (uap->fd >= 0) {
991 		fp = holdfp(p->p_fd, uap->fd, -1);
992 		if (fp == NULL) {
993 			error = EBADF;
994 			goto done;
995 		}
996 	} else {
997 		fp = NULL;
998 	}
999 
1000 	/*
1001 	 * Execute the internal kernel function and clean up.
1002 	 */
1003 	get_mplock();
1004 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1005 	rel_mplock();
1006 	if (fp)
1007 		fdrop(fp);
1008 	if (error == 0 && uap->sysmsg_result > 0)
1009 		error = copyout(buf, uap->buf, uap->sysmsg_result);
1010 done:
1011 	if (path)
1012 		objcache_put(namei_oc, path);
1013 	if (ctl)
1014 		kfree(ctl, M_TEMP);
1015 	if (buf)
1016 		kfree(buf, M_TEMP);
1017 	return (error);
1018 }
1019 
1020 /*
1021  * Execute a mount control operation by resolving the path to a mount point
1022  * and calling vop_mountctl().
1023  *
1024  * Use the mount point from the nch instead of the vnode so nullfs mounts
1025  * can properly spike the VOP.
1026  */
1027 int
1028 kern_mountctl(const char *path, int op, struct file *fp,
1029 		const void *ctl, int ctllen,
1030 		void *buf, int buflen, int *res)
1031 {
1032 	struct vnode *vp;
1033 	struct mount *mp;
1034 	struct nlookupdata nd;
1035 	int error;
1036 
1037 	*res = 0;
1038 	vp = NULL;
1039 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1040 	if (error == 0)
1041 		error = nlookup(&nd);
1042 	if (error == 0)
1043 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1044 	mp = nd.nl_nch.mount;
1045 	nlookup_done(&nd);
1046 	if (error)
1047 		return (error);
1048 	vn_unlock(vp);
1049 
1050 	/*
1051 	 * Must be the root of the filesystem
1052 	 */
1053 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1054 		vrele(vp);
1055 		return (EINVAL);
1056 	}
1057 	error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1058 			     buf, buflen, res);
1059 	vrele(vp);
1060 	return (error);
1061 }
1062 
1063 int
1064 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1065 {
1066 	struct thread *td = curthread;
1067 	struct proc *p = td->td_proc;
1068 	struct mount *mp;
1069 	struct statfs *sp;
1070 	char *fullpath, *freepath;
1071 	int error;
1072 
1073 	if ((error = nlookup(nd)) != 0)
1074 		return (error);
1075 	mp = nd->nl_nch.mount;
1076 	sp = &mp->mnt_stat;
1077 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1078 		return (error);
1079 
1080 	error = mount_path(p, mp, &fullpath, &freepath);
1081 	if (error)
1082 		return(error);
1083 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1084 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1085 	kfree(freepath, M_TEMP);
1086 
1087 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1088 	bcopy(sp, buf, sizeof(*buf));
1089 	/* Only root should have access to the fsid's. */
1090 	if (priv_check(td, PRIV_ROOT))
1091 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1092 	return (0);
1093 }
1094 
1095 /*
1096  * statfs_args(char *path, struct statfs *buf)
1097  *
1098  * Get filesystem statistics.
1099  *
1100  * MPALMOSTSAFE
1101  */
1102 int
1103 sys_statfs(struct statfs_args *uap)
1104 {
1105 	struct nlookupdata nd;
1106 	struct statfs buf;
1107 	int error;
1108 
1109 	get_mplock();
1110 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1111 	if (error == 0)
1112 		error = kern_statfs(&nd, &buf);
1113 	nlookup_done(&nd);
1114 	if (error == 0)
1115 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1116 	rel_mplock();
1117 	return (error);
1118 }
1119 
1120 /*
1121  * MPALMOSTSAFE
1122  */
1123 int
1124 kern_fstatfs(int fd, struct statfs *buf)
1125 {
1126 	struct thread *td = curthread;
1127 	struct proc *p = td->td_proc;
1128 	struct file *fp;
1129 	struct mount *mp;
1130 	struct statfs *sp;
1131 	char *fullpath, *freepath;
1132 	int error;
1133 
1134 	KKASSERT(p);
1135 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1136 		return (error);
1137 	get_mplock();
1138 	mp = ((struct vnode *)fp->f_data)->v_mount;
1139 	if (mp == NULL) {
1140 		error = EBADF;
1141 		goto done;
1142 	}
1143 	if (fp->f_cred == NULL) {
1144 		error = EINVAL;
1145 		goto done;
1146 	}
1147 	sp = &mp->mnt_stat;
1148 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1149 		goto done;
1150 
1151 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1152 		goto done;
1153 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1154 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1155 	kfree(freepath, M_TEMP);
1156 
1157 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1158 	bcopy(sp, buf, sizeof(*buf));
1159 
1160 	/* Only root should have access to the fsid's. */
1161 	if (priv_check(td, PRIV_ROOT))
1162 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1163 	error = 0;
1164 done:
1165 	rel_mplock();
1166 	fdrop(fp);
1167 	return (error);
1168 }
1169 
1170 /*
1171  * fstatfs_args(int fd, struct statfs *buf)
1172  *
1173  * Get filesystem statistics.
1174  *
1175  * MPSAFE
1176  */
1177 int
1178 sys_fstatfs(struct fstatfs_args *uap)
1179 {
1180 	struct statfs buf;
1181 	int error;
1182 
1183 	error = kern_fstatfs(uap->fd, &buf);
1184 
1185 	if (error == 0)
1186 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1187 	return (error);
1188 }
1189 
1190 int
1191 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1192 {
1193 	struct mount *mp;
1194 	struct statvfs *sp;
1195 	int error;
1196 
1197 	if ((error = nlookup(nd)) != 0)
1198 		return (error);
1199 	mp = nd->nl_nch.mount;
1200 	sp = &mp->mnt_vstat;
1201 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1202 		return (error);
1203 
1204 	sp->f_flag = 0;
1205 	if (mp->mnt_flag & MNT_RDONLY)
1206 		sp->f_flag |= ST_RDONLY;
1207 	if (mp->mnt_flag & MNT_NOSUID)
1208 		sp->f_flag |= ST_NOSUID;
1209 	bcopy(sp, buf, sizeof(*buf));
1210 	return (0);
1211 }
1212 
1213 /*
1214  * statfs_args(char *path, struct statfs *buf)
1215  *
1216  * Get filesystem statistics.
1217  *
1218  * MPALMOSTSAFE
1219  */
1220 int
1221 sys_statvfs(struct statvfs_args *uap)
1222 {
1223 	struct nlookupdata nd;
1224 	struct statvfs buf;
1225 	int error;
1226 
1227 	get_mplock();
1228 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1229 	if (error == 0)
1230 		error = kern_statvfs(&nd, &buf);
1231 	nlookup_done(&nd);
1232 	if (error == 0)
1233 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1234 	rel_mplock();
1235 	return (error);
1236 }
1237 
1238 int
1239 kern_fstatvfs(int fd, struct statvfs *buf)
1240 {
1241 	struct thread *td = curthread;
1242 	struct proc *p = td->td_proc;
1243 	struct file *fp;
1244 	struct mount *mp;
1245 	struct statvfs *sp;
1246 	int error;
1247 
1248 	KKASSERT(p);
1249 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1250 		return (error);
1251 	mp = ((struct vnode *)fp->f_data)->v_mount;
1252 	if (mp == NULL) {
1253 		error = EBADF;
1254 		goto done;
1255 	}
1256 	if (fp->f_cred == NULL) {
1257 		error = EINVAL;
1258 		goto done;
1259 	}
1260 	sp = &mp->mnt_vstat;
1261 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1262 		goto done;
1263 
1264 	sp->f_flag = 0;
1265 	if (mp->mnt_flag & MNT_RDONLY)
1266 		sp->f_flag |= ST_RDONLY;
1267 	if (mp->mnt_flag & MNT_NOSUID)
1268 		sp->f_flag |= ST_NOSUID;
1269 
1270 	bcopy(sp, buf, sizeof(*buf));
1271 	error = 0;
1272 done:
1273 	fdrop(fp);
1274 	return (error);
1275 }
1276 
1277 /*
1278  * fstatfs_args(int fd, struct statfs *buf)
1279  *
1280  * Get filesystem statistics.
1281  *
1282  * MPALMOSTSAFE
1283  */
1284 int
1285 sys_fstatvfs(struct fstatvfs_args *uap)
1286 {
1287 	struct statvfs buf;
1288 	int error;
1289 
1290 	get_mplock();
1291 	error = kern_fstatvfs(uap->fd, &buf);
1292 	rel_mplock();
1293 
1294 	if (error == 0)
1295 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1296 	return (error);
1297 }
1298 
1299 /*
1300  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1301  *
1302  * Get statistics on all filesystems.
1303  */
1304 
1305 struct getfsstat_info {
1306 	struct statfs *sfsp;
1307 	long count;
1308 	long maxcount;
1309 	int error;
1310 	int flags;
1311 	struct thread *td;
1312 };
1313 
1314 static int getfsstat_callback(struct mount *, void *);
1315 
1316 /*
1317  * MPALMOSTSAFE
1318  */
1319 int
1320 sys_getfsstat(struct getfsstat_args *uap)
1321 {
1322 	struct thread *td = curthread;
1323 	struct getfsstat_info info;
1324 
1325 	bzero(&info, sizeof(info));
1326 
1327 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1328 	info.sfsp = uap->buf;
1329 	info.count = 0;
1330 	info.flags = uap->flags;
1331 	info.td = td;
1332 
1333 	get_mplock();
1334 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1335 	rel_mplock();
1336 	if (info.sfsp && info.count > info.maxcount)
1337 		uap->sysmsg_result = info.maxcount;
1338 	else
1339 		uap->sysmsg_result = info.count;
1340 	return (info.error);
1341 }
1342 
1343 static int
1344 getfsstat_callback(struct mount *mp, void *data)
1345 {
1346 	struct getfsstat_info *info = data;
1347 	struct statfs *sp;
1348 	char *freepath;
1349 	char *fullpath;
1350 	int error;
1351 
1352 	if (info->sfsp && info->count < info->maxcount) {
1353 		if (info->td->td_proc &&
1354 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1355 			return(0);
1356 		}
1357 		sp = &mp->mnt_stat;
1358 
1359 		/*
1360 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1361 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1362 		 * overrides MNT_WAIT.
1363 		 */
1364 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1365 		    (info->flags & MNT_WAIT)) &&
1366 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1367 			return(0);
1368 		}
1369 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1370 
1371 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1372 		if (error) {
1373 			info->error = error;
1374 			return(-1);
1375 		}
1376 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1377 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1378 		kfree(freepath, M_TEMP);
1379 
1380 		error = copyout(sp, info->sfsp, sizeof(*sp));
1381 		if (error) {
1382 			info->error = error;
1383 			return (-1);
1384 		}
1385 		++info->sfsp;
1386 	}
1387 	info->count++;
1388 	return(0);
1389 }
1390 
1391 /*
1392  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1393 		   long bufsize, int flags)
1394  *
1395  * Get statistics on all filesystems.
1396  */
1397 
1398 struct getvfsstat_info {
1399 	struct statfs *sfsp;
1400 	struct statvfs *vsfsp;
1401 	long count;
1402 	long maxcount;
1403 	int error;
1404 	int flags;
1405 	struct thread *td;
1406 };
1407 
1408 static int getvfsstat_callback(struct mount *, void *);
1409 
1410 /*
1411  * MPALMOSTSAFE
1412  */
1413 int
1414 sys_getvfsstat(struct getvfsstat_args *uap)
1415 {
1416 	struct thread *td = curthread;
1417 	struct getvfsstat_info info;
1418 
1419 	bzero(&info, sizeof(info));
1420 
1421 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1422 	info.sfsp = uap->buf;
1423 	info.vsfsp = uap->vbuf;
1424 	info.count = 0;
1425 	info.flags = uap->flags;
1426 	info.td = td;
1427 
1428 	get_mplock();
1429 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1430 	if (info.vsfsp && info.count > info.maxcount)
1431 		uap->sysmsg_result = info.maxcount;
1432 	else
1433 		uap->sysmsg_result = info.count;
1434 	rel_mplock();
1435 	return (info.error);
1436 }
1437 
1438 static int
1439 getvfsstat_callback(struct mount *mp, void *data)
1440 {
1441 	struct getvfsstat_info *info = data;
1442 	struct statfs *sp;
1443 	struct statvfs *vsp;
1444 	char *freepath;
1445 	char *fullpath;
1446 	int error;
1447 
1448 	if (info->vsfsp && info->count < info->maxcount) {
1449 		if (info->td->td_proc &&
1450 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1451 			return(0);
1452 		}
1453 		sp = &mp->mnt_stat;
1454 		vsp = &mp->mnt_vstat;
1455 
1456 		/*
1457 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1458 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1459 		 * overrides MNT_WAIT.
1460 		 */
1461 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1462 		    (info->flags & MNT_WAIT)) &&
1463 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1464 			return(0);
1465 		}
1466 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1467 
1468 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1469 		    (info->flags & MNT_WAIT)) &&
1470 		    (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1471 			return(0);
1472 		}
1473 		vsp->f_flag = 0;
1474 		if (mp->mnt_flag & MNT_RDONLY)
1475 			vsp->f_flag |= ST_RDONLY;
1476 		if (mp->mnt_flag & MNT_NOSUID)
1477 			vsp->f_flag |= ST_NOSUID;
1478 
1479 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1480 		if (error) {
1481 			info->error = error;
1482 			return(-1);
1483 		}
1484 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1485 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1486 		kfree(freepath, M_TEMP);
1487 
1488 		error = copyout(sp, info->sfsp, sizeof(*sp));
1489 		if (error == 0)
1490 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1491 		if (error) {
1492 			info->error = error;
1493 			return (-1);
1494 		}
1495 		++info->sfsp;
1496 		++info->vsfsp;
1497 	}
1498 	info->count++;
1499 	return(0);
1500 }
1501 
1502 
1503 /*
1504  * fchdir_args(int fd)
1505  *
1506  * Change current working directory to a given file descriptor.
1507  *
1508  * MPALMOSTSAFE
1509  */
1510 int
1511 sys_fchdir(struct fchdir_args *uap)
1512 {
1513 	struct thread *td = curthread;
1514 	struct proc *p = td->td_proc;
1515 	struct filedesc *fdp = p->p_fd;
1516 	struct vnode *vp, *ovp;
1517 	struct mount *mp;
1518 	struct file *fp;
1519 	struct nchandle nch, onch, tnch;
1520 	int error;
1521 
1522 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1523 		return (error);
1524 	get_mplock();
1525 	vp = (struct vnode *)fp->f_data;
1526 	vref(vp);
1527 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1528 	if (fp->f_nchandle.ncp == NULL)
1529 		error = ENOTDIR;
1530 	else
1531 		error = checkvp_chdir(vp, td);
1532 	if (error) {
1533 		vput(vp);
1534 		goto done;
1535 	}
1536 	cache_copy(&fp->f_nchandle, &nch);
1537 
1538 	/*
1539 	 * If the ncp has become a mount point, traverse through
1540 	 * the mount point.
1541 	 */
1542 
1543 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1544 	       (mp = cache_findmount(&nch)) != NULL
1545 	) {
1546 		error = nlookup_mp(mp, &tnch);
1547 		if (error == 0) {
1548 			cache_unlock(&tnch);	/* leave ref intact */
1549 			vput(vp);
1550 			vp = tnch.ncp->nc_vp;
1551 			error = vget(vp, LK_SHARED);
1552 			KKASSERT(error == 0);
1553 			cache_drop(&nch);
1554 			nch = tnch;
1555 		}
1556 	}
1557 	if (error == 0) {
1558 		ovp = fdp->fd_cdir;
1559 		onch = fdp->fd_ncdir;
1560 		vn_unlock(vp);		/* leave ref intact */
1561 		fdp->fd_cdir = vp;
1562 		fdp->fd_ncdir = nch;
1563 		cache_drop(&onch);
1564 		vrele(ovp);
1565 	} else {
1566 		cache_drop(&nch);
1567 		vput(vp);
1568 	}
1569 	fdrop(fp);
1570 done:
1571 	rel_mplock();
1572 	return (error);
1573 }
1574 
1575 int
1576 kern_chdir(struct nlookupdata *nd)
1577 {
1578 	struct thread *td = curthread;
1579 	struct proc *p = td->td_proc;
1580 	struct filedesc *fdp = p->p_fd;
1581 	struct vnode *vp, *ovp;
1582 	struct nchandle onch;
1583 	int error;
1584 
1585 	if ((error = nlookup(nd)) != 0)
1586 		return (error);
1587 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1588 		return (ENOENT);
1589 	if ((error = vget(vp, LK_SHARED)) != 0)
1590 		return (error);
1591 
1592 	error = checkvp_chdir(vp, td);
1593 	vn_unlock(vp);
1594 	if (error == 0) {
1595 		ovp = fdp->fd_cdir;
1596 		onch = fdp->fd_ncdir;
1597 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1598 		fdp->fd_ncdir = nd->nl_nch;
1599 		fdp->fd_cdir = vp;
1600 		cache_drop(&onch);
1601 		vrele(ovp);
1602 		cache_zero(&nd->nl_nch);
1603 	} else {
1604 		vrele(vp);
1605 	}
1606 	return (error);
1607 }
1608 
1609 /*
1610  * chdir_args(char *path)
1611  *
1612  * Change current working directory (``.'').
1613  *
1614  * MPALMOSTSAFE
1615  */
1616 int
1617 sys_chdir(struct chdir_args *uap)
1618 {
1619 	struct nlookupdata nd;
1620 	int error;
1621 
1622 	get_mplock();
1623 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1624 	if (error == 0)
1625 		error = kern_chdir(&nd);
1626 	nlookup_done(&nd);
1627 	rel_mplock();
1628 	return (error);
1629 }
1630 
1631 /*
1632  * Helper function for raised chroot(2) security function:  Refuse if
1633  * any filedescriptors are open directories.
1634  */
1635 static int
1636 chroot_refuse_vdir_fds(struct filedesc *fdp)
1637 {
1638 	struct vnode *vp;
1639 	struct file *fp;
1640 	int error;
1641 	int fd;
1642 
1643 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1644 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1645 			continue;
1646 		vp = (struct vnode *)fp->f_data;
1647 		if (vp->v_type != VDIR) {
1648 			fdrop(fp);
1649 			continue;
1650 		}
1651 		fdrop(fp);
1652 		return(EPERM);
1653 	}
1654 	return (0);
1655 }
1656 
1657 /*
1658  * This sysctl determines if we will allow a process to chroot(2) if it
1659  * has a directory open:
1660  *	0: disallowed for all processes.
1661  *	1: allowed for processes that were not already chroot(2)'ed.
1662  *	2: allowed for all processes.
1663  */
1664 
1665 static int chroot_allow_open_directories = 1;
1666 
1667 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1668      &chroot_allow_open_directories, 0, "");
1669 
1670 /*
1671  * chroot to the specified namecache entry.  We obtain the vp from the
1672  * namecache data.  The passed ncp must be locked and referenced and will
1673  * remain locked and referenced on return.
1674  */
1675 int
1676 kern_chroot(struct nchandle *nch)
1677 {
1678 	struct thread *td = curthread;
1679 	struct proc *p = td->td_proc;
1680 	struct filedesc *fdp = p->p_fd;
1681 	struct vnode *vp;
1682 	int error;
1683 
1684 	/*
1685 	 * Only privileged user can chroot
1686 	 */
1687 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1688 	if (error)
1689 		return (error);
1690 
1691 	/*
1692 	 * Disallow open directory descriptors (fchdir() breakouts).
1693 	 */
1694 	if (chroot_allow_open_directories == 0 ||
1695 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1696 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1697 			return (error);
1698 	}
1699 	if ((vp = nch->ncp->nc_vp) == NULL)
1700 		return (ENOENT);
1701 
1702 	if ((error = vget(vp, LK_SHARED)) != 0)
1703 		return (error);
1704 
1705 	/*
1706 	 * Check the validity of vp as a directory to change to and
1707 	 * associate it with rdir/jdir.
1708 	 */
1709 	error = checkvp_chdir(vp, td);
1710 	vn_unlock(vp);			/* leave reference intact */
1711 	if (error == 0) {
1712 		vrele(fdp->fd_rdir);
1713 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1714 		cache_drop(&fdp->fd_nrdir);
1715 		cache_copy(nch, &fdp->fd_nrdir);
1716 		if (fdp->fd_jdir == NULL) {
1717 			fdp->fd_jdir = vp;
1718 			vref(fdp->fd_jdir);
1719 			cache_copy(nch, &fdp->fd_njdir);
1720 		}
1721 	} else {
1722 		vrele(vp);
1723 	}
1724 	return (error);
1725 }
1726 
1727 /*
1728  * chroot_args(char *path)
1729  *
1730  * Change notion of root (``/'') directory.
1731  *
1732  * MPALMOSTSAFE
1733  */
1734 int
1735 sys_chroot(struct chroot_args *uap)
1736 {
1737 	struct thread *td __debugvar = curthread;
1738 	struct nlookupdata nd;
1739 	int error;
1740 
1741 	KKASSERT(td->td_proc);
1742 	get_mplock();
1743 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1744 	if (error == 0) {
1745 		nd.nl_flags |= NLC_EXEC;
1746 		error = nlookup(&nd);
1747 		if (error == 0)
1748 			error = kern_chroot(&nd.nl_nch);
1749 	}
1750 	nlookup_done(&nd);
1751 	rel_mplock();
1752 	return(error);
1753 }
1754 
1755 int
1756 sys_chroot_kernel(struct chroot_kernel_args *uap)
1757 {
1758 	struct thread *td = curthread;
1759 	struct nlookupdata nd;
1760 	struct nchandle *nch;
1761 	struct vnode *vp;
1762 	int error;
1763 
1764 	get_mplock();
1765 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1766 	if (error)
1767 		goto error_nond;
1768 
1769 	error = nlookup(&nd);
1770 	if (error)
1771 		goto error_out;
1772 
1773 	nch = &nd.nl_nch;
1774 
1775 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1776 	if (error)
1777 		goto error_out;
1778 
1779 	if ((vp = nch->ncp->nc_vp) == NULL) {
1780 		error = ENOENT;
1781 		goto error_out;
1782 	}
1783 
1784 	if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1785 		goto error_out;
1786 
1787 	kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1788 	vfs_cache_setroot(vp, cache_hold(nch));
1789 
1790 error_out:
1791 	nlookup_done(&nd);
1792 error_nond:
1793 	rel_mplock();
1794 	return(error);
1795 }
1796 
1797 /*
1798  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1799  * determine whether it is legal to chdir to the vnode.  The vnode's state
1800  * is not changed by this call.
1801  */
1802 int
1803 checkvp_chdir(struct vnode *vp, struct thread *td)
1804 {
1805 	int error;
1806 
1807 	if (vp->v_type != VDIR)
1808 		error = ENOTDIR;
1809 	else
1810 		error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1811 	return (error);
1812 }
1813 
1814 /*
1815  * MPSAFE
1816  */
1817 int
1818 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1819 {
1820 	struct thread *td = curthread;
1821 	struct proc *p = td->td_proc;
1822 	struct lwp *lp = td->td_lwp;
1823 	struct filedesc *fdp = p->p_fd;
1824 	int cmode, flags;
1825 	struct file *nfp;
1826 	struct file *fp;
1827 	struct vnode *vp;
1828 	int type, indx, error;
1829 	struct flock lf;
1830 
1831 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1832 		return (EINVAL);
1833 	flags = FFLAGS(oflags);
1834 	error = falloc(lp, &nfp, NULL);
1835 	if (error)
1836 		return (error);
1837 	fp = nfp;
1838 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1839 
1840 	/*
1841 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1842 	 * file descriptor to be duplicated rather then doing the open
1843 	 * itself.
1844 	 */
1845 	lp->lwp_dupfd = -1;
1846 
1847 	/*
1848 	 * Call vn_open() to do the lookup and assign the vnode to the
1849 	 * file pointer.  vn_open() does not change the ref count on fp
1850 	 * and the vnode, on success, will be inherited by the file pointer
1851 	 * and unlocked.
1852 	 */
1853 	nd->nl_flags |= NLC_LOCKVP;
1854 	error = vn_open(nd, fp, flags, cmode);
1855 	nlookup_done(nd);
1856 	if (error) {
1857 		/*
1858 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1859 		 * responsible for dropping the old contents of ofiles[indx]
1860 		 * if it succeeds.
1861 		 *
1862 		 * Note that fsetfd() will add a ref to fp which represents
1863 		 * the fd_files[] assignment.  We must still drop our
1864 		 * reference.
1865 		 */
1866 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1867 			if (fdalloc(p, 0, &indx) == 0) {
1868 				error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1869 				if (error == 0) {
1870 					*res = indx;
1871 					fdrop(fp);	/* our ref */
1872 					return (0);
1873 				}
1874 				fsetfd(fdp, NULL, indx);
1875 			}
1876 		}
1877 		fdrop(fp);	/* our ref */
1878 		if (error == ERESTART)
1879 			error = EINTR;
1880 		return (error);
1881 	}
1882 
1883 	/*
1884 	 * ref the vnode for ourselves so it can't be ripped out from under
1885 	 * is.  XXX need an ND flag to request that the vnode be returned
1886 	 * anyway.
1887 	 *
1888 	 * Reserve a file descriptor but do not assign it until the open
1889 	 * succeeds.
1890 	 */
1891 	vp = (struct vnode *)fp->f_data;
1892 	vref(vp);
1893 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1894 		fdrop(fp);
1895 		vrele(vp);
1896 		return (error);
1897 	}
1898 
1899 	/*
1900 	 * If no error occurs the vp will have been assigned to the file
1901 	 * pointer.
1902 	 */
1903 	lp->lwp_dupfd = 0;
1904 
1905 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1906 		lf.l_whence = SEEK_SET;
1907 		lf.l_start = 0;
1908 		lf.l_len = 0;
1909 		if (flags & O_EXLOCK)
1910 			lf.l_type = F_WRLCK;
1911 		else
1912 			lf.l_type = F_RDLCK;
1913 		if (flags & FNONBLOCK)
1914 			type = 0;
1915 		else
1916 			type = F_WAIT;
1917 
1918 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1919 			/*
1920 			 * lock request failed.  Clean up the reserved
1921 			 * descriptor.
1922 			 */
1923 			vrele(vp);
1924 			fsetfd(fdp, NULL, indx);
1925 			fdrop(fp);
1926 			return (error);
1927 		}
1928 		fp->f_flag |= FHASLOCK;
1929 	}
1930 #if 0
1931 	/*
1932 	 * Assert that all regular file vnodes were created with a object.
1933 	 */
1934 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1935 		("open: regular file has no backing object after vn_open"));
1936 #endif
1937 
1938 	vrele(vp);
1939 
1940 	/*
1941 	 * release our private reference, leaving the one associated with the
1942 	 * descriptor table intact.
1943 	 */
1944 	fsetfd(fdp, fp, indx);
1945 	fdrop(fp);
1946 	*res = indx;
1947 	return (0);
1948 }
1949 
1950 /*
1951  * open_args(char *path, int flags, int mode)
1952  *
1953  * Check permissions, allocate an open file structure,
1954  * and call the device open routine if any.
1955  *
1956  * MPALMOSTSAFE
1957  */
1958 int
1959 sys_open(struct open_args *uap)
1960 {
1961 	CACHE_MPLOCK_DECLARE;
1962 	struct nlookupdata nd;
1963 	int error;
1964 
1965 	CACHE_GETMPLOCK1();
1966 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1967 	if (error == 0) {
1968 		error = kern_open(&nd, uap->flags,
1969 				    uap->mode, &uap->sysmsg_result);
1970 	}
1971 	nlookup_done(&nd);
1972 	CACHE_RELMPLOCK();
1973 	return (error);
1974 }
1975 
1976 /*
1977  * openat_args(int fd, char *path, int flags, int mode)
1978  *
1979  * MPALMOSTSAFE
1980  */
1981 int
1982 sys_openat(struct openat_args *uap)
1983 {
1984 	CACHE_MPLOCK_DECLARE;
1985 	struct nlookupdata nd;
1986 	int error;
1987 	struct file *fp;
1988 
1989 	CACHE_GETMPLOCK1();
1990 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
1991 	if (error == 0) {
1992 		error = kern_open(&nd, uap->flags, uap->mode,
1993 					&uap->sysmsg_result);
1994 	}
1995 	nlookup_done_at(&nd, fp);
1996 	CACHE_RELMPLOCK();
1997 	return (error);
1998 }
1999 
2000 int
2001 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2002 {
2003 	struct thread *td = curthread;
2004 	struct proc *p = td->td_proc;
2005 	struct vnode *vp;
2006 	struct vattr vattr;
2007 	int error;
2008 	int whiteout = 0;
2009 
2010 	KKASSERT(p);
2011 
2012 	VATTR_NULL(&vattr);
2013 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2014 	vattr.va_rmajor = rmajor;
2015 	vattr.va_rminor = rminor;
2016 
2017 	switch (mode & S_IFMT) {
2018 	case S_IFMT:	/* used by badsect to flag bad sectors */
2019 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2020 		vattr.va_type = VBAD;
2021 		break;
2022 	case S_IFCHR:
2023 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2024 		vattr.va_type = VCHR;
2025 		break;
2026 	case S_IFBLK:
2027 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2028 		vattr.va_type = VBLK;
2029 		break;
2030 	case S_IFWHT:
2031 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2032 		whiteout = 1;
2033 		break;
2034 	case S_IFDIR:	/* special directories support for HAMMER */
2035 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2036 		vattr.va_type = VDIR;
2037 		break;
2038 	default:
2039 		error = EINVAL;
2040 		break;
2041 	}
2042 
2043 	if (error)
2044 		return (error);
2045 
2046 	bwillinode(1);
2047 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2048 	if ((error = nlookup(nd)) != 0)
2049 		return (error);
2050 	if (nd->nl_nch.ncp->nc_vp)
2051 		return (EEXIST);
2052 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2053 		return (error);
2054 
2055 	if (whiteout) {
2056 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2057 				      nd->nl_cred, NAMEI_CREATE);
2058 	} else {
2059 		vp = NULL;
2060 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2061 				   &vp, nd->nl_cred, &vattr);
2062 		if (error == 0)
2063 			vput(vp);
2064 	}
2065 	return (error);
2066 }
2067 
2068 /*
2069  * mknod_args(char *path, int mode, int dev)
2070  *
2071  * Create a special file.
2072  *
2073  * MPALMOSTSAFE
2074  */
2075 int
2076 sys_mknod(struct mknod_args *uap)
2077 {
2078 	struct nlookupdata nd;
2079 	int error;
2080 
2081 	get_mplock();
2082 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2083 	if (error == 0) {
2084 		error = kern_mknod(&nd, uap->mode,
2085 				   umajor(uap->dev), uminor(uap->dev));
2086 	}
2087 	nlookup_done(&nd);
2088 	rel_mplock();
2089 	return (error);
2090 }
2091 
2092 int
2093 kern_mkfifo(struct nlookupdata *nd, int mode)
2094 {
2095 	struct thread *td = curthread;
2096 	struct proc *p = td->td_proc;
2097 	struct vattr vattr;
2098 	struct vnode *vp;
2099 	int error;
2100 
2101 	bwillinode(1);
2102 
2103 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2104 	if ((error = nlookup(nd)) != 0)
2105 		return (error);
2106 	if (nd->nl_nch.ncp->nc_vp)
2107 		return (EEXIST);
2108 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2109 		return (error);
2110 
2111 	VATTR_NULL(&vattr);
2112 	vattr.va_type = VFIFO;
2113 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2114 	vp = NULL;
2115 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2116 	if (error == 0)
2117 		vput(vp);
2118 	return (error);
2119 }
2120 
2121 /*
2122  * mkfifo_args(char *path, int mode)
2123  *
2124  * Create a named pipe.
2125  *
2126  * MPALMOSTSAFE
2127  */
2128 int
2129 sys_mkfifo(struct mkfifo_args *uap)
2130 {
2131 	struct nlookupdata nd;
2132 	int error;
2133 
2134 	get_mplock();
2135 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2136 	if (error == 0)
2137 		error = kern_mkfifo(&nd, uap->mode);
2138 	nlookup_done(&nd);
2139 	rel_mplock();
2140 	return (error);
2141 }
2142 
2143 static int hardlink_check_uid = 0;
2144 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2145     &hardlink_check_uid, 0,
2146     "Unprivileged processes cannot create hard links to files owned by other "
2147     "users");
2148 static int hardlink_check_gid = 0;
2149 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2150     &hardlink_check_gid, 0,
2151     "Unprivileged processes cannot create hard links to files owned by other "
2152     "groups");
2153 
2154 static int
2155 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2156 {
2157 	struct vattr va;
2158 	int error;
2159 
2160 	/*
2161 	 * Shortcut if disabled
2162 	 */
2163 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2164 		return (0);
2165 
2166 	/*
2167 	 * Privileged user can always hardlink
2168 	 */
2169 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2170 		return (0);
2171 
2172 	/*
2173 	 * Otherwise only if the originating file is owned by the
2174 	 * same user or group.  Note that any group is allowed if
2175 	 * the file is owned by the caller.
2176 	 */
2177 	error = VOP_GETATTR(vp, &va);
2178 	if (error != 0)
2179 		return (error);
2180 
2181 	if (hardlink_check_uid) {
2182 		if (cred->cr_uid != va.va_uid)
2183 			return (EPERM);
2184 	}
2185 
2186 	if (hardlink_check_gid) {
2187 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2188 			return (EPERM);
2189 	}
2190 
2191 	return (0);
2192 }
2193 
2194 int
2195 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2196 {
2197 	struct thread *td = curthread;
2198 	struct vnode *vp;
2199 	int error;
2200 
2201 	/*
2202 	 * Lookup the source and obtained a locked vnode.
2203 	 *
2204 	 * You may only hardlink a file which you have write permission
2205 	 * on or which you own.
2206 	 *
2207 	 * XXX relookup on vget failure / race ?
2208 	 */
2209 	bwillinode(1);
2210 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2211 	if ((error = nlookup(nd)) != 0)
2212 		return (error);
2213 	vp = nd->nl_nch.ncp->nc_vp;
2214 	KKASSERT(vp != NULL);
2215 	if (vp->v_type == VDIR)
2216 		return (EPERM);		/* POSIX */
2217 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2218 		return (error);
2219 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2220 		return (error);
2221 
2222 	/*
2223 	 * Unlock the source so we can lookup the target without deadlocking
2224 	 * (XXX vp is locked already, possible other deadlock?).  The target
2225 	 * must not exist.
2226 	 */
2227 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2228 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2229 	cache_unlock(&nd->nl_nch);
2230 	vn_unlock(vp);
2231 
2232 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2233 	if ((error = nlookup(linknd)) != 0) {
2234 		vrele(vp);
2235 		return (error);
2236 	}
2237 	if (linknd->nl_nch.ncp->nc_vp) {
2238 		vrele(vp);
2239 		return (EEXIST);
2240 	}
2241 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2242 		vrele(vp);
2243 		return (error);
2244 	}
2245 
2246 	/*
2247 	 * Finally run the new API VOP.
2248 	 */
2249 	error = can_hardlink(vp, td, td->td_ucred);
2250 	if (error == 0) {
2251 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2252 				  vp, linknd->nl_cred);
2253 	}
2254 	vput(vp);
2255 	return (error);
2256 }
2257 
2258 /*
2259  * link_args(char *path, char *link)
2260  *
2261  * Make a hard file link.
2262  *
2263  * MPALMOSTSAFE
2264  */
2265 int
2266 sys_link(struct link_args *uap)
2267 {
2268 	struct nlookupdata nd, linknd;
2269 	int error;
2270 
2271 	get_mplock();
2272 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2273 	if (error == 0) {
2274 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2275 		if (error == 0)
2276 			error = kern_link(&nd, &linknd);
2277 		nlookup_done(&linknd);
2278 	}
2279 	nlookup_done(&nd);
2280 	rel_mplock();
2281 	return (error);
2282 }
2283 
2284 int
2285 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2286 {
2287 	struct vattr vattr;
2288 	struct vnode *vp;
2289 	struct vnode *dvp;
2290 	int error;
2291 
2292 	bwillinode(1);
2293 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2294 	if ((error = nlookup(nd)) != 0)
2295 		return (error);
2296 	if (nd->nl_nch.ncp->nc_vp)
2297 		return (EEXIST);
2298 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2299 		return (error);
2300 	dvp = nd->nl_dvp;
2301 	VATTR_NULL(&vattr);
2302 	vattr.va_mode = mode;
2303 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2304 	if (error == 0)
2305 		vput(vp);
2306 	return (error);
2307 }
2308 
2309 /*
2310  * symlink(char *path, char *link)
2311  *
2312  * Make a symbolic link.
2313  *
2314  * MPALMOSTSAFE
2315  */
2316 int
2317 sys_symlink(struct symlink_args *uap)
2318 {
2319 	struct thread *td = curthread;
2320 	struct nlookupdata nd;
2321 	char *path;
2322 	int error;
2323 	int mode;
2324 
2325 	path = objcache_get(namei_oc, M_WAITOK);
2326 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2327 	if (error == 0) {
2328 		get_mplock();
2329 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2330 		if (error == 0) {
2331 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2332 			error = kern_symlink(&nd, path, mode);
2333 		}
2334 		nlookup_done(&nd);
2335 		rel_mplock();
2336 	}
2337 	objcache_put(namei_oc, path);
2338 	return (error);
2339 }
2340 
2341 /*
2342  * undelete_args(char *path)
2343  *
2344  * Delete a whiteout from the filesystem.
2345  *
2346  * MPALMOSTSAFE
2347  */
2348 int
2349 sys_undelete(struct undelete_args *uap)
2350 {
2351 	struct nlookupdata nd;
2352 	int error;
2353 
2354 	get_mplock();
2355 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2356 	bwillinode(1);
2357 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2358 	if (error == 0)
2359 		error = nlookup(&nd);
2360 	if (error == 0)
2361 		error = ncp_writechk(&nd.nl_nch);
2362 	if (error == 0) {
2363 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2364 				      NAMEI_DELETE);
2365 	}
2366 	nlookup_done(&nd);
2367 	rel_mplock();
2368 	return (error);
2369 }
2370 
2371 int
2372 kern_unlink(struct nlookupdata *nd)
2373 {
2374 	int error;
2375 
2376 	bwillinode(1);
2377 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2378 	if ((error = nlookup(nd)) != 0)
2379 		return (error);
2380 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2381 		return (error);
2382 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2383 	return (error);
2384 }
2385 
2386 /*
2387  * unlink_args(char *path)
2388  *
2389  * Delete a name from the filesystem.
2390  *
2391  * MPALMOSTSAFE
2392  */
2393 int
2394 sys_unlink(struct unlink_args *uap)
2395 {
2396 	struct nlookupdata nd;
2397 	int error;
2398 
2399 	get_mplock();
2400 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2401 	if (error == 0)
2402 		error = kern_unlink(&nd);
2403 	nlookup_done(&nd);
2404 	rel_mplock();
2405 	return (error);
2406 }
2407 
2408 
2409 /*
2410  * unlinkat_args(int fd, char *path, int flags)
2411  *
2412  * Delete the file or directory entry pointed to by fd/path.
2413  *
2414  * MPALMOSTSAFE
2415  */
2416 int
2417 sys_unlinkat(struct unlinkat_args *uap)
2418 {
2419 	struct nlookupdata nd;
2420 	struct file *fp;
2421 	int error;
2422 
2423 	if (uap->flags & ~AT_REMOVEDIR)
2424 		return (EINVAL);
2425 
2426 	get_mplock();
2427 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2428 	if (error == 0) {
2429 		if (uap->flags & AT_REMOVEDIR)
2430 			error = kern_rmdir(&nd);
2431 		else
2432 			error = kern_unlink(&nd);
2433 	}
2434 	nlookup_done_at(&nd, fp);
2435 	rel_mplock();
2436 	return (error);
2437 }
2438 
2439 /*
2440  * MPALMOSTSAFE
2441  */
2442 int
2443 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2444 {
2445 	struct thread *td = curthread;
2446 	struct proc *p = td->td_proc;
2447 	struct file *fp;
2448 	struct vnode *vp;
2449 	struct vattr vattr;
2450 	off_t new_offset;
2451 	int error;
2452 
2453 	fp = holdfp(p->p_fd, fd, -1);
2454 	if (fp == NULL)
2455 		return (EBADF);
2456 	if (fp->f_type != DTYPE_VNODE) {
2457 		error = ESPIPE;
2458 		goto done;
2459 	}
2460 	vp = (struct vnode *)fp->f_data;
2461 
2462 	switch (whence) {
2463 	case L_INCR:
2464 		spin_lock_wr(&fp->f_spin);
2465 		new_offset = fp->f_offset + offset;
2466 		error = 0;
2467 		break;
2468 	case L_XTND:
2469 		get_mplock();
2470 		error = VOP_GETATTR(vp, &vattr);
2471 		rel_mplock();
2472 		spin_lock_wr(&fp->f_spin);
2473 		new_offset = offset + vattr.va_size;
2474 		break;
2475 	case L_SET:
2476 		new_offset = offset;
2477 		error = 0;
2478 		spin_lock_wr(&fp->f_spin);
2479 		break;
2480 	default:
2481 		new_offset = 0;
2482 		error = EINVAL;
2483 		spin_lock_wr(&fp->f_spin);
2484 		break;
2485 	}
2486 
2487 	/*
2488 	 * Validate the seek position.  Negative offsets are not allowed
2489 	 * for regular files or directories.
2490 	 *
2491 	 * Normally we would also not want to allow negative offsets for
2492 	 * character and block-special devices.  However kvm addresses
2493 	 * on 64 bit architectures might appear to be negative and must
2494 	 * be allowed.
2495 	 */
2496 	if (error == 0) {
2497 		if (new_offset < 0 &&
2498 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2499 			error = EINVAL;
2500 		} else {
2501 			fp->f_offset = new_offset;
2502 		}
2503 	}
2504 	*res = fp->f_offset;
2505 	spin_unlock_wr(&fp->f_spin);
2506 done:
2507 	fdrop(fp);
2508 	return (error);
2509 }
2510 
2511 /*
2512  * lseek_args(int fd, int pad, off_t offset, int whence)
2513  *
2514  * Reposition read/write file offset.
2515  *
2516  * MPSAFE
2517  */
2518 int
2519 sys_lseek(struct lseek_args *uap)
2520 {
2521 	int error;
2522 
2523 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2524 			   &uap->sysmsg_offset);
2525 
2526 	return (error);
2527 }
2528 
2529 /*
2530  * Check if current process can access given file.  amode is a bitmask of *_OK
2531  * access bits.  flags is a bitmask of AT_* flags.
2532  */
2533 int
2534 kern_access(struct nlookupdata *nd, int amode, int flags)
2535 {
2536 	struct vnode *vp;
2537 	int error, mode;
2538 
2539 	if (flags & ~AT_EACCESS)
2540 		return (EINVAL);
2541 	if ((error = nlookup(nd)) != 0)
2542 		return (error);
2543 retry:
2544 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2545 	if (error)
2546 		return (error);
2547 
2548 	/* Flags == 0 means only check for existence. */
2549 	if (amode) {
2550 		mode = 0;
2551 		if (amode & R_OK)
2552 			mode |= VREAD;
2553 		if (amode & W_OK)
2554 			mode |= VWRITE;
2555 		if (amode & X_OK)
2556 			mode |= VEXEC;
2557 		if ((mode & VWRITE) == 0 ||
2558 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2559 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2560 
2561 		/*
2562 		 * If the file handle is stale we have to re-resolve the
2563 		 * entry.  This is a hack at the moment.
2564 		 */
2565 		if (error == ESTALE) {
2566 			vput(vp);
2567 			cache_setunresolved(&nd->nl_nch);
2568 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2569 			if (error == 0) {
2570 				vp = NULL;
2571 				goto retry;
2572 			}
2573 			return(error);
2574 		}
2575 	}
2576 	vput(vp);
2577 	return (error);
2578 }
2579 
2580 /*
2581  * access_args(char *path, int flags)
2582  *
2583  * Check access permissions.
2584  *
2585  * MPALMOSTSAFE
2586  */
2587 int
2588 sys_access(struct access_args *uap)
2589 {
2590 	struct nlookupdata nd;
2591 	int error;
2592 
2593 	get_mplock();
2594 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2595 	if (error == 0)
2596 		error = kern_access(&nd, uap->flags, 0);
2597 	nlookup_done(&nd);
2598 	rel_mplock();
2599 	return (error);
2600 }
2601 
2602 
2603 /*
2604  * faccessat_args(int fd, char *path, int amode, int flags)
2605  *
2606  * Check access permissions.
2607  *
2608  * MPALMOSTSAFE
2609  */
2610 int
2611 sys_faccessat(struct faccessat_args *uap)
2612 {
2613 	struct nlookupdata nd;
2614 	struct file *fp;
2615 	int error;
2616 
2617 	get_mplock();
2618 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2619 				NLC_FOLLOW);
2620 	if (error == 0)
2621 		error = kern_access(&nd, uap->amode, uap->flags);
2622 	nlookup_done_at(&nd, fp);
2623 	rel_mplock();
2624 	return (error);
2625 }
2626 
2627 
2628 /*
2629  * MPSAFE
2630  */
2631 int
2632 kern_stat(struct nlookupdata *nd, struct stat *st)
2633 {
2634 	int error;
2635 	struct vnode *vp;
2636 	thread_t td;
2637 
2638 	if ((error = nlookup(nd)) != 0)
2639 		return (error);
2640 again:
2641 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2642 		return (ENOENT);
2643 
2644 	td = curthread;
2645 	if ((error = vget(vp, LK_SHARED)) != 0)
2646 		return (error);
2647 	error = vn_stat(vp, st, nd->nl_cred);
2648 
2649 	/*
2650 	 * If the file handle is stale we have to re-resolve the entry.  This
2651 	 * is a hack at the moment.
2652 	 */
2653 	if (error == ESTALE) {
2654 		vput(vp);
2655 		cache_setunresolved(&nd->nl_nch);
2656 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2657 		if (error == 0)
2658 			goto again;
2659 	} else {
2660 		vput(vp);
2661 	}
2662 	return (error);
2663 }
2664 
2665 /*
2666  * stat_args(char *path, struct stat *ub)
2667  *
2668  * Get file status; this version follows links.
2669  *
2670  * MPSAFE
2671  */
2672 int
2673 sys_stat(struct stat_args *uap)
2674 {
2675 	CACHE_MPLOCK_DECLARE;
2676 	struct nlookupdata nd;
2677 	struct stat st;
2678 	int error;
2679 
2680 	CACHE_GETMPLOCK1();
2681 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2682 	if (error == 0) {
2683 		error = kern_stat(&nd, &st);
2684 		if (error == 0)
2685 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2686 	}
2687 	nlookup_done(&nd);
2688 	CACHE_RELMPLOCK();
2689 	return (error);
2690 }
2691 
2692 /*
2693  * lstat_args(char *path, struct stat *ub)
2694  *
2695  * Get file status; this version does not follow links.
2696  *
2697  * MPALMOSTSAFE
2698  */
2699 int
2700 sys_lstat(struct lstat_args *uap)
2701 {
2702 	CACHE_MPLOCK_DECLARE;
2703 	struct nlookupdata nd;
2704 	struct stat st;
2705 	int error;
2706 
2707 	CACHE_GETMPLOCK1();
2708 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2709 	if (error == 0) {
2710 		error = kern_stat(&nd, &st);
2711 		if (error == 0)
2712 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2713 	}
2714 	nlookup_done(&nd);
2715 	CACHE_RELMPLOCK();
2716 	return (error);
2717 }
2718 
2719 /*
2720  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2721  *
2722  * Get status of file pointed to by fd/path.
2723  *
2724  * MPALMOSTSAFE
2725  */
2726 int
2727 sys_fstatat(struct fstatat_args *uap)
2728 {
2729 	CACHE_MPLOCK_DECLARE;
2730 	struct nlookupdata nd;
2731 	struct stat st;
2732 	int error;
2733 	int flags;
2734 	struct file *fp;
2735 
2736 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2737 		return (EINVAL);
2738 
2739 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2740 
2741 	CACHE_GETMPLOCK1();
2742 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2743 				UIO_USERSPACE, flags);
2744 	if (error == 0) {
2745 		error = kern_stat(&nd, &st);
2746 		if (error == 0)
2747 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
2748 	}
2749 	nlookup_done_at(&nd, fp);
2750 	CACHE_RELMPLOCK();
2751 	return (error);
2752 }
2753 
2754 /*
2755  * pathconf_Args(char *path, int name)
2756  *
2757  * Get configurable pathname variables.
2758  *
2759  * MPALMOSTSAFE
2760  */
2761 int
2762 sys_pathconf(struct pathconf_args *uap)
2763 {
2764 	struct nlookupdata nd;
2765 	struct vnode *vp;
2766 	int error;
2767 
2768 	vp = NULL;
2769 	get_mplock();
2770 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2771 	if (error == 0)
2772 		error = nlookup(&nd);
2773 	if (error == 0)
2774 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2775 	nlookup_done(&nd);
2776 	if (error == 0) {
2777 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2778 		vput(vp);
2779 	}
2780 	rel_mplock();
2781 	return (error);
2782 }
2783 
2784 /*
2785  * XXX: daver
2786  * kern_readlink isn't properly split yet.  There is a copyin burried
2787  * in VOP_READLINK().
2788  */
2789 int
2790 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2791 {
2792 	struct thread *td = curthread;
2793 	struct vnode *vp;
2794 	struct iovec aiov;
2795 	struct uio auio;
2796 	int error;
2797 
2798 	if ((error = nlookup(nd)) != 0)
2799 		return (error);
2800 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2801 	if (error)
2802 		return (error);
2803 	if (vp->v_type != VLNK) {
2804 		error = EINVAL;
2805 	} else {
2806 		aiov.iov_base = buf;
2807 		aiov.iov_len = count;
2808 		auio.uio_iov = &aiov;
2809 		auio.uio_iovcnt = 1;
2810 		auio.uio_offset = 0;
2811 		auio.uio_rw = UIO_READ;
2812 		auio.uio_segflg = UIO_USERSPACE;
2813 		auio.uio_td = td;
2814 		auio.uio_resid = count;
2815 		error = VOP_READLINK(vp, &auio, td->td_ucred);
2816 	}
2817 	vput(vp);
2818 	*res = count - auio.uio_resid;
2819 	return (error);
2820 }
2821 
2822 /*
2823  * readlink_args(char *path, char *buf, int count)
2824  *
2825  * Return target name of a symbolic link.
2826  *
2827  * MPALMOSTSAFE
2828  */
2829 int
2830 sys_readlink(struct readlink_args *uap)
2831 {
2832 	struct nlookupdata nd;
2833 	int error;
2834 
2835 	get_mplock();
2836 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2837 	if (error == 0) {
2838 		error = kern_readlink(&nd, uap->buf, uap->count,
2839 					&uap->sysmsg_result);
2840 	}
2841 	nlookup_done(&nd);
2842 	rel_mplock();
2843 	return (error);
2844 }
2845 
2846 static int
2847 setfflags(struct vnode *vp, int flags)
2848 {
2849 	struct thread *td = curthread;
2850 	int error;
2851 	struct vattr vattr;
2852 
2853 	/*
2854 	 * Prevent non-root users from setting flags on devices.  When
2855 	 * a device is reused, users can retain ownership of the device
2856 	 * if they are allowed to set flags and programs assume that
2857 	 * chown can't fail when done as root.
2858 	 */
2859 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2860 	    ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2861 		return (error);
2862 
2863 	/*
2864 	 * note: vget is required for any operation that might mod the vnode
2865 	 * so VINACTIVE is properly cleared.
2866 	 */
2867 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2868 		VATTR_NULL(&vattr);
2869 		vattr.va_flags = flags;
2870 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2871 		vput(vp);
2872 	}
2873 	return (error);
2874 }
2875 
2876 /*
2877  * chflags(char *path, int flags)
2878  *
2879  * Change flags of a file given a path name.
2880  *
2881  * MPALMOSTSAFE
2882  */
2883 int
2884 sys_chflags(struct chflags_args *uap)
2885 {
2886 	struct nlookupdata nd;
2887 	struct vnode *vp;
2888 	int error;
2889 
2890 	vp = NULL;
2891 	get_mplock();
2892 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2893 	if (error == 0)
2894 		error = nlookup(&nd);
2895 	if (error == 0)
2896 		error = ncp_writechk(&nd.nl_nch);
2897 	if (error == 0)
2898 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2899 	nlookup_done(&nd);
2900 	if (error == 0) {
2901 		error = setfflags(vp, uap->flags);
2902 		vrele(vp);
2903 	}
2904 	rel_mplock();
2905 	return (error);
2906 }
2907 
2908 /*
2909  * lchflags(char *path, int flags)
2910  *
2911  * Change flags of a file given a path name, but don't follow symlinks.
2912  *
2913  * MPALMOSTSAFE
2914  */
2915 int
2916 sys_lchflags(struct lchflags_args *uap)
2917 {
2918 	struct nlookupdata nd;
2919 	struct vnode *vp;
2920 	int error;
2921 
2922 	vp = NULL;
2923 	get_mplock();
2924 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2925 	if (error == 0)
2926 		error = nlookup(&nd);
2927 	if (error == 0)
2928 		error = ncp_writechk(&nd.nl_nch);
2929 	if (error == 0)
2930 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2931 	nlookup_done(&nd);
2932 	if (error == 0) {
2933 		error = setfflags(vp, uap->flags);
2934 		vrele(vp);
2935 	}
2936 	rel_mplock();
2937 	return (error);
2938 }
2939 
2940 /*
2941  * fchflags_args(int fd, int flags)
2942  *
2943  * Change flags of a file given a file descriptor.
2944  *
2945  * MPALMOSTSAFE
2946  */
2947 int
2948 sys_fchflags(struct fchflags_args *uap)
2949 {
2950 	struct thread *td = curthread;
2951 	struct proc *p = td->td_proc;
2952 	struct file *fp;
2953 	int error;
2954 
2955 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2956 		return (error);
2957 	get_mplock();
2958 	if (fp->f_nchandle.ncp)
2959 		error = ncp_writechk(&fp->f_nchandle);
2960 	if (error == 0)
2961 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2962 	rel_mplock();
2963 	fdrop(fp);
2964 	return (error);
2965 }
2966 
2967 static int
2968 setfmode(struct vnode *vp, int mode)
2969 {
2970 	struct thread *td = curthread;
2971 	int error;
2972 	struct vattr vattr;
2973 
2974 	/*
2975 	 * note: vget is required for any operation that might mod the vnode
2976 	 * so VINACTIVE is properly cleared.
2977 	 */
2978 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2979 		VATTR_NULL(&vattr);
2980 		vattr.va_mode = mode & ALLPERMS;
2981 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2982 		vput(vp);
2983 	}
2984 	return error;
2985 }
2986 
2987 int
2988 kern_chmod(struct nlookupdata *nd, int mode)
2989 {
2990 	struct vnode *vp;
2991 	int error;
2992 
2993 	if ((error = nlookup(nd)) != 0)
2994 		return (error);
2995 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2996 		return (error);
2997 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2998 		error = setfmode(vp, mode);
2999 	vrele(vp);
3000 	return (error);
3001 }
3002 
3003 /*
3004  * chmod_args(char *path, int mode)
3005  *
3006  * Change mode of a file given path name.
3007  *
3008  * MPALMOSTSAFE
3009  */
3010 int
3011 sys_chmod(struct chmod_args *uap)
3012 {
3013 	struct nlookupdata nd;
3014 	int error;
3015 
3016 	get_mplock();
3017 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3018 	if (error == 0)
3019 		error = kern_chmod(&nd, uap->mode);
3020 	nlookup_done(&nd);
3021 	rel_mplock();
3022 	return (error);
3023 }
3024 
3025 /*
3026  * lchmod_args(char *path, int mode)
3027  *
3028  * Change mode of a file given path name (don't follow links.)
3029  *
3030  * MPALMOSTSAFE
3031  */
3032 int
3033 sys_lchmod(struct lchmod_args *uap)
3034 {
3035 	struct nlookupdata nd;
3036 	int error;
3037 
3038 	get_mplock();
3039 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3040 	if (error == 0)
3041 		error = kern_chmod(&nd, uap->mode);
3042 	nlookup_done(&nd);
3043 	rel_mplock();
3044 	return (error);
3045 }
3046 
3047 /*
3048  * fchmod_args(int fd, int mode)
3049  *
3050  * Change mode of a file given a file descriptor.
3051  *
3052  * MPALMOSTSAFE
3053  */
3054 int
3055 sys_fchmod(struct fchmod_args *uap)
3056 {
3057 	struct thread *td = curthread;
3058 	struct proc *p = td->td_proc;
3059 	struct file *fp;
3060 	int error;
3061 
3062 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3063 		return (error);
3064 	get_mplock();
3065 	if (fp->f_nchandle.ncp)
3066 		error = ncp_writechk(&fp->f_nchandle);
3067 	if (error == 0)
3068 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
3069 	rel_mplock();
3070 	fdrop(fp);
3071 	return (error);
3072 }
3073 
3074 /*
3075  * fchmodat_args(char *path, int mode)
3076  *
3077  * Change mode of a file pointed to by fd/path.
3078  *
3079  * MPALMOSTSAFE
3080  */
3081 int
3082 sys_fchmodat(struct fchmodat_args *uap)
3083 {
3084 	struct nlookupdata nd;
3085 	struct file *fp;
3086 	int error;
3087 	int flags;
3088 
3089 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3090 		return (EINVAL);
3091 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3092 
3093 	get_mplock();
3094 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3095 				UIO_USERSPACE, flags);
3096 	if (error == 0)
3097 		error = kern_chmod(&nd, uap->mode);
3098 	nlookup_done_at(&nd, fp);
3099 	rel_mplock();
3100 	return (error);
3101 }
3102 
3103 static int
3104 setfown(struct vnode *vp, uid_t uid, gid_t gid)
3105 {
3106 	struct thread *td = curthread;
3107 	int error;
3108 	struct vattr vattr;
3109 
3110 	/*
3111 	 * note: vget is required for any operation that might mod the vnode
3112 	 * so VINACTIVE is properly cleared.
3113 	 */
3114 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3115 		VATTR_NULL(&vattr);
3116 		vattr.va_uid = uid;
3117 		vattr.va_gid = gid;
3118 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3119 		vput(vp);
3120 	}
3121 	return error;
3122 }
3123 
3124 int
3125 kern_chown(struct nlookupdata *nd, int uid, int gid)
3126 {
3127 	struct vnode *vp;
3128 	int error;
3129 
3130 	if ((error = nlookup(nd)) != 0)
3131 		return (error);
3132 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3133 		return (error);
3134 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3135 		error = setfown(vp, uid, gid);
3136 	vrele(vp);
3137 	return (error);
3138 }
3139 
3140 /*
3141  * chown(char *path, int uid, int gid)
3142  *
3143  * Set ownership given a path name.
3144  *
3145  * MPALMOSTSAFE
3146  */
3147 int
3148 sys_chown(struct chown_args *uap)
3149 {
3150 	struct nlookupdata nd;
3151 	int error;
3152 
3153 	get_mplock();
3154 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3155 	if (error == 0)
3156 		error = kern_chown(&nd, uap->uid, uap->gid);
3157 	nlookup_done(&nd);
3158 	rel_mplock();
3159 	return (error);
3160 }
3161 
3162 /*
3163  * lchown_args(char *path, int uid, int gid)
3164  *
3165  * Set ownership given a path name, do not cross symlinks.
3166  *
3167  * MPALMOSTSAFE
3168  */
3169 int
3170 sys_lchown(struct lchown_args *uap)
3171 {
3172 	struct nlookupdata nd;
3173 	int error;
3174 
3175 	get_mplock();
3176 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3177 	if (error == 0)
3178 		error = kern_chown(&nd, uap->uid, uap->gid);
3179 	nlookup_done(&nd);
3180 	rel_mplock();
3181 	return (error);
3182 }
3183 
3184 /*
3185  * fchown_args(int fd, int uid, int gid)
3186  *
3187  * Set ownership given a file descriptor.
3188  *
3189  * MPALMOSTSAFE
3190  */
3191 int
3192 sys_fchown(struct fchown_args *uap)
3193 {
3194 	struct thread *td = curthread;
3195 	struct proc *p = td->td_proc;
3196 	struct file *fp;
3197 	int error;
3198 
3199 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3200 		return (error);
3201 	get_mplock();
3202 	if (fp->f_nchandle.ncp)
3203 		error = ncp_writechk(&fp->f_nchandle);
3204 	if (error == 0)
3205 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
3206 	rel_mplock();
3207 	fdrop(fp);
3208 	return (error);
3209 }
3210 
3211 /*
3212  * fchownat(int fd, char *path, int uid, int gid, int flags)
3213  *
3214  * Set ownership of file pointed to by fd/path.
3215  *
3216  * MPALMOSTSAFE
3217  */
3218 int
3219 sys_fchownat(struct fchownat_args *uap)
3220 {
3221 	struct nlookupdata nd;
3222 	struct file *fp;
3223 	int error;
3224 	int flags;
3225 
3226 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3227 		return (EINVAL);
3228 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3229 
3230 	get_mplock();
3231 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3232 				UIO_USERSPACE, flags);
3233 	if (error == 0)
3234 		error = kern_chown(&nd, uap->uid, uap->gid);
3235 	nlookup_done_at(&nd, fp);
3236 	rel_mplock();
3237 	return (error);
3238 }
3239 
3240 
3241 static int
3242 getutimes(const struct timeval *tvp, struct timespec *tsp)
3243 {
3244 	struct timeval tv[2];
3245 
3246 	if (tvp == NULL) {
3247 		microtime(&tv[0]);
3248 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3249 		tsp[1] = tsp[0];
3250 	} else {
3251 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3252 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3253 	}
3254 	return 0;
3255 }
3256 
3257 static int
3258 setutimes(struct vnode *vp, struct vattr *vattr,
3259 	  const struct timespec *ts, int nullflag)
3260 {
3261 	struct thread *td = curthread;
3262 	int error;
3263 
3264 	VATTR_NULL(vattr);
3265 	vattr->va_atime = ts[0];
3266 	vattr->va_mtime = ts[1];
3267 	if (nullflag)
3268 		vattr->va_vaflags |= VA_UTIMES_NULL;
3269 	error = VOP_SETATTR(vp, vattr, td->td_ucred);
3270 
3271 	return error;
3272 }
3273 
3274 int
3275 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3276 {
3277 	struct timespec ts[2];
3278 	struct vnode *vp;
3279 	struct vattr vattr;
3280 	int error;
3281 
3282 	if ((error = getutimes(tptr, ts)) != 0)
3283 		return (error);
3284 
3285 	/*
3286 	 * NOTE: utimes() succeeds for the owner even if the file
3287 	 * is not user-writable.
3288 	 */
3289 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3290 
3291 	if ((error = nlookup(nd)) != 0)
3292 		return (error);
3293 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3294 		return (error);
3295 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3296 		return (error);
3297 
3298 	/*
3299 	 * note: vget is required for any operation that might mod the vnode
3300 	 * so VINACTIVE is properly cleared.
3301 	 */
3302 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3303 		error = vget(vp, LK_EXCLUSIVE);
3304 		if (error == 0) {
3305 			error = setutimes(vp, &vattr, ts, (tptr == NULL));
3306 			vput(vp);
3307 		}
3308 	}
3309 	vrele(vp);
3310 	return (error);
3311 }
3312 
3313 /*
3314  * utimes_args(char *path, struct timeval *tptr)
3315  *
3316  * Set the access and modification times of a file.
3317  *
3318  * MPALMOSTSAFE
3319  */
3320 int
3321 sys_utimes(struct utimes_args *uap)
3322 {
3323 	struct timeval tv[2];
3324 	struct nlookupdata nd;
3325 	int error;
3326 
3327 	if (uap->tptr) {
3328  		error = copyin(uap->tptr, tv, sizeof(tv));
3329 		if (error)
3330 			return (error);
3331 	}
3332 	get_mplock();
3333 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3334 	if (error == 0)
3335 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3336 	nlookup_done(&nd);
3337 	rel_mplock();
3338 	return (error);
3339 }
3340 
3341 /*
3342  * lutimes_args(char *path, struct timeval *tptr)
3343  *
3344  * Set the access and modification times of a file.
3345  *
3346  * MPALMOSTSAFE
3347  */
3348 int
3349 sys_lutimes(struct lutimes_args *uap)
3350 {
3351 	struct timeval tv[2];
3352 	struct nlookupdata nd;
3353 	int error;
3354 
3355 	if (uap->tptr) {
3356 		error = copyin(uap->tptr, tv, sizeof(tv));
3357 		if (error)
3358 			return (error);
3359 	}
3360 	get_mplock();
3361 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3362 	if (error == 0)
3363 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3364 	nlookup_done(&nd);
3365 	rel_mplock();
3366 	return (error);
3367 }
3368 
3369 /*
3370  * Set utimes on a file descriptor.  The creds used to open the
3371  * file are used to determine whether the operation is allowed
3372  * or not.
3373  */
3374 int
3375 kern_futimes(int fd, struct timeval *tptr)
3376 {
3377 	struct thread *td = curthread;
3378 	struct proc *p = td->td_proc;
3379 	struct timespec ts[2];
3380 	struct file *fp;
3381 	struct vnode *vp;
3382 	struct vattr vattr;
3383 	int error;
3384 
3385 	error = getutimes(tptr, ts);
3386 	if (error)
3387 		return (error);
3388 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3389 		return (error);
3390 	if (fp->f_nchandle.ncp)
3391 		error = ncp_writechk(&fp->f_nchandle);
3392 	if (error == 0) {
3393 		vp = fp->f_data;
3394 		error = vget(vp, LK_EXCLUSIVE);
3395 		if (error == 0) {
3396 			error = VOP_GETATTR(vp, &vattr);
3397 			if (error == 0) {
3398 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3399 						   fp->f_cred);
3400 			}
3401 			if (error == 0) {
3402 				error = setutimes(vp, &vattr, ts,
3403 						  (tptr == NULL));
3404 			}
3405 			vput(vp);
3406 		}
3407 	}
3408 	fdrop(fp);
3409 	return (error);
3410 }
3411 
3412 /*
3413  * futimes_args(int fd, struct timeval *tptr)
3414  *
3415  * Set the access and modification times of a file.
3416  *
3417  * MPALMOSTSAFE
3418  */
3419 int
3420 sys_futimes(struct futimes_args *uap)
3421 {
3422 	struct timeval tv[2];
3423 	int error;
3424 
3425 	if (uap->tptr) {
3426 		error = copyin(uap->tptr, tv, sizeof(tv));
3427 		if (error)
3428 			return (error);
3429 	}
3430 	get_mplock();
3431 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3432 	rel_mplock();
3433 
3434 	return (error);
3435 }
3436 
3437 int
3438 kern_truncate(struct nlookupdata *nd, off_t length)
3439 {
3440 	struct vnode *vp;
3441 	struct vattr vattr;
3442 	int error;
3443 
3444 	if (length < 0)
3445 		return(EINVAL);
3446 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3447 	if ((error = nlookup(nd)) != 0)
3448 		return (error);
3449 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3450 		return (error);
3451 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3452 		return (error);
3453 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3454 		vrele(vp);
3455 		return (error);
3456 	}
3457 	if (vp->v_type == VDIR) {
3458 		error = EISDIR;
3459 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3460 		VATTR_NULL(&vattr);
3461 		vattr.va_size = length;
3462 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3463 	}
3464 	vput(vp);
3465 	return (error);
3466 }
3467 
3468 /*
3469  * truncate(char *path, int pad, off_t length)
3470  *
3471  * Truncate a file given its path name.
3472  *
3473  * MPALMOSTSAFE
3474  */
3475 int
3476 sys_truncate(struct truncate_args *uap)
3477 {
3478 	struct nlookupdata nd;
3479 	int error;
3480 
3481 	get_mplock();
3482 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3483 	if (error == 0)
3484 		error = kern_truncate(&nd, uap->length);
3485 	nlookup_done(&nd);
3486 	rel_mplock();
3487 	return error;
3488 }
3489 
3490 int
3491 kern_ftruncate(int fd, off_t length)
3492 {
3493 	struct thread *td = curthread;
3494 	struct proc *p = td->td_proc;
3495 	struct vattr vattr;
3496 	struct vnode *vp;
3497 	struct file *fp;
3498 	int error;
3499 
3500 	if (length < 0)
3501 		return(EINVAL);
3502 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3503 		return (error);
3504 	if (fp->f_nchandle.ncp) {
3505 		error = ncp_writechk(&fp->f_nchandle);
3506 		if (error)
3507 			goto done;
3508 	}
3509 	if ((fp->f_flag & FWRITE) == 0) {
3510 		error = EINVAL;
3511 		goto done;
3512 	}
3513 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3514 		error = EINVAL;
3515 		goto done;
3516 	}
3517 	vp = (struct vnode *)fp->f_data;
3518 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3519 	if (vp->v_type == VDIR) {
3520 		error = EISDIR;
3521 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
3522 		VATTR_NULL(&vattr);
3523 		vattr.va_size = length;
3524 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3525 	}
3526 	vn_unlock(vp);
3527 done:
3528 	fdrop(fp);
3529 	return (error);
3530 }
3531 
3532 /*
3533  * ftruncate_args(int fd, int pad, off_t length)
3534  *
3535  * Truncate a file given a file descriptor.
3536  *
3537  * MPALMOSTSAFE
3538  */
3539 int
3540 sys_ftruncate(struct ftruncate_args *uap)
3541 {
3542 	int error;
3543 
3544 	get_mplock();
3545 	error = kern_ftruncate(uap->fd, uap->length);
3546 	rel_mplock();
3547 
3548 	return (error);
3549 }
3550 
3551 /*
3552  * fsync(int fd)
3553  *
3554  * Sync an open file.
3555  *
3556  * MPALMOSTSAFE
3557  */
3558 int
3559 sys_fsync(struct fsync_args *uap)
3560 {
3561 	struct thread *td = curthread;
3562 	struct proc *p = td->td_proc;
3563 	struct vnode *vp;
3564 	struct file *fp;
3565 	vm_object_t obj;
3566 	int error;
3567 
3568 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3569 		return (error);
3570 	get_mplock();
3571 	vp = (struct vnode *)fp->f_data;
3572 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3573 	if ((obj = vp->v_object) != NULL)
3574 		vm_object_page_clean(obj, 0, 0, 0);
3575 	error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3576 	if (error == 0 && vp->v_mount)
3577 		error = buf_fsync(vp);
3578 	vn_unlock(vp);
3579 	rel_mplock();
3580 	fdrop(fp);
3581 
3582 	return (error);
3583 }
3584 
3585 int
3586 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3587 {
3588 	struct nchandle fnchd;
3589 	struct nchandle tnchd;
3590 	struct namecache *ncp;
3591 	struct vnode *fdvp;
3592 	struct vnode *tdvp;
3593 	struct mount *mp;
3594 	int error;
3595 
3596 	bwillinode(1);
3597 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3598 	if ((error = nlookup(fromnd)) != 0)
3599 		return (error);
3600 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3601 		return (ENOENT);
3602 	fnchd.mount = fromnd->nl_nch.mount;
3603 	cache_hold(&fnchd);
3604 
3605 	/*
3606 	 * unlock the source nch so we can lookup the target nch without
3607 	 * deadlocking.  The target may or may not exist so we do not check
3608 	 * for a target vp like kern_mkdir() and other creation functions do.
3609 	 *
3610 	 * The source and target directories are ref'd and rechecked after
3611 	 * everything is relocked to determine if the source or target file
3612 	 * has been renamed.
3613 	 */
3614 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3615 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3616 	cache_unlock(&fromnd->nl_nch);
3617 
3618 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3619 	if ((error = nlookup(tond)) != 0) {
3620 		cache_drop(&fnchd);
3621 		return (error);
3622 	}
3623 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3624 		cache_drop(&fnchd);
3625 		return (ENOENT);
3626 	}
3627 	tnchd.mount = tond->nl_nch.mount;
3628 	cache_hold(&tnchd);
3629 
3630 	/*
3631 	 * If the source and target are the same there is nothing to do
3632 	 */
3633 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3634 		cache_drop(&fnchd);
3635 		cache_drop(&tnchd);
3636 		return (0);
3637 	}
3638 
3639 	/*
3640 	 * Mount points cannot be renamed or overwritten
3641 	 */
3642 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3643 	    NCF_ISMOUNTPT
3644 	) {
3645 		cache_drop(&fnchd);
3646 		cache_drop(&tnchd);
3647 		return (EINVAL);
3648 	}
3649 
3650 	/*
3651 	 * Relock the source ncp.  cache_relock() will deal with any
3652 	 * deadlocks against the already-locked tond and will also
3653 	 * make sure both are resolved.
3654 	 *
3655 	 * NOTE AFTER RELOCKING: The source or target ncp may have become
3656 	 * invalid while they were unlocked, nc_vp and nc_mount could
3657 	 * be NULL.
3658 	 */
3659 	cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
3660 		     &tond->nl_nch, tond->nl_cred);
3661 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3662 
3663 	/*
3664 	 * make sure the parent directories linkages are the same
3665 	 */
3666 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3667 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3668 		cache_drop(&fnchd);
3669 		cache_drop(&tnchd);
3670 		return (ENOENT);
3671 	}
3672 
3673 	/*
3674 	 * Both the source and target must be within the same filesystem and
3675 	 * in the same filesystem as their parent directories within the
3676 	 * namecache topology.
3677 	 *
3678 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3679 	 */
3680 	mp = fnchd.mount;
3681 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3682 	    mp != tond->nl_nch.mount) {
3683 		cache_drop(&fnchd);
3684 		cache_drop(&tnchd);
3685 		return (EXDEV);
3686 	}
3687 
3688 	/*
3689 	 * Make sure the mount point is writable
3690 	 */
3691 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3692 		cache_drop(&fnchd);
3693 		cache_drop(&tnchd);
3694 		return (error);
3695 	}
3696 
3697 	/*
3698 	 * If the target exists and either the source or target is a directory,
3699 	 * then both must be directories.
3700 	 *
3701 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3702 	 * have become NULL.
3703 	 */
3704 	if (tond->nl_nch.ncp->nc_vp) {
3705 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3706 			error = ENOENT;
3707 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3708 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3709 				error = ENOTDIR;
3710 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3711 			error = EISDIR;
3712 		}
3713 	}
3714 
3715 	/*
3716 	 * You cannot rename a source into itself or a subdirectory of itself.
3717 	 * We check this by travsersing the target directory upwards looking
3718 	 * for a match against the source.
3719 	 *
3720 	 * XXX MPSAFE
3721 	 */
3722 	if (error == 0) {
3723 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3724 			if (fromnd->nl_nch.ncp == ncp) {
3725 				error = EINVAL;
3726 				break;
3727 			}
3728 		}
3729 	}
3730 
3731 	cache_drop(&fnchd);
3732 	cache_drop(&tnchd);
3733 
3734 	/*
3735 	 * Even though the namespaces are different, they may still represent
3736 	 * hardlinks to the same file.  The filesystem might have a hard time
3737 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3738 	 * when we detect the situation.
3739 	 */
3740 	if (error == 0) {
3741 		fdvp = fromnd->nl_dvp;
3742 		tdvp = tond->nl_dvp;
3743 		if (fdvp == NULL || tdvp == NULL) {
3744 			error = EPERM;
3745 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3746 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3747 					    fromnd->nl_cred);
3748 		} else {
3749 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3750 					    fdvp, tdvp, tond->nl_cred);
3751 		}
3752 	}
3753 	return (error);
3754 }
3755 
3756 /*
3757  * rename_args(char *from, char *to)
3758  *
3759  * Rename files.  Source and destination must either both be directories,
3760  * or both not be directories.  If target is a directory, it must be empty.
3761  *
3762  * MPALMOSTSAFE
3763  */
3764 int
3765 sys_rename(struct rename_args *uap)
3766 {
3767 	struct nlookupdata fromnd, tond;
3768 	int error;
3769 
3770 	get_mplock();
3771 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3772 	if (error == 0) {
3773 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3774 		if (error == 0)
3775 			error = kern_rename(&fromnd, &tond);
3776 		nlookup_done(&tond);
3777 	}
3778 	nlookup_done(&fromnd);
3779 	rel_mplock();
3780 	return (error);
3781 }
3782 
3783 int
3784 kern_mkdir(struct nlookupdata *nd, int mode)
3785 {
3786 	struct thread *td = curthread;
3787 	struct proc *p = td->td_proc;
3788 	struct vnode *vp;
3789 	struct vattr vattr;
3790 	int error;
3791 
3792 	bwillinode(1);
3793 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3794 	if ((error = nlookup(nd)) != 0)
3795 		return (error);
3796 
3797 	if (nd->nl_nch.ncp->nc_vp)
3798 		return (EEXIST);
3799 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3800 		return (error);
3801 	VATTR_NULL(&vattr);
3802 	vattr.va_type = VDIR;
3803 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3804 
3805 	vp = NULL;
3806 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
3807 	if (error == 0)
3808 		vput(vp);
3809 	return (error);
3810 }
3811 
3812 /*
3813  * mkdir_args(char *path, int mode)
3814  *
3815  * Make a directory file.
3816  *
3817  * MPALMOSTSAFE
3818  */
3819 int
3820 sys_mkdir(struct mkdir_args *uap)
3821 {
3822 	struct nlookupdata nd;
3823 	int error;
3824 
3825 	get_mplock();
3826 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3827 	if (error == 0)
3828 		error = kern_mkdir(&nd, uap->mode);
3829 	nlookup_done(&nd);
3830 	rel_mplock();
3831 	return (error);
3832 }
3833 
3834 int
3835 kern_rmdir(struct nlookupdata *nd)
3836 {
3837 	int error;
3838 
3839 	bwillinode(1);
3840 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3841 	if ((error = nlookup(nd)) != 0)
3842 		return (error);
3843 
3844 	/*
3845 	 * Do not allow directories representing mount points to be
3846 	 * deleted, even if empty.  Check write perms on mount point
3847 	 * in case the vnode is aliased (aka nullfs).
3848 	 */
3849 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3850 		return (EINVAL);
3851 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3852 		return (error);
3853 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3854 	return (error);
3855 }
3856 
3857 /*
3858  * rmdir_args(char *path)
3859  *
3860  * Remove a directory file.
3861  *
3862  * MPALMOSTSAFE
3863  */
3864 int
3865 sys_rmdir(struct rmdir_args *uap)
3866 {
3867 	struct nlookupdata nd;
3868 	int error;
3869 
3870 	get_mplock();
3871 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3872 	if (error == 0)
3873 		error = kern_rmdir(&nd);
3874 	nlookup_done(&nd);
3875 	rel_mplock();
3876 	return (error);
3877 }
3878 
3879 int
3880 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3881     enum uio_seg direction)
3882 {
3883 	struct thread *td = curthread;
3884 	struct proc *p = td->td_proc;
3885 	struct vnode *vp;
3886 	struct file *fp;
3887 	struct uio auio;
3888 	struct iovec aiov;
3889 	off_t loff;
3890 	int error, eofflag;
3891 
3892 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3893 		return (error);
3894 	if ((fp->f_flag & FREAD) == 0) {
3895 		error = EBADF;
3896 		goto done;
3897 	}
3898 	vp = (struct vnode *)fp->f_data;
3899 unionread:
3900 	if (vp->v_type != VDIR) {
3901 		error = EINVAL;
3902 		goto done;
3903 	}
3904 	aiov.iov_base = buf;
3905 	aiov.iov_len = count;
3906 	auio.uio_iov = &aiov;
3907 	auio.uio_iovcnt = 1;
3908 	auio.uio_rw = UIO_READ;
3909 	auio.uio_segflg = direction;
3910 	auio.uio_td = td;
3911 	auio.uio_resid = count;
3912 	loff = auio.uio_offset = fp->f_offset;
3913 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3914 	fp->f_offset = auio.uio_offset;
3915 	if (error)
3916 		goto done;
3917 	if (count == auio.uio_resid) {
3918 		if (union_dircheckp) {
3919 			error = union_dircheckp(td, &vp, fp);
3920 			if (error == -1)
3921 				goto unionread;
3922 			if (error)
3923 				goto done;
3924 		}
3925 #if 0
3926 		if ((vp->v_flag & VROOT) &&
3927 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3928 			struct vnode *tvp = vp;
3929 			vp = vp->v_mount->mnt_vnodecovered;
3930 			vref(vp);
3931 			fp->f_data = vp;
3932 			fp->f_offset = 0;
3933 			vrele(tvp);
3934 			goto unionread;
3935 		}
3936 #endif
3937 	}
3938 
3939 	/*
3940 	 * WARNING!  *basep may not be wide enough to accomodate the
3941 	 * seek offset.   XXX should we hack this to return the upper 32 bits
3942 	 * for offsets greater then 4G?
3943 	 */
3944 	if (basep) {
3945 		*basep = (long)loff;
3946 	}
3947 	*res = count - auio.uio_resid;
3948 done:
3949 	fdrop(fp);
3950 	return (error);
3951 }
3952 
3953 /*
3954  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3955  *
3956  * Read a block of directory entries in a file system independent format.
3957  *
3958  * MPALMOSTSAFE
3959  */
3960 int
3961 sys_getdirentries(struct getdirentries_args *uap)
3962 {
3963 	long base;
3964 	int error;
3965 
3966 	get_mplock();
3967 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3968 				   &uap->sysmsg_result, UIO_USERSPACE);
3969 	rel_mplock();
3970 
3971 	if (error == 0 && uap->basep)
3972 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3973 	return (error);
3974 }
3975 
3976 /*
3977  * getdents_args(int fd, char *buf, size_t count)
3978  *
3979  * MPALMOSTSAFE
3980  */
3981 int
3982 sys_getdents(struct getdents_args *uap)
3983 {
3984 	int error;
3985 
3986 	get_mplock();
3987 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3988 				   &uap->sysmsg_result, UIO_USERSPACE);
3989 	rel_mplock();
3990 
3991 	return (error);
3992 }
3993 
3994 /*
3995  * Set the mode mask for creation of filesystem nodes.
3996  *
3997  * umask(int newmask)
3998  *
3999  * MPSAFE
4000  */
4001 int
4002 sys_umask(struct umask_args *uap)
4003 {
4004 	struct thread *td = curthread;
4005 	struct proc *p = td->td_proc;
4006 	struct filedesc *fdp;
4007 
4008 	fdp = p->p_fd;
4009 	uap->sysmsg_result = fdp->fd_cmask;
4010 	fdp->fd_cmask = uap->newmask & ALLPERMS;
4011 	return (0);
4012 }
4013 
4014 /*
4015  * revoke(char *path)
4016  *
4017  * Void all references to file by ripping underlying filesystem
4018  * away from vnode.
4019  *
4020  * MPALMOSTSAFE
4021  */
4022 int
4023 sys_revoke(struct revoke_args *uap)
4024 {
4025 	struct nlookupdata nd;
4026 	struct vattr vattr;
4027 	struct vnode *vp;
4028 	struct ucred *cred;
4029 	int error;
4030 
4031 	vp = NULL;
4032 	get_mplock();
4033 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4034 	if (error == 0)
4035 		error = nlookup(&nd);
4036 	if (error == 0)
4037 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4038 	cred = crhold(nd.nl_cred);
4039 	nlookup_done(&nd);
4040 	if (error == 0) {
4041 		if (error == 0)
4042 			error = VOP_GETATTR(vp, &vattr);
4043 		if (error == 0 && cred->cr_uid != vattr.va_uid)
4044 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4045 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4046 			if (vcount(vp) > 0)
4047 				error = vrevoke(vp, cred);
4048 		} else if (error == 0) {
4049 			error = vrevoke(vp, cred);
4050 		}
4051 		vrele(vp);
4052 	}
4053 	if (cred)
4054 		crfree(cred);
4055 	rel_mplock();
4056 	return (error);
4057 }
4058 
4059 /*
4060  * getfh_args(char *fname, fhandle_t *fhp)
4061  *
4062  * Get (NFS) file handle
4063  *
4064  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4065  * mount.  This allows nullfs mounts to be explicitly exported.
4066  *
4067  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4068  *
4069  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
4070  *	    work, but you do not really have protection against access to
4071  *	    the related parent directories.
4072  *
4073  * MPALMOSTSAFE
4074  */
4075 int
4076 sys_getfh(struct getfh_args *uap)
4077 {
4078 	struct thread *td = curthread;
4079 	struct nlookupdata nd;
4080 	fhandle_t fh;
4081 	struct vnode *vp;
4082 	struct mount *mp;
4083 	int error;
4084 
4085 	/*
4086 	 * Must be super user
4087 	 */
4088 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
4089 		return (error);
4090 
4091 	vp = NULL;
4092 	get_mplock();
4093 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4094 	if (error == 0)
4095 		error = nlookup(&nd);
4096 	if (error == 0)
4097 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4098 	mp = nd.nl_nch.mount;
4099 	nlookup_done(&nd);
4100 	if (error == 0) {
4101 		bzero(&fh, sizeof(fh));
4102 		fh.fh_fsid = mp->mnt_stat.f_fsid;
4103 		error = VFS_VPTOFH(vp, &fh.fh_fid);
4104 		vput(vp);
4105 		if (error == 0)
4106 			error = copyout(&fh, uap->fhp, sizeof(fh));
4107 	}
4108 	rel_mplock();
4109 	return (error);
4110 }
4111 
4112 /*
4113  * fhopen_args(const struct fhandle *u_fhp, int flags)
4114  *
4115  * syscall for the rpc.lockd to use to translate a NFS file handle into
4116  * an open descriptor.
4117  *
4118  * warning: do not remove the priv_check() call or this becomes one giant
4119  * security hole.
4120  *
4121  * MPALMOSTSAFE
4122  */
4123 int
4124 sys_fhopen(struct fhopen_args *uap)
4125 {
4126 	struct thread *td = curthread;
4127 	struct filedesc *fdp = td->td_proc->p_fd;
4128 	struct mount *mp;
4129 	struct vnode *vp;
4130 	struct fhandle fhp;
4131 	struct vattr vat;
4132 	struct vattr *vap = &vat;
4133 	struct flock lf;
4134 	int fmode, mode, error, type;
4135 	struct file *nfp;
4136 	struct file *fp;
4137 	int indx;
4138 
4139 	/*
4140 	 * Must be super user
4141 	 */
4142 	error = priv_check(td, PRIV_ROOT);
4143 	if (error)
4144 		return (error);
4145 
4146 	fmode = FFLAGS(uap->flags);
4147 
4148 	/*
4149 	 * Why not allow a non-read/write open for our lockd?
4150 	 */
4151 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4152 		return (EINVAL);
4153 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4154 	if (error)
4155 		return(error);
4156 
4157 	/*
4158 	 * Find the mount point
4159 	 */
4160 	get_mplock();
4161 	mp = vfs_getvfs(&fhp.fh_fsid);
4162 	if (mp == NULL) {
4163 		error = ESTALE;
4164 		goto  done;
4165 	}
4166 	/* now give me my vnode, it gets returned to me locked */
4167 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4168 	if (error)
4169 		goto done;
4170  	/*
4171 	 * from now on we have to make sure not
4172 	 * to forget about the vnode
4173 	 * any error that causes an abort must vput(vp)
4174 	 * just set error = err and 'goto bad;'.
4175 	 */
4176 
4177 	/*
4178 	 * from vn_open
4179 	 */
4180 	if (vp->v_type == VLNK) {
4181 		error = EMLINK;
4182 		goto bad;
4183 	}
4184 	if (vp->v_type == VSOCK) {
4185 		error = EOPNOTSUPP;
4186 		goto bad;
4187 	}
4188 	mode = 0;
4189 	if (fmode & (FWRITE | O_TRUNC)) {
4190 		if (vp->v_type == VDIR) {
4191 			error = EISDIR;
4192 			goto bad;
4193 		}
4194 		error = vn_writechk(vp, NULL);
4195 		if (error)
4196 			goto bad;
4197 		mode |= VWRITE;
4198 	}
4199 	if (fmode & FREAD)
4200 		mode |= VREAD;
4201 	if (mode) {
4202 		error = VOP_ACCESS(vp, mode, td->td_ucred);
4203 		if (error)
4204 			goto bad;
4205 	}
4206 	if (fmode & O_TRUNC) {
4207 		vn_unlock(vp);				/* XXX */
4208 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
4209 		VATTR_NULL(vap);
4210 		vap->va_size = 0;
4211 		error = VOP_SETATTR(vp, vap, td->td_ucred);
4212 		if (error)
4213 			goto bad;
4214 	}
4215 
4216 	/*
4217 	 * VOP_OPEN needs the file pointer so it can potentially override
4218 	 * it.
4219 	 *
4220 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
4221 	 * directory.  XXX
4222 	 */
4223 	if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4224 		goto bad;
4225 	fp = nfp;
4226 
4227 	error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4228 	if (error) {
4229 		/*
4230 		 * setting f_ops this way prevents VOP_CLOSE from being
4231 		 * called or fdrop() releasing the vp from v_data.   Since
4232 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4233 		 */
4234 		fp->f_ops = &badfileops;
4235 		fp->f_data = NULL;
4236 		goto bad_drop;
4237 	}
4238 
4239 	/*
4240 	 * The fp is given its own reference, we still have our ref and lock.
4241 	 *
4242 	 * Assert that all regular files must be created with a VM object.
4243 	 */
4244 	if (vp->v_type == VREG && vp->v_object == NULL) {
4245 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4246 		goto bad_drop;
4247 	}
4248 
4249 	/*
4250 	 * The open was successful.  Handle any locking requirements.
4251 	 */
4252 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
4253 		lf.l_whence = SEEK_SET;
4254 		lf.l_start = 0;
4255 		lf.l_len = 0;
4256 		if (fmode & O_EXLOCK)
4257 			lf.l_type = F_WRLCK;
4258 		else
4259 			lf.l_type = F_RDLCK;
4260 		if (fmode & FNONBLOCK)
4261 			type = 0;
4262 		else
4263 			type = F_WAIT;
4264 		vn_unlock(vp);
4265 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4266 			/*
4267 			 * release our private reference.
4268 			 */
4269 			fsetfd(fdp, NULL, indx);
4270 			fdrop(fp);
4271 			vrele(vp);
4272 			goto done;
4273 		}
4274 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4275 		fp->f_flag |= FHASLOCK;
4276 	}
4277 
4278 	/*
4279 	 * Clean up.  Associate the file pointer with the previously
4280 	 * reserved descriptor and return it.
4281 	 */
4282 	vput(vp);
4283 	rel_mplock();
4284 	fsetfd(fdp, fp, indx);
4285 	fdrop(fp);
4286 	uap->sysmsg_result = indx;
4287 	return (0);
4288 
4289 bad_drop:
4290 	fsetfd(fdp, NULL, indx);
4291 	fdrop(fp);
4292 bad:
4293 	vput(vp);
4294 done:
4295 	rel_mplock();
4296 	return (error);
4297 }
4298 
4299 /*
4300  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4301  *
4302  * MPALMOSTSAFE
4303  */
4304 int
4305 sys_fhstat(struct fhstat_args *uap)
4306 {
4307 	struct thread *td = curthread;
4308 	struct stat sb;
4309 	fhandle_t fh;
4310 	struct mount *mp;
4311 	struct vnode *vp;
4312 	int error;
4313 
4314 	/*
4315 	 * Must be super user
4316 	 */
4317 	error = priv_check(td, PRIV_ROOT);
4318 	if (error)
4319 		return (error);
4320 
4321 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4322 	if (error)
4323 		return (error);
4324 
4325 	get_mplock();
4326 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4327 		error = ESTALE;
4328 	if (error == 0) {
4329 		if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4330 			error = vn_stat(vp, &sb, td->td_ucred);
4331 			vput(vp);
4332 		}
4333 	}
4334 	rel_mplock();
4335 	if (error == 0)
4336 		error = copyout(&sb, uap->sb, sizeof(sb));
4337 	return (error);
4338 }
4339 
4340 /*
4341  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4342  *
4343  * MPALMOSTSAFE
4344  */
4345 int
4346 sys_fhstatfs(struct fhstatfs_args *uap)
4347 {
4348 	struct thread *td = curthread;
4349 	struct proc *p = td->td_proc;
4350 	struct statfs *sp;
4351 	struct mount *mp;
4352 	struct vnode *vp;
4353 	struct statfs sb;
4354 	char *fullpath, *freepath;
4355 	fhandle_t fh;
4356 	int error;
4357 
4358 	/*
4359 	 * Must be super user
4360 	 */
4361 	if ((error = priv_check(td, PRIV_ROOT)))
4362 		return (error);
4363 
4364 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4365 		return (error);
4366 
4367 	get_mplock();
4368 
4369 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4370 		error = ESTALE;
4371 		goto done;
4372 	}
4373 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4374 		error = ESTALE;
4375 		goto done;
4376 	}
4377 
4378 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4379 		goto done;
4380 	mp = vp->v_mount;
4381 	sp = &mp->mnt_stat;
4382 	vput(vp);
4383 	if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4384 		goto done;
4385 
4386 	error = mount_path(p, mp, &fullpath, &freepath);
4387 	if (error)
4388 		goto done;
4389 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4390 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4391 	kfree(freepath, M_TEMP);
4392 
4393 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4394 	if (priv_check(td, PRIV_ROOT)) {
4395 		bcopy(sp, &sb, sizeof(sb));
4396 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4397 		sp = &sb;
4398 	}
4399 	error = copyout(sp, uap->buf, sizeof(*sp));
4400 done:
4401 	rel_mplock();
4402 	return (error);
4403 }
4404 
4405 /*
4406  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4407  *
4408  * MPALMOSTSAFE
4409  */
4410 int
4411 sys_fhstatvfs(struct fhstatvfs_args *uap)
4412 {
4413 	struct thread *td = curthread;
4414 	struct proc *p = td->td_proc;
4415 	struct statvfs *sp;
4416 	struct mount *mp;
4417 	struct vnode *vp;
4418 	fhandle_t fh;
4419 	int error;
4420 
4421 	/*
4422 	 * Must be super user
4423 	 */
4424 	if ((error = priv_check(td, PRIV_ROOT)))
4425 		return (error);
4426 
4427 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4428 		return (error);
4429 
4430 	get_mplock();
4431 
4432 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4433 		error = ESTALE;
4434 		goto done;
4435 	}
4436 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4437 		error = ESTALE;
4438 		goto done;
4439 	}
4440 
4441 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4442 		goto done;
4443 	mp = vp->v_mount;
4444 	sp = &mp->mnt_vstat;
4445 	vput(vp);
4446 	if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4447 		goto done;
4448 
4449 	sp->f_flag = 0;
4450 	if (mp->mnt_flag & MNT_RDONLY)
4451 		sp->f_flag |= ST_RDONLY;
4452 	if (mp->mnt_flag & MNT_NOSUID)
4453 		sp->f_flag |= ST_NOSUID;
4454 	error = copyout(sp, uap->buf, sizeof(*sp));
4455 done:
4456 	rel_mplock();
4457 	return (error);
4458 }
4459 
4460 
4461 /*
4462  * Syscall to push extended attribute configuration information into the
4463  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4464  * a command (int cmd), and attribute name and misc data.  For now, the
4465  * attribute name is left in userspace for consumption by the VFS_op.
4466  * It will probably be changed to be copied into sysspace by the
4467  * syscall in the future, once issues with various consumers of the
4468  * attribute code have raised their hands.
4469  *
4470  * Currently this is used only by UFS Extended Attributes.
4471  *
4472  * MPALMOSTSAFE
4473  */
4474 int
4475 sys_extattrctl(struct extattrctl_args *uap)
4476 {
4477 	struct nlookupdata nd;
4478 	struct vnode *vp;
4479 	char attrname[EXTATTR_MAXNAMELEN];
4480 	int error;
4481 	size_t size;
4482 
4483 	get_mplock();
4484 
4485 	attrname[0] = 0;
4486 	vp = NULL;
4487 	error = 0;
4488 
4489 	if (error == 0 && uap->filename) {
4490 		error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4491 				     NLC_FOLLOW);
4492 		if (error == 0)
4493 			error = nlookup(&nd);
4494 		if (error == 0)
4495 			error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4496 		nlookup_done(&nd);
4497 	}
4498 
4499 	if (error == 0 && uap->attrname) {
4500 		error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4501 				  &size);
4502 	}
4503 
4504 	if (error == 0) {
4505 		error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4506 		if (error == 0)
4507 			error = nlookup(&nd);
4508 		if (error == 0)
4509 			error = ncp_writechk(&nd.nl_nch);
4510 		if (error == 0) {
4511 			error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4512 					       uap->attrnamespace,
4513 					       uap->attrname, nd.nl_cred);
4514 		}
4515 		nlookup_done(&nd);
4516 	}
4517 
4518 	rel_mplock();
4519 
4520 	return (error);
4521 }
4522 
4523 /*
4524  * Syscall to get a named extended attribute on a file or directory.
4525  *
4526  * MPALMOSTSAFE
4527  */
4528 int
4529 sys_extattr_set_file(struct extattr_set_file_args *uap)
4530 {
4531 	char attrname[EXTATTR_MAXNAMELEN];
4532 	struct nlookupdata nd;
4533 	struct vnode *vp;
4534 	struct uio auio;
4535 	struct iovec aiov;
4536 	int error;
4537 
4538 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4539 	if (error)
4540 		return (error);
4541 
4542 	vp = NULL;
4543 	get_mplock();
4544 
4545 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4546 	if (error == 0)
4547 		error = nlookup(&nd);
4548 	if (error == 0)
4549 		error = ncp_writechk(&nd.nl_nch);
4550 	if (error == 0)
4551 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4552 	if (error) {
4553 		nlookup_done(&nd);
4554 		rel_mplock();
4555 		return (error);
4556 	}
4557 
4558 	bzero(&auio, sizeof(auio));
4559 	aiov.iov_base = uap->data;
4560 	aiov.iov_len = uap->nbytes;
4561 	auio.uio_iov = &aiov;
4562 	auio.uio_iovcnt = 1;
4563 	auio.uio_offset = 0;
4564 	auio.uio_resid = uap->nbytes;
4565 	auio.uio_rw = UIO_WRITE;
4566 	auio.uio_td = curthread;
4567 
4568 	error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4569 			       &auio, nd.nl_cred);
4570 
4571 	vput(vp);
4572 	nlookup_done(&nd);
4573 	rel_mplock();
4574 	return (error);
4575 }
4576 
4577 /*
4578  * Syscall to get a named extended attribute on a file or directory.
4579  *
4580  * MPALMOSTSAFE
4581  */
4582 int
4583 sys_extattr_get_file(struct extattr_get_file_args *uap)
4584 {
4585 	char attrname[EXTATTR_MAXNAMELEN];
4586 	struct nlookupdata nd;
4587 	struct uio auio;
4588 	struct iovec aiov;
4589 	struct vnode *vp;
4590 	int error;
4591 
4592 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4593 	if (error)
4594 		return (error);
4595 
4596 	vp = NULL;
4597 	get_mplock();
4598 
4599 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4600 	if (error == 0)
4601 		error = nlookup(&nd);
4602 	if (error == 0)
4603 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4604 	if (error) {
4605 		nlookup_done(&nd);
4606 		rel_mplock();
4607 		return (error);
4608 	}
4609 
4610 	bzero(&auio, sizeof(auio));
4611 	aiov.iov_base = uap->data;
4612 	aiov.iov_len = uap->nbytes;
4613 	auio.uio_iov = &aiov;
4614 	auio.uio_iovcnt = 1;
4615 	auio.uio_offset = 0;
4616 	auio.uio_resid = uap->nbytes;
4617 	auio.uio_rw = UIO_READ;
4618 	auio.uio_td = curthread;
4619 
4620 	error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4621 				&auio, nd.nl_cred);
4622 	uap->sysmsg_result = uap->nbytes - auio.uio_resid;
4623 
4624 	vput(vp);
4625 	nlookup_done(&nd);
4626 	rel_mplock();
4627 	return(error);
4628 }
4629 
4630 /*
4631  * Syscall to delete a named extended attribute from a file or directory.
4632  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4633  *
4634  * MPALMOSTSAFE
4635  */
4636 int
4637 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4638 {
4639 	char attrname[EXTATTR_MAXNAMELEN];
4640 	struct nlookupdata nd;
4641 	struct vnode *vp;
4642 	int error;
4643 
4644 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4645 	if (error)
4646 		return(error);
4647 
4648 	get_mplock();
4649 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4650 	if (error == 0)
4651 		error = nlookup(&nd);
4652 	if (error == 0)
4653 		error = ncp_writechk(&nd.nl_nch);
4654 	if (error == 0) {
4655 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4656 		if (error == 0) {
4657 			error = VOP_SETEXTATTR(vp, uap->attrnamespace,
4658 					       attrname, NULL, nd.nl_cred);
4659 			vput(vp);
4660 		}
4661 	}
4662 	nlookup_done(&nd);
4663 	rel_mplock();
4664 	return(error);
4665 }
4666 
4667 /*
4668  * Determine if the mount is visible to the process.
4669  */
4670 static int
4671 chroot_visible_mnt(struct mount *mp, struct proc *p)
4672 {
4673 	struct nchandle nch;
4674 
4675 	/*
4676 	 * Traverse from the mount point upwards.  If we hit the process
4677 	 * root then the mount point is visible to the process.
4678 	 */
4679 	nch = mp->mnt_ncmountpt;
4680 	while (nch.ncp) {
4681 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4682 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4683 			return(1);
4684 		}
4685 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4686 			nch = nch.mount->mnt_ncmounton;
4687 		} else {
4688 			nch.ncp = nch.ncp->nc_parent;
4689 		}
4690 	}
4691 
4692 	/*
4693 	 * If the mount point is not visible to the process, but the
4694 	 * process root is in a subdirectory of the mount, return
4695 	 * TRUE anyway.
4696 	 */
4697 	if (p->p_fd->fd_nrdir.mount == mp)
4698 		return(1);
4699 
4700 	return(0);
4701 }
4702 
4703