xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 6ea70f7669242fe9fd042834f455f06a4adfaadf)
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 1.110 2006/12/23 00:35:04 swildner Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/linker.h>
57 #include <sys/stat.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/proc.h>
61 #include <sys/namei.h>
62 #include <sys/nlookup.h>
63 #include <sys/dirent.h>
64 #include <sys/extattr.h>
65 #include <sys/spinlock.h>
66 #include <sys/kern_syscall.h>
67 #include <sys/objcache.h>
68 #include <sys/sysctl.h>
69 #include <sys/file2.h>
70 #include <sys/spinlock2.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 
76 #include <machine/limits.h>
77 #include <machine/stdarg.h>
78 
79 #include <vfs/union/union.h>
80 
81 static void mount_warning(struct mount *mp, const char *ctl, ...);
82 static int checkvp_chdir (struct vnode *vn, struct thread *td);
83 static void checkdirs (struct vnode *olddp, struct nchandle *nch);
84 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
85 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
86 static int getutimes (const struct timeval *, struct timespec *);
87 static int setfown (struct vnode *, uid_t, gid_t);
88 static int setfmode (struct vnode *, int);
89 static int setfflags (struct vnode *, int);
90 static int setutimes (struct vnode *, const struct timespec *, int);
91 static int	usermount = 0;	/* if 1, non-root can mount fs. */
92 
93 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
94 
95 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
96 
97 /*
98  * Virtual File System System Calls
99  */
100 
101 /*
102  * Mount a file system.
103  */
104 /*
105  * mount_args(char *type, char *path, int flags, caddr_t data)
106  */
107 /* ARGSUSED */
108 int
109 sys_mount(struct mount_args *uap)
110 {
111 	struct thread *td = curthread;
112 	struct proc *p = td->td_proc;
113 	struct vnode *vp;
114 	struct nchandle nch;
115 	struct mount *mp;
116 	struct vfsconf *vfsp;
117 	int error, flag = 0, flag2 = 0;
118 	int hasmount;
119 	struct vattr va;
120 	struct nlookupdata nd;
121 	char fstypename[MFSNAMELEN];
122 	struct ucred *cred = p->p_ucred;
123 
124 	KKASSERT(p);
125 	if (cred->cr_prison != NULL)
126 		return (EPERM);
127 	if (usermount == 0 && (error = suser(td)))
128 		return (error);
129 	/*
130 	 * Do not allow NFS export by non-root users.
131 	 */
132 	if (uap->flags & MNT_EXPORTED) {
133 		error = suser(td);
134 		if (error)
135 			return (error);
136 	}
137 	/*
138 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
139 	 */
140 	if (suser(td))
141 		uap->flags |= MNT_NOSUID | MNT_NODEV;
142 
143 	/*
144 	 * Lookup the requested path and extract the nch and vnode.
145 	 */
146 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
147 	if (error == 0) {
148 		if ((error = nlookup(&nd)) == 0) {
149 			if (nd.nl_nch.ncp->nc_vp == NULL)
150 				error = ENOENT;
151 		}
152 	}
153 	if (error) {
154 		nlookup_done(&nd);
155 		return (error);
156 	}
157 
158 	/*
159 	 * Extract the locked+refd ncp and cleanup the nd structure
160 	 */
161 	nch = nd.nl_nch;
162 	cache_zero(&nd.nl_nch);
163 	nlookup_done(&nd);
164 
165 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
166 		hasmount = 1;
167 	else
168 		hasmount = 0;
169 
170 
171 	/*
172 	 * now we have the locked ref'd nch and unreferenced vnode.
173 	 */
174 	vp = nch.ncp->nc_vp;
175 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
176 		cache_put(&nch);
177 		return (error);
178 	}
179 	cache_unlock(&nch);
180 
181 	/*
182 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
183 	 */
184 	if (uap->flags & MNT_UPDATE) {
185 		if ((vp->v_flag & VROOT) == 0) {
186 			cache_drop(&nch);
187 			vput(vp);
188 			return (EINVAL);
189 		}
190 		mp = vp->v_mount;
191 		flag = mp->mnt_flag;
192 		flag2 = mp->mnt_kern_flag;
193 		/*
194 		 * We only allow the filesystem to be reloaded if it
195 		 * is currently mounted read-only.
196 		 */
197 		if ((uap->flags & MNT_RELOAD) &&
198 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
199 			cache_drop(&nch);
200 			vput(vp);
201 			return (EOPNOTSUPP);	/* Needs translation */
202 		}
203 		/*
204 		 * Only root, or the user that did the original mount is
205 		 * permitted to update it.
206 		 */
207 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
208 		    (error = suser(td))) {
209 			cache_drop(&nch);
210 			vput(vp);
211 			return (error);
212 		}
213 		if (vfs_busy(mp, LK_NOWAIT)) {
214 			cache_drop(&nch);
215 			vput(vp);
216 			return (EBUSY);
217 		}
218 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
219 			cache_drop(&nch);
220 			vfs_unbusy(mp);
221 			vput(vp);
222 			return (EBUSY);
223 		}
224 		vp->v_flag |= VMOUNT;
225 		mp->mnt_flag |=
226 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
227 		vn_unlock(vp);
228 		goto update;
229 	}
230 	/*
231 	 * If the user is not root, ensure that they own the directory
232 	 * onto which we are attempting to mount.
233 	 */
234 	if ((error = VOP_GETATTR(vp, &va)) ||
235 	    (va.va_uid != cred->cr_uid && (error = suser(td)))) {
236 		cache_drop(&nch);
237 		vput(vp);
238 		return (error);
239 	}
240 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
241 		cache_drop(&nch);
242 		vput(vp);
243 		return (error);
244 	}
245 	if (vp->v_type != VDIR) {
246 		cache_drop(&nch);
247 		vput(vp);
248 		return (ENOTDIR);
249 	}
250 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
251 		cache_drop(&nch);
252 		vput(vp);
253 		return (error);
254 	}
255 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
256 		if (!strcmp(vfsp->vfc_name, fstypename))
257 			break;
258 	}
259 	if (vfsp == NULL) {
260 		linker_file_t lf;
261 
262 		/* Only load modules for root (very important!) */
263 		if ((error = suser(td)) != 0) {
264 			cache_drop(&nch);
265 			vput(vp);
266 			return error;
267 		}
268 		error = linker_load_file(fstypename, &lf);
269 		if (error || lf == NULL) {
270 			cache_drop(&nch);
271 			vput(vp);
272 			if (lf == NULL)
273 				error = ENODEV;
274 			return error;
275 		}
276 		lf->userrefs++;
277 		/* lookup again, see if the VFS was loaded */
278 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
279 			if (!strcmp(vfsp->vfc_name, fstypename))
280 				break;
281 		}
282 		if (vfsp == NULL) {
283 			lf->userrefs--;
284 			linker_file_unload(lf);
285 			cache_drop(&nch);
286 			vput(vp);
287 			return (ENODEV);
288 		}
289 	}
290 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
291 		cache_drop(&nch);
292 		vput(vp);
293 		return (EBUSY);
294 	}
295 	vp->v_flag |= VMOUNT;
296 
297 	/*
298 	 * Allocate and initialize the filesystem.
299 	 */
300 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
301 	TAILQ_INIT(&mp->mnt_nvnodelist);
302 	TAILQ_INIT(&mp->mnt_reservedvnlist);
303 	TAILQ_INIT(&mp->mnt_jlist);
304 	mp->mnt_nvnodelistsize = 0;
305 	lockinit(&mp->mnt_lock, "vfslock", 0, 0);
306 	vfs_busy(mp, LK_NOWAIT);
307 	mp->mnt_op = vfsp->vfc_vfsops;
308 	mp->mnt_vfc = vfsp;
309 	vfsp->vfc_refcount++;
310 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
311 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
312 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
313 	mp->mnt_stat.f_owner = cred->cr_uid;
314 	mp->mnt_iosize_max = DFLTPHYS;
315 	vn_unlock(vp);
316 update:
317 	/*
318 	 * Set the mount level flags.
319 	 */
320 	if (uap->flags & MNT_RDONLY)
321 		mp->mnt_flag |= MNT_RDONLY;
322 	else if (mp->mnt_flag & MNT_RDONLY)
323 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
324 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
325 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
326 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
327 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
328 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
329 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
330 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
331 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
332 	/*
333 	 * Mount the filesystem.
334 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
335 	 * get.
336 	 */
337 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
338 	if (mp->mnt_flag & MNT_UPDATE) {
339 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
340 			mp->mnt_flag &= ~MNT_RDONLY;
341 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
342 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
343 		if (error) {
344 			mp->mnt_flag = flag;
345 			mp->mnt_kern_flag = flag2;
346 		}
347 		vfs_unbusy(mp);
348 		vp->v_flag &= ~VMOUNT;
349 		vrele(vp);
350 		cache_drop(&nch);
351 		return (error);
352 	}
353 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
354 	/*
355 	 * Put the new filesystem on the mount list after root.  The mount
356 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
357 	 * up) which represents the root of the mount.  The lookup code
358 	 * detects the mount point going forward and checks the root of
359 	 * the mount going backwards.
360 	 *
361 	 * It is not necessary to invalidate or purge the vnode underneath
362 	 * because elements under the mount will be given their own glue
363 	 * namecache record.
364 	 */
365 	if (!error) {
366 		if (mp->mnt_ncmountpt.ncp == NULL) {
367 			/*
368 			 * allocate, then unlock, but leave the ref intact
369 			 */
370 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
371 			cache_unlock(&mp->mnt_ncmountpt);
372 		}
373 		mp->mnt_ncmounton = nch;		/* inherits ref */
374 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
375 
376 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
377 		vp->v_flag &= ~VMOUNT;
378 		mountlist_insert(mp, MNTINS_LAST);
379 		checkdirs(vp, &mp->mnt_ncmounton);
380 		vn_unlock(vp);
381 		error = vfs_allocate_syncvnode(mp);
382 		vfs_unbusy(mp);
383 		error = VFS_START(mp, 0);
384 		vrele(vp);
385 	} else {
386 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
387 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
388 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
389 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
390 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
391 		vp->v_flag &= ~VMOUNT;
392 		mp->mnt_vfc->vfc_refcount--;
393 		vfs_unbusy(mp);
394 		kfree(mp, M_MOUNT);
395 		cache_drop(&nch);
396 		vput(vp);
397 	}
398 	return (error);
399 }
400 
401 /*
402  * Scan all active processes to see if any of them have a current
403  * or root directory onto which the new filesystem has just been
404  * mounted. If so, replace them with the new mount point.
405  *
406  * The passed ncp is ref'd and locked (from the mount code) and
407  * must be associated with the vnode representing the root of the
408  * mount point.
409  */
410 struct checkdirs_info {
411 	struct vnode *olddp;
412 	struct vnode *newdp;
413 	struct nchandle nch;
414 };
415 
416 static int checkdirs_callback(struct proc *p, void *data);
417 
418 static void
419 checkdirs(struct vnode *olddp, struct nchandle *nch)
420 {
421 	struct checkdirs_info info;
422 	struct vnode *newdp;
423 	struct mount *mp;
424 
425 	if (olddp->v_usecount == 1)
426 		return;
427 	mp = nch->mount;
428 	if (VFS_ROOT(mp, &newdp))
429 		panic("mount: lost mount");
430 	cache_setunresolved(nch);
431 	cache_setvp(nch, newdp);
432 
433 	if (rootvnode == olddp) {
434 		vref(newdp);
435 		vfs_cache_setroot(newdp, cache_hold(nch));
436 	}
437 
438 	info.olddp = olddp;
439 	info.newdp = newdp;
440 	info.nch = *nch;
441 	allproc_scan(checkdirs_callback, &info);
442 	vput(newdp);
443 }
444 
445 /*
446  * NOTE: callback is not MP safe because the scanned process's filedesc
447  * structure can be ripped out from under us, amoung other things.
448  */
449 static int
450 checkdirs_callback(struct proc *p, void *data)
451 {
452 	struct checkdirs_info *info = data;
453 	struct filedesc *fdp;
454 	struct nchandle ncdrop1;
455 	struct nchandle ncdrop2;
456 	struct vnode *vprele1;
457 	struct vnode *vprele2;
458 
459 	if ((fdp = p->p_fd) != NULL) {
460 		cache_zero(&ncdrop1);
461 		cache_zero(&ncdrop2);
462 		vprele1 = NULL;
463 		vprele2 = NULL;
464 
465 		/*
466 		 * MPUNSAFE - XXX fdp can be pulled out from under a
467 		 * foreign process.
468 		 *
469 		 * A shared filedesc is ok, we don't have to copy it
470 		 * because we are making this change globally.
471 		 */
472 		spin_lock_wr(&fdp->fd_spin);
473 		if (fdp->fd_cdir == info->olddp) {
474 			vprele1 = fdp->fd_cdir;
475 			vref(info->newdp);
476 			fdp->fd_cdir = info->newdp;
477 			ncdrop1 = fdp->fd_ncdir;
478 			cache_copy(&info->nch, &fdp->fd_ncdir);
479 		}
480 		if (fdp->fd_rdir == info->olddp) {
481 			vprele2 = fdp->fd_rdir;
482 			vref(info->newdp);
483 			fdp->fd_rdir = info->newdp;
484 			ncdrop2 = fdp->fd_nrdir;
485 			cache_copy(&info->nch, &fdp->fd_nrdir);
486 		}
487 		spin_unlock_wr(&fdp->fd_spin);
488 		if (ncdrop1.ncp)
489 			cache_drop(&ncdrop1);
490 		if (ncdrop2.ncp)
491 			cache_drop(&ncdrop2);
492 		if (vprele1)
493 			vrele(vprele1);
494 		if (vprele2)
495 			vrele(vprele2);
496 	}
497 	return(0);
498 }
499 
500 /*
501  * Unmount a file system.
502  *
503  * Note: unmount takes a path to the vnode mounted on as argument,
504  * not special file (as before).
505  */
506 /*
507  * umount_args(char *path, int flags)
508  */
509 /* ARGSUSED */
510 int
511 sys_unmount(struct unmount_args *uap)
512 {
513 	struct thread *td = curthread;
514 	struct proc *p = td->td_proc;
515 	struct mount *mp = NULL;
516 	int error;
517 	struct nlookupdata nd;
518 
519 	KKASSERT(p);
520 	if (p->p_ucred->cr_prison != NULL)
521 		return (EPERM);
522 	if (usermount == 0 && (error = suser(td)))
523 		return (error);
524 
525 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
526 	if (error == 0)
527 		error = nlookup(&nd);
528 	if (error)
529 		goto out;
530 
531 	mp = nd.nl_nch.mount;
532 
533 	/*
534 	 * Only root, or the user that did the original mount is
535 	 * permitted to unmount this filesystem.
536 	 */
537 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
538 	    (error = suser(td)))
539 		goto out;
540 
541 	/*
542 	 * Don't allow unmounting the root file system.
543 	 */
544 	if (mp->mnt_flag & MNT_ROOTFS) {
545 		error = EINVAL;
546 		goto out;
547 	}
548 
549 	/*
550 	 * Must be the root of the filesystem
551 	 */
552 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
553 		error = EINVAL;
554 		goto out;
555 	}
556 
557 out:
558 	nlookup_done(&nd);
559 	if (error)
560 		return (error);
561 	return (dounmount(mp, uap->flags));
562 }
563 
564 /*
565  * Do the actual file system unmount.
566  */
567 static int
568 dounmount_interlock(struct mount *mp)
569 {
570 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
571 		return (EBUSY);
572 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
573 	return(0);
574 }
575 
576 int
577 dounmount(struct mount *mp, int flags)
578 {
579 	struct namecache *ncp;
580 	struct nchandle nch;
581 	int error;
582 	int async_flag;
583 	int lflags;
584 	int freeok = 1;
585 
586 	/*
587 	 * Exclusive access for unmounting purposes
588 	 */
589 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
590 		return (error);
591 
592 	/*
593 	 * Allow filesystems to detect that a forced unmount is in progress.
594 	 */
595 	if (flags & MNT_FORCE)
596 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
597 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
598 	error = lockmgr(&mp->mnt_lock, lflags);
599 	if (error) {
600 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
601 		if (mp->mnt_kern_flag & MNTK_MWAIT)
602 			wakeup(mp);
603 		return (error);
604 	}
605 
606 	if (mp->mnt_flag & MNT_EXPUBLIC)
607 		vfs_setpublicfs(NULL, NULL, NULL);
608 
609 	vfs_msync(mp, MNT_WAIT);
610 	async_flag = mp->mnt_flag & MNT_ASYNC;
611 	mp->mnt_flag &=~ MNT_ASYNC;
612 
613 	/*
614 	 * If this filesystem isn't aliasing other filesystems,
615 	 * try to invalidate any remaining namecache entries and
616 	 * check the count afterwords.
617 	 */
618 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
619 		cache_lock(&mp->mnt_ncmountpt);
620 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
621 		cache_unlock(&mp->mnt_ncmountpt);
622 
623 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
624 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
625 
626 			if ((flags & MNT_FORCE) == 0) {
627 				error = EBUSY;
628 				mount_warning(mp, "Cannot unmount: "
629 						  "%d namecache "
630 						  "references still "
631 						  "present",
632 						  ncp->nc_refs - 1);
633 			} else {
634 				mount_warning(mp, "Forced unmount: "
635 						  "%d namecache "
636 						  "references still "
637 						  "present",
638 						  ncp->nc_refs - 1);
639 				freeok = 0;
640 			}
641 		}
642 	}
643 
644 	/*
645 	 * nchandle records ref the mount structure.  Expect a count of 1
646 	 * (our mount->mnt_ncmountpt).
647 	 */
648 	if (mp->mnt_refs != 1) {
649 		if ((flags & MNT_FORCE) == 0) {
650 			mount_warning(mp, "Cannot unmount: "
651 					  "%d process references still "
652 					  "present", mp->mnt_refs);
653 			error = EBUSY;
654 		} else {
655 			mount_warning(mp, "Forced unmount: "
656 					  "%d process references still "
657 					  "present", mp->mnt_refs);
658 			freeok = 0;
659 		}
660 	}
661 
662 	if (error == 0) {
663 		if (mp->mnt_syncer != NULL)
664 			vrele(mp->mnt_syncer);
665 		if (((mp->mnt_flag & MNT_RDONLY) ||
666 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
667 		    (flags & MNT_FORCE)) {
668 			error = VFS_UNMOUNT(mp, flags);
669 		}
670 	}
671 	if (error) {
672 		if (mp->mnt_syncer == NULL)
673 			vfs_allocate_syncvnode(mp);
674 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
675 		mp->mnt_flag |= async_flag;
676 		lockmgr(&mp->mnt_lock, LK_RELEASE);
677 		if (mp->mnt_kern_flag & MNTK_MWAIT)
678 			wakeup(mp);
679 		return (error);
680 	}
681 	/*
682 	 * Clean up any journals still associated with the mount after
683 	 * filesystem activity has ceased.
684 	 */
685 	journal_remove_all_journals(mp,
686 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
687 
688 	mountlist_remove(mp);
689 
690 	/*
691 	 * Remove any installed vnode ops here so the individual VFSs don't
692 	 * have to.
693 	 */
694 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
695 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
696 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
697 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
698 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
699 
700 	if (mp->mnt_ncmountpt.ncp != NULL) {
701 		nch = mp->mnt_ncmountpt;
702 		cache_zero(&mp->mnt_ncmountpt);
703 		cache_clrmountpt(&nch);
704 		cache_drop(&nch);
705 	}
706 	if (mp->mnt_ncmounton.ncp != NULL) {
707 		nch = mp->mnt_ncmounton;
708 		cache_zero(&mp->mnt_ncmounton);
709 		cache_clrmountpt(&nch);
710 		cache_drop(&nch);
711 	}
712 
713 	mp->mnt_vfc->vfc_refcount--;
714 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
715 		panic("unmount: dangling vnode");
716 	lockmgr(&mp->mnt_lock, LK_RELEASE);
717 	if (mp->mnt_kern_flag & MNTK_MWAIT)
718 		wakeup(mp);
719 	if (freeok)
720 		kfree(mp, M_MOUNT);
721 	return (0);
722 }
723 
724 static
725 void
726 mount_warning(struct mount *mp, const char *ctl, ...)
727 {
728 	char *ptr;
729 	char *buf;
730 	__va_list va;
731 
732 	__va_start(va, ctl);
733 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
734 		kprintf("unmount(%s): ", ptr);
735 		kvprintf(ctl, va);
736 		kprintf("\n");
737 		kfree(buf, M_TEMP);
738 	} else {
739 		kprintf("unmount(%p): ", mp);
740 		kvprintf(ctl, va);
741 		kprintf("\n");
742 	}
743 	__va_end(va);
744 }
745 
746 /*
747  * Sync each mounted filesystem.
748  */
749 
750 #ifdef DEBUG
751 static int syncprt = 0;
752 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
753 #endif /* DEBUG */
754 
755 static int sync_callback(struct mount *mp, void *data);
756 
757 /* ARGSUSED */
758 int
759 sys_sync(struct sync_args *uap)
760 {
761 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
762 #ifdef DEBUG
763 	/*
764 	 * print out buffer pool stat information on each sync() call.
765 	 */
766 	if (syncprt)
767 		vfs_bufstats();
768 #endif /* DEBUG */
769 	return (0);
770 }
771 
772 static
773 int
774 sync_callback(struct mount *mp, void *data __unused)
775 {
776 	int asyncflag;
777 
778 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
779 		asyncflag = mp->mnt_flag & MNT_ASYNC;
780 		mp->mnt_flag &= ~MNT_ASYNC;
781 		vfs_msync(mp, MNT_NOWAIT);
782 		VFS_SYNC(mp, MNT_NOWAIT);
783 		mp->mnt_flag |= asyncflag;
784 	}
785 	return(0);
786 }
787 
788 /* XXX PRISON: could be per prison flag */
789 static int prison_quotas;
790 #if 0
791 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
792 #endif
793 
794 /*
795  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
796  *
797  * Change filesystem quotas.
798  */
799 /* ARGSUSED */
800 int
801 sys_quotactl(struct quotactl_args *uap)
802 {
803 	struct nlookupdata nd;
804 	struct thread *td;
805 	struct proc *p;
806 	struct mount *mp;
807 	int error;
808 
809 	td = curthread;
810 	p = td->td_proc;
811 	if (p->p_ucred->cr_prison && !prison_quotas)
812 		return (EPERM);
813 
814 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
815 	if (error == 0)
816 		error = nlookup(&nd);
817 	if (error == 0) {
818 		mp = nd.nl_nch.mount;
819 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
820 				    uap->arg, nd.nl_cred);
821 	}
822 	nlookup_done(&nd);
823 	return (error);
824 }
825 
826 /*
827  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
828  *		void *buf, int buflen)
829  *
830  * This function operates on a mount point and executes the specified
831  * operation using the specified control data, and possibly returns data.
832  *
833  * The actual number of bytes stored in the result buffer is returned, 0
834  * if none, otherwise an error is returned.
835  */
836 /* ARGSUSED */
837 int
838 sys_mountctl(struct mountctl_args *uap)
839 {
840 	struct thread *td = curthread;
841 	struct proc *p = td->td_proc;
842 	struct file *fp;
843 	void *ctl = NULL;
844 	void *buf = NULL;
845 	char *path = NULL;
846 	int error;
847 
848 	/*
849 	 * Sanity and permissions checks.  We must be root.
850 	 */
851 	KKASSERT(p);
852 	if (p->p_ucred->cr_prison != NULL)
853 		return (EPERM);
854 	if ((error = suser(td)) != 0)
855 		return (error);
856 
857 	/*
858 	 * Argument length checks
859 	 */
860 	if (uap->ctllen < 0 || uap->ctllen > 1024)
861 		return (EINVAL);
862 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
863 		return (EINVAL);
864 	if (uap->path == NULL)
865 		return (EINVAL);
866 
867 	/*
868 	 * Allocate the necessary buffers and copyin data
869 	 */
870 	path = objcache_get(namei_oc, M_WAITOK);
871 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
872 	if (error)
873 		goto done;
874 
875 	if (uap->ctllen) {
876 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
877 		error = copyin(uap->ctl, ctl, uap->ctllen);
878 		if (error)
879 			goto done;
880 	}
881 	if (uap->buflen)
882 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
883 
884 	/*
885 	 * Validate the descriptor
886 	 */
887 	fp = holdfp(p->p_fd, uap->fd, -1);
888 	if (fp == NULL) {
889 		error = EBADF;
890 		goto done;
891 	}
892 
893 	/*
894 	 * Execute the internal kernel function and clean up.
895 	 */
896 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
897 	if (fp)
898 		fdrop(fp);
899 	if (error == 0 && uap->sysmsg_result > 0)
900 		error = copyout(buf, uap->buf, uap->sysmsg_result);
901 done:
902 	if (path)
903 		objcache_put(namei_oc, path);
904 	if (ctl)
905 		kfree(ctl, M_TEMP);
906 	if (buf)
907 		kfree(buf, M_TEMP);
908 	return (error);
909 }
910 
911 /*
912  * Execute a mount control operation by resolving the path to a mount point
913  * and calling vop_mountctl().
914  */
915 int
916 kern_mountctl(const char *path, int op, struct file *fp,
917 		const void *ctl, int ctllen,
918 		void *buf, int buflen, int *res)
919 {
920 	struct vnode *vp;
921 	struct mount *mp;
922 	struct nlookupdata nd;
923 	int error;
924 
925 	*res = 0;
926 	vp = NULL;
927 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
928 	if (error == 0)
929 		error = nlookup(&nd);
930 	if (error == 0)
931 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
932 	nlookup_done(&nd);
933 	if (error)
934 		return (error);
935 
936 	mp = vp->v_mount;
937 
938 	/*
939 	 * Must be the root of the filesystem
940 	 */
941 	if ((vp->v_flag & VROOT) == 0) {
942 		vput(vp);
943 		return (EINVAL);
944 	}
945 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
946 				buf, buflen, res);
947 	vput(vp);
948 	return (error);
949 }
950 
951 int
952 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
953 {
954 	struct thread *td = curthread;
955 	struct proc *p = td->td_proc;
956 	struct mount *mp;
957 	struct statfs *sp;
958 	char *fullpath, *freepath;
959 	int error;
960 
961 	if ((error = nlookup(nd)) != 0)
962 		return (error);
963 	mp = nd->nl_nch.mount;
964 	sp = &mp->mnt_stat;
965 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
966 		return (error);
967 
968 	error = cache_fullpath(p, &mp->mnt_ncmountpt, &fullpath, &freepath);
969 	if (error)
970 		return(error);
971 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
972 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
973 	kfree(freepath, M_TEMP);
974 
975 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
976 	bcopy(sp, buf, sizeof(*buf));
977 	/* Only root should have access to the fsid's. */
978 	if (suser(td))
979 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
980 	return (0);
981 }
982 
983 /*
984  * statfs_args(char *path, struct statfs *buf)
985  *
986  * Get filesystem statistics.
987  */
988 int
989 sys_statfs(struct statfs_args *uap)
990 {
991 	struct nlookupdata nd;
992 	struct statfs buf;
993 	int error;
994 
995 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
996 	if (error == 0)
997 		error = kern_statfs(&nd, &buf);
998 	nlookup_done(&nd);
999 	if (error == 0)
1000 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1001 	return (error);
1002 }
1003 
1004 int
1005 kern_fstatfs(int fd, struct statfs *buf)
1006 {
1007 	struct thread *td = curthread;
1008 	struct proc *p = td->td_proc;
1009 	struct file *fp;
1010 	struct mount *mp;
1011 	struct statfs *sp;
1012 	char *fullpath, *freepath;
1013 	int error;
1014 
1015 	KKASSERT(p);
1016 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1017 		return (error);
1018 	mp = ((struct vnode *)fp->f_data)->v_mount;
1019 	if (mp == NULL) {
1020 		error = EBADF;
1021 		goto done;
1022 	}
1023 	if (fp->f_cred == NULL) {
1024 		error = EINVAL;
1025 		goto done;
1026 	}
1027 	sp = &mp->mnt_stat;
1028 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1029 		goto done;
1030 
1031 	if ((error = cache_fullpath(p, &mp->mnt_ncmountpt, &fullpath, &freepath)) != 0)
1032 		goto done;
1033 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1034 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1035 	kfree(freepath, M_TEMP);
1036 
1037 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1038 	bcopy(sp, buf, sizeof(*buf));
1039 
1040 	/* Only root should have access to the fsid's. */
1041 	if (suser(td))
1042 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1043 	error = 0;
1044 done:
1045 	fdrop(fp);
1046 	return (error);
1047 }
1048 
1049 /*
1050  * fstatfs_args(int fd, struct statfs *buf)
1051  *
1052  * Get filesystem statistics.
1053  */
1054 int
1055 sys_fstatfs(struct fstatfs_args *uap)
1056 {
1057 	struct statfs buf;
1058 	int error;
1059 
1060 	error = kern_fstatfs(uap->fd, &buf);
1061 
1062 	if (error == 0)
1063 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1064 	return (error);
1065 }
1066 
1067 /*
1068  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1069  *
1070  * Get statistics on all filesystems.
1071  */
1072 
1073 struct getfsstat_info {
1074 	struct statfs *sfsp;
1075 	long count;
1076 	long maxcount;
1077 	int error;
1078 	int flags;
1079 	struct proc *p;
1080 };
1081 
1082 static int getfsstat_callback(struct mount *, void *);
1083 
1084 /* ARGSUSED */
1085 int
1086 sys_getfsstat(struct getfsstat_args *uap)
1087 {
1088 	struct thread *td = curthread;
1089 	struct proc *p = td->td_proc;
1090 	struct getfsstat_info info;
1091 
1092 	bzero(&info, sizeof(info));
1093 
1094 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1095 	info.sfsp = uap->buf;
1096 	info.count = 0;
1097 	info.flags = uap->flags;
1098 	info.p = p;
1099 
1100 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1101 	if (info.sfsp && info.count > info.maxcount)
1102 		uap->sysmsg_result = info.maxcount;
1103 	else
1104 		uap->sysmsg_result = info.count;
1105 	return (info.error);
1106 }
1107 
1108 static int
1109 getfsstat_callback(struct mount *mp, void *data)
1110 {
1111 	struct getfsstat_info *info = data;
1112 	struct statfs *sp;
1113 	char *freepath;
1114 	char *fullpath;
1115 	int error;
1116 
1117 	if (info->sfsp && info->count < info->maxcount) {
1118 		if (info->p && !chroot_visible_mnt(mp, info->p))
1119 			return(0);
1120 		sp = &mp->mnt_stat;
1121 
1122 		/*
1123 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1124 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1125 		 * overrides MNT_WAIT.
1126 		 */
1127 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1128 		    (info->flags & MNT_WAIT)) &&
1129 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1130 			return(0);
1131 		}
1132 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1133 
1134 		error = cache_fullpath(info->p, &mp->mnt_ncmountpt,
1135 					&fullpath, &freepath);
1136 		if (error) {
1137 			info->error = error;
1138 			return(-1);
1139 		}
1140 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1141 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1142 		kfree(freepath, M_TEMP);
1143 
1144 		error = copyout(sp, info->sfsp, sizeof(*sp));
1145 		if (error) {
1146 			info->error = error;
1147 			return (-1);
1148 		}
1149 		++info->sfsp;
1150 	}
1151 	info->count++;
1152 	return(0);
1153 }
1154 
1155 /*
1156  * fchdir_args(int fd)
1157  *
1158  * Change current working directory to a given file descriptor.
1159  */
1160 /* ARGSUSED */
1161 int
1162 sys_fchdir(struct fchdir_args *uap)
1163 {
1164 	struct thread *td = curthread;
1165 	struct proc *p = td->td_proc;
1166 	struct filedesc *fdp = p->p_fd;
1167 	struct vnode *vp, *ovp;
1168 	struct mount *mp;
1169 	struct file *fp;
1170 	struct nchandle nch, onch, tnch;
1171 	int error;
1172 
1173 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1174 		return (error);
1175 	vp = (struct vnode *)fp->f_data;
1176 	vref(vp);
1177 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1178 	if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1179 		error = ENOTDIR;
1180 	else
1181 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1182 	if (error) {
1183 		vput(vp);
1184 		fdrop(fp);
1185 		return (error);
1186 	}
1187 	cache_copy(&fp->f_nchandle, &nch);
1188 
1189 	/*
1190 	 * If the ncp has become a mount point, traverse through
1191 	 * the mount point.
1192 	 */
1193 
1194 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1195 	       (mp = cache_findmount(&nch)) != NULL
1196 	) {
1197 		error = nlookup_mp(mp, &tnch);
1198 		if (error == 0) {
1199 			cache_unlock(&tnch);	/* leave ref intact */
1200 			vput(vp);
1201 			vp = tnch.ncp->nc_vp;
1202 			error = vget(vp, LK_SHARED);
1203 			KKASSERT(error == 0);
1204 			cache_drop(&nch);
1205 			nch = tnch;
1206 		}
1207 	}
1208 	if (error == 0) {
1209 		ovp = fdp->fd_cdir;
1210 		onch = fdp->fd_ncdir;
1211 		vn_unlock(vp);		/* leave ref intact */
1212 		fdp->fd_cdir = vp;
1213 		fdp->fd_ncdir = nch;
1214 		cache_drop(&onch);
1215 		vrele(ovp);
1216 	} else {
1217 		cache_drop(&nch);
1218 		vput(vp);
1219 	}
1220 	fdrop(fp);
1221 	return (error);
1222 }
1223 
1224 int
1225 kern_chdir(struct nlookupdata *nd)
1226 {
1227 	struct thread *td = curthread;
1228 	struct proc *p = td->td_proc;
1229 	struct filedesc *fdp = p->p_fd;
1230 	struct vnode *vp, *ovp;
1231 	struct nchandle onch;
1232 	int error;
1233 
1234 	if ((error = nlookup(nd)) != 0)
1235 		return (error);
1236 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1237 		return (ENOENT);
1238 	if ((error = vget(vp, LK_SHARED)) != 0)
1239 		return (error);
1240 
1241 	error = checkvp_chdir(vp, td);
1242 	vn_unlock(vp);
1243 	if (error == 0) {
1244 		ovp = fdp->fd_cdir;
1245 		onch = fdp->fd_ncdir;
1246 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1247 		fdp->fd_ncdir = nd->nl_nch;
1248 		fdp->fd_cdir = vp;
1249 		cache_drop(&onch);
1250 		vrele(ovp);
1251 		cache_zero(&nd->nl_nch);
1252 	} else {
1253 		vrele(vp);
1254 	}
1255 	return (error);
1256 }
1257 
1258 /*
1259  * chdir_args(char *path)
1260  *
1261  * Change current working directory (``.'').
1262  */
1263 int
1264 sys_chdir(struct chdir_args *uap)
1265 {
1266 	struct nlookupdata nd;
1267 	int error;
1268 
1269 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1270 	if (error == 0)
1271 		error = kern_chdir(&nd);
1272 	nlookup_done(&nd);
1273 	return (error);
1274 }
1275 
1276 /*
1277  * Helper function for raised chroot(2) security function:  Refuse if
1278  * any filedescriptors are open directories.
1279  */
1280 static int
1281 chroot_refuse_vdir_fds(fdp)
1282 	struct filedesc *fdp;
1283 {
1284 	struct vnode *vp;
1285 	struct file *fp;
1286 	int error;
1287 	int fd;
1288 
1289 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1290 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1291 			continue;
1292 		vp = (struct vnode *)fp->f_data;
1293 		if (vp->v_type != VDIR) {
1294 			fdrop(fp);
1295 			continue;
1296 		}
1297 		fdrop(fp);
1298 		return(EPERM);
1299 	}
1300 	return (0);
1301 }
1302 
1303 /*
1304  * This sysctl determines if we will allow a process to chroot(2) if it
1305  * has a directory open:
1306  *	0: disallowed for all processes.
1307  *	1: allowed for processes that were not already chroot(2)'ed.
1308  *	2: allowed for all processes.
1309  */
1310 
1311 static int chroot_allow_open_directories = 1;
1312 
1313 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1314      &chroot_allow_open_directories, 0, "");
1315 
1316 /*
1317  * chroot to the specified namecache entry.  We obtain the vp from the
1318  * namecache data.  The passed ncp must be locked and referenced and will
1319  * remain locked and referenced on return.
1320  */
1321 int
1322 kern_chroot(struct nchandle *nch)
1323 {
1324 	struct thread *td = curthread;
1325 	struct proc *p = td->td_proc;
1326 	struct filedesc *fdp = p->p_fd;
1327 	struct vnode *vp;
1328 	int error;
1329 
1330 	/*
1331 	 * Only root can chroot
1332 	 */
1333 	if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1334 		return (error);
1335 
1336 	/*
1337 	 * Disallow open directory descriptors (fchdir() breakouts).
1338 	 */
1339 	if (chroot_allow_open_directories == 0 ||
1340 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1341 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1342 			return (error);
1343 	}
1344 	if ((vp = nch->ncp->nc_vp) == NULL)
1345 		return (ENOENT);
1346 
1347 	if ((error = vget(vp, LK_SHARED)) != 0)
1348 		return (error);
1349 
1350 	/*
1351 	 * Check the validity of vp as a directory to change to and
1352 	 * associate it with rdir/jdir.
1353 	 */
1354 	error = checkvp_chdir(vp, td);
1355 	vn_unlock(vp);			/* leave reference intact */
1356 	if (error == 0) {
1357 		vrele(fdp->fd_rdir);
1358 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1359 		cache_drop(&fdp->fd_nrdir);
1360 		cache_copy(nch, &fdp->fd_nrdir);
1361 		if (fdp->fd_jdir == NULL) {
1362 			fdp->fd_jdir = vp;
1363 			vref(fdp->fd_jdir);
1364 			cache_copy(nch, &fdp->fd_njdir);
1365 		}
1366 	} else {
1367 		vrele(vp);
1368 	}
1369 	return (error);
1370 }
1371 
1372 /*
1373  * chroot_args(char *path)
1374  *
1375  * Change notion of root (``/'') directory.
1376  */
1377 /* ARGSUSED */
1378 int
1379 sys_chroot(struct chroot_args *uap)
1380 {
1381 	struct thread *td = curthread;
1382 	struct nlookupdata nd;
1383 	int error;
1384 
1385 	KKASSERT(td->td_proc);
1386 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1387 	if (error) {
1388 		nlookup_done(&nd);
1389 		return(error);
1390 	}
1391 	error = nlookup(&nd);
1392 	if (error == 0)
1393 		error = kern_chroot(&nd.nl_nch);
1394 	nlookup_done(&nd);
1395 	return(error);
1396 }
1397 
1398 /*
1399  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1400  * determine whether it is legal to chdir to the vnode.  The vnode's state
1401  * is not changed by this call.
1402  */
1403 int
1404 checkvp_chdir(struct vnode *vp, struct thread *td)
1405 {
1406 	int error;
1407 
1408 	if (vp->v_type != VDIR)
1409 		error = ENOTDIR;
1410 	else
1411 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1412 	return (error);
1413 }
1414 
1415 int
1416 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1417 {
1418 	struct thread *td = curthread;
1419 	struct proc *p = td->td_proc;
1420 	struct lwp *lp = td->td_lwp;
1421 	struct filedesc *fdp = p->p_fd;
1422 	int cmode, flags;
1423 	struct file *nfp;
1424 	struct file *fp;
1425 	struct vnode *vp;
1426 	int type, indx, error;
1427 	struct flock lf;
1428 
1429 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1430 		return (EINVAL);
1431 	flags = FFLAGS(oflags);
1432 	error = falloc(p, &nfp, NULL);
1433 	if (error)
1434 		return (error);
1435 	fp = nfp;
1436 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1437 
1438 	/*
1439 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1440 	 * file descriptor to be duplicated rather then doing the open
1441 	 * itself.
1442 	 */
1443 	lp->lwp_dupfd = -1;
1444 
1445 	/*
1446 	 * Call vn_open() to do the lookup and assign the vnode to the
1447 	 * file pointer.  vn_open() does not change the ref count on fp
1448 	 * and the vnode, on success, will be inherited by the file pointer
1449 	 * and unlocked.
1450 	 */
1451 	nd->nl_flags |= NLC_LOCKVP;
1452 	error = vn_open(nd, fp, flags, cmode);
1453 	nlookup_done(nd);
1454 	if (error) {
1455 		/*
1456 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1457 		 * responsible for dropping the old contents of ofiles[indx]
1458 		 * if it succeeds.
1459 		 *
1460 		 * Note that fsetfd() will add a ref to fp which represents
1461 		 * the fd_files[] assignment.  We must still drop our
1462 		 * reference.
1463 		 */
1464 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1465 			if (fdalloc(p, 0, &indx) == 0) {
1466 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1467 				if (error == 0) {
1468 					*res = indx;
1469 					fdrop(fp);	/* our ref */
1470 					return (0);
1471 				}
1472 				fsetfd(p, NULL, indx);
1473 			}
1474 		}
1475 		fdrop(fp);	/* our ref */
1476 		if (error == ERESTART)
1477 			error = EINTR;
1478 		return (error);
1479 	}
1480 
1481 	/*
1482 	 * ref the vnode for ourselves so it can't be ripped out from under
1483 	 * is.  XXX need an ND flag to request that the vnode be returned
1484 	 * anyway.
1485 	 *
1486 	 * Reserve a file descriptor but do not assign it until the open
1487 	 * succeeds.
1488 	 */
1489 	vp = (struct vnode *)fp->f_data;
1490 	vref(vp);
1491 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1492 		fdrop(fp);
1493 		vrele(vp);
1494 		return (error);
1495 	}
1496 
1497 	/*
1498 	 * If no error occurs the vp will have been assigned to the file
1499 	 * pointer.
1500 	 */
1501 	lp->lwp_dupfd = 0;
1502 
1503 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1504 		lf.l_whence = SEEK_SET;
1505 		lf.l_start = 0;
1506 		lf.l_len = 0;
1507 		if (flags & O_EXLOCK)
1508 			lf.l_type = F_WRLCK;
1509 		else
1510 			lf.l_type = F_RDLCK;
1511 		if (flags & FNONBLOCK)
1512 			type = 0;
1513 		else
1514 			type = F_WAIT;
1515 
1516 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1517 			/*
1518 			 * lock request failed.  Clean up the reserved
1519 			 * descriptor.
1520 			 */
1521 			vrele(vp);
1522 			fsetfd(p, NULL, indx);
1523 			fdrop(fp);
1524 			return (error);
1525 		}
1526 		fp->f_flag |= FHASLOCK;
1527 	}
1528 #if 0
1529 	/*
1530 	 * Assert that all regular file vnodes were created with a object.
1531 	 */
1532 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1533 		("open: regular file has no backing object after vn_open"));
1534 #endif
1535 
1536 	vrele(vp);
1537 
1538 	/*
1539 	 * release our private reference, leaving the one associated with the
1540 	 * descriptor table intact.
1541 	 */
1542 	fsetfd(p, fp, indx);
1543 	fdrop(fp);
1544 	*res = indx;
1545 	return (0);
1546 }
1547 
1548 /*
1549  * open_args(char *path, int flags, int mode)
1550  *
1551  * Check permissions, allocate an open file structure,
1552  * and call the device open routine if any.
1553  */
1554 int
1555 sys_open(struct open_args *uap)
1556 {
1557 	struct nlookupdata nd;
1558 	int error;
1559 
1560 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1561 	if (error == 0) {
1562 		error = kern_open(&nd, uap->flags,
1563 				    uap->mode, &uap->sysmsg_result);
1564 	}
1565 	nlookup_done(&nd);
1566 	return (error);
1567 }
1568 
1569 int
1570 kern_mknod(struct nlookupdata *nd, int mode, int dev)
1571 {
1572 	struct thread *td = curthread;
1573 	struct proc *p = td->td_proc;
1574 	struct vnode *vp;
1575 	struct vattr vattr;
1576 	int error;
1577 	int whiteout = 0;
1578 
1579 	KKASSERT(p);
1580 
1581 	switch (mode & S_IFMT) {
1582 	case S_IFCHR:
1583 	case S_IFBLK:
1584 		error = suser(td);
1585 		break;
1586 	default:
1587 		error = suser_cred(p->p_ucred, PRISON_ROOT);
1588 		break;
1589 	}
1590 	if (error)
1591 		return (error);
1592 
1593 	bwillwrite();
1594 	nd->nl_flags |= NLC_CREATE;
1595 	if ((error = nlookup(nd)) != 0)
1596 		return (error);
1597 	if (nd->nl_nch.ncp->nc_vp)
1598 		return (EEXIST);
1599 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1600 		return (error);
1601 
1602 	VATTR_NULL(&vattr);
1603 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1604 	vattr.va_rdev = dev;
1605 	whiteout = 0;
1606 
1607 	switch (mode & S_IFMT) {
1608 	case S_IFMT:	/* used by badsect to flag bad sectors */
1609 		vattr.va_type = VBAD;
1610 		break;
1611 	case S_IFCHR:
1612 		vattr.va_type = VCHR;
1613 		break;
1614 	case S_IFBLK:
1615 		vattr.va_type = VBLK;
1616 		break;
1617 	case S_IFWHT:
1618 		whiteout = 1;
1619 		break;
1620 	default:
1621 		error = EINVAL;
1622 		break;
1623 	}
1624 	if (error == 0) {
1625 		if (whiteout) {
1626 			error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_cred, NAMEI_CREATE);
1627 		} else {
1628 			vp = NULL;
1629 			error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1630 			if (error == 0)
1631 				vput(vp);
1632 		}
1633 	}
1634 	return (error);
1635 }
1636 
1637 /*
1638  * mknod_args(char *path, int mode, int dev)
1639  *
1640  * Create a special file.
1641  */
1642 int
1643 sys_mknod(struct mknod_args *uap)
1644 {
1645 	struct nlookupdata nd;
1646 	int error;
1647 
1648 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1649 	if (error == 0)
1650 		error = kern_mknod(&nd, uap->mode, uap->dev);
1651 	nlookup_done(&nd);
1652 	return (error);
1653 }
1654 
1655 int
1656 kern_mkfifo(struct nlookupdata *nd, int mode)
1657 {
1658 	struct thread *td = curthread;
1659 	struct proc *p = td->td_proc;
1660 	struct vattr vattr;
1661 	struct vnode *vp;
1662 	int error;
1663 
1664 	bwillwrite();
1665 
1666 	nd->nl_flags |= NLC_CREATE;
1667 	if ((error = nlookup(nd)) != 0)
1668 		return (error);
1669 	if (nd->nl_nch.ncp->nc_vp)
1670 		return (EEXIST);
1671 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1672 		return (error);
1673 
1674 	VATTR_NULL(&vattr);
1675 	vattr.va_type = VFIFO;
1676 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1677 	vp = NULL;
1678 	error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1679 	if (error == 0)
1680 		vput(vp);
1681 	return (error);
1682 }
1683 
1684 /*
1685  * mkfifo_args(char *path, int mode)
1686  *
1687  * Create a named pipe.
1688  */
1689 int
1690 sys_mkfifo(struct mkfifo_args *uap)
1691 {
1692 	struct nlookupdata nd;
1693 	int error;
1694 
1695 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1696 	if (error == 0)
1697 		error = kern_mkfifo(&nd, uap->mode);
1698 	nlookup_done(&nd);
1699 	return (error);
1700 }
1701 
1702 static int hardlink_check_uid = 0;
1703 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1704     &hardlink_check_uid, 0,
1705     "Unprivileged processes cannot create hard links to files owned by other "
1706     "users");
1707 static int hardlink_check_gid = 0;
1708 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1709     &hardlink_check_gid, 0,
1710     "Unprivileged processes cannot create hard links to files owned by other "
1711     "groups");
1712 
1713 static int
1714 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1715 {
1716 	struct vattr va;
1717 	int error;
1718 
1719 	/*
1720 	 * Shortcut if disabled
1721 	 */
1722 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1723 		return (0);
1724 
1725 	/*
1726 	 * root cred can always hardlink
1727 	 */
1728 	if (suser_cred(cred, PRISON_ROOT) == 0)
1729 		return (0);
1730 
1731 	/*
1732 	 * Otherwise only if the originating file is owned by the
1733 	 * same user or group.  Note that any group is allowed if
1734 	 * the file is owned by the caller.
1735 	 */
1736 	error = VOP_GETATTR(vp, &va);
1737 	if (error != 0)
1738 		return (error);
1739 
1740 	if (hardlink_check_uid) {
1741 		if (cred->cr_uid != va.va_uid)
1742 			return (EPERM);
1743 	}
1744 
1745 	if (hardlink_check_gid) {
1746 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1747 			return (EPERM);
1748 	}
1749 
1750 	return (0);
1751 }
1752 
1753 int
1754 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1755 {
1756 	struct thread *td = curthread;
1757 	struct vnode *vp;
1758 	int error;
1759 
1760 	/*
1761 	 * Lookup the source and obtained a locked vnode.
1762 	 *
1763 	 * XXX relookup on vget failure / race ?
1764 	 */
1765 	bwillwrite();
1766 	if ((error = nlookup(nd)) != 0)
1767 		return (error);
1768 	vp = nd->nl_nch.ncp->nc_vp;
1769 	KKASSERT(vp != NULL);
1770 	if (vp->v_type == VDIR)
1771 		return (EPERM);		/* POSIX */
1772 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1773 		return (error);
1774 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1775 		return (error);
1776 
1777 	/*
1778 	 * Unlock the source so we can lookup the target without deadlocking
1779 	 * (XXX vp is locked already, possible other deadlock?).  The target
1780 	 * must not exist.
1781 	 */
1782 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1783 	nd->nl_flags &= ~NLC_NCPISLOCKED;
1784 	cache_unlock(&nd->nl_nch);
1785 
1786 	linknd->nl_flags |= NLC_CREATE;
1787 	if ((error = nlookup(linknd)) != 0) {
1788 		vput(vp);
1789 		return (error);
1790 	}
1791 	if (linknd->nl_nch.ncp->nc_vp) {
1792 		vput(vp);
1793 		return (EEXIST);
1794 	}
1795 
1796 	/*
1797 	 * Finally run the new API VOP.
1798 	 */
1799 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
1800 	if (error == 0)
1801 		error = VOP_NLINK(&linknd->nl_nch, vp, linknd->nl_cred);
1802 	vput(vp);
1803 	return (error);
1804 }
1805 
1806 /*
1807  * link_args(char *path, char *link)
1808  *
1809  * Make a hard file link.
1810  */
1811 int
1812 sys_link(struct link_args *uap)
1813 {
1814 	struct nlookupdata nd, linknd;
1815 	int error;
1816 
1817 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1818 	if (error == 0) {
1819 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1820 		if (error == 0)
1821 			error = kern_link(&nd, &linknd);
1822 		nlookup_done(&linknd);
1823 	}
1824 	nlookup_done(&nd);
1825 	return (error);
1826 }
1827 
1828 int
1829 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1830 {
1831 	struct vattr vattr;
1832 	struct vnode *vp;
1833 	int error;
1834 
1835 	bwillwrite();
1836 	nd->nl_flags |= NLC_CREATE;
1837 	if ((error = nlookup(nd)) != 0)
1838 		return (error);
1839 	if (nd->nl_nch.ncp->nc_vp)
1840 		return (EEXIST);
1841 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1842 		return (error);
1843 	VATTR_NULL(&vattr);
1844 	vattr.va_mode = mode;
1845 	error = VOP_NSYMLINK(&nd->nl_nch, &vp, nd->nl_cred, &vattr, path);
1846 	if (error == 0)
1847 		vput(vp);
1848 	return (error);
1849 }
1850 
1851 /*
1852  * symlink(char *path, char *link)
1853  *
1854  * Make a symbolic link.
1855  */
1856 int
1857 sys_symlink(struct symlink_args *uap)
1858 {
1859 	struct thread *td = curthread;
1860 	struct nlookupdata nd;
1861 	char *path;
1862 	int error;
1863 	int mode;
1864 
1865 	path = objcache_get(namei_oc, M_WAITOK);
1866 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1867 	if (error == 0) {
1868 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1869 		if (error == 0) {
1870 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1871 			error = kern_symlink(&nd, path, mode);
1872 		}
1873 		nlookup_done(&nd);
1874 	}
1875 	objcache_put(namei_oc, path);
1876 	return (error);
1877 }
1878 
1879 /*
1880  * undelete_args(char *path)
1881  *
1882  * Delete a whiteout from the filesystem.
1883  */
1884 /* ARGSUSED */
1885 int
1886 sys_undelete(struct undelete_args *uap)
1887 {
1888 	struct nlookupdata nd;
1889 	int error;
1890 
1891 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1892 	bwillwrite();
1893 	nd.nl_flags |= NLC_DELETE;
1894 	if (error == 0)
1895 		error = nlookup(&nd);
1896 	if (error == 0)
1897 		error = ncp_writechk(&nd.nl_nch);
1898 	if (error == 0)
1899 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_cred, NAMEI_DELETE);
1900 	nlookup_done(&nd);
1901 	return (error);
1902 }
1903 
1904 int
1905 kern_unlink(struct nlookupdata *nd)
1906 {
1907 	int error;
1908 
1909 	bwillwrite();
1910 	nd->nl_flags |= NLC_DELETE;
1911 	if ((error = nlookup(nd)) != 0)
1912 		return (error);
1913 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1914 		return (error);
1915 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_cred);
1916 	return (error);
1917 }
1918 
1919 /*
1920  * unlink_args(char *path)
1921  *
1922  * Delete a name from the filesystem.
1923  */
1924 int
1925 sys_unlink(struct unlink_args *uap)
1926 {
1927 	struct nlookupdata nd;
1928 	int error;
1929 
1930 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1931 	if (error == 0)
1932 		error = kern_unlink(&nd);
1933 	nlookup_done(&nd);
1934 	return (error);
1935 }
1936 
1937 int
1938 kern_lseek(int fd, off_t offset, int whence, off_t *res)
1939 {
1940 	struct thread *td = curthread;
1941 	struct proc *p = td->td_proc;
1942 	struct file *fp;
1943 	struct vattr vattr;
1944 	int error;
1945 
1946 	fp = holdfp(p->p_fd, fd, -1);
1947 	if (fp == NULL)
1948 		return (EBADF);
1949 	if (fp->f_type != DTYPE_VNODE) {
1950 		error = ESPIPE;
1951 		goto done;
1952 	}
1953 
1954 	switch (whence) {
1955 	case L_INCR:
1956 		fp->f_offset += offset;
1957 		error = 0;
1958 		break;
1959 	case L_XTND:
1960 		error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
1961 		if (error == 0)
1962 			fp->f_offset = offset + vattr.va_size;
1963 		break;
1964 	case L_SET:
1965 		fp->f_offset = offset;
1966 		error = 0;
1967 		break;
1968 	default:
1969 		error = EINVAL;
1970 		break;
1971 	}
1972 	*res = fp->f_offset;
1973 done:
1974 	fdrop(fp);
1975 	return (error);
1976 }
1977 
1978 /*
1979  * lseek_args(int fd, int pad, off_t offset, int whence)
1980  *
1981  * Reposition read/write file offset.
1982  */
1983 int
1984 sys_lseek(struct lseek_args *uap)
1985 {
1986 	int error;
1987 
1988 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
1989 	    &uap->sysmsg_offset);
1990 
1991 	return (error);
1992 }
1993 
1994 int
1995 kern_access(struct nlookupdata *nd, int aflags)
1996 {
1997 	struct vnode *vp;
1998 	int error, flags;
1999 
2000 	if ((error = nlookup(nd)) != 0)
2001 		return (error);
2002 retry:
2003 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2004 	if (error)
2005 		return (error);
2006 
2007 	/* Flags == 0 means only check for existence. */
2008 	if (aflags) {
2009 		flags = 0;
2010 		if (aflags & R_OK)
2011 			flags |= VREAD;
2012 		if (aflags & W_OK)
2013 			flags |= VWRITE;
2014 		if (aflags & X_OK)
2015 			flags |= VEXEC;
2016 		if ((flags & VWRITE) == 0 ||
2017 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2018 			error = VOP_ACCESS(vp, flags, nd->nl_cred);
2019 
2020 		/*
2021 		 * If the file handle is stale we have to re-resolve the
2022 		 * entry.  This is a hack at the moment.
2023 		 */
2024 		if (error == ESTALE) {
2025 			cache_setunresolved(&nd->nl_nch);
2026 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2027 			if (error == 0) {
2028 				vput(vp);
2029 				vp = NULL;
2030 				goto retry;
2031 			}
2032 		}
2033 	}
2034 	vput(vp);
2035 	return (error);
2036 }
2037 
2038 /*
2039  * access_args(char *path, int flags)
2040  *
2041  * Check access permissions.
2042  */
2043 int
2044 sys_access(struct access_args *uap)
2045 {
2046 	struct nlookupdata nd;
2047 	int error;
2048 
2049 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2050 	if (error == 0)
2051 		error = kern_access(&nd, uap->flags);
2052 	nlookup_done(&nd);
2053 	return (error);
2054 }
2055 
2056 int
2057 kern_stat(struct nlookupdata *nd, struct stat *st)
2058 {
2059 	int error;
2060 	struct vnode *vp;
2061 	thread_t td;
2062 
2063 	if ((error = nlookup(nd)) != 0)
2064 		return (error);
2065 again:
2066 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2067 		return (ENOENT);
2068 
2069 	td = curthread;
2070 	if ((error = vget(vp, LK_SHARED)) != 0)
2071 		return (error);
2072 	error = vn_stat(vp, st, nd->nl_cred);
2073 
2074 	/*
2075 	 * If the file handle is stale we have to re-resolve the entry.  This
2076 	 * is a hack at the moment.
2077 	 */
2078 	if (error == ESTALE) {
2079 		cache_setunresolved(&nd->nl_nch);
2080 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2081 		if (error == 0) {
2082 			vput(vp);
2083 			goto again;
2084 		}
2085 	}
2086 	vput(vp);
2087 	return (error);
2088 }
2089 
2090 /*
2091  * stat_args(char *path, struct stat *ub)
2092  *
2093  * Get file status; this version follows links.
2094  */
2095 int
2096 sys_stat(struct stat_args *uap)
2097 {
2098 	struct nlookupdata nd;
2099 	struct stat st;
2100 	int error;
2101 
2102 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2103 	if (error == 0) {
2104 		error = kern_stat(&nd, &st);
2105 		if (error == 0)
2106 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2107 	}
2108 	nlookup_done(&nd);
2109 	return (error);
2110 }
2111 
2112 /*
2113  * lstat_args(char *path, struct stat *ub)
2114  *
2115  * Get file status; this version does not follow links.
2116  */
2117 int
2118 sys_lstat(struct lstat_args *uap)
2119 {
2120 	struct nlookupdata nd;
2121 	struct stat st;
2122 	int error;
2123 
2124 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2125 	if (error == 0) {
2126 		error = kern_stat(&nd, &st);
2127 		if (error == 0)
2128 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2129 	}
2130 	nlookup_done(&nd);
2131 	return (error);
2132 }
2133 
2134 /*
2135  * pathconf_Args(char *path, int name)
2136  *
2137  * Get configurable pathname variables.
2138  */
2139 /* ARGSUSED */
2140 int
2141 sys_pathconf(struct pathconf_args *uap)
2142 {
2143 	struct nlookupdata nd;
2144 	struct vnode *vp;
2145 	int error;
2146 
2147 	vp = NULL;
2148 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2149 	if (error == 0)
2150 		error = nlookup(&nd);
2151 	if (error == 0)
2152 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2153 	nlookup_done(&nd);
2154 	if (error == 0) {
2155 		error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2156 		vput(vp);
2157 	}
2158 	return (error);
2159 }
2160 
2161 /*
2162  * XXX: daver
2163  * kern_readlink isn't properly split yet.  There is a copyin burried
2164  * in VOP_READLINK().
2165  */
2166 int
2167 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2168 {
2169 	struct thread *td = curthread;
2170 	struct proc *p = td->td_proc;
2171 	struct vnode *vp;
2172 	struct iovec aiov;
2173 	struct uio auio;
2174 	int error;
2175 
2176 	if ((error = nlookup(nd)) != 0)
2177 		return (error);
2178 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2179 	if (error)
2180 		return (error);
2181 	if (vp->v_type != VLNK) {
2182 		error = EINVAL;
2183 	} else {
2184 		aiov.iov_base = buf;
2185 		aiov.iov_len = count;
2186 		auio.uio_iov = &aiov;
2187 		auio.uio_iovcnt = 1;
2188 		auio.uio_offset = 0;
2189 		auio.uio_rw = UIO_READ;
2190 		auio.uio_segflg = UIO_USERSPACE;
2191 		auio.uio_td = td;
2192 		auio.uio_resid = count;
2193 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2194 	}
2195 	vput(vp);
2196 	*res = count - auio.uio_resid;
2197 	return (error);
2198 }
2199 
2200 /*
2201  * readlink_args(char *path, char *buf, int count)
2202  *
2203  * Return target name of a symbolic link.
2204  */
2205 int
2206 sys_readlink(struct readlink_args *uap)
2207 {
2208 	struct nlookupdata nd;
2209 	int error;
2210 
2211 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2212 	if (error == 0) {
2213 		error = kern_readlink(&nd, uap->buf, uap->count,
2214 					&uap->sysmsg_result);
2215 	}
2216 	nlookup_done(&nd);
2217 	return (error);
2218 }
2219 
2220 static int
2221 setfflags(struct vnode *vp, int flags)
2222 {
2223 	struct thread *td = curthread;
2224 	struct proc *p = td->td_proc;
2225 	int error;
2226 	struct vattr vattr;
2227 
2228 	/*
2229 	 * Prevent non-root users from setting flags on devices.  When
2230 	 * a device is reused, users can retain ownership of the device
2231 	 * if they are allowed to set flags and programs assume that
2232 	 * chown can't fail when done as root.
2233 	 */
2234 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2235 	    ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2236 		return (error);
2237 
2238 	/*
2239 	 * note: vget is required for any operation that might mod the vnode
2240 	 * so VINACTIVE is properly cleared.
2241 	 */
2242 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2243 		VATTR_NULL(&vattr);
2244 		vattr.va_flags = flags;
2245 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2246 		vput(vp);
2247 	}
2248 	return (error);
2249 }
2250 
2251 /*
2252  * chflags(char *path, int flags)
2253  *
2254  * Change flags of a file given a path name.
2255  */
2256 /* ARGSUSED */
2257 int
2258 sys_chflags(struct chflags_args *uap)
2259 {
2260 	struct nlookupdata nd;
2261 	struct vnode *vp;
2262 	int error;
2263 
2264 	vp = NULL;
2265 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2266 	/* XXX Add NLC flag indicating modifying operation? */
2267 	if (error == 0)
2268 		error = nlookup(&nd);
2269 	if (error == 0)
2270 		error = ncp_writechk(&nd.nl_nch);
2271 	if (error == 0)
2272 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2273 	nlookup_done(&nd);
2274 	if (error == 0) {
2275 		error = setfflags(vp, uap->flags);
2276 		vrele(vp);
2277 	}
2278 	return (error);
2279 }
2280 
2281 /*
2282  * fchflags_args(int fd, int flags)
2283  *
2284  * Change flags of a file given a file descriptor.
2285  */
2286 /* ARGSUSED */
2287 int
2288 sys_fchflags(struct fchflags_args *uap)
2289 {
2290 	struct thread *td = curthread;
2291 	struct proc *p = td->td_proc;
2292 	struct file *fp;
2293 	int error;
2294 
2295 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2296 		return (error);
2297 	if (fp->f_nchandle.ncp)
2298 		error = ncp_writechk(&fp->f_nchandle);
2299 	if (error == 0)
2300 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2301 	fdrop(fp);
2302 	return (error);
2303 }
2304 
2305 static int
2306 setfmode(struct vnode *vp, int mode)
2307 {
2308 	struct thread *td = curthread;
2309 	struct proc *p = td->td_proc;
2310 	int error;
2311 	struct vattr vattr;
2312 
2313 	/*
2314 	 * note: vget is required for any operation that might mod the vnode
2315 	 * so VINACTIVE is properly cleared.
2316 	 */
2317 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2318 		VATTR_NULL(&vattr);
2319 		vattr.va_mode = mode & ALLPERMS;
2320 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2321 		vput(vp);
2322 	}
2323 	return error;
2324 }
2325 
2326 int
2327 kern_chmod(struct nlookupdata *nd, int mode)
2328 {
2329 	struct vnode *vp;
2330 	int error;
2331 
2332 	/* XXX Add NLC flag indicating modifying operation? */
2333 	if ((error = nlookup(nd)) != 0)
2334 		return (error);
2335 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2336 		return (error);
2337 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2338 		error = setfmode(vp, mode);
2339 	vrele(vp);
2340 	return (error);
2341 }
2342 
2343 /*
2344  * chmod_args(char *path, int mode)
2345  *
2346  * Change mode of a file given path name.
2347  */
2348 /* ARGSUSED */
2349 int
2350 sys_chmod(struct chmod_args *uap)
2351 {
2352 	struct nlookupdata nd;
2353 	int error;
2354 
2355 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2356 	if (error == 0)
2357 		error = kern_chmod(&nd, uap->mode);
2358 	nlookup_done(&nd);
2359 	return (error);
2360 }
2361 
2362 /*
2363  * lchmod_args(char *path, int mode)
2364  *
2365  * Change mode of a file given path name (don't follow links.)
2366  */
2367 /* ARGSUSED */
2368 int
2369 sys_lchmod(struct lchmod_args *uap)
2370 {
2371 	struct nlookupdata nd;
2372 	int error;
2373 
2374 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2375 	if (error == 0)
2376 		error = kern_chmod(&nd, uap->mode);
2377 	nlookup_done(&nd);
2378 	return (error);
2379 }
2380 
2381 /*
2382  * fchmod_args(int fd, int mode)
2383  *
2384  * Change mode of a file given a file descriptor.
2385  */
2386 /* ARGSUSED */
2387 int
2388 sys_fchmod(struct fchmod_args *uap)
2389 {
2390 	struct thread *td = curthread;
2391 	struct proc *p = td->td_proc;
2392 	struct file *fp;
2393 	int error;
2394 
2395 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2396 		return (error);
2397 	if (fp->f_nchandle.ncp)
2398 		error = ncp_writechk(&fp->f_nchandle);
2399 	if (error == 0)
2400 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
2401 	fdrop(fp);
2402 	return (error);
2403 }
2404 
2405 static int
2406 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2407 {
2408 	struct thread *td = curthread;
2409 	struct proc *p = td->td_proc;
2410 	int error;
2411 	struct vattr vattr;
2412 
2413 	/*
2414 	 * note: vget is required for any operation that might mod the vnode
2415 	 * so VINACTIVE is properly cleared.
2416 	 */
2417 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2418 		VATTR_NULL(&vattr);
2419 		vattr.va_uid = uid;
2420 		vattr.va_gid = gid;
2421 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2422 		vput(vp);
2423 	}
2424 	return error;
2425 }
2426 
2427 int
2428 kern_chown(struct nlookupdata *nd, int uid, int gid)
2429 {
2430 	struct vnode *vp;
2431 	int error;
2432 
2433 	/* XXX Add NLC flag indicating modifying operation? */
2434 	if ((error = nlookup(nd)) != 0)
2435 		return (error);
2436 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2437 		return (error);
2438 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2439 		error = setfown(vp, uid, gid);
2440 	vrele(vp);
2441 	return (error);
2442 }
2443 
2444 /*
2445  * chown(char *path, int uid, int gid)
2446  *
2447  * Set ownership given a path name.
2448  */
2449 int
2450 sys_chown(struct chown_args *uap)
2451 {
2452 	struct nlookupdata nd;
2453 	int error;
2454 
2455 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2456 	if (error == 0)
2457 		error = kern_chown(&nd, uap->uid, uap->gid);
2458 	nlookup_done(&nd);
2459 	return (error);
2460 }
2461 
2462 /*
2463  * lchown_args(char *path, int uid, int gid)
2464  *
2465  * Set ownership given a path name, do not cross symlinks.
2466  */
2467 int
2468 sys_lchown(struct lchown_args *uap)
2469 {
2470 	struct nlookupdata nd;
2471 	int error;
2472 
2473 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2474 	if (error == 0)
2475 		error = kern_chown(&nd, uap->uid, uap->gid);
2476 	nlookup_done(&nd);
2477 	return (error);
2478 }
2479 
2480 /*
2481  * fchown_args(int fd, int uid, int gid)
2482  *
2483  * Set ownership given a file descriptor.
2484  */
2485 /* ARGSUSED */
2486 int
2487 sys_fchown(struct fchown_args *uap)
2488 {
2489 	struct thread *td = curthread;
2490 	struct proc *p = td->td_proc;
2491 	struct file *fp;
2492 	int error;
2493 
2494 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2495 		return (error);
2496 	if (fp->f_nchandle.ncp)
2497 		error = ncp_writechk(&fp->f_nchandle);
2498 	if (error == 0)
2499 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2500 	fdrop(fp);
2501 	return (error);
2502 }
2503 
2504 static int
2505 getutimes(const struct timeval *tvp, struct timespec *tsp)
2506 {
2507 	struct timeval tv[2];
2508 
2509 	if (tvp == NULL) {
2510 		microtime(&tv[0]);
2511 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2512 		tsp[1] = tsp[0];
2513 	} else {
2514 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2515 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2516 	}
2517 	return 0;
2518 }
2519 
2520 static int
2521 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2522 {
2523 	struct thread *td = curthread;
2524 	struct proc *p = td->td_proc;
2525 	int error;
2526 	struct vattr vattr;
2527 
2528 	/*
2529 	 * note: vget is required for any operation that might mod the vnode
2530 	 * so VINACTIVE is properly cleared.
2531 	 */
2532 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2533 		VATTR_NULL(&vattr);
2534 		vattr.va_atime = ts[0];
2535 		vattr.va_mtime = ts[1];
2536 		if (nullflag)
2537 			vattr.va_vaflags |= VA_UTIMES_NULL;
2538 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2539 		vput(vp);
2540 	}
2541 	return error;
2542 }
2543 
2544 int
2545 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2546 {
2547 	struct timespec ts[2];
2548 	struct vnode *vp;
2549 	int error;
2550 
2551 	if ((error = getutimes(tptr, ts)) != 0)
2552 		return (error);
2553 	/* XXX Add NLC flag indicating modifying operation? */
2554 	if ((error = nlookup(nd)) != 0)
2555 		return (error);
2556 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2557 		return (error);
2558 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2559 		return (error);
2560 	error = setutimes(vp, ts, tptr == NULL);
2561 	vrele(vp);
2562 	return (error);
2563 }
2564 
2565 /*
2566  * utimes_args(char *path, struct timeval *tptr)
2567  *
2568  * Set the access and modification times of a file.
2569  */
2570 int
2571 sys_utimes(struct utimes_args *uap)
2572 {
2573 	struct timeval tv[2];
2574 	struct nlookupdata nd;
2575 	int error;
2576 
2577 	if (uap->tptr) {
2578  		error = copyin(uap->tptr, tv, sizeof(tv));
2579 		if (error)
2580 			return (error);
2581 	}
2582 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2583 	if (error == 0)
2584 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2585 	nlookup_done(&nd);
2586 	return (error);
2587 }
2588 
2589 /*
2590  * lutimes_args(char *path, struct timeval *tptr)
2591  *
2592  * Set the access and modification times of a file.
2593  */
2594 int
2595 sys_lutimes(struct lutimes_args *uap)
2596 {
2597 	struct timeval tv[2];
2598 	struct nlookupdata nd;
2599 	int error;
2600 
2601 	if (uap->tptr) {
2602 		error = copyin(uap->tptr, tv, sizeof(tv));
2603 		if (error)
2604 			return (error);
2605 	}
2606 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2607 	if (error == 0)
2608 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2609 	nlookup_done(&nd);
2610 	return (error);
2611 }
2612 
2613 int
2614 kern_futimes(int fd, struct timeval *tptr)
2615 {
2616 	struct thread *td = curthread;
2617 	struct proc *p = td->td_proc;
2618 	struct timespec ts[2];
2619 	struct file *fp;
2620 	int error;
2621 
2622 	error = getutimes(tptr, ts);
2623 	if (error)
2624 		return (error);
2625 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2626 		return (error);
2627 	if (fp->f_nchandle.ncp)
2628 		error = ncp_writechk(&fp->f_nchandle);
2629 	if (error == 0)
2630 		error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2631 	fdrop(fp);
2632 	return (error);
2633 }
2634 
2635 /*
2636  * futimes_args(int fd, struct timeval *tptr)
2637  *
2638  * Set the access and modification times of a file.
2639  */
2640 int
2641 sys_futimes(struct futimes_args *uap)
2642 {
2643 	struct timeval tv[2];
2644 	int error;
2645 
2646 	if (uap->tptr) {
2647 		error = copyin(uap->tptr, tv, sizeof(tv));
2648 		if (error)
2649 			return (error);
2650 	}
2651 
2652 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2653 
2654 	return (error);
2655 }
2656 
2657 int
2658 kern_truncate(struct nlookupdata *nd, off_t length)
2659 {
2660 	struct vnode *vp;
2661 	struct vattr vattr;
2662 	int error;
2663 
2664 	if (length < 0)
2665 		return(EINVAL);
2666 	/* XXX Add NLC flag indicating modifying operation? */
2667 	if ((error = nlookup(nd)) != 0)
2668 		return (error);
2669 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2670 		return (error);
2671 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2672 		return (error);
2673 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2674 		vrele(vp);
2675 		return (error);
2676 	}
2677 	if (vp->v_type == VDIR) {
2678 		error = EISDIR;
2679 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2680 	    (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2681 		VATTR_NULL(&vattr);
2682 		vattr.va_size = length;
2683 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2684 	}
2685 	vput(vp);
2686 	return (error);
2687 }
2688 
2689 /*
2690  * truncate(char *path, int pad, off_t length)
2691  *
2692  * Truncate a file given its path name.
2693  */
2694 int
2695 sys_truncate(struct truncate_args *uap)
2696 {
2697 	struct nlookupdata nd;
2698 	int error;
2699 
2700 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2701 	if (error == 0)
2702 		error = kern_truncate(&nd, uap->length);
2703 	nlookup_done(&nd);
2704 	return error;
2705 }
2706 
2707 int
2708 kern_ftruncate(int fd, off_t length)
2709 {
2710 	struct thread *td = curthread;
2711 	struct proc *p = td->td_proc;
2712 	struct vattr vattr;
2713 	struct vnode *vp;
2714 	struct file *fp;
2715 	int error;
2716 
2717 	if (length < 0)
2718 		return(EINVAL);
2719 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2720 		return (error);
2721 	if (fp->f_nchandle.ncp) {
2722 		error = ncp_writechk(&fp->f_nchandle);
2723 		if (error)
2724 			goto done;
2725 	}
2726 	if ((fp->f_flag & FWRITE) == 0) {
2727 		error = EINVAL;
2728 		goto done;
2729 	}
2730 	vp = (struct vnode *)fp->f_data;
2731 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2732 	if (vp->v_type == VDIR) {
2733 		error = EISDIR;
2734 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
2735 		VATTR_NULL(&vattr);
2736 		vattr.va_size = length;
2737 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2738 	}
2739 	vn_unlock(vp);
2740 done:
2741 	fdrop(fp);
2742 	return (error);
2743 }
2744 
2745 /*
2746  * ftruncate_args(int fd, int pad, off_t length)
2747  *
2748  * Truncate a file given a file descriptor.
2749  */
2750 int
2751 sys_ftruncate(struct ftruncate_args *uap)
2752 {
2753 	int error;
2754 
2755 	error = kern_ftruncate(uap->fd, uap->length);
2756 
2757 	return (error);
2758 }
2759 
2760 /*
2761  * fsync(int fd)
2762  *
2763  * Sync an open file.
2764  */
2765 /* ARGSUSED */
2766 int
2767 sys_fsync(struct fsync_args *uap)
2768 {
2769 	struct thread *td = curthread;
2770 	struct proc *p = td->td_proc;
2771 	struct vnode *vp;
2772 	struct file *fp;
2773 	vm_object_t obj;
2774 	int error;
2775 
2776 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2777 		return (error);
2778 	vp = (struct vnode *)fp->f_data;
2779 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2780 	if ((obj = vp->v_object) != NULL)
2781 		vm_object_page_clean(obj, 0, 0, 0);
2782 	if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 &&
2783 	    vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2784 	    bioops.io_fsync) {
2785 		error = (*bioops.io_fsync)(vp);
2786 	}
2787 	vn_unlock(vp);
2788 	fdrop(fp);
2789 	return (error);
2790 }
2791 
2792 int
2793 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2794 {
2795 	struct nchandle fnchd;
2796 	struct nchandle tnchd;
2797 	struct namecache *ncp;
2798 	struct mount *mp;
2799 	int error;
2800 
2801 	bwillwrite();
2802 	if ((error = nlookup(fromnd)) != 0)
2803 		return (error);
2804 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
2805 		return (ENOENT);
2806 	fnchd.mount = fromnd->nl_nch.mount;
2807 	cache_hold(&fnchd);
2808 
2809 	/*
2810 	 * unlock the source nch so we can lookup the target nch without
2811 	 * deadlocking.  The target may or may not exist so we do not check
2812 	 * for a target vp like kern_mkdir() and other creation functions do.
2813 	 *
2814 	 * The source and target directories are ref'd and rechecked after
2815 	 * everything is relocked to determine if the source or target file
2816 	 * has been renamed.
2817 	 */
2818 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2819 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2820 	cache_unlock(&fromnd->nl_nch);
2821 
2822 	tond->nl_flags |= NLC_CREATE;
2823 	if ((error = nlookup(tond)) != 0) {
2824 		cache_drop(&fnchd);
2825 		return (error);
2826 	}
2827 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
2828 		cache_drop(&fnchd);
2829 		return (ENOENT);
2830 	}
2831 	tnchd.mount = tond->nl_nch.mount;
2832 	cache_hold(&tnchd);
2833 
2834 	/*
2835 	 * If the source and target are the same there is nothing to do
2836 	 */
2837 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
2838 		cache_drop(&fnchd);
2839 		cache_drop(&tnchd);
2840 		return (0);
2841 	}
2842 
2843 	/*
2844 	 * Mount points cannot be renamed or overwritten
2845 	 */
2846 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
2847 	    NCF_ISMOUNTPT
2848 	) {
2849 		cache_drop(&fnchd);
2850 		cache_drop(&tnchd);
2851 		return (EINVAL);
2852 	}
2853 
2854 	/*
2855 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2856 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
2857 	 * could be NULL.
2858 	 */
2859 	if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
2860 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2861 	} else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
2862 		cache_lock(&fromnd->nl_nch);
2863 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2864 	} else {
2865 		cache_unlock(&tond->nl_nch);
2866 		cache_lock(&fromnd->nl_nch);
2867 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2868 		cache_lock(&tond->nl_nch);
2869 		cache_resolve(&tond->nl_nch, tond->nl_cred);
2870 	}
2871 	fromnd->nl_flags |= NLC_NCPISLOCKED;
2872 
2873 	/*
2874 	 * make sure the parent directories linkages are the same
2875 	 */
2876 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
2877 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
2878 		cache_drop(&fnchd);
2879 		cache_drop(&tnchd);
2880 		return (ENOENT);
2881 	}
2882 
2883 	/*
2884 	 * Both the source and target must be within the same filesystem and
2885 	 * in the same filesystem as their parent directories within the
2886 	 * namecache topology.
2887 	 *
2888 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2889 	 */
2890 	mp = fnchd.mount;
2891 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
2892 	    mp != tond->nl_nch.mount) {
2893 		cache_drop(&fnchd);
2894 		cache_drop(&tnchd);
2895 		return (EXDEV);
2896 	}
2897 
2898 	/*
2899 	 * Make sure the mount point is writable
2900 	 */
2901 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
2902 		cache_drop(&fnchd);
2903 		cache_drop(&tnchd);
2904 		return (error);
2905 	}
2906 
2907 	/*
2908 	 * If the target exists and either the source or target is a directory,
2909 	 * then both must be directories.
2910 	 *
2911 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
2912 	 * have become NULL.
2913 	 */
2914 	if (tond->nl_nch.ncp->nc_vp) {
2915 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
2916 			error = ENOENT;
2917 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
2918 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
2919 				error = ENOTDIR;
2920 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
2921 			error = EISDIR;
2922 		}
2923 	}
2924 
2925 	/*
2926 	 * You cannot rename a source into itself or a subdirectory of itself.
2927 	 * We check this by travsersing the target directory upwards looking
2928 	 * for a match against the source.
2929 	 */
2930 	if (error == 0) {
2931 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
2932 			if (fromnd->nl_nch.ncp == ncp) {
2933 				error = EINVAL;
2934 				break;
2935 			}
2936 		}
2937 	}
2938 
2939 	cache_drop(&fnchd);
2940 	cache_drop(&tnchd);
2941 
2942 	/*
2943 	 * Even though the namespaces are different, they may still represent
2944 	 * hardlinks to the same file.  The filesystem might have a hard time
2945 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
2946 	 * when we detect the situation.
2947 	 */
2948 	if (error == 0) {
2949 		if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
2950 			error = VOP_NREMOVE(&fromnd->nl_nch, fromnd->nl_cred);
2951 		} else {
2952 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
2953 					    tond->nl_cred);
2954 		}
2955 	}
2956 	return (error);
2957 }
2958 
2959 /*
2960  * rename_args(char *from, char *to)
2961  *
2962  * Rename files.  Source and destination must either both be directories,
2963  * or both not be directories.  If target is a directory, it must be empty.
2964  */
2965 int
2966 sys_rename(struct rename_args *uap)
2967 {
2968 	struct nlookupdata fromnd, tond;
2969 	int error;
2970 
2971 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
2972 	if (error == 0) {
2973 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
2974 		if (error == 0)
2975 			error = kern_rename(&fromnd, &tond);
2976 		nlookup_done(&tond);
2977 	}
2978 	nlookup_done(&fromnd);
2979 	return (error);
2980 }
2981 
2982 int
2983 kern_mkdir(struct nlookupdata *nd, int mode)
2984 {
2985 	struct thread *td = curthread;
2986 	struct proc *p = td->td_proc;
2987 	struct vnode *vp;
2988 	struct vattr vattr;
2989 	int error;
2990 
2991 	bwillwrite();
2992 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
2993 	if ((error = nlookup(nd)) != 0)
2994 		return (error);
2995 
2996 	if (nd->nl_nch.ncp->nc_vp)
2997 		return (EEXIST);
2998 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2999 		return (error);
3000 
3001 	VATTR_NULL(&vattr);
3002 	vattr.va_type = VDIR;
3003 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3004 
3005 	vp = NULL;
3006 	error = VOP_NMKDIR(&nd->nl_nch, &vp, p->p_ucred, &vattr);
3007 	if (error == 0)
3008 		vput(vp);
3009 	return (error);
3010 }
3011 
3012 /*
3013  * mkdir_args(char *path, int mode)
3014  *
3015  * Make a directory file.
3016  */
3017 /* ARGSUSED */
3018 int
3019 sys_mkdir(struct mkdir_args *uap)
3020 {
3021 	struct nlookupdata nd;
3022 	int error;
3023 
3024 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3025 	if (error == 0)
3026 		error = kern_mkdir(&nd, uap->mode);
3027 	nlookup_done(&nd);
3028 	return (error);
3029 }
3030 
3031 int
3032 kern_rmdir(struct nlookupdata *nd)
3033 {
3034 	int error;
3035 
3036 	bwillwrite();
3037 	nd->nl_flags |= NLC_DELETE;
3038 	if ((error = nlookup(nd)) != 0)
3039 		return (error);
3040 
3041 	/*
3042 	 * Do not allow directories representing mount points to be
3043 	 * deleted, even if empty.  Check write perms on mount point
3044 	 * in case the vnode is aliased (aka nullfs).
3045 	 */
3046 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3047 		return (EINVAL);
3048 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3049 		return (error);
3050 
3051 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_cred);
3052 	return (error);
3053 }
3054 
3055 /*
3056  * rmdir_args(char *path)
3057  *
3058  * Remove a directory file.
3059  */
3060 /* ARGSUSED */
3061 int
3062 sys_rmdir(struct rmdir_args *uap)
3063 {
3064 	struct nlookupdata nd;
3065 	int error;
3066 
3067 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3068 	if (error == 0)
3069 		error = kern_rmdir(&nd);
3070 	nlookup_done(&nd);
3071 	return (error);
3072 }
3073 
3074 int
3075 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3076     enum uio_seg direction)
3077 {
3078 	struct thread *td = curthread;
3079 	struct proc *p = td->td_proc;
3080 	struct vnode *vp;
3081 	struct file *fp;
3082 	struct uio auio;
3083 	struct iovec aiov;
3084 	long loff;
3085 	int error, eofflag;
3086 
3087 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3088 		return (error);
3089 	if ((fp->f_flag & FREAD) == 0) {
3090 		error = EBADF;
3091 		goto done;
3092 	}
3093 	vp = (struct vnode *)fp->f_data;
3094 unionread:
3095 	if (vp->v_type != VDIR) {
3096 		error = EINVAL;
3097 		goto done;
3098 	}
3099 	aiov.iov_base = buf;
3100 	aiov.iov_len = count;
3101 	auio.uio_iov = &aiov;
3102 	auio.uio_iovcnt = 1;
3103 	auio.uio_rw = UIO_READ;
3104 	auio.uio_segflg = direction;
3105 	auio.uio_td = td;
3106 	auio.uio_resid = count;
3107 	loff = auio.uio_offset = fp->f_offset;
3108 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3109 	fp->f_offset = auio.uio_offset;
3110 	if (error)
3111 		goto done;
3112 	if (count == auio.uio_resid) {
3113 		if (union_dircheckp) {
3114 			error = union_dircheckp(td, &vp, fp);
3115 			if (error == -1)
3116 				goto unionread;
3117 			if (error)
3118 				goto done;
3119 		}
3120 #if 0
3121 		if ((vp->v_flag & VROOT) &&
3122 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3123 			struct vnode *tvp = vp;
3124 			vp = vp->v_mount->mnt_vnodecovered;
3125 			vref(vp);
3126 			fp->f_data = vp;
3127 			fp->f_offset = 0;
3128 			vrele(tvp);
3129 			goto unionread;
3130 		}
3131 #endif
3132 	}
3133 	if (basep) {
3134 		*basep = loff;
3135 	}
3136 	*res = count - auio.uio_resid;
3137 done:
3138 	fdrop(fp);
3139 	return (error);
3140 }
3141 
3142 /*
3143  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3144  *
3145  * Read a block of directory entries in a file system independent format.
3146  */
3147 int
3148 sys_getdirentries(struct getdirentries_args *uap)
3149 {
3150 	long base;
3151 	int error;
3152 
3153 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3154 	    &uap->sysmsg_result, UIO_USERSPACE);
3155 
3156 	if (error == 0)
3157 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3158 	return (error);
3159 }
3160 
3161 /*
3162  * getdents_args(int fd, char *buf, size_t count)
3163  */
3164 int
3165 sys_getdents(struct getdents_args *uap)
3166 {
3167 	int error;
3168 
3169 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3170 	    &uap->sysmsg_result, UIO_USERSPACE);
3171 
3172 	return (error);
3173 }
3174 
3175 /*
3176  * umask(int newmask)
3177  *
3178  * Set the mode mask for creation of filesystem nodes.
3179  *
3180  * MP SAFE
3181  */
3182 int
3183 sys_umask(struct umask_args *uap)
3184 {
3185 	struct thread *td = curthread;
3186 	struct proc *p = td->td_proc;
3187 	struct filedesc *fdp;
3188 
3189 	fdp = p->p_fd;
3190 	uap->sysmsg_result = fdp->fd_cmask;
3191 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3192 	return (0);
3193 }
3194 
3195 /*
3196  * revoke(char *path)
3197  *
3198  * Void all references to file by ripping underlying filesystem
3199  * away from vnode.
3200  */
3201 /* ARGSUSED */
3202 int
3203 sys_revoke(struct revoke_args *uap)
3204 {
3205 	struct nlookupdata nd;
3206 	struct vattr vattr;
3207 	struct vnode *vp;
3208 	struct ucred *cred;
3209 	int error;
3210 
3211 	vp = NULL;
3212 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3213 	if (error == 0)
3214 		error = nlookup(&nd);
3215 	if (error == 0)
3216 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3217 	cred = crhold(nd.nl_cred);
3218 	nlookup_done(&nd);
3219 	if (error == 0) {
3220 		if (vp->v_type != VCHR && vp->v_type != VBLK)
3221 			error = EINVAL;
3222 		if (error == 0)
3223 			error = VOP_GETATTR(vp, &vattr);
3224 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3225 			error = suser_cred(cred, PRISON_ROOT);
3226 		if (error == 0 && count_udev(vp->v_udev) > 0) {
3227 			error = 0;
3228 			vx_lock(vp);
3229 			VOP_REVOKE(vp, REVOKEALL);
3230 			vx_unlock(vp);
3231 		}
3232 		vrele(vp);
3233 	}
3234 	if (cred)
3235 		crfree(cred);
3236 	return (error);
3237 }
3238 
3239 /*
3240  * getfh_args(char *fname, fhandle_t *fhp)
3241  *
3242  * Get (NFS) file handle
3243  */
3244 int
3245 sys_getfh(struct getfh_args *uap)
3246 {
3247 	struct thread *td = curthread;
3248 	struct nlookupdata nd;
3249 	fhandle_t fh;
3250 	struct vnode *vp;
3251 	int error;
3252 
3253 	/*
3254 	 * Must be super user
3255 	 */
3256 	if ((error = suser(td)) != 0)
3257 		return (error);
3258 
3259 	vp = NULL;
3260 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3261 	if (error == 0)
3262 		error = nlookup(&nd);
3263 	if (error == 0)
3264 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3265 	nlookup_done(&nd);
3266 	if (error == 0) {
3267 		bzero(&fh, sizeof(fh));
3268 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3269 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3270 		vput(vp);
3271 		if (error == 0)
3272 			error = copyout(&fh, uap->fhp, sizeof(fh));
3273 	}
3274 	return (error);
3275 }
3276 
3277 /*
3278  * fhopen_args(const struct fhandle *u_fhp, int flags)
3279  *
3280  * syscall for the rpc.lockd to use to translate a NFS file handle into
3281  * an open descriptor.
3282  *
3283  * warning: do not remove the suser() call or this becomes one giant
3284  * security hole.
3285  */
3286 int
3287 sys_fhopen(struct fhopen_args *uap)
3288 {
3289 	struct thread *td = curthread;
3290 	struct proc *p = td->td_proc;
3291 	struct mount *mp;
3292 	struct vnode *vp;
3293 	struct fhandle fhp;
3294 	struct vattr vat;
3295 	struct vattr *vap = &vat;
3296 	struct flock lf;
3297 	int fmode, mode, error, type;
3298 	struct file *nfp;
3299 	struct file *fp;
3300 	int indx;
3301 
3302 	/*
3303 	 * Must be super user
3304 	 */
3305 	error = suser(td);
3306 	if (error)
3307 		return (error);
3308 
3309 	fmode = FFLAGS(uap->flags);
3310 	/* why not allow a non-read/write open for our lockd? */
3311 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3312 		return (EINVAL);
3313 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3314 	if (error)
3315 		return(error);
3316 	/* find the mount point */
3317 	mp = vfs_getvfs(&fhp.fh_fsid);
3318 	if (mp == NULL)
3319 		return (ESTALE);
3320 	/* now give me my vnode, it gets returned to me locked */
3321 	error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3322 	if (error)
3323 		return (error);
3324  	/*
3325 	 * from now on we have to make sure not
3326 	 * to forget about the vnode
3327 	 * any error that causes an abort must vput(vp)
3328 	 * just set error = err and 'goto bad;'.
3329 	 */
3330 
3331 	/*
3332 	 * from vn_open
3333 	 */
3334 	if (vp->v_type == VLNK) {
3335 		error = EMLINK;
3336 		goto bad;
3337 	}
3338 	if (vp->v_type == VSOCK) {
3339 		error = EOPNOTSUPP;
3340 		goto bad;
3341 	}
3342 	mode = 0;
3343 	if (fmode & (FWRITE | O_TRUNC)) {
3344 		if (vp->v_type == VDIR) {
3345 			error = EISDIR;
3346 			goto bad;
3347 		}
3348 		error = vn_writechk(vp, NULL);
3349 		if (error)
3350 			goto bad;
3351 		mode |= VWRITE;
3352 	}
3353 	if (fmode & FREAD)
3354 		mode |= VREAD;
3355 	if (mode) {
3356 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3357 		if (error)
3358 			goto bad;
3359 	}
3360 	if (fmode & O_TRUNC) {
3361 		vn_unlock(vp);				/* XXX */
3362 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3363 		VATTR_NULL(vap);
3364 		vap->va_size = 0;
3365 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3366 		if (error)
3367 			goto bad;
3368 	}
3369 
3370 	/*
3371 	 * VOP_OPEN needs the file pointer so it can potentially override
3372 	 * it.
3373 	 *
3374 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
3375 	 * directory.  XXX
3376 	 */
3377 	if ((error = falloc(p, &nfp, &indx)) != 0)
3378 		goto bad;
3379 	fp = nfp;
3380 
3381 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3382 	if (error) {
3383 		/*
3384 		 * setting f_ops this way prevents VOP_CLOSE from being
3385 		 * called or fdrop() releasing the vp from v_data.   Since
3386 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3387 		 */
3388 		fp->f_ops = &badfileops;
3389 		fp->f_data = NULL;
3390 		goto bad_drop;
3391 	}
3392 
3393 	/*
3394 	 * The fp is given its own reference, we still have our ref and lock.
3395 	 *
3396 	 * Assert that all regular files must be created with a VM object.
3397 	 */
3398 	if (vp->v_type == VREG && vp->v_object == NULL) {
3399 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3400 		goto bad_drop;
3401 	}
3402 
3403 	/*
3404 	 * The open was successful.  Handle any locking requirements.
3405 	 */
3406 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3407 		lf.l_whence = SEEK_SET;
3408 		lf.l_start = 0;
3409 		lf.l_len = 0;
3410 		if (fmode & O_EXLOCK)
3411 			lf.l_type = F_WRLCK;
3412 		else
3413 			lf.l_type = F_RDLCK;
3414 		if (fmode & FNONBLOCK)
3415 			type = 0;
3416 		else
3417 			type = F_WAIT;
3418 		vn_unlock(vp);
3419 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3420 			/*
3421 			 * release our private reference.
3422 			 */
3423 			fsetfd(p, NULL, indx);
3424 			fdrop(fp);
3425 			vrele(vp);
3426 			return (error);
3427 		}
3428 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3429 		fp->f_flag |= FHASLOCK;
3430 	}
3431 
3432 	/*
3433 	 * Clean up.  Associate the file pointer with the previously
3434 	 * reserved descriptor and return it.
3435 	 */
3436 	vput(vp);
3437 	fsetfd(p, fp, indx);
3438 	fdrop(fp);
3439 	uap->sysmsg_result = indx;
3440 	return (0);
3441 
3442 bad_drop:
3443 	fsetfd(p, NULL, indx);
3444 	fdrop(fp);
3445 bad:
3446 	vput(vp);
3447 	return (error);
3448 }
3449 
3450 /*
3451  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3452  */
3453 int
3454 sys_fhstat(struct fhstat_args *uap)
3455 {
3456 	struct thread *td = curthread;
3457 	struct stat sb;
3458 	fhandle_t fh;
3459 	struct mount *mp;
3460 	struct vnode *vp;
3461 	int error;
3462 
3463 	/*
3464 	 * Must be super user
3465 	 */
3466 	error = suser(td);
3467 	if (error)
3468 		return (error);
3469 
3470 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3471 	if (error)
3472 		return (error);
3473 
3474 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3475 		return (ESTALE);
3476 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3477 		return (error);
3478 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3479 	vput(vp);
3480 	if (error)
3481 		return (error);
3482 	error = copyout(&sb, uap->sb, sizeof(sb));
3483 	return (error);
3484 }
3485 
3486 /*
3487  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3488  */
3489 int
3490 sys_fhstatfs(struct fhstatfs_args *uap)
3491 {
3492 	struct thread *td = curthread;
3493 	struct proc *p = td->td_proc;
3494 	struct statfs *sp;
3495 	struct mount *mp;
3496 	struct vnode *vp;
3497 	struct statfs sb;
3498 	char *fullpath, *freepath;
3499 	fhandle_t fh;
3500 	int error;
3501 
3502 	/*
3503 	 * Must be super user
3504 	 */
3505 	if ((error = suser(td)))
3506 		return (error);
3507 
3508 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3509 		return (error);
3510 
3511 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3512 		return (ESTALE);
3513 
3514 	if (p != NULL && !chroot_visible_mnt(mp, p))
3515 		return (ESTALE);
3516 
3517 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3518 		return (error);
3519 	mp = vp->v_mount;
3520 	sp = &mp->mnt_stat;
3521 	vput(vp);
3522 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3523 		return (error);
3524 
3525 	error = cache_fullpath(p, &mp->mnt_ncmountpt, &fullpath, &freepath);
3526 	if (error)
3527 		return(error);
3528 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3529 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3530 	kfree(freepath, M_TEMP);
3531 
3532 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3533 	if (suser(td)) {
3534 		bcopy(sp, &sb, sizeof(sb));
3535 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3536 		sp = &sb;
3537 	}
3538 	return (copyout(sp, uap->buf, sizeof(*sp)));
3539 }
3540 
3541 /*
3542  * Syscall to push extended attribute configuration information into the
3543  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3544  * a command (int cmd), and attribute name and misc data.  For now, the
3545  * attribute name is left in userspace for consumption by the VFS_op.
3546  * It will probably be changed to be copied into sysspace by the
3547  * syscall in the future, once issues with various consumers of the
3548  * attribute code have raised their hands.
3549  *
3550  * Currently this is used only by UFS Extended Attributes.
3551  */
3552 int
3553 sys_extattrctl(struct extattrctl_args *uap)
3554 {
3555 	struct nlookupdata nd;
3556 	struct mount *mp;
3557 	struct vnode *vp;
3558 	int error;
3559 
3560 	vp = NULL;
3561 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3562 	if (error == 0)
3563 		error = nlookup(&nd);
3564 	if (error == 0) {
3565 		mp = nd.nl_nch.mount;
3566 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3567 				uap->attrname, uap->arg,
3568 				nd.nl_cred);
3569 	}
3570 	nlookup_done(&nd);
3571 	return (error);
3572 }
3573 
3574 /*
3575  * Syscall to set a named extended attribute on a file or directory.
3576  * Accepts attribute name, and a uio structure pointing to the data to set.
3577  * The uio is consumed in the style of writev().  The real work happens
3578  * in VOP_SETEXTATTR().
3579  */
3580 int
3581 sys_extattr_set_file(struct extattr_set_file_args *uap)
3582 {
3583 	char attrname[EXTATTR_MAXNAMELEN];
3584 	struct iovec aiov[UIO_SMALLIOV];
3585 	struct iovec *needfree;
3586 	struct nlookupdata nd;
3587 	struct iovec *iov;
3588 	struct vnode *vp;
3589 	struct uio auio;
3590 	u_int iovlen;
3591 	u_int cnt;
3592 	int error;
3593 	int i;
3594 
3595 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3596 	if (error)
3597 		return (error);
3598 
3599 	vp = NULL;
3600 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3601 	if (error == 0)
3602 		error = nlookup(&nd);
3603 	if (error == 0)
3604 		error = ncp_writechk(&nd.nl_nch);
3605 	if (error == 0)
3606 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3607 	if (error) {
3608 		nlookup_done(&nd);
3609 		return (error);
3610 	}
3611 
3612 	needfree = NULL;
3613 	iovlen = uap->iovcnt * sizeof(struct iovec);
3614 	if (uap->iovcnt > UIO_SMALLIOV) {
3615 		if (uap->iovcnt > UIO_MAXIOV) {
3616 			error = EINVAL;
3617 			goto done;
3618 		}
3619 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3620 		needfree = iov;
3621 	} else {
3622 		iov = aiov;
3623 	}
3624 	auio.uio_iov = iov;
3625 	auio.uio_iovcnt = uap->iovcnt;
3626 	auio.uio_rw = UIO_WRITE;
3627 	auio.uio_segflg = UIO_USERSPACE;
3628 	auio.uio_td = nd.nl_td;
3629 	auio.uio_offset = 0;
3630 	if ((error = copyin(uap->iovp, iov, iovlen)))
3631 		goto done;
3632 	auio.uio_resid = 0;
3633 	for (i = 0; i < uap->iovcnt; i++) {
3634 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3635 			error = EINVAL;
3636 			goto done;
3637 		}
3638 		auio.uio_resid += iov->iov_len;
3639 		iov++;
3640 	}
3641 	cnt = auio.uio_resid;
3642 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3643 	cnt -= auio.uio_resid;
3644 	uap->sysmsg_result = cnt;
3645 done:
3646 	vput(vp);
3647 	nlookup_done(&nd);
3648 	if (needfree)
3649 		FREE(needfree, M_IOV);
3650 	return (error);
3651 }
3652 
3653 /*
3654  * Syscall to get a named extended attribute on a file or directory.
3655  * Accepts attribute name, and a uio structure pointing to a buffer for the
3656  * data.  The uio is consumed in the style of readv().  The real work
3657  * happens in VOP_GETEXTATTR();
3658  */
3659 int
3660 sys_extattr_get_file(struct extattr_get_file_args *uap)
3661 {
3662 	char attrname[EXTATTR_MAXNAMELEN];
3663 	struct iovec aiov[UIO_SMALLIOV];
3664 	struct iovec *needfree;
3665 	struct nlookupdata nd;
3666 	struct iovec *iov;
3667 	struct vnode *vp;
3668 	struct uio auio;
3669 	u_int iovlen;
3670 	u_int cnt;
3671 	int error;
3672 	int i;
3673 
3674 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3675 	if (error)
3676 		return (error);
3677 
3678 	vp = NULL;
3679 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3680 	if (error == 0)
3681 		error = nlookup(&nd);
3682 	if (error == 0)
3683 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3684 	if (error) {
3685 		nlookup_done(&nd);
3686 		return (error);
3687 	}
3688 
3689 	iovlen = uap->iovcnt * sizeof (struct iovec);
3690 	needfree = NULL;
3691 	if (uap->iovcnt > UIO_SMALLIOV) {
3692 		if (uap->iovcnt > UIO_MAXIOV) {
3693 			error = EINVAL;
3694 			goto done;
3695 		}
3696 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3697 		needfree = iov;
3698 	} else {
3699 		iov = aiov;
3700 	}
3701 	auio.uio_iov = iov;
3702 	auio.uio_iovcnt = uap->iovcnt;
3703 	auio.uio_rw = UIO_READ;
3704 	auio.uio_segflg = UIO_USERSPACE;
3705 	auio.uio_td = nd.nl_td;
3706 	auio.uio_offset = 0;
3707 	if ((error = copyin(uap->iovp, iov, iovlen)))
3708 		goto done;
3709 	auio.uio_resid = 0;
3710 	for (i = 0; i < uap->iovcnt; i++) {
3711 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3712 			error = EINVAL;
3713 			goto done;
3714 		}
3715 		auio.uio_resid += iov->iov_len;
3716 		iov++;
3717 	}
3718 	cnt = auio.uio_resid;
3719 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3720 	cnt -= auio.uio_resid;
3721 	uap->sysmsg_result = cnt;
3722 done:
3723 	vput(vp);
3724 	nlookup_done(&nd);
3725 	if (needfree)
3726 		FREE(needfree, M_IOV);
3727 	return(error);
3728 }
3729 
3730 /*
3731  * Syscall to delete a named extended attribute from a file or directory.
3732  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3733  */
3734 int
3735 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3736 {
3737 	char attrname[EXTATTR_MAXNAMELEN];
3738 	struct nlookupdata nd;
3739 	struct vnode *vp;
3740 	int error;
3741 
3742 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3743 	if (error)
3744 		return(error);
3745 
3746 	vp = NULL;
3747 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3748 	if (error == 0)
3749 		error = nlookup(&nd);
3750 	if (error == 0)
3751 		error = ncp_writechk(&nd.nl_nch);
3752 	if (error == 0)
3753 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3754 	if (error) {
3755 		nlookup_done(&nd);
3756 		return (error);
3757 	}
3758 
3759 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3760 	vput(vp);
3761 	nlookup_done(&nd);
3762 	return(error);
3763 }
3764 
3765 /*
3766  * Determine if the mount is visible to the process.
3767  */
3768 static int
3769 chroot_visible_mnt(struct mount *mp, struct proc *p)
3770 {
3771 	struct nchandle nch;
3772 
3773 	/*
3774 	 * Traverse from the mount point upwards.  If we hit the process
3775 	 * root then the mount point is visible to the process.
3776 	 */
3777 	nch = mp->mnt_ncmountpt;
3778 	while (nch.ncp) {
3779 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
3780 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
3781 			return(1);
3782 		}
3783 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
3784 			nch = nch.mount->mnt_ncmounton;
3785 		} else {
3786 			nch.ncp = nch.ncp->nc_parent;
3787 		}
3788 	}
3789 	return(0);
3790 }
3791 
3792