xref: /dragonfly/sys/kern/vfs_syscalls.c (revision 89a89091)
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  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/buf.h>
45 #include <sys/conf.h>
46 #include <sys/sysent.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/sysproto.h>
51 #include <sys/filedesc.h>
52 #include <sys/kernel.h>
53 #include <sys/fcntl.h>
54 #include <sys/file.h>
55 #include <sys/linker.h>
56 #include <sys/stat.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
59 #include <sys/proc.h>
60 #include <sys/priv.h>
61 #include <sys/jail.h>
62 #include <sys/namei.h>
63 #include <sys/nlookup.h>
64 #include <sys/dirent.h>
65 #include <sys/extattr.h>
66 #include <sys/spinlock.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/objcache.h>
69 #include <sys/sysctl.h>
70 
71 #include <sys/buf2.h>
72 #include <sys/file2.h>
73 #include <sys/spinlock2.h>
74 #include <sys/mplock2.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 
80 #include <machine/limits.h>
81 #include <machine/stdarg.h>
82 
83 #include <vfs/union/union.h>
84 
85 static void mount_warning(struct mount *mp, const char *ctl, ...)
86 		__printflike(2, 3);
87 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
88 static int checkvp_chdir (struct vnode *vn, struct thread *td);
89 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
90 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
91 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
92 static int getutimes (const struct timeval *, struct timespec *);
93 static int setfown (struct mount *, struct vnode *, uid_t, gid_t);
94 static int setfmode (struct vnode *, int);
95 static int setfflags (struct vnode *, int);
96 static int setutimes (struct vnode *, struct vattr *,
97 			const struct timespec *, int);
98 static int	usermount = 0;	/* if 1, non-root can mount fs. */
99 
100 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
101 
102 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
103     "Allow non-root users to mount filesystems");
104 
105 /*
106  * Virtual File System System Calls
107  */
108 
109 /*
110  * Mount a file system.
111  *
112  * mount_args(char *type, char *path, int flags, caddr_t data)
113  *
114  * MPALMOSTSAFE
115  */
116 int
117 sys_mount(struct mount_args *uap)
118 {
119 	struct thread *td = curthread;
120 	struct vnode *vp;
121 	struct nchandle nch;
122 	struct mount *mp, *nullmp;
123 	struct vfsconf *vfsp;
124 	int error, flag = 0, flag2 = 0;
125 	int hasmount;
126 	struct vattr va;
127 	struct nlookupdata nd;
128 	char fstypename[MFSNAMELEN];
129 	struct ucred *cred;
130 
131 	get_mplock();
132 	cred = td->td_ucred;
133 	if (jailed(cred)) {
134 		error = EPERM;
135 		goto done;
136 	}
137 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
138 		goto done;
139 
140 	/*
141 	 * Do not allow NFS export by non-root users.
142 	 */
143 	if (uap->flags & MNT_EXPORTED) {
144 		error = priv_check(td, PRIV_ROOT);
145 		if (error)
146 			goto done;
147 	}
148 	/*
149 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
150 	 */
151 	if (priv_check(td, PRIV_ROOT))
152 		uap->flags |= MNT_NOSUID | MNT_NODEV;
153 
154 	/*
155 	 * Lookup the requested path and extract the nch and vnode.
156 	 */
157 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
158 	if (error == 0) {
159 		if ((error = nlookup(&nd)) == 0) {
160 			if (nd.nl_nch.ncp->nc_vp == NULL)
161 				error = ENOENT;
162 		}
163 	}
164 	if (error) {
165 		nlookup_done(&nd);
166 		goto done;
167 	}
168 
169 	/*
170 	 * If the target filesystem is resolved via a nullfs mount, then
171 	 * nd.nl_nch.mount will be pointing to the nullfs mount structure
172 	 * instead of the target file system. We need it in case we are
173 	 * doing an update.
174 	 */
175 	nullmp = nd.nl_nch.mount;
176 
177 	/*
178 	 * Extract the locked+refd ncp and cleanup the nd structure
179 	 */
180 	nch = nd.nl_nch;
181 	cache_zero(&nd.nl_nch);
182 	nlookup_done(&nd);
183 
184 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
185 	    (mp = cache_findmount(&nch)) != NULL) {
186 		cache_dropmount(mp);
187 		hasmount = 1;
188 	} else {
189 		hasmount = 0;
190 	}
191 
192 
193 	/*
194 	 * now we have the locked ref'd nch and unreferenced vnode.
195 	 */
196 	vp = nch.ncp->nc_vp;
197 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
198 		cache_put(&nch);
199 		goto done;
200 	}
201 	cache_unlock(&nch);
202 
203 	/*
204 	 * Extract the file system type. We need to know this early, to take
205 	 * appropriate actions if we are dealing with a nullfs.
206 	 */
207         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
208                 cache_drop(&nch);
209                 vput(vp);
210 		goto done;
211         }
212 
213 	/*
214 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
215 	 */
216 	if (uap->flags & MNT_UPDATE) {
217 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
218 			cache_drop(&nch);
219 			vput(vp);
220 			error = EINVAL;
221 			goto done;
222 		}
223 
224 		if (strncmp(fstypename, "null", 5) == 0) {
225 			KKASSERT(nullmp);
226 			mp = nullmp;
227 		} else {
228 			mp = vp->v_mount;
229 		}
230 
231 		flag = mp->mnt_flag;
232 		flag2 = mp->mnt_kern_flag;
233 		/*
234 		 * We only allow the filesystem to be reloaded if it
235 		 * is currently mounted read-only.
236 		 */
237 		if ((uap->flags & MNT_RELOAD) &&
238 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
239 			cache_drop(&nch);
240 			vput(vp);
241 			error = EOPNOTSUPP;	/* Needs translation */
242 			goto done;
243 		}
244 		/*
245 		 * Only root, or the user that did the original mount is
246 		 * permitted to update it.
247 		 */
248 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
249 		    (error = priv_check(td, PRIV_ROOT))) {
250 			cache_drop(&nch);
251 			vput(vp);
252 			goto done;
253 		}
254 		if (vfs_busy(mp, LK_NOWAIT)) {
255 			cache_drop(&nch);
256 			vput(vp);
257 			error = EBUSY;
258 			goto done;
259 		}
260 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
261 			cache_drop(&nch);
262 			vfs_unbusy(mp);
263 			vput(vp);
264 			error = EBUSY;
265 			goto done;
266 		}
267 		vsetflags(vp, VMOUNT);
268 		mp->mnt_flag |=
269 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
270 		vn_unlock(vp);
271 		goto update;
272 	}
273 	/*
274 	 * If the user is not root, ensure that they own the directory
275 	 * onto which we are attempting to mount.
276 	 */
277 	if ((error = VOP_GETATTR(vp, &va)) ||
278 	    (va.va_uid != cred->cr_uid && (error = priv_check(td, PRIV_ROOT)))) {
279 		cache_drop(&nch);
280 		vput(vp);
281 		goto done;
282 	}
283 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
284 		cache_drop(&nch);
285 		vput(vp);
286 		goto done;
287 	}
288 	if (vp->v_type != VDIR) {
289 		cache_drop(&nch);
290 		vput(vp);
291 		error = ENOTDIR;
292 		goto done;
293 	}
294 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
295 		cache_drop(&nch);
296 		vput(vp);
297 		error = EPERM;
298 		goto done;
299 	}
300 	vfsp = vfsconf_find_by_name(fstypename);
301 	if (vfsp == NULL) {
302 		linker_file_t lf;
303 
304 		/* Only load modules for root (very important!) */
305 		if ((error = priv_check(td, PRIV_ROOT)) != 0) {
306 			cache_drop(&nch);
307 			vput(vp);
308 			goto done;
309 		}
310 		error = linker_load_file(fstypename, &lf);
311 		if (error || lf == NULL) {
312 			cache_drop(&nch);
313 			vput(vp);
314 			if (lf == NULL)
315 				error = ENODEV;
316 			goto done;
317 		}
318 		lf->userrefs++;
319 		/* lookup again, see if the VFS was loaded */
320 		vfsp = vfsconf_find_by_name(fstypename);
321 		if (vfsp == NULL) {
322 			lf->userrefs--;
323 			linker_file_unload(lf);
324 			cache_drop(&nch);
325 			vput(vp);
326 			error = ENODEV;
327 			goto done;
328 		}
329 	}
330 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
331 		cache_drop(&nch);
332 		vput(vp);
333 		error = EBUSY;
334 		goto done;
335 	}
336 	vsetflags(vp, VMOUNT);
337 
338 	/*
339 	 * Allocate and initialize the filesystem.
340 	 */
341 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
342 	mount_init(mp);
343 	vfs_busy(mp, LK_NOWAIT);
344 	mp->mnt_op = vfsp->vfc_vfsops;
345 	mp->mnt_vfc = vfsp;
346 	vfsp->vfc_refcount++;
347 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
348 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
349 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
350 	mp->mnt_stat.f_owner = cred->cr_uid;
351 	vn_unlock(vp);
352 update:
353 	/*
354 	 * Set the mount level flags.
355 	 */
356 	if (uap->flags & MNT_RDONLY)
357 		mp->mnt_flag |= MNT_RDONLY;
358 	else if (mp->mnt_flag & MNT_RDONLY)
359 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
360 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
361 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
362 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
363 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
364 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
365 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
366 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
367 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
368 	/*
369 	 * Mount the filesystem.
370 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
371 	 * get.
372 	 */
373 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
374 	if (mp->mnt_flag & MNT_UPDATE) {
375 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
376 			mp->mnt_flag &= ~MNT_RDONLY;
377 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
378 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
379 		if (error) {
380 			mp->mnt_flag = flag;
381 			mp->mnt_kern_flag = flag2;
382 		}
383 		vfs_unbusy(mp);
384 		vclrflags(vp, VMOUNT);
385 		vrele(vp);
386 		cache_drop(&nch);
387 		goto done;
388 	}
389 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
390 	/*
391 	 * Put the new filesystem on the mount list after root.  The mount
392 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
393 	 * up) which represents the root of the mount.  The lookup code
394 	 * detects the mount point going forward and checks the root of
395 	 * the mount going backwards.
396 	 *
397 	 * It is not necessary to invalidate or purge the vnode underneath
398 	 * because elements under the mount will be given their own glue
399 	 * namecache record.
400 	 */
401 	if (!error) {
402 		if (mp->mnt_ncmountpt.ncp == NULL) {
403 			/*
404 			 * allocate, then unlock, but leave the ref intact
405 			 */
406 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
407 			cache_unlock(&mp->mnt_ncmountpt);
408 		}
409 		mp->mnt_ncmounton = nch;		/* inherits ref */
410 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
411 
412 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
413 		vclrflags(vp, VMOUNT);
414 		mountlist_insert(mp, MNTINS_LAST);
415 		vn_unlock(vp);
416 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
417 		error = vfs_allocate_syncvnode(mp);
418 		vfs_unbusy(mp);
419 		error = VFS_START(mp, 0);
420 		vrele(vp);
421 	} else {
422 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
423 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
424 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
425 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
426 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
427 		vclrflags(vp, VMOUNT);
428 		mp->mnt_vfc->vfc_refcount--;
429 		vfs_unbusy(mp);
430 		kfree(mp, M_MOUNT);
431 		cache_drop(&nch);
432 		vput(vp);
433 	}
434 done:
435 	rel_mplock();
436 	return (error);
437 }
438 
439 /*
440  * Scan all active processes to see if any of them have a current
441  * or root directory onto which the new filesystem has just been
442  * mounted. If so, replace them with the new mount point.
443  *
444  * The passed ncp is ref'd and locked (from the mount code) and
445  * must be associated with the vnode representing the root of the
446  * mount point.
447  */
448 struct checkdirs_info {
449 	struct nchandle old_nch;
450 	struct nchandle new_nch;
451 	struct vnode *old_vp;
452 	struct vnode *new_vp;
453 };
454 
455 static int checkdirs_callback(struct proc *p, void *data);
456 
457 static void
458 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
459 {
460 	struct checkdirs_info info;
461 	struct vnode *olddp;
462 	struct vnode *newdp;
463 	struct mount *mp;
464 
465 	/*
466 	 * If the old mount point's vnode has a usecount of 1, it is not
467 	 * being held as a descriptor anywhere.
468 	 */
469 	olddp = old_nch->ncp->nc_vp;
470 	if (olddp == NULL || olddp->v_sysref.refcnt == 1)
471 		return;
472 
473 	/*
474 	 * Force the root vnode of the new mount point to be resolved
475 	 * so we can update any matching processes.
476 	 */
477 	mp = new_nch->mount;
478 	if (VFS_ROOT(mp, &newdp))
479 		panic("mount: lost mount");
480 	cache_setunresolved(new_nch);
481 	cache_setvp(new_nch, newdp);
482 
483 	/*
484 	 * Special handling of the root node
485 	 */
486 	if (rootvnode == olddp) {
487 		vref(newdp);
488 		vfs_cache_setroot(newdp, cache_hold(new_nch));
489 	}
490 
491 	/*
492 	 * Pass newdp separately so the callback does not have to access
493 	 * it via new_nch->ncp->nc_vp.
494 	 */
495 	info.old_nch = *old_nch;
496 	info.new_nch = *new_nch;
497 	info.new_vp = newdp;
498 	allproc_scan(checkdirs_callback, &info);
499 	vput(newdp);
500 }
501 
502 /*
503  * NOTE: callback is not MP safe because the scanned process's filedesc
504  * structure can be ripped out from under us, amoung other things.
505  */
506 static int
507 checkdirs_callback(struct proc *p, void *data)
508 {
509 	struct checkdirs_info *info = data;
510 	struct filedesc *fdp;
511 	struct nchandle ncdrop1;
512 	struct nchandle ncdrop2;
513 	struct vnode *vprele1;
514 	struct vnode *vprele2;
515 
516 	if ((fdp = p->p_fd) != NULL) {
517 		cache_zero(&ncdrop1);
518 		cache_zero(&ncdrop2);
519 		vprele1 = NULL;
520 		vprele2 = NULL;
521 
522 		/*
523 		 * MPUNSAFE - XXX fdp can be pulled out from under a
524 		 * foreign process.
525 		 *
526 		 * A shared filedesc is ok, we don't have to copy it
527 		 * because we are making this change globally.
528 		 */
529 		spin_lock(&fdp->fd_spin);
530 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
531 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
532 			vprele1 = fdp->fd_cdir;
533 			vref(info->new_vp);
534 			fdp->fd_cdir = info->new_vp;
535 			ncdrop1 = fdp->fd_ncdir;
536 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
537 		}
538 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
539 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
540 			vprele2 = fdp->fd_rdir;
541 			vref(info->new_vp);
542 			fdp->fd_rdir = info->new_vp;
543 			ncdrop2 = fdp->fd_nrdir;
544 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
545 		}
546 		spin_unlock(&fdp->fd_spin);
547 		if (ncdrop1.ncp)
548 			cache_drop(&ncdrop1);
549 		if (ncdrop2.ncp)
550 			cache_drop(&ncdrop2);
551 		if (vprele1)
552 			vrele(vprele1);
553 		if (vprele2)
554 			vrele(vprele2);
555 	}
556 	return(0);
557 }
558 
559 /*
560  * Unmount a file system.
561  *
562  * Note: unmount takes a path to the vnode mounted on as argument,
563  * not special file (as before).
564  *
565  * umount_args(char *path, int flags)
566  *
567  * MPALMOSTSAFE
568  */
569 int
570 sys_unmount(struct unmount_args *uap)
571 {
572 	struct thread *td = curthread;
573 	struct proc *p __debugvar = td->td_proc;
574 	struct mount *mp = NULL;
575 	struct nlookupdata nd;
576 	int error;
577 
578 	KKASSERT(p);
579 	get_mplock();
580 	if (td->td_ucred->cr_prison != NULL) {
581 		error = EPERM;
582 		goto done;
583 	}
584 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
585 		goto done;
586 
587 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
588 	if (error == 0)
589 		error = nlookup(&nd);
590 	if (error)
591 		goto out;
592 
593 	mp = nd.nl_nch.mount;
594 
595 	/*
596 	 * Only root, or the user that did the original mount is
597 	 * permitted to unmount this filesystem.
598 	 */
599 	if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
600 	    (error = priv_check(td, PRIV_ROOT)))
601 		goto out;
602 
603 	/*
604 	 * Don't allow unmounting the root file system.
605 	 */
606 	if (mp->mnt_flag & MNT_ROOTFS) {
607 		error = EINVAL;
608 		goto out;
609 	}
610 
611 	/*
612 	 * Must be the root of the filesystem
613 	 */
614 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
615 		error = EINVAL;
616 		goto out;
617 	}
618 
619 out:
620 	nlookup_done(&nd);
621 	if (error == 0)
622 		error = dounmount(mp, uap->flags);
623 done:
624 	rel_mplock();
625 	return (error);
626 }
627 
628 /*
629  * Do the actual file system unmount.
630  */
631 static int
632 dounmount_interlock(struct mount *mp)
633 {
634 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
635 		return (EBUSY);
636 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
637 	return(0);
638 }
639 
640 static int
641 unmount_allproc_cb(struct proc *p, void *arg)
642 {
643 	struct mount *mp;
644 
645 	if (p->p_textnch.ncp == NULL)
646 		return 0;
647 
648 	mp = (struct mount *)arg;
649 	if (p->p_textnch.mount == mp)
650 		cache_drop(&p->p_textnch);
651 
652 	return 0;
653 }
654 
655 int
656 dounmount(struct mount *mp, int flags)
657 {
658 	struct namecache *ncp;
659 	struct nchandle nch;
660 	struct vnode *vp;
661 	int error;
662 	int async_flag;
663 	int lflags;
664 	int freeok = 1;
665 
666 	/*
667 	 * Exclusive access for unmounting purposes
668 	 */
669 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
670 		return (error);
671 
672 	/*
673 	 * Allow filesystems to detect that a forced unmount is in progress.
674 	 */
675 	if (flags & MNT_FORCE)
676 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
677 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
678 	error = lockmgr(&mp->mnt_lock, lflags);
679 	if (error) {
680 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
681 		if (mp->mnt_kern_flag & MNTK_MWAIT)
682 			wakeup(mp);
683 		return (error);
684 	}
685 
686 	if (mp->mnt_flag & MNT_EXPUBLIC)
687 		vfs_setpublicfs(NULL, NULL, NULL);
688 
689 	vfs_msync(mp, MNT_WAIT);
690 	async_flag = mp->mnt_flag & MNT_ASYNC;
691 	mp->mnt_flag &=~ MNT_ASYNC;
692 
693 	/*
694 	 * If this filesystem isn't aliasing other filesystems,
695 	 * try to invalidate any remaining namecache entries and
696 	 * check the count afterwords.
697 	 */
698 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
699 		cache_lock(&mp->mnt_ncmountpt);
700 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
701 		cache_unlock(&mp->mnt_ncmountpt);
702 
703 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
704 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
705 			allproc_scan(&unmount_allproc_cb, mp);
706 		}
707 
708 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
709 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
710 
711 			if ((flags & MNT_FORCE) == 0) {
712 				error = EBUSY;
713 				mount_warning(mp, "Cannot unmount: "
714 						  "%d namecache "
715 						  "references still "
716 						  "present",
717 						  ncp->nc_refs - 1);
718 			} else {
719 				mount_warning(mp, "Forced unmount: "
720 						  "%d namecache "
721 						  "references still "
722 						  "present",
723 						  ncp->nc_refs - 1);
724 				freeok = 0;
725 			}
726 		}
727 	}
728 
729 	/*
730 	 * nchandle records ref the mount structure.  Expect a count of 1
731 	 * (our mount->mnt_ncmountpt).
732 	 */
733 	if (mp->mnt_refs != 1) {
734 		if ((flags & MNT_FORCE) == 0) {
735 			mount_warning(mp, "Cannot unmount: "
736 					  "%d process references still "
737 					  "present", mp->mnt_refs);
738 			error = EBUSY;
739 		} else {
740 			mount_warning(mp, "Forced unmount: "
741 					  "%d process references still "
742 					  "present", mp->mnt_refs);
743 			freeok = 0;
744 		}
745 	}
746 
747 	/*
748 	 * Decomission our special mnt_syncer vnode.  This also stops
749 	 * the vnlru code.  If we are unable to unmount we recommission
750 	 * the vnode.
751 	 */
752 	if (error == 0) {
753 		if ((vp = mp->mnt_syncer) != NULL) {
754 			mp->mnt_syncer = NULL;
755 			vrele(vp);
756 		}
757 		if (((mp->mnt_flag & MNT_RDONLY) ||
758 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
759 		    (flags & MNT_FORCE)) {
760 			error = VFS_UNMOUNT(mp, flags);
761 		}
762 	}
763 	if (error) {
764 		if (mp->mnt_syncer == NULL)
765 			vfs_allocate_syncvnode(mp);
766 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
767 		mp->mnt_flag |= async_flag;
768 		lockmgr(&mp->mnt_lock, LK_RELEASE);
769 		if (mp->mnt_kern_flag & MNTK_MWAIT)
770 			wakeup(mp);
771 		return (error);
772 	}
773 	/*
774 	 * Clean up any journals still associated with the mount after
775 	 * filesystem activity has ceased.
776 	 */
777 	journal_remove_all_journals(mp,
778 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
779 
780 	mountlist_remove(mp);
781 
782 	/*
783 	 * Remove any installed vnode ops here so the individual VFSs don't
784 	 * have to.
785 	 */
786 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
787 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
788 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
789 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
790 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
791 
792 	if (mp->mnt_ncmountpt.ncp != NULL) {
793 		nch = mp->mnt_ncmountpt;
794 		cache_zero(&mp->mnt_ncmountpt);
795 		cache_clrmountpt(&nch);
796 		cache_drop(&nch);
797 	}
798 	if (mp->mnt_ncmounton.ncp != NULL) {
799 		nch = mp->mnt_ncmounton;
800 		cache_zero(&mp->mnt_ncmounton);
801 		cache_clrmountpt(&nch);
802 		cache_drop(&nch);
803 	}
804 
805 	mp->mnt_vfc->vfc_refcount--;
806 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
807 		panic("unmount: dangling vnode");
808 	lockmgr(&mp->mnt_lock, LK_RELEASE);
809 	if (mp->mnt_kern_flag & MNTK_MWAIT)
810 		wakeup(mp);
811 	if (freeok)
812 		kfree(mp, M_MOUNT);
813 	return (0);
814 }
815 
816 static
817 void
818 mount_warning(struct mount *mp, const char *ctl, ...)
819 {
820 	char *ptr;
821 	char *buf;
822 	__va_list va;
823 
824 	__va_start(va, ctl);
825 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf, 0) == 0) {
826 		kprintf("unmount(%s): ", ptr);
827 		kvprintf(ctl, va);
828 		kprintf("\n");
829 		kfree(buf, M_TEMP);
830 	} else {
831 		kprintf("unmount(%p", mp);
832 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
833 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
834 		kprintf("): ");
835 		kvprintf(ctl, va);
836 		kprintf("\n");
837 	}
838 	__va_end(va);
839 }
840 
841 /*
842  * Shim cache_fullpath() to handle the case where a process is chrooted into
843  * a subdirectory of a mount.  In this case if the root mount matches the
844  * process root directory's mount we have to specify the process's root
845  * directory instead of the mount point, because the mount point might
846  * be above the root directory.
847  */
848 static
849 int
850 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
851 {
852 	struct nchandle *nch;
853 
854 	if (p && p->p_fd->fd_nrdir.mount == mp)
855 		nch = &p->p_fd->fd_nrdir;
856 	else
857 		nch = &mp->mnt_ncmountpt;
858 	return(cache_fullpath(p, nch, rb, fb, 0));
859 }
860 
861 /*
862  * Sync each mounted filesystem.
863  */
864 
865 #ifdef DEBUG
866 static int syncprt = 0;
867 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
868 #endif /* DEBUG */
869 
870 static int sync_callback(struct mount *mp, void *data);
871 
872 int
873 sys_sync(struct sync_args *uap)
874 {
875 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
876 #ifdef DEBUG
877 	/*
878 	 * print out buffer pool stat information on each sync() call.
879 	 */
880 	if (syncprt)
881 		vfs_bufstats();
882 #endif /* DEBUG */
883 	return (0);
884 }
885 
886 static
887 int
888 sync_callback(struct mount *mp, void *data __unused)
889 {
890 	int asyncflag;
891 
892 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
893 		asyncflag = mp->mnt_flag & MNT_ASYNC;
894 		mp->mnt_flag &= ~MNT_ASYNC;
895 		vfs_msync(mp, MNT_NOWAIT);
896 		VFS_SYNC(mp, MNT_NOWAIT | MNT_LAZY);
897 		mp->mnt_flag |= asyncflag;
898 	}
899 	return(0);
900 }
901 
902 /* XXX PRISON: could be per prison flag */
903 static int prison_quotas;
904 #if 0
905 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
906 #endif
907 
908 /*
909  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
910  *
911  * Change filesystem quotas.
912  *
913  * MPALMOSTSAFE
914  */
915 int
916 sys_quotactl(struct quotactl_args *uap)
917 {
918 	struct nlookupdata nd;
919 	struct thread *td;
920 	struct mount *mp;
921 	int error;
922 
923 	get_mplock();
924 	td = curthread;
925 	if (td->td_ucred->cr_prison && !prison_quotas) {
926 		error = EPERM;
927 		goto done;
928 	}
929 
930 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
931 	if (error == 0)
932 		error = nlookup(&nd);
933 	if (error == 0) {
934 		mp = nd.nl_nch.mount;
935 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
936 				    uap->arg, nd.nl_cred);
937 	}
938 	nlookup_done(&nd);
939 done:
940 	rel_mplock();
941 	return (error);
942 }
943 
944 /*
945  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
946  *		void *buf, int buflen)
947  *
948  * This function operates on a mount point and executes the specified
949  * operation using the specified control data, and possibly returns data.
950  *
951  * The actual number of bytes stored in the result buffer is returned, 0
952  * if none, otherwise an error is returned.
953  *
954  * MPALMOSTSAFE
955  */
956 int
957 sys_mountctl(struct mountctl_args *uap)
958 {
959 	struct thread *td = curthread;
960 	struct proc *p = td->td_proc;
961 	struct file *fp;
962 	void *ctl = NULL;
963 	void *buf = NULL;
964 	char *path = NULL;
965 	int error;
966 
967 	/*
968 	 * Sanity and permissions checks.  We must be root.
969 	 */
970 	KKASSERT(p);
971 	if (td->td_ucred->cr_prison != NULL)
972 		return (EPERM);
973 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
974 	    (error = priv_check(td, PRIV_ROOT)) != 0)
975 		return (error);
976 
977 	/*
978 	 * Argument length checks
979 	 */
980 	if (uap->ctllen < 0 || uap->ctllen > 1024)
981 		return (EINVAL);
982 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
983 		return (EINVAL);
984 	if (uap->path == NULL)
985 		return (EINVAL);
986 
987 	/*
988 	 * Allocate the necessary buffers and copyin data
989 	 */
990 	path = objcache_get(namei_oc, M_WAITOK);
991 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
992 	if (error)
993 		goto done;
994 
995 	if (uap->ctllen) {
996 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
997 		error = copyin(uap->ctl, ctl, uap->ctllen);
998 		if (error)
999 			goto done;
1000 	}
1001 	if (uap->buflen)
1002 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1003 
1004 	/*
1005 	 * Validate the descriptor
1006 	 */
1007 	if (uap->fd >= 0) {
1008 		fp = holdfp(p->p_fd, uap->fd, -1);
1009 		if (fp == NULL) {
1010 			error = EBADF;
1011 			goto done;
1012 		}
1013 	} else {
1014 		fp = NULL;
1015 	}
1016 
1017 	/*
1018 	 * Execute the internal kernel function and clean up.
1019 	 */
1020 	get_mplock();
1021 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1022 	rel_mplock();
1023 	if (fp)
1024 		fdrop(fp);
1025 	if (error == 0 && uap->sysmsg_result > 0)
1026 		error = copyout(buf, uap->buf, uap->sysmsg_result);
1027 done:
1028 	if (path)
1029 		objcache_put(namei_oc, path);
1030 	if (ctl)
1031 		kfree(ctl, M_TEMP);
1032 	if (buf)
1033 		kfree(buf, M_TEMP);
1034 	return (error);
1035 }
1036 
1037 /*
1038  * Execute a mount control operation by resolving the path to a mount point
1039  * and calling vop_mountctl().
1040  *
1041  * Use the mount point from the nch instead of the vnode so nullfs mounts
1042  * can properly spike the VOP.
1043  */
1044 int
1045 kern_mountctl(const char *path, int op, struct file *fp,
1046 		const void *ctl, int ctllen,
1047 		void *buf, int buflen, int *res)
1048 {
1049 	struct vnode *vp;
1050 	struct mount *mp;
1051 	struct nlookupdata nd;
1052 	int error;
1053 
1054 	*res = 0;
1055 	vp = NULL;
1056 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1057 	if (error == 0)
1058 		error = nlookup(&nd);
1059 	if (error == 0)
1060 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1061 	mp = nd.nl_nch.mount;
1062 	nlookup_done(&nd);
1063 	if (error)
1064 		return (error);
1065 	vn_unlock(vp);
1066 
1067 	/*
1068 	 * Must be the root of the filesystem
1069 	 */
1070 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1071 		vrele(vp);
1072 		return (EINVAL);
1073 	}
1074 	error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1075 			     buf, buflen, res);
1076 	vrele(vp);
1077 	return (error);
1078 }
1079 
1080 int
1081 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1082 {
1083 	struct thread *td = curthread;
1084 	struct proc *p = td->td_proc;
1085 	struct mount *mp;
1086 	struct statfs *sp;
1087 	char *fullpath, *freepath;
1088 	int error;
1089 
1090 	if ((error = nlookup(nd)) != 0)
1091 		return (error);
1092 	mp = nd->nl_nch.mount;
1093 	sp = &mp->mnt_stat;
1094 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1095 		return (error);
1096 
1097 	error = mount_path(p, mp, &fullpath, &freepath);
1098 	if (error)
1099 		return(error);
1100 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1101 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1102 	kfree(freepath, M_TEMP);
1103 
1104 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1105 	bcopy(sp, buf, sizeof(*buf));
1106 	/* Only root should have access to the fsid's. */
1107 	if (priv_check(td, PRIV_ROOT))
1108 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1109 	return (0);
1110 }
1111 
1112 /*
1113  * statfs_args(char *path, struct statfs *buf)
1114  *
1115  * Get filesystem statistics.
1116  */
1117 int
1118 sys_statfs(struct statfs_args *uap)
1119 {
1120 	struct nlookupdata nd;
1121 	struct statfs buf;
1122 	int error;
1123 
1124 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1125 	if (error == 0)
1126 		error = kern_statfs(&nd, &buf);
1127 	nlookup_done(&nd);
1128 	if (error == 0)
1129 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1130 	return (error);
1131 }
1132 
1133 int
1134 kern_fstatfs(int fd, struct statfs *buf)
1135 {
1136 	struct thread *td = curthread;
1137 	struct proc *p = td->td_proc;
1138 	struct file *fp;
1139 	struct mount *mp;
1140 	struct statfs *sp;
1141 	char *fullpath, *freepath;
1142 	int error;
1143 
1144 	KKASSERT(p);
1145 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1146 		return (error);
1147 
1148 	/*
1149 	 * Try to use mount info from any overlays rather than the
1150 	 * mount info for the underlying vnode, otherwise we will
1151 	 * fail when operating on null-mounted paths inside a chroot.
1152 	 */
1153 	if ((mp = fp->f_nchandle.mount) == NULL)
1154 		mp = ((struct vnode *)fp->f_data)->v_mount;
1155 	if (mp == NULL) {
1156 		error = EBADF;
1157 		goto done;
1158 	}
1159 	if (fp->f_cred == NULL) {
1160 		error = EINVAL;
1161 		goto done;
1162 	}
1163 	sp = &mp->mnt_stat;
1164 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1165 		goto done;
1166 
1167 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1168 		goto done;
1169 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1170 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1171 	kfree(freepath, M_TEMP);
1172 
1173 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1174 	bcopy(sp, buf, sizeof(*buf));
1175 
1176 	/* Only root should have access to the fsid's. */
1177 	if (priv_check(td, PRIV_ROOT))
1178 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1179 	error = 0;
1180 done:
1181 	fdrop(fp);
1182 	return (error);
1183 }
1184 
1185 /*
1186  * fstatfs_args(int fd, struct statfs *buf)
1187  *
1188  * Get filesystem statistics.
1189  */
1190 int
1191 sys_fstatfs(struct fstatfs_args *uap)
1192 {
1193 	struct statfs buf;
1194 	int error;
1195 
1196 	error = kern_fstatfs(uap->fd, &buf);
1197 
1198 	if (error == 0)
1199 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1200 	return (error);
1201 }
1202 
1203 int
1204 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1205 {
1206 	struct mount *mp;
1207 	struct statvfs *sp;
1208 	int error;
1209 
1210 	if ((error = nlookup(nd)) != 0)
1211 		return (error);
1212 	mp = nd->nl_nch.mount;
1213 	sp = &mp->mnt_vstat;
1214 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1215 		return (error);
1216 
1217 	sp->f_flag = 0;
1218 	if (mp->mnt_flag & MNT_RDONLY)
1219 		sp->f_flag |= ST_RDONLY;
1220 	if (mp->mnt_flag & MNT_NOSUID)
1221 		sp->f_flag |= ST_NOSUID;
1222 	bcopy(sp, buf, sizeof(*buf));
1223 	return (0);
1224 }
1225 
1226 /*
1227  * statfs_args(char *path, struct statfs *buf)
1228  *
1229  * Get filesystem statistics.
1230  */
1231 int
1232 sys_statvfs(struct statvfs_args *uap)
1233 {
1234 	struct nlookupdata nd;
1235 	struct statvfs buf;
1236 	int error;
1237 
1238 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1239 	if (error == 0)
1240 		error = kern_statvfs(&nd, &buf);
1241 	nlookup_done(&nd);
1242 	if (error == 0)
1243 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1244 	return (error);
1245 }
1246 
1247 int
1248 kern_fstatvfs(int fd, struct statvfs *buf)
1249 {
1250 	struct thread *td = curthread;
1251 	struct proc *p = td->td_proc;
1252 	struct file *fp;
1253 	struct mount *mp;
1254 	struct statvfs *sp;
1255 	int error;
1256 
1257 	KKASSERT(p);
1258 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1259 		return (error);
1260 	if ((mp = fp->f_nchandle.mount) == NULL)
1261 		mp = ((struct vnode *)fp->f_data)->v_mount;
1262 	if (mp == NULL) {
1263 		error = EBADF;
1264 		goto done;
1265 	}
1266 	if (fp->f_cred == NULL) {
1267 		error = EINVAL;
1268 		goto done;
1269 	}
1270 	sp = &mp->mnt_vstat;
1271 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1272 		goto done;
1273 
1274 	sp->f_flag = 0;
1275 	if (mp->mnt_flag & MNT_RDONLY)
1276 		sp->f_flag |= ST_RDONLY;
1277 	if (mp->mnt_flag & MNT_NOSUID)
1278 		sp->f_flag |= ST_NOSUID;
1279 
1280 	bcopy(sp, buf, sizeof(*buf));
1281 	error = 0;
1282 done:
1283 	fdrop(fp);
1284 	return (error);
1285 }
1286 
1287 /*
1288  * fstatfs_args(int fd, struct statfs *buf)
1289  *
1290  * Get filesystem statistics.
1291  */
1292 int
1293 sys_fstatvfs(struct fstatvfs_args *uap)
1294 {
1295 	struct statvfs buf;
1296 	int error;
1297 
1298 	error = kern_fstatvfs(uap->fd, &buf);
1299 
1300 	if (error == 0)
1301 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1302 	return (error);
1303 }
1304 
1305 /*
1306  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1307  *
1308  * Get statistics on all filesystems.
1309  */
1310 
1311 struct getfsstat_info {
1312 	struct statfs *sfsp;
1313 	long count;
1314 	long maxcount;
1315 	int error;
1316 	int flags;
1317 	struct thread *td;
1318 };
1319 
1320 static int getfsstat_callback(struct mount *, void *);
1321 
1322 int
1323 sys_getfsstat(struct getfsstat_args *uap)
1324 {
1325 	struct thread *td = curthread;
1326 	struct getfsstat_info info;
1327 
1328 	bzero(&info, sizeof(info));
1329 
1330 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1331 	info.sfsp = uap->buf;
1332 	info.count = 0;
1333 	info.flags = uap->flags;
1334 	info.td = td;
1335 
1336 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1337 	if (info.sfsp && info.count > info.maxcount)
1338 		uap->sysmsg_result = info.maxcount;
1339 	else
1340 		uap->sysmsg_result = info.count;
1341 	return (info.error);
1342 }
1343 
1344 static int
1345 getfsstat_callback(struct mount *mp, void *data)
1346 {
1347 	struct getfsstat_info *info = data;
1348 	struct statfs *sp;
1349 	char *freepath;
1350 	char *fullpath;
1351 	int error;
1352 
1353 	if (info->sfsp && info->count < info->maxcount) {
1354 		if (info->td->td_proc &&
1355 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1356 			return(0);
1357 		}
1358 		sp = &mp->mnt_stat;
1359 
1360 		/*
1361 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1362 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1363 		 * overrides MNT_WAIT.
1364 		 */
1365 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1366 		    (info->flags & MNT_WAIT)) &&
1367 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1368 			return(0);
1369 		}
1370 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1371 
1372 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1373 		if (error) {
1374 			info->error = error;
1375 			return(-1);
1376 		}
1377 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1378 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1379 		kfree(freepath, M_TEMP);
1380 
1381 		error = copyout(sp, info->sfsp, sizeof(*sp));
1382 		if (error) {
1383 			info->error = error;
1384 			return (-1);
1385 		}
1386 		++info->sfsp;
1387 	}
1388 	info->count++;
1389 	return(0);
1390 }
1391 
1392 /*
1393  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1394 		   long bufsize, int flags)
1395  *
1396  * Get statistics on all filesystems.
1397  */
1398 
1399 struct getvfsstat_info {
1400 	struct statfs *sfsp;
1401 	struct statvfs *vsfsp;
1402 	long count;
1403 	long maxcount;
1404 	int error;
1405 	int flags;
1406 	struct thread *td;
1407 };
1408 
1409 static int getvfsstat_callback(struct mount *, void *);
1410 
1411 int
1412 sys_getvfsstat(struct getvfsstat_args *uap)
1413 {
1414 	struct thread *td = curthread;
1415 	struct getvfsstat_info info;
1416 
1417 	bzero(&info, sizeof(info));
1418 
1419 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1420 	info.sfsp = uap->buf;
1421 	info.vsfsp = uap->vbuf;
1422 	info.count = 0;
1423 	info.flags = uap->flags;
1424 	info.td = td;
1425 
1426 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1427 	if (info.vsfsp && info.count > info.maxcount)
1428 		uap->sysmsg_result = info.maxcount;
1429 	else
1430 		uap->sysmsg_result = info.count;
1431 	return (info.error);
1432 }
1433 
1434 static int
1435 getvfsstat_callback(struct mount *mp, void *data)
1436 {
1437 	struct getvfsstat_info *info = data;
1438 	struct statfs *sp;
1439 	struct statvfs *vsp;
1440 	char *freepath;
1441 	char *fullpath;
1442 	int error;
1443 
1444 	if (info->vsfsp && info->count < info->maxcount) {
1445 		if (info->td->td_proc &&
1446 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1447 			return(0);
1448 		}
1449 		sp = &mp->mnt_stat;
1450 		vsp = &mp->mnt_vstat;
1451 
1452 		/*
1453 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1454 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1455 		 * overrides MNT_WAIT.
1456 		 */
1457 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1458 		    (info->flags & MNT_WAIT)) &&
1459 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1460 			return(0);
1461 		}
1462 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1463 
1464 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1465 		    (info->flags & MNT_WAIT)) &&
1466 		    (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1467 			return(0);
1468 		}
1469 		vsp->f_flag = 0;
1470 		if (mp->mnt_flag & MNT_RDONLY)
1471 			vsp->f_flag |= ST_RDONLY;
1472 		if (mp->mnt_flag & MNT_NOSUID)
1473 			vsp->f_flag |= ST_NOSUID;
1474 
1475 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1476 		if (error) {
1477 			info->error = error;
1478 			return(-1);
1479 		}
1480 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1481 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1482 		kfree(freepath, M_TEMP);
1483 
1484 		error = copyout(sp, info->sfsp, sizeof(*sp));
1485 		if (error == 0)
1486 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1487 		if (error) {
1488 			info->error = error;
1489 			return (-1);
1490 		}
1491 		++info->sfsp;
1492 		++info->vsfsp;
1493 	}
1494 	info->count++;
1495 	return(0);
1496 }
1497 
1498 
1499 /*
1500  * fchdir_args(int fd)
1501  *
1502  * Change current working directory to a given file descriptor.
1503  */
1504 int
1505 sys_fchdir(struct fchdir_args *uap)
1506 {
1507 	struct thread *td = curthread;
1508 	struct proc *p = td->td_proc;
1509 	struct filedesc *fdp = p->p_fd;
1510 	struct vnode *vp, *ovp;
1511 	struct mount *mp;
1512 	struct file *fp;
1513 	struct nchandle nch, onch, tnch;
1514 	int error;
1515 
1516 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1517 		return (error);
1518 	lwkt_gettoken(&p->p_token);
1519 	vp = (struct vnode *)fp->f_data;
1520 	vref(vp);
1521 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1522 	if (fp->f_nchandle.ncp == NULL)
1523 		error = ENOTDIR;
1524 	else
1525 		error = checkvp_chdir(vp, td);
1526 	if (error) {
1527 		vput(vp);
1528 		goto done;
1529 	}
1530 	cache_copy(&fp->f_nchandle, &nch);
1531 
1532 	/*
1533 	 * If the ncp has become a mount point, traverse through
1534 	 * the mount point.
1535 	 */
1536 
1537 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1538 	       (mp = cache_findmount(&nch)) != NULL
1539 	) {
1540 		error = nlookup_mp(mp, &tnch);
1541 		if (error == 0) {
1542 			cache_unlock(&tnch);	/* leave ref intact */
1543 			vput(vp);
1544 			vp = tnch.ncp->nc_vp;
1545 			error = vget(vp, LK_SHARED);
1546 			KKASSERT(error == 0);
1547 			cache_drop(&nch);
1548 			nch = tnch;
1549 		}
1550 		cache_dropmount(mp);
1551 	}
1552 	if (error == 0) {
1553 		ovp = fdp->fd_cdir;
1554 		onch = fdp->fd_ncdir;
1555 		vn_unlock(vp);		/* leave ref intact */
1556 		fdp->fd_cdir = vp;
1557 		fdp->fd_ncdir = nch;
1558 		cache_drop(&onch);
1559 		vrele(ovp);
1560 	} else {
1561 		cache_drop(&nch);
1562 		vput(vp);
1563 	}
1564 	fdrop(fp);
1565 done:
1566 	lwkt_reltoken(&p->p_token);
1567 	return (error);
1568 }
1569 
1570 int
1571 kern_chdir(struct nlookupdata *nd)
1572 {
1573 	struct thread *td = curthread;
1574 	struct proc *p = td->td_proc;
1575 	struct filedesc *fdp = p->p_fd;
1576 	struct vnode *vp, *ovp;
1577 	struct nchandle onch;
1578 	int error;
1579 
1580 	if ((error = nlookup(nd)) != 0)
1581 		return (error);
1582 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1583 		return (ENOENT);
1584 	if ((error = vget(vp, LK_SHARED)) != 0)
1585 		return (error);
1586 
1587 	lwkt_gettoken(&p->p_token);
1588 	error = checkvp_chdir(vp, td);
1589 	vn_unlock(vp);
1590 	if (error == 0) {
1591 		ovp = fdp->fd_cdir;
1592 		onch = fdp->fd_ncdir;
1593 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1594 		fdp->fd_ncdir = nd->nl_nch;
1595 		fdp->fd_cdir = vp;
1596 		cache_drop(&onch);
1597 		vrele(ovp);
1598 		cache_zero(&nd->nl_nch);
1599 	} else {
1600 		vrele(vp);
1601 	}
1602 	lwkt_reltoken(&p->p_token);
1603 	return (error);
1604 }
1605 
1606 /*
1607  * chdir_args(char *path)
1608  *
1609  * Change current working directory (``.'').
1610  */
1611 int
1612 sys_chdir(struct chdir_args *uap)
1613 {
1614 	struct nlookupdata nd;
1615 	int error;
1616 
1617 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1618 	if (error == 0)
1619 		error = kern_chdir(&nd);
1620 	nlookup_done(&nd);
1621 	return (error);
1622 }
1623 
1624 /*
1625  * Helper function for raised chroot(2) security function:  Refuse if
1626  * any filedescriptors are open directories.
1627  */
1628 static int
1629 chroot_refuse_vdir_fds(struct filedesc *fdp)
1630 {
1631 	struct vnode *vp;
1632 	struct file *fp;
1633 	int error;
1634 	int fd;
1635 
1636 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1637 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1638 			continue;
1639 		vp = (struct vnode *)fp->f_data;
1640 		if (vp->v_type != VDIR) {
1641 			fdrop(fp);
1642 			continue;
1643 		}
1644 		fdrop(fp);
1645 		return(EPERM);
1646 	}
1647 	return (0);
1648 }
1649 
1650 /*
1651  * This sysctl determines if we will allow a process to chroot(2) if it
1652  * has a directory open:
1653  *	0: disallowed for all processes.
1654  *	1: allowed for processes that were not already chroot(2)'ed.
1655  *	2: allowed for all processes.
1656  */
1657 
1658 static int chroot_allow_open_directories = 1;
1659 
1660 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1661      &chroot_allow_open_directories, 0, "");
1662 
1663 /*
1664  * chroot to the specified namecache entry.  We obtain the vp from the
1665  * namecache data.  The passed ncp must be locked and referenced and will
1666  * remain locked and referenced on return.
1667  */
1668 int
1669 kern_chroot(struct nchandle *nch)
1670 {
1671 	struct thread *td = curthread;
1672 	struct proc *p = td->td_proc;
1673 	struct filedesc *fdp = p->p_fd;
1674 	struct vnode *vp;
1675 	int error;
1676 
1677 	/*
1678 	 * Only privileged user can chroot
1679 	 */
1680 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1681 	if (error)
1682 		return (error);
1683 
1684 	/*
1685 	 * Disallow open directory descriptors (fchdir() breakouts).
1686 	 */
1687 	if (chroot_allow_open_directories == 0 ||
1688 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1689 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1690 			return (error);
1691 	}
1692 	if ((vp = nch->ncp->nc_vp) == NULL)
1693 		return (ENOENT);
1694 
1695 	if ((error = vget(vp, LK_SHARED)) != 0)
1696 		return (error);
1697 
1698 	/*
1699 	 * Check the validity of vp as a directory to change to and
1700 	 * associate it with rdir/jdir.
1701 	 */
1702 	error = checkvp_chdir(vp, td);
1703 	vn_unlock(vp);			/* leave reference intact */
1704 	if (error == 0) {
1705 		vrele(fdp->fd_rdir);
1706 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1707 		cache_drop(&fdp->fd_nrdir);
1708 		cache_copy(nch, &fdp->fd_nrdir);
1709 		if (fdp->fd_jdir == NULL) {
1710 			fdp->fd_jdir = vp;
1711 			vref(fdp->fd_jdir);
1712 			cache_copy(nch, &fdp->fd_njdir);
1713 		}
1714 	} else {
1715 		vrele(vp);
1716 	}
1717 	return (error);
1718 }
1719 
1720 /*
1721  * chroot_args(char *path)
1722  *
1723  * Change notion of root (``/'') directory.
1724  */
1725 int
1726 sys_chroot(struct chroot_args *uap)
1727 {
1728 	struct thread *td __debugvar = curthread;
1729 	struct nlookupdata nd;
1730 	int error;
1731 
1732 	KKASSERT(td->td_proc);
1733 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1734 	if (error == 0) {
1735 		nd.nl_flags |= NLC_EXEC;
1736 		error = nlookup(&nd);
1737 		if (error == 0)
1738 			error = kern_chroot(&nd.nl_nch);
1739 	}
1740 	nlookup_done(&nd);
1741 	return(error);
1742 }
1743 
1744 int
1745 sys_chroot_kernel(struct chroot_kernel_args *uap)
1746 {
1747 	struct thread *td = curthread;
1748 	struct nlookupdata nd;
1749 	struct nchandle *nch;
1750 	struct vnode *vp;
1751 	int error;
1752 
1753 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1754 	if (error)
1755 		goto error_nond;
1756 
1757 	error = nlookup(&nd);
1758 	if (error)
1759 		goto error_out;
1760 
1761 	nch = &nd.nl_nch;
1762 
1763 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1764 	if (error)
1765 		goto error_out;
1766 
1767 	if ((vp = nch->ncp->nc_vp) == NULL) {
1768 		error = ENOENT;
1769 		goto error_out;
1770 	}
1771 
1772 	if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1773 		goto error_out;
1774 
1775 	kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1776 	get_mplock();
1777 	vfs_cache_setroot(vp, cache_hold(nch));
1778 	rel_mplock();
1779 
1780 error_out:
1781 	nlookup_done(&nd);
1782 error_nond:
1783 	return(error);
1784 }
1785 
1786 /*
1787  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1788  * determine whether it is legal to chdir to the vnode.  The vnode's state
1789  * is not changed by this call.
1790  */
1791 int
1792 checkvp_chdir(struct vnode *vp, struct thread *td)
1793 {
1794 	int error;
1795 
1796 	if (vp->v_type != VDIR)
1797 		error = ENOTDIR;
1798 	else
1799 		error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1800 	return (error);
1801 }
1802 
1803 int
1804 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1805 {
1806 	struct thread *td = curthread;
1807 	struct proc *p = td->td_proc;
1808 	struct lwp *lp = td->td_lwp;
1809 	struct filedesc *fdp = p->p_fd;
1810 	int cmode, flags;
1811 	struct file *nfp;
1812 	struct file *fp;
1813 	struct vnode *vp;
1814 	int type, indx, error = 0;
1815 	struct flock lf;
1816 
1817 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1818 		return (EINVAL);
1819 	flags = FFLAGS(oflags);
1820 	error = falloc(lp, &nfp, NULL);
1821 	if (error)
1822 		return (error);
1823 	fp = nfp;
1824 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1825 
1826 	/*
1827 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1828 	 * file descriptor to be duplicated rather then doing the open
1829 	 * itself.
1830 	 */
1831 	lp->lwp_dupfd = -1;
1832 
1833 	/*
1834 	 * Call vn_open() to do the lookup and assign the vnode to the
1835 	 * file pointer.  vn_open() does not change the ref count on fp
1836 	 * and the vnode, on success, will be inherited by the file pointer
1837 	 * and unlocked.
1838 	 */
1839 	nd->nl_flags |= NLC_LOCKVP;
1840 	error = vn_open(nd, fp, flags, cmode);
1841 	nlookup_done(nd);
1842 	if (error) {
1843 		/*
1844 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1845 		 * responsible for dropping the old contents of ofiles[indx]
1846 		 * if it succeeds.
1847 		 *
1848 		 * Note that fsetfd() will add a ref to fp which represents
1849 		 * the fd_files[] assignment.  We must still drop our
1850 		 * reference.
1851 		 */
1852 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1853 			if (fdalloc(p, 0, &indx) == 0) {
1854 				error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1855 				if (error == 0) {
1856 					*res = indx;
1857 					fdrop(fp);	/* our ref */
1858 					return (0);
1859 				}
1860 				fsetfd(fdp, NULL, indx);
1861 			}
1862 		}
1863 		fdrop(fp);	/* our ref */
1864 		if (error == ERESTART)
1865 			error = EINTR;
1866 		return (error);
1867 	}
1868 
1869 	/*
1870 	 * ref the vnode for ourselves so it can't be ripped out from under
1871 	 * is.  XXX need an ND flag to request that the vnode be returned
1872 	 * anyway.
1873 	 *
1874 	 * Reserve a file descriptor but do not assign it until the open
1875 	 * succeeds.
1876 	 */
1877 	vp = (struct vnode *)fp->f_data;
1878 	vref(vp);
1879 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1880 		fdrop(fp);
1881 		vrele(vp);
1882 		return (error);
1883 	}
1884 
1885 	/*
1886 	 * If no error occurs the vp will have been assigned to the file
1887 	 * pointer.
1888 	 */
1889 	lp->lwp_dupfd = 0;
1890 
1891 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1892 		lf.l_whence = SEEK_SET;
1893 		lf.l_start = 0;
1894 		lf.l_len = 0;
1895 		if (flags & O_EXLOCK)
1896 			lf.l_type = F_WRLCK;
1897 		else
1898 			lf.l_type = F_RDLCK;
1899 		if (flags & FNONBLOCK)
1900 			type = 0;
1901 		else
1902 			type = F_WAIT;
1903 
1904 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1905 			/*
1906 			 * lock request failed.  Clean up the reserved
1907 			 * descriptor.
1908 			 */
1909 			vrele(vp);
1910 			fsetfd(fdp, NULL, indx);
1911 			fdrop(fp);
1912 			return (error);
1913 		}
1914 		fp->f_flag |= FHASLOCK;
1915 	}
1916 #if 0
1917 	/*
1918 	 * Assert that all regular file vnodes were created with a object.
1919 	 */
1920 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1921 		("open: regular file has no backing object after vn_open"));
1922 #endif
1923 
1924 	vrele(vp);
1925 
1926 	/*
1927 	 * release our private reference, leaving the one associated with the
1928 	 * descriptor table intact.
1929 	 */
1930 	fsetfd(fdp, fp, indx);
1931 	fdrop(fp);
1932 	*res = indx;
1933 	if (oflags & O_CLOEXEC)
1934 		error = fsetfdflags(fdp, *res, UF_EXCLOSE);
1935 	return (error);
1936 }
1937 
1938 /*
1939  * open_args(char *path, int flags, int mode)
1940  *
1941  * Check permissions, allocate an open file structure,
1942  * and call the device open routine if any.
1943  */
1944 int
1945 sys_open(struct open_args *uap)
1946 {
1947 	struct nlookupdata nd;
1948 	int error;
1949 
1950 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1951 	if (error == 0) {
1952 		error = kern_open(&nd, uap->flags,
1953 				    uap->mode, &uap->sysmsg_result);
1954 	}
1955 	nlookup_done(&nd);
1956 	return (error);
1957 }
1958 
1959 /*
1960  * openat_args(int fd, char *path, int flags, int mode)
1961  */
1962 int
1963 sys_openat(struct openat_args *uap)
1964 {
1965 	struct nlookupdata nd;
1966 	int error;
1967 	struct file *fp;
1968 
1969 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
1970 	if (error == 0) {
1971 		error = kern_open(&nd, uap->flags, uap->mode,
1972 					&uap->sysmsg_result);
1973 	}
1974 	nlookup_done_at(&nd, fp);
1975 	return (error);
1976 }
1977 
1978 int
1979 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1980 {
1981 	struct thread *td = curthread;
1982 	struct proc *p = td->td_proc;
1983 	struct vnode *vp;
1984 	struct vattr vattr;
1985 	int error;
1986 	int whiteout = 0;
1987 
1988 	KKASSERT(p);
1989 
1990 	VATTR_NULL(&vattr);
1991 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1992 	vattr.va_rmajor = rmajor;
1993 	vattr.va_rminor = rminor;
1994 
1995 	switch (mode & S_IFMT) {
1996 	case S_IFMT:	/* used by badsect to flag bad sectors */
1997 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
1998 		vattr.va_type = VBAD;
1999 		break;
2000 	case S_IFCHR:
2001 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2002 		vattr.va_type = VCHR;
2003 		break;
2004 	case S_IFBLK:
2005 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2006 		vattr.va_type = VBLK;
2007 		break;
2008 	case S_IFWHT:
2009 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2010 		whiteout = 1;
2011 		break;
2012 	case S_IFDIR:	/* special directories support for HAMMER */
2013 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2014 		vattr.va_type = VDIR;
2015 		break;
2016 	default:
2017 		error = EINVAL;
2018 		break;
2019 	}
2020 
2021 	if (error)
2022 		return (error);
2023 
2024 	bwillinode(1);
2025 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2026 	if ((error = nlookup(nd)) != 0)
2027 		return (error);
2028 	if (nd->nl_nch.ncp->nc_vp)
2029 		return (EEXIST);
2030 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2031 		return (error);
2032 
2033 	if (whiteout) {
2034 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2035 				      nd->nl_cred, NAMEI_CREATE);
2036 	} else {
2037 		vp = NULL;
2038 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2039 				   &vp, nd->nl_cred, &vattr);
2040 		if (error == 0)
2041 			vput(vp);
2042 	}
2043 	return (error);
2044 }
2045 
2046 /*
2047  * mknod_args(char *path, int mode, int dev)
2048  *
2049  * Create a special file.
2050  */
2051 int
2052 sys_mknod(struct mknod_args *uap)
2053 {
2054 	struct nlookupdata nd;
2055 	int error;
2056 
2057 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2058 	if (error == 0) {
2059 		error = kern_mknod(&nd, uap->mode,
2060 				   umajor(uap->dev), uminor(uap->dev));
2061 	}
2062 	nlookup_done(&nd);
2063 	return (error);
2064 }
2065 
2066 /*
2067  * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2068  *
2069  * Create a special file.  The path is relative to the directory associated
2070  * with fd.
2071  */
2072 int
2073 sys_mknodat(struct mknodat_args *uap)
2074 {
2075 	struct nlookupdata nd;
2076 	struct file *fp;
2077 	int error;
2078 
2079 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2080 	if (error == 0) {
2081 		error = kern_mknod(&nd, uap->mode,
2082 				   umajor(uap->dev), uminor(uap->dev));
2083 	}
2084 	nlookup_done_at(&nd, fp);
2085 	return (error);
2086 }
2087 
2088 int
2089 kern_mkfifo(struct nlookupdata *nd, int mode)
2090 {
2091 	struct thread *td = curthread;
2092 	struct proc *p = td->td_proc;
2093 	struct vattr vattr;
2094 	struct vnode *vp;
2095 	int error;
2096 
2097 	bwillinode(1);
2098 
2099 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2100 	if ((error = nlookup(nd)) != 0)
2101 		return (error);
2102 	if (nd->nl_nch.ncp->nc_vp)
2103 		return (EEXIST);
2104 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2105 		return (error);
2106 
2107 	VATTR_NULL(&vattr);
2108 	vattr.va_type = VFIFO;
2109 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2110 	vp = NULL;
2111 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2112 	if (error == 0)
2113 		vput(vp);
2114 	return (error);
2115 }
2116 
2117 /*
2118  * mkfifo_args(char *path, int mode)
2119  *
2120  * Create a named pipe.
2121  */
2122 int
2123 sys_mkfifo(struct mkfifo_args *uap)
2124 {
2125 	struct nlookupdata nd;
2126 	int error;
2127 
2128 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2129 	if (error == 0)
2130 		error = kern_mkfifo(&nd, uap->mode);
2131 	nlookup_done(&nd);
2132 	return (error);
2133 }
2134 
2135 /*
2136  * mkfifoat_args(int fd, char *path, mode_t mode)
2137  *
2138  * Create a named pipe.  The path is relative to the directory associated
2139  * with fd.
2140  */
2141 int
2142 sys_mkfifoat(struct mkfifoat_args *uap)
2143 {
2144 	struct nlookupdata nd;
2145 	struct file *fp;
2146 	int error;
2147 
2148 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2149 	if (error == 0)
2150 		error = kern_mkfifo(&nd, uap->mode);
2151 	nlookup_done_at(&nd, fp);
2152 	return (error);
2153 }
2154 
2155 static int hardlink_check_uid = 0;
2156 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2157     &hardlink_check_uid, 0,
2158     "Unprivileged processes cannot create hard links to files owned by other "
2159     "users");
2160 static int hardlink_check_gid = 0;
2161 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2162     &hardlink_check_gid, 0,
2163     "Unprivileged processes cannot create hard links to files owned by other "
2164     "groups");
2165 
2166 static int
2167 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2168 {
2169 	struct vattr va;
2170 	int error;
2171 
2172 	/*
2173 	 * Shortcut if disabled
2174 	 */
2175 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2176 		return (0);
2177 
2178 	/*
2179 	 * Privileged user can always hardlink
2180 	 */
2181 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2182 		return (0);
2183 
2184 	/*
2185 	 * Otherwise only if the originating file is owned by the
2186 	 * same user or group.  Note that any group is allowed if
2187 	 * the file is owned by the caller.
2188 	 */
2189 	error = VOP_GETATTR(vp, &va);
2190 	if (error != 0)
2191 		return (error);
2192 
2193 	if (hardlink_check_uid) {
2194 		if (cred->cr_uid != va.va_uid)
2195 			return (EPERM);
2196 	}
2197 
2198 	if (hardlink_check_gid) {
2199 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2200 			return (EPERM);
2201 	}
2202 
2203 	return (0);
2204 }
2205 
2206 int
2207 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2208 {
2209 	struct thread *td = curthread;
2210 	struct vnode *vp;
2211 	int error;
2212 
2213 	/*
2214 	 * Lookup the source and obtained a locked vnode.
2215 	 *
2216 	 * You may only hardlink a file which you have write permission
2217 	 * on or which you own.
2218 	 *
2219 	 * XXX relookup on vget failure / race ?
2220 	 */
2221 	bwillinode(1);
2222 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2223 	if ((error = nlookup(nd)) != 0)
2224 		return (error);
2225 	vp = nd->nl_nch.ncp->nc_vp;
2226 	KKASSERT(vp != NULL);
2227 	if (vp->v_type == VDIR)
2228 		return (EPERM);		/* POSIX */
2229 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2230 		return (error);
2231 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2232 		return (error);
2233 
2234 	/*
2235 	 * Unlock the source so we can lookup the target without deadlocking
2236 	 * (XXX vp is locked already, possible other deadlock?).  The target
2237 	 * must not exist.
2238 	 */
2239 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2240 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2241 	cache_unlock(&nd->nl_nch);
2242 	vn_unlock(vp);
2243 
2244 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2245 	if ((error = nlookup(linknd)) != 0) {
2246 		vrele(vp);
2247 		return (error);
2248 	}
2249 	if (linknd->nl_nch.ncp->nc_vp) {
2250 		vrele(vp);
2251 		return (EEXIST);
2252 	}
2253 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2254 		vrele(vp);
2255 		return (error);
2256 	}
2257 
2258 	/*
2259 	 * Finally run the new API VOP.
2260 	 */
2261 	error = can_hardlink(vp, td, td->td_ucred);
2262 	if (error == 0) {
2263 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2264 				  vp, linknd->nl_cred);
2265 	}
2266 	vput(vp);
2267 	return (error);
2268 }
2269 
2270 /*
2271  * link_args(char *path, char *link)
2272  *
2273  * Make a hard file link.
2274  */
2275 int
2276 sys_link(struct link_args *uap)
2277 {
2278 	struct nlookupdata nd, linknd;
2279 	int error;
2280 
2281 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2282 	if (error == 0) {
2283 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2284 		if (error == 0)
2285 			error = kern_link(&nd, &linknd);
2286 		nlookup_done(&linknd);
2287 	}
2288 	nlookup_done(&nd);
2289 	return (error);
2290 }
2291 
2292 /*
2293  * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags)
2294  *
2295  * Make a hard file link. The path1 argument is relative to the directory
2296  * associated with fd1, and similarly the path2 argument is relative to
2297  * the directory associated with fd2.
2298  */
2299 int
2300 sys_linkat(struct linkat_args *uap)
2301 {
2302 	struct nlookupdata nd, linknd;
2303 	struct file *fp1, *fp2;
2304 	int error;
2305 
2306 	error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE,
2307 	    (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0);
2308 	if (error == 0) {
2309 		error = nlookup_init_at(&linknd, &fp2, uap->fd2,
2310 		    uap->path2, UIO_USERSPACE, 0);
2311 		if (error == 0)
2312 			error = kern_link(&nd, &linknd);
2313 		nlookup_done_at(&linknd, fp2);
2314 	}
2315 	nlookup_done_at(&nd, fp1);
2316 	return (error);
2317 }
2318 
2319 int
2320 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2321 {
2322 	struct vattr vattr;
2323 	struct vnode *vp;
2324 	struct vnode *dvp;
2325 	int error;
2326 
2327 	bwillinode(1);
2328 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2329 	if ((error = nlookup(nd)) != 0)
2330 		return (error);
2331 	if (nd->nl_nch.ncp->nc_vp)
2332 		return (EEXIST);
2333 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2334 		return (error);
2335 	dvp = nd->nl_dvp;
2336 	VATTR_NULL(&vattr);
2337 	vattr.va_mode = mode;
2338 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2339 	if (error == 0)
2340 		vput(vp);
2341 	return (error);
2342 }
2343 
2344 /*
2345  * symlink(char *path, char *link)
2346  *
2347  * Make a symbolic link.
2348  */
2349 int
2350 sys_symlink(struct symlink_args *uap)
2351 {
2352 	struct thread *td = curthread;
2353 	struct nlookupdata nd;
2354 	char *path;
2355 	int error;
2356 	int mode;
2357 
2358 	path = objcache_get(namei_oc, M_WAITOK);
2359 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2360 	if (error == 0) {
2361 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2362 		if (error == 0) {
2363 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2364 			error = kern_symlink(&nd, path, mode);
2365 		}
2366 		nlookup_done(&nd);
2367 	}
2368 	objcache_put(namei_oc, path);
2369 	return (error);
2370 }
2371 
2372 /*
2373  * symlinkat_args(char *path1, int fd, char *path2)
2374  *
2375  * Make a symbolic link.  The path2 argument is relative to the directory
2376  * associated with fd.
2377  */
2378 int
2379 sys_symlinkat(struct symlinkat_args *uap)
2380 {
2381 	struct thread *td = curthread;
2382 	struct nlookupdata nd;
2383 	struct file *fp;
2384 	char *path1;
2385 	int error;
2386 	int mode;
2387 
2388 	path1 = objcache_get(namei_oc, M_WAITOK);
2389 	error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2390 	if (error == 0) {
2391 		error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2392 		    UIO_USERSPACE, 0);
2393 		if (error == 0) {
2394 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2395 			error = kern_symlink(&nd, path1, mode);
2396 		}
2397 		nlookup_done_at(&nd, fp);
2398 	}
2399 	objcache_put(namei_oc, path1);
2400 	return (error);
2401 }
2402 
2403 /*
2404  * undelete_args(char *path)
2405  *
2406  * Delete a whiteout from the filesystem.
2407  */
2408 int
2409 sys_undelete(struct undelete_args *uap)
2410 {
2411 	struct nlookupdata nd;
2412 	int error;
2413 
2414 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2415 	bwillinode(1);
2416 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2417 	if (error == 0)
2418 		error = nlookup(&nd);
2419 	if (error == 0)
2420 		error = ncp_writechk(&nd.nl_nch);
2421 	if (error == 0) {
2422 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2423 				      NAMEI_DELETE);
2424 	}
2425 	nlookup_done(&nd);
2426 	return (error);
2427 }
2428 
2429 int
2430 kern_unlink(struct nlookupdata *nd)
2431 {
2432 	int error;
2433 
2434 	bwillinode(1);
2435 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2436 	if ((error = nlookup(nd)) != 0)
2437 		return (error);
2438 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2439 		return (error);
2440 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2441 	return (error);
2442 }
2443 
2444 /*
2445  * unlink_args(char *path)
2446  *
2447  * Delete a name from the filesystem.
2448  */
2449 int
2450 sys_unlink(struct unlink_args *uap)
2451 {
2452 	struct nlookupdata nd;
2453 	int error;
2454 
2455 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2456 	if (error == 0)
2457 		error = kern_unlink(&nd);
2458 	nlookup_done(&nd);
2459 	return (error);
2460 }
2461 
2462 
2463 /*
2464  * unlinkat_args(int fd, char *path, int flags)
2465  *
2466  * Delete the file or directory entry pointed to by fd/path.
2467  */
2468 int
2469 sys_unlinkat(struct unlinkat_args *uap)
2470 {
2471 	struct nlookupdata nd;
2472 	struct file *fp;
2473 	int error;
2474 
2475 	if (uap->flags & ~AT_REMOVEDIR)
2476 		return (EINVAL);
2477 
2478 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2479 	if (error == 0) {
2480 		if (uap->flags & AT_REMOVEDIR)
2481 			error = kern_rmdir(&nd);
2482 		else
2483 			error = kern_unlink(&nd);
2484 	}
2485 	nlookup_done_at(&nd, fp);
2486 	return (error);
2487 }
2488 
2489 int
2490 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2491 {
2492 	struct thread *td = curthread;
2493 	struct proc *p = td->td_proc;
2494 	struct file *fp;
2495 	struct vnode *vp;
2496 	struct vattr vattr;
2497 	off_t new_offset;
2498 	int error;
2499 
2500 	fp = holdfp(p->p_fd, fd, -1);
2501 	if (fp == NULL)
2502 		return (EBADF);
2503 	if (fp->f_type != DTYPE_VNODE) {
2504 		error = ESPIPE;
2505 		goto done;
2506 	}
2507 	vp = (struct vnode *)fp->f_data;
2508 
2509 	switch (whence) {
2510 	case L_INCR:
2511 		spin_lock(&fp->f_spin);
2512 		new_offset = fp->f_offset + offset;
2513 		error = 0;
2514 		break;
2515 	case L_XTND:
2516 		error = VOP_GETATTR(vp, &vattr);
2517 		spin_lock(&fp->f_spin);
2518 		new_offset = offset + vattr.va_size;
2519 		break;
2520 	case L_SET:
2521 		new_offset = offset;
2522 		error = 0;
2523 		spin_lock(&fp->f_spin);
2524 		break;
2525 	default:
2526 		new_offset = 0;
2527 		error = EINVAL;
2528 		spin_lock(&fp->f_spin);
2529 		break;
2530 	}
2531 
2532 	/*
2533 	 * Validate the seek position.  Negative offsets are not allowed
2534 	 * for regular files or directories.
2535 	 *
2536 	 * Normally we would also not want to allow negative offsets for
2537 	 * character and block-special devices.  However kvm addresses
2538 	 * on 64 bit architectures might appear to be negative and must
2539 	 * be allowed.
2540 	 */
2541 	if (error == 0) {
2542 		if (new_offset < 0 &&
2543 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2544 			error = EINVAL;
2545 		} else {
2546 			fp->f_offset = new_offset;
2547 		}
2548 	}
2549 	*res = fp->f_offset;
2550 	spin_unlock(&fp->f_spin);
2551 done:
2552 	fdrop(fp);
2553 	return (error);
2554 }
2555 
2556 /*
2557  * lseek_args(int fd, int pad, off_t offset, int whence)
2558  *
2559  * Reposition read/write file offset.
2560  */
2561 int
2562 sys_lseek(struct lseek_args *uap)
2563 {
2564 	int error;
2565 
2566 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2567 			   &uap->sysmsg_offset);
2568 
2569 	return (error);
2570 }
2571 
2572 /*
2573  * Check if current process can access given file.  amode is a bitmask of *_OK
2574  * access bits.  flags is a bitmask of AT_* flags.
2575  */
2576 int
2577 kern_access(struct nlookupdata *nd, int amode, int flags)
2578 {
2579 	struct vnode *vp;
2580 	int error, mode;
2581 
2582 	if (flags & ~AT_EACCESS)
2583 		return (EINVAL);
2584 	if ((error = nlookup(nd)) != 0)
2585 		return (error);
2586 retry:
2587 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2588 	if (error)
2589 		return (error);
2590 
2591 	/* Flags == 0 means only check for existence. */
2592 	if (amode) {
2593 		mode = 0;
2594 		if (amode & R_OK)
2595 			mode |= VREAD;
2596 		if (amode & W_OK)
2597 			mode |= VWRITE;
2598 		if (amode & X_OK)
2599 			mode |= VEXEC;
2600 		if ((mode & VWRITE) == 0 ||
2601 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2602 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2603 
2604 		/*
2605 		 * If the file handle is stale we have to re-resolve the
2606 		 * entry.  This is a hack at the moment.
2607 		 */
2608 		if (error == ESTALE) {
2609 			vput(vp);
2610 			cache_setunresolved(&nd->nl_nch);
2611 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2612 			if (error == 0) {
2613 				vp = NULL;
2614 				goto retry;
2615 			}
2616 			return(error);
2617 		}
2618 	}
2619 	vput(vp);
2620 	return (error);
2621 }
2622 
2623 /*
2624  * access_args(char *path, int flags)
2625  *
2626  * Check access permissions.
2627  */
2628 int
2629 sys_access(struct access_args *uap)
2630 {
2631 	struct nlookupdata nd;
2632 	int error;
2633 
2634 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2635 	if (error == 0)
2636 		error = kern_access(&nd, uap->flags, 0);
2637 	nlookup_done(&nd);
2638 	return (error);
2639 }
2640 
2641 
2642 /*
2643  * faccessat_args(int fd, char *path, int amode, int flags)
2644  *
2645  * Check access permissions.
2646  */
2647 int
2648 sys_faccessat(struct faccessat_args *uap)
2649 {
2650 	struct nlookupdata nd;
2651 	struct file *fp;
2652 	int error;
2653 
2654 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2655 				NLC_FOLLOW);
2656 	if (error == 0)
2657 		error = kern_access(&nd, uap->amode, uap->flags);
2658 	nlookup_done_at(&nd, fp);
2659 	return (error);
2660 }
2661 
2662 
2663 int
2664 kern_stat(struct nlookupdata *nd, struct stat *st)
2665 {
2666 	int error;
2667 	struct vnode *vp;
2668 
2669 	if ((error = nlookup(nd)) != 0)
2670 		return (error);
2671 again:
2672 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2673 		return (ENOENT);
2674 
2675 	if ((error = vget(vp, LK_SHARED)) != 0)
2676 		return (error);
2677 	error = vn_stat(vp, st, nd->nl_cred);
2678 
2679 	/*
2680 	 * If the file handle is stale we have to re-resolve the entry.  This
2681 	 * is a hack at the moment.
2682 	 */
2683 	if (error == ESTALE) {
2684 		vput(vp);
2685 		cache_setunresolved(&nd->nl_nch);
2686 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2687 		if (error == 0)
2688 			goto again;
2689 	} else {
2690 		vput(vp);
2691 	}
2692 	return (error);
2693 }
2694 
2695 /*
2696  * stat_args(char *path, struct stat *ub)
2697  *
2698  * Get file status; this version follows links.
2699  */
2700 int
2701 sys_stat(struct stat_args *uap)
2702 {
2703 	struct nlookupdata nd;
2704 	struct stat st;
2705 	int error;
2706 
2707 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2708 	if (error == 0) {
2709 		error = kern_stat(&nd, &st);
2710 		if (error == 0)
2711 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2712 	}
2713 	nlookup_done(&nd);
2714 	return (error);
2715 }
2716 
2717 /*
2718  * lstat_args(char *path, struct stat *ub)
2719  *
2720  * Get file status; this version does not follow links.
2721  */
2722 int
2723 sys_lstat(struct lstat_args *uap)
2724 {
2725 	struct nlookupdata nd;
2726 	struct stat st;
2727 	int error;
2728 
2729 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2730 	if (error == 0) {
2731 		error = kern_stat(&nd, &st);
2732 		if (error == 0)
2733 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2734 	}
2735 	nlookup_done(&nd);
2736 	return (error);
2737 }
2738 
2739 /*
2740  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2741  *
2742  * Get status of file pointed to by fd/path.
2743  */
2744 int
2745 sys_fstatat(struct fstatat_args *uap)
2746 {
2747 	struct nlookupdata nd;
2748 	struct stat st;
2749 	int error;
2750 	int flags;
2751 	struct file *fp;
2752 
2753 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2754 		return (EINVAL);
2755 
2756 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2757 
2758 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2759 				UIO_USERSPACE, flags);
2760 	if (error == 0) {
2761 		error = kern_stat(&nd, &st);
2762 		if (error == 0)
2763 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
2764 	}
2765 	nlookup_done_at(&nd, fp);
2766 	return (error);
2767 }
2768 
2769 /*
2770  * pathconf_Args(char *path, int name)
2771  *
2772  * Get configurable pathname variables.
2773  */
2774 int
2775 sys_pathconf(struct pathconf_args *uap)
2776 {
2777 	struct nlookupdata nd;
2778 	struct vnode *vp;
2779 	int error;
2780 
2781 	vp = NULL;
2782 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2783 	if (error == 0)
2784 		error = nlookup(&nd);
2785 	if (error == 0)
2786 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2787 	nlookup_done(&nd);
2788 	if (error == 0) {
2789 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2790 		vput(vp);
2791 	}
2792 	return (error);
2793 }
2794 
2795 /*
2796  * XXX: daver
2797  * kern_readlink isn't properly split yet.  There is a copyin burried
2798  * in VOP_READLINK().
2799  */
2800 int
2801 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2802 {
2803 	struct thread *td = curthread;
2804 	struct vnode *vp;
2805 	struct iovec aiov;
2806 	struct uio auio;
2807 	int error;
2808 
2809 	if ((error = nlookup(nd)) != 0)
2810 		return (error);
2811 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2812 	if (error)
2813 		return (error);
2814 	if (vp->v_type != VLNK) {
2815 		error = EINVAL;
2816 	} else {
2817 		aiov.iov_base = buf;
2818 		aiov.iov_len = count;
2819 		auio.uio_iov = &aiov;
2820 		auio.uio_iovcnt = 1;
2821 		auio.uio_offset = 0;
2822 		auio.uio_rw = UIO_READ;
2823 		auio.uio_segflg = UIO_USERSPACE;
2824 		auio.uio_td = td;
2825 		auio.uio_resid = count;
2826 		error = VOP_READLINK(vp, &auio, td->td_ucred);
2827 	}
2828 	vput(vp);
2829 	*res = count - auio.uio_resid;
2830 	return (error);
2831 }
2832 
2833 /*
2834  * readlink_args(char *path, char *buf, int count)
2835  *
2836  * Return target name of a symbolic link.
2837  */
2838 int
2839 sys_readlink(struct readlink_args *uap)
2840 {
2841 	struct nlookupdata nd;
2842 	int error;
2843 
2844 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2845 	if (error == 0) {
2846 		error = kern_readlink(&nd, uap->buf, uap->count,
2847 					&uap->sysmsg_result);
2848 	}
2849 	nlookup_done(&nd);
2850 	return (error);
2851 }
2852 
2853 /*
2854  * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
2855  *
2856  * Return target name of a symbolic link.  The path is relative to the
2857  * directory associated with fd.
2858  */
2859 int
2860 sys_readlinkat(struct readlinkat_args *uap)
2861 {
2862 	struct nlookupdata nd;
2863 	struct file *fp;
2864 	int error;
2865 
2866 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2867 	if (error == 0) {
2868 		error = kern_readlink(&nd, uap->buf, uap->bufsize,
2869 					&uap->sysmsg_result);
2870 	}
2871 	nlookup_done_at(&nd, fp);
2872 	return (error);
2873 }
2874 
2875 static int
2876 setfflags(struct vnode *vp, int flags)
2877 {
2878 	struct thread *td = curthread;
2879 	int error;
2880 	struct vattr vattr;
2881 
2882 	/*
2883 	 * Prevent non-root users from setting flags on devices.  When
2884 	 * a device is reused, users can retain ownership of the device
2885 	 * if they are allowed to set flags and programs assume that
2886 	 * chown can't fail when done as root.
2887 	 */
2888 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2889 	    ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2890 		return (error);
2891 
2892 	/*
2893 	 * note: vget is required for any operation that might mod the vnode
2894 	 * so VINACTIVE is properly cleared.
2895 	 */
2896 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2897 		VATTR_NULL(&vattr);
2898 		vattr.va_flags = flags;
2899 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2900 		vput(vp);
2901 	}
2902 	return (error);
2903 }
2904 
2905 /*
2906  * chflags(char *path, int flags)
2907  *
2908  * Change flags of a file given a path name.
2909  */
2910 int
2911 sys_chflags(struct chflags_args *uap)
2912 {
2913 	struct nlookupdata nd;
2914 	struct vnode *vp;
2915 	int error;
2916 
2917 	vp = NULL;
2918 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2919 	if (error == 0)
2920 		error = nlookup(&nd);
2921 	if (error == 0)
2922 		error = ncp_writechk(&nd.nl_nch);
2923 	if (error == 0)
2924 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2925 	nlookup_done(&nd);
2926 	if (error == 0) {
2927 		error = setfflags(vp, uap->flags);
2928 		vrele(vp);
2929 	}
2930 	return (error);
2931 }
2932 
2933 /*
2934  * lchflags(char *path, int flags)
2935  *
2936  * Change flags of a file given a path name, but don't follow symlinks.
2937  */
2938 int
2939 sys_lchflags(struct lchflags_args *uap)
2940 {
2941 	struct nlookupdata nd;
2942 	struct vnode *vp;
2943 	int error;
2944 
2945 	vp = NULL;
2946 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2947 	if (error == 0)
2948 		error = nlookup(&nd);
2949 	if (error == 0)
2950 		error = ncp_writechk(&nd.nl_nch);
2951 	if (error == 0)
2952 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2953 	nlookup_done(&nd);
2954 	if (error == 0) {
2955 		error = setfflags(vp, uap->flags);
2956 		vrele(vp);
2957 	}
2958 	return (error);
2959 }
2960 
2961 /*
2962  * fchflags_args(int fd, int flags)
2963  *
2964  * Change flags of a file given a file descriptor.
2965  */
2966 int
2967 sys_fchflags(struct fchflags_args *uap)
2968 {
2969 	struct thread *td = curthread;
2970 	struct proc *p = td->td_proc;
2971 	struct file *fp;
2972 	int error;
2973 
2974 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2975 		return (error);
2976 	if (fp->f_nchandle.ncp)
2977 		error = ncp_writechk(&fp->f_nchandle);
2978 	if (error == 0)
2979 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2980 	fdrop(fp);
2981 	return (error);
2982 }
2983 
2984 static int
2985 setfmode(struct vnode *vp, int mode)
2986 {
2987 	struct thread *td = curthread;
2988 	int error;
2989 	struct vattr vattr;
2990 
2991 	/*
2992 	 * note: vget is required for any operation that might mod the vnode
2993 	 * so VINACTIVE is properly cleared.
2994 	 */
2995 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2996 		VATTR_NULL(&vattr);
2997 		vattr.va_mode = mode & ALLPERMS;
2998 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2999 		vput(vp);
3000 	}
3001 	return error;
3002 }
3003 
3004 int
3005 kern_chmod(struct nlookupdata *nd, int mode)
3006 {
3007 	struct vnode *vp;
3008 	int error;
3009 
3010 	if ((error = nlookup(nd)) != 0)
3011 		return (error);
3012 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3013 		return (error);
3014 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3015 		error = setfmode(vp, mode);
3016 	vrele(vp);
3017 	return (error);
3018 }
3019 
3020 /*
3021  * chmod_args(char *path, int mode)
3022  *
3023  * Change mode of a file given path name.
3024  */
3025 int
3026 sys_chmod(struct chmod_args *uap)
3027 {
3028 	struct nlookupdata nd;
3029 	int error;
3030 
3031 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3032 	if (error == 0)
3033 		error = kern_chmod(&nd, uap->mode);
3034 	nlookup_done(&nd);
3035 	return (error);
3036 }
3037 
3038 /*
3039  * lchmod_args(char *path, int mode)
3040  *
3041  * Change mode of a file given path name (don't follow links.)
3042  */
3043 int
3044 sys_lchmod(struct lchmod_args *uap)
3045 {
3046 	struct nlookupdata nd;
3047 	int error;
3048 
3049 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3050 	if (error == 0)
3051 		error = kern_chmod(&nd, uap->mode);
3052 	nlookup_done(&nd);
3053 	return (error);
3054 }
3055 
3056 /*
3057  * fchmod_args(int fd, int mode)
3058  *
3059  * Change mode of a file given a file descriptor.
3060  */
3061 int
3062 sys_fchmod(struct fchmod_args *uap)
3063 {
3064 	struct thread *td = curthread;
3065 	struct proc *p = td->td_proc;
3066 	struct file *fp;
3067 	int error;
3068 
3069 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3070 		return (error);
3071 	if (fp->f_nchandle.ncp)
3072 		error = ncp_writechk(&fp->f_nchandle);
3073 	if (error == 0)
3074 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
3075 	fdrop(fp);
3076 	return (error);
3077 }
3078 
3079 /*
3080  * fchmodat_args(char *path, int mode)
3081  *
3082  * Change mode of a file pointed to by fd/path.
3083  */
3084 int
3085 sys_fchmodat(struct fchmodat_args *uap)
3086 {
3087 	struct nlookupdata nd;
3088 	struct file *fp;
3089 	int error;
3090 	int flags;
3091 
3092 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3093 		return (EINVAL);
3094 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3095 
3096 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3097 				UIO_USERSPACE, flags);
3098 	if (error == 0)
3099 		error = kern_chmod(&nd, uap->mode);
3100 	nlookup_done_at(&nd, fp);
3101 	return (error);
3102 }
3103 
3104 static int
3105 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid)
3106 {
3107 	struct thread *td = curthread;
3108 	int error;
3109 	struct vattr vattr;
3110 	uid_t o_uid;
3111 	gid_t o_gid;
3112 	uint64_t size;
3113 
3114 	/*
3115 	 * note: vget is required for any operation that might mod the vnode
3116 	 * so VINACTIVE is properly cleared.
3117 	 */
3118 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3119 		if ((error = VOP_GETATTR(vp, &vattr)) != 0)
3120 			return error;
3121 		o_uid = vattr.va_uid;
3122 		o_gid = vattr.va_gid;
3123 		size = vattr.va_size;
3124 
3125 		VATTR_NULL(&vattr);
3126 		vattr.va_uid = uid;
3127 		vattr.va_gid = gid;
3128 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3129 		vput(vp);
3130 	}
3131 
3132 	if (error == 0) {
3133 		if (uid == -1)
3134 			uid = o_uid;
3135 		if (gid == -1)
3136 			gid = o_gid;
3137 		VFS_ACCOUNT(mp, o_uid, o_gid, -size);
3138 		VFS_ACCOUNT(mp,   uid,   gid,  size);
3139 	}
3140 
3141 	return error;
3142 }
3143 
3144 int
3145 kern_chown(struct nlookupdata *nd, int uid, int gid)
3146 {
3147 	struct vnode *vp;
3148 	int error;
3149 
3150 	if ((error = nlookup(nd)) != 0)
3151 		return (error);
3152 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3153 		return (error);
3154 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3155 		error = setfown(nd->nl_nch.mount, vp, uid, gid);
3156 	vrele(vp);
3157 	return (error);
3158 }
3159 
3160 /*
3161  * chown(char *path, int uid, int gid)
3162  *
3163  * Set ownership given a path name.
3164  */
3165 int
3166 sys_chown(struct chown_args *uap)
3167 {
3168 	struct nlookupdata nd;
3169 	int error;
3170 
3171 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3172 	if (error == 0)
3173 		error = kern_chown(&nd, uap->uid, uap->gid);
3174 	nlookup_done(&nd);
3175 	return (error);
3176 }
3177 
3178 /*
3179  * lchown_args(char *path, int uid, int gid)
3180  *
3181  * Set ownership given a path name, do not cross symlinks.
3182  */
3183 int
3184 sys_lchown(struct lchown_args *uap)
3185 {
3186 	struct nlookupdata nd;
3187 	int error;
3188 
3189 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3190 	if (error == 0)
3191 		error = kern_chown(&nd, uap->uid, uap->gid);
3192 	nlookup_done(&nd);
3193 	return (error);
3194 }
3195 
3196 /*
3197  * fchown_args(int fd, int uid, int gid)
3198  *
3199  * Set ownership given a file descriptor.
3200  */
3201 int
3202 sys_fchown(struct fchown_args *uap)
3203 {
3204 	struct thread *td = curthread;
3205 	struct proc *p = td->td_proc;
3206 	struct file *fp;
3207 	int error;
3208 
3209 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3210 		return (error);
3211 	if (fp->f_nchandle.ncp)
3212 		error = ncp_writechk(&fp->f_nchandle);
3213 	if (error == 0)
3214 		error = setfown(p->p_fd->fd_ncdir.mount,
3215 			(struct vnode *)fp->f_data, uap->uid, uap->gid);
3216 	fdrop(fp);
3217 	return (error);
3218 }
3219 
3220 /*
3221  * fchownat(int fd, char *path, int uid, int gid, int flags)
3222  *
3223  * Set ownership of file pointed to by fd/path.
3224  */
3225 int
3226 sys_fchownat(struct fchownat_args *uap)
3227 {
3228 	struct nlookupdata nd;
3229 	struct file *fp;
3230 	int error;
3231 	int flags;
3232 
3233 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3234 		return (EINVAL);
3235 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3236 
3237 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3238 				UIO_USERSPACE, flags);
3239 	if (error == 0)
3240 		error = kern_chown(&nd, uap->uid, uap->gid);
3241 	nlookup_done_at(&nd, fp);
3242 	return (error);
3243 }
3244 
3245 
3246 static int
3247 getutimes(const struct timeval *tvp, struct timespec *tsp)
3248 {
3249 	struct timeval tv[2];
3250 
3251 	if (tvp == NULL) {
3252 		microtime(&tv[0]);
3253 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3254 		tsp[1] = tsp[0];
3255 	} else {
3256 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3257 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3258 	}
3259 	return 0;
3260 }
3261 
3262 static int
3263 setutimes(struct vnode *vp, struct vattr *vattr,
3264 	  const struct timespec *ts, int nullflag)
3265 {
3266 	struct thread *td = curthread;
3267 	int error;
3268 
3269 	VATTR_NULL(vattr);
3270 	vattr->va_atime = ts[0];
3271 	vattr->va_mtime = ts[1];
3272 	if (nullflag)
3273 		vattr->va_vaflags |= VA_UTIMES_NULL;
3274 	error = VOP_SETATTR(vp, vattr, td->td_ucred);
3275 
3276 	return error;
3277 }
3278 
3279 int
3280 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3281 {
3282 	struct timespec ts[2];
3283 	struct vnode *vp;
3284 	struct vattr vattr;
3285 	int error;
3286 
3287 	if ((error = getutimes(tptr, ts)) != 0)
3288 		return (error);
3289 
3290 	/*
3291 	 * NOTE: utimes() succeeds for the owner even if the file
3292 	 * is not user-writable.
3293 	 */
3294 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3295 
3296 	if ((error = nlookup(nd)) != 0)
3297 		return (error);
3298 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3299 		return (error);
3300 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3301 		return (error);
3302 
3303 	/*
3304 	 * note: vget is required for any operation that might mod the vnode
3305 	 * so VINACTIVE is properly cleared.
3306 	 */
3307 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3308 		error = vget(vp, LK_EXCLUSIVE);
3309 		if (error == 0) {
3310 			error = setutimes(vp, &vattr, ts, (tptr == NULL));
3311 			vput(vp);
3312 		}
3313 	}
3314 	vrele(vp);
3315 	return (error);
3316 }
3317 
3318 /*
3319  * utimes_args(char *path, struct timeval *tptr)
3320  *
3321  * Set the access and modification times of a file.
3322  */
3323 int
3324 sys_utimes(struct utimes_args *uap)
3325 {
3326 	struct timeval tv[2];
3327 	struct nlookupdata nd;
3328 	int error;
3329 
3330 	if (uap->tptr) {
3331  		error = copyin(uap->tptr, tv, sizeof(tv));
3332 		if (error)
3333 			return (error);
3334 	}
3335 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3336 	if (error == 0)
3337 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3338 	nlookup_done(&nd);
3339 	return (error);
3340 }
3341 
3342 /*
3343  * lutimes_args(char *path, struct timeval *tptr)
3344  *
3345  * Set the access and modification times of a file.
3346  */
3347 int
3348 sys_lutimes(struct lutimes_args *uap)
3349 {
3350 	struct timeval tv[2];
3351 	struct nlookupdata nd;
3352 	int error;
3353 
3354 	if (uap->tptr) {
3355 		error = copyin(uap->tptr, tv, sizeof(tv));
3356 		if (error)
3357 			return (error);
3358 	}
3359 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3360 	if (error == 0)
3361 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3362 	nlookup_done(&nd);
3363 	return (error);
3364 }
3365 
3366 /*
3367  * Set utimes on a file descriptor.  The creds used to open the
3368  * file are used to determine whether the operation is allowed
3369  * or not.
3370  */
3371 int
3372 kern_futimes(int fd, struct timeval *tptr)
3373 {
3374 	struct thread *td = curthread;
3375 	struct proc *p = td->td_proc;
3376 	struct timespec ts[2];
3377 	struct file *fp;
3378 	struct vnode *vp;
3379 	struct vattr vattr;
3380 	int error;
3381 
3382 	error = getutimes(tptr, ts);
3383 	if (error)
3384 		return (error);
3385 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3386 		return (error);
3387 	if (fp->f_nchandle.ncp)
3388 		error = ncp_writechk(&fp->f_nchandle);
3389 	if (error == 0) {
3390 		vp = fp->f_data;
3391 		error = vget(vp, LK_EXCLUSIVE);
3392 		if (error == 0) {
3393 			error = VOP_GETATTR(vp, &vattr);
3394 			if (error == 0) {
3395 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3396 						   fp->f_cred);
3397 			}
3398 			if (error == 0) {
3399 				error = setutimes(vp, &vattr, ts,
3400 						  (tptr == NULL));
3401 			}
3402 			vput(vp);
3403 		}
3404 	}
3405 	fdrop(fp);
3406 	return (error);
3407 }
3408 
3409 /*
3410  * futimes_args(int fd, struct timeval *tptr)
3411  *
3412  * Set the access and modification times of a file.
3413  */
3414 int
3415 sys_futimes(struct futimes_args *uap)
3416 {
3417 	struct timeval tv[2];
3418 	int error;
3419 
3420 	if (uap->tptr) {
3421 		error = copyin(uap->tptr, tv, sizeof(tv));
3422 		if (error)
3423 			return (error);
3424 	}
3425 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3426 
3427 	return (error);
3428 }
3429 
3430 int
3431 kern_truncate(struct nlookupdata *nd, off_t length)
3432 {
3433 	struct vnode *vp;
3434 	struct vattr vattr;
3435 	int error;
3436 	uid_t uid = 0;
3437 	gid_t gid = 0;
3438 	uint64_t old_size = 0;
3439 
3440 	if (length < 0)
3441 		return(EINVAL);
3442 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3443 	if ((error = nlookup(nd)) != 0)
3444 		return (error);
3445 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3446 		return (error);
3447 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3448 		return (error);
3449 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3450 		vrele(vp);
3451 		return (error);
3452 	}
3453 	if (vp->v_type == VDIR) {
3454 		error = EISDIR;
3455 		goto done;
3456 	}
3457 	if (vfs_quota_enabled) {
3458 		error = VOP_GETATTR(vp, &vattr);
3459 		KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0"));
3460 		uid = vattr.va_uid;
3461 		gid = vattr.va_gid;
3462 		old_size = vattr.va_size;
3463 	}
3464 
3465 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3466 		VATTR_NULL(&vattr);
3467 		vattr.va_size = length;
3468 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3469 		VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size);
3470 	}
3471 done:
3472 	vput(vp);
3473 	return (error);
3474 }
3475 
3476 /*
3477  * truncate(char *path, int pad, off_t length)
3478  *
3479  * Truncate a file given its path name.
3480  */
3481 int
3482 sys_truncate(struct truncate_args *uap)
3483 {
3484 	struct nlookupdata nd;
3485 	int error;
3486 
3487 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3488 	if (error == 0)
3489 		error = kern_truncate(&nd, uap->length);
3490 	nlookup_done(&nd);
3491 	return error;
3492 }
3493 
3494 int
3495 kern_ftruncate(int fd, off_t length)
3496 {
3497 	struct thread *td = curthread;
3498 	struct proc *p = td->td_proc;
3499 	struct vattr vattr;
3500 	struct vnode *vp;
3501 	struct file *fp;
3502 	int error;
3503 	uid_t uid = 0;
3504 	gid_t gid = 0;
3505 	uint64_t old_size = 0;
3506 	struct mount *mp;
3507 
3508 	if (length < 0)
3509 		return(EINVAL);
3510 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3511 		return (error);
3512 	if (fp->f_nchandle.ncp) {
3513 		error = ncp_writechk(&fp->f_nchandle);
3514 		if (error)
3515 			goto done;
3516 	}
3517 	if ((fp->f_flag & FWRITE) == 0) {
3518 		error = EINVAL;
3519 		goto done;
3520 	}
3521 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3522 		error = EINVAL;
3523 		goto done;
3524 	}
3525 	vp = (struct vnode *)fp->f_data;
3526 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3527 	if (vp->v_type == VDIR) {
3528 		error = EISDIR;
3529 		goto done;
3530 	}
3531 
3532 	if (vfs_quota_enabled) {
3533 		error = VOP_GETATTR(vp, &vattr);
3534 		KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0"));
3535 		uid = vattr.va_uid;
3536 		gid = vattr.va_gid;
3537 		old_size = vattr.va_size;
3538 	}
3539 
3540 	if ((error = vn_writechk(vp, NULL)) == 0) {
3541 		VATTR_NULL(&vattr);
3542 		vattr.va_size = length;
3543 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3544 		mp = vq_vptomp(vp);
3545 		VFS_ACCOUNT(mp, uid, gid, length - old_size);
3546 	}
3547 	vn_unlock(vp);
3548 done:
3549 	fdrop(fp);
3550 	return (error);
3551 }
3552 
3553 /*
3554  * ftruncate_args(int fd, int pad, off_t length)
3555  *
3556  * Truncate a file given a file descriptor.
3557  */
3558 int
3559 sys_ftruncate(struct ftruncate_args *uap)
3560 {
3561 	int error;
3562 
3563 	error = kern_ftruncate(uap->fd, uap->length);
3564 
3565 	return (error);
3566 }
3567 
3568 /*
3569  * fsync(int fd)
3570  *
3571  * Sync an open file.
3572  */
3573 int
3574 sys_fsync(struct fsync_args *uap)
3575 {
3576 	struct thread *td = curthread;
3577 	struct proc *p = td->td_proc;
3578 	struct vnode *vp;
3579 	struct file *fp;
3580 	vm_object_t obj;
3581 	int error;
3582 
3583 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3584 		return (error);
3585 	vp = (struct vnode *)fp->f_data;
3586 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3587 	if ((obj = vp->v_object) != NULL)
3588 		vm_object_page_clean(obj, 0, 0, 0);
3589 	error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3590 	if (error == 0 && vp->v_mount)
3591 		error = buf_fsync(vp);
3592 	vn_unlock(vp);
3593 	fdrop(fp);
3594 
3595 	return (error);
3596 }
3597 
3598 int
3599 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3600 {
3601 	struct nchandle fnchd;
3602 	struct nchandle tnchd;
3603 	struct namecache *ncp;
3604 	struct vnode *fdvp;
3605 	struct vnode *tdvp;
3606 	struct mount *mp;
3607 	int error;
3608 
3609 	bwillinode(1);
3610 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3611 	if ((error = nlookup(fromnd)) != 0)
3612 		return (error);
3613 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3614 		return (ENOENT);
3615 	fnchd.mount = fromnd->nl_nch.mount;
3616 	cache_hold(&fnchd);
3617 
3618 	/*
3619 	 * unlock the source nch so we can lookup the target nch without
3620 	 * deadlocking.  The target may or may not exist so we do not check
3621 	 * for a target vp like kern_mkdir() and other creation functions do.
3622 	 *
3623 	 * The source and target directories are ref'd and rechecked after
3624 	 * everything is relocked to determine if the source or target file
3625 	 * has been renamed.
3626 	 */
3627 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3628 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3629 	cache_unlock(&fromnd->nl_nch);
3630 
3631 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3632 	if ((error = nlookup(tond)) != 0) {
3633 		cache_drop(&fnchd);
3634 		return (error);
3635 	}
3636 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3637 		cache_drop(&fnchd);
3638 		return (ENOENT);
3639 	}
3640 	tnchd.mount = tond->nl_nch.mount;
3641 	cache_hold(&tnchd);
3642 
3643 	/*
3644 	 * If the source and target are the same there is nothing to do
3645 	 */
3646 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3647 		cache_drop(&fnchd);
3648 		cache_drop(&tnchd);
3649 		return (0);
3650 	}
3651 
3652 	/*
3653 	 * Mount points cannot be renamed or overwritten
3654 	 */
3655 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3656 	    NCF_ISMOUNTPT
3657 	) {
3658 		cache_drop(&fnchd);
3659 		cache_drop(&tnchd);
3660 		return (EINVAL);
3661 	}
3662 
3663 	/*
3664 	 * Relock the source ncp.  cache_relock() will deal with any
3665 	 * deadlocks against the already-locked tond and will also
3666 	 * make sure both are resolved.
3667 	 *
3668 	 * NOTE AFTER RELOCKING: The source or target ncp may have become
3669 	 * invalid while they were unlocked, nc_vp and nc_mount could
3670 	 * be NULL.
3671 	 */
3672 	cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
3673 		     &tond->nl_nch, tond->nl_cred);
3674 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3675 
3676 	/*
3677 	 * make sure the parent directories linkages are the same
3678 	 */
3679 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3680 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3681 		cache_drop(&fnchd);
3682 		cache_drop(&tnchd);
3683 		return (ENOENT);
3684 	}
3685 
3686 	/*
3687 	 * Both the source and target must be within the same filesystem and
3688 	 * in the same filesystem as their parent directories within the
3689 	 * namecache topology.
3690 	 *
3691 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3692 	 */
3693 	mp = fnchd.mount;
3694 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3695 	    mp != tond->nl_nch.mount) {
3696 		cache_drop(&fnchd);
3697 		cache_drop(&tnchd);
3698 		return (EXDEV);
3699 	}
3700 
3701 	/*
3702 	 * Make sure the mount point is writable
3703 	 */
3704 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3705 		cache_drop(&fnchd);
3706 		cache_drop(&tnchd);
3707 		return (error);
3708 	}
3709 
3710 	/*
3711 	 * If the target exists and either the source or target is a directory,
3712 	 * then both must be directories.
3713 	 *
3714 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3715 	 * have become NULL.
3716 	 */
3717 	if (tond->nl_nch.ncp->nc_vp) {
3718 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3719 			error = ENOENT;
3720 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3721 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3722 				error = ENOTDIR;
3723 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3724 			error = EISDIR;
3725 		}
3726 	}
3727 
3728 	/*
3729 	 * You cannot rename a source into itself or a subdirectory of itself.
3730 	 * We check this by travsersing the target directory upwards looking
3731 	 * for a match against the source.
3732 	 *
3733 	 * XXX MPSAFE
3734 	 */
3735 	if (error == 0) {
3736 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3737 			if (fromnd->nl_nch.ncp == ncp) {
3738 				error = EINVAL;
3739 				break;
3740 			}
3741 		}
3742 	}
3743 
3744 	cache_drop(&fnchd);
3745 	cache_drop(&tnchd);
3746 
3747 	/*
3748 	 * Even though the namespaces are different, they may still represent
3749 	 * hardlinks to the same file.  The filesystem might have a hard time
3750 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3751 	 * when we detect the situation.
3752 	 */
3753 	if (error == 0) {
3754 		fdvp = fromnd->nl_dvp;
3755 		tdvp = tond->nl_dvp;
3756 		if (fdvp == NULL || tdvp == NULL) {
3757 			error = EPERM;
3758 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3759 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3760 					    fromnd->nl_cred);
3761 		} else {
3762 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3763 					    fdvp, tdvp, tond->nl_cred);
3764 		}
3765 	}
3766 	return (error);
3767 }
3768 
3769 /*
3770  * rename_args(char *from, char *to)
3771  *
3772  * Rename files.  Source and destination must either both be directories,
3773  * or both not be directories.  If target is a directory, it must be empty.
3774  */
3775 int
3776 sys_rename(struct rename_args *uap)
3777 {
3778 	struct nlookupdata fromnd, tond;
3779 	int error;
3780 
3781 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3782 	if (error == 0) {
3783 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3784 		if (error == 0)
3785 			error = kern_rename(&fromnd, &tond);
3786 		nlookup_done(&tond);
3787 	}
3788 	nlookup_done(&fromnd);
3789 	return (error);
3790 }
3791 
3792 /*
3793  * renameat_args(int oldfd, char *old, int newfd, char *new)
3794  *
3795  * Rename files using paths relative to the directories associated with
3796  * oldfd and newfd.  Source and destination must either both be directories,
3797  * or both not be directories.  If target is a directory, it must be empty.
3798  */
3799 int
3800 sys_renameat(struct renameat_args *uap)
3801 {
3802 	struct nlookupdata oldnd, newnd;
3803 	struct file *oldfp, *newfp;
3804 	int error;
3805 
3806 	error = nlookup_init_at(&oldnd, &oldfp, uap->oldfd, uap->old,
3807 	    UIO_USERSPACE, 0);
3808 	if (error == 0) {
3809 		error = nlookup_init_at(&newnd, &newfp, uap->newfd, uap->new,
3810 		    UIO_USERSPACE, 0);
3811 		if (error == 0)
3812 			error = kern_rename(&oldnd, &newnd);
3813 		nlookup_done_at(&newnd, newfp);
3814 	}
3815 	nlookup_done_at(&oldnd, oldfp);
3816 	return (error);
3817 }
3818 
3819 int
3820 kern_mkdir(struct nlookupdata *nd, int mode)
3821 {
3822 	struct thread *td = curthread;
3823 	struct proc *p = td->td_proc;
3824 	struct vnode *vp;
3825 	struct vattr vattr;
3826 	int error;
3827 
3828 	bwillinode(1);
3829 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3830 	if ((error = nlookup(nd)) != 0)
3831 		return (error);
3832 
3833 	if (nd->nl_nch.ncp->nc_vp)
3834 		return (EEXIST);
3835 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3836 		return (error);
3837 	VATTR_NULL(&vattr);
3838 	vattr.va_type = VDIR;
3839 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3840 
3841 	vp = NULL;
3842 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
3843 	if (error == 0)
3844 		vput(vp);
3845 	return (error);
3846 }
3847 
3848 /*
3849  * mkdir_args(char *path, int mode)
3850  *
3851  * Make a directory file.
3852  */
3853 int
3854 sys_mkdir(struct mkdir_args *uap)
3855 {
3856 	struct nlookupdata nd;
3857 	int error;
3858 
3859 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3860 	if (error == 0)
3861 		error = kern_mkdir(&nd, uap->mode);
3862 	nlookup_done(&nd);
3863 	return (error);
3864 }
3865 
3866 /*
3867  * mkdirat_args(int fd, char *path, mode_t mode)
3868  *
3869  * Make a directory file.  The path is relative to the directory associated
3870  * with fd.
3871  */
3872 int
3873 sys_mkdirat(struct mkdirat_args *uap)
3874 {
3875 	struct nlookupdata nd;
3876 	struct file *fp;
3877 	int error;
3878 
3879 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
3880 	if (error == 0)
3881 		error = kern_mkdir(&nd, uap->mode);
3882 	nlookup_done_at(&nd, fp);
3883 	return (error);
3884 }
3885 
3886 int
3887 kern_rmdir(struct nlookupdata *nd)
3888 {
3889 	int error;
3890 
3891 	bwillinode(1);
3892 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3893 	if ((error = nlookup(nd)) != 0)
3894 		return (error);
3895 
3896 	/*
3897 	 * Do not allow directories representing mount points to be
3898 	 * deleted, even if empty.  Check write perms on mount point
3899 	 * in case the vnode is aliased (aka nullfs).
3900 	 */
3901 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3902 		return (EINVAL);
3903 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3904 		return (error);
3905 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3906 	return (error);
3907 }
3908 
3909 /*
3910  * rmdir_args(char *path)
3911  *
3912  * Remove a directory file.
3913  */
3914 int
3915 sys_rmdir(struct rmdir_args *uap)
3916 {
3917 	struct nlookupdata nd;
3918 	int error;
3919 
3920 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3921 	if (error == 0)
3922 		error = kern_rmdir(&nd);
3923 	nlookup_done(&nd);
3924 	return (error);
3925 }
3926 
3927 int
3928 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3929 		   enum uio_seg direction)
3930 {
3931 	struct thread *td = curthread;
3932 	struct proc *p = td->td_proc;
3933 	struct vnode *vp;
3934 	struct file *fp;
3935 	struct uio auio;
3936 	struct iovec aiov;
3937 	off_t loff;
3938 	int error, eofflag;
3939 
3940 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3941 		return (error);
3942 	if ((fp->f_flag & FREAD) == 0) {
3943 		error = EBADF;
3944 		goto done;
3945 	}
3946 	vp = (struct vnode *)fp->f_data;
3947 unionread:
3948 	if (vp->v_type != VDIR) {
3949 		error = EINVAL;
3950 		goto done;
3951 	}
3952 	aiov.iov_base = buf;
3953 	aiov.iov_len = count;
3954 	auio.uio_iov = &aiov;
3955 	auio.uio_iovcnt = 1;
3956 	auio.uio_rw = UIO_READ;
3957 	auio.uio_segflg = direction;
3958 	auio.uio_td = td;
3959 	auio.uio_resid = count;
3960 	loff = auio.uio_offset = fp->f_offset;
3961 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3962 	fp->f_offset = auio.uio_offset;
3963 	if (error)
3964 		goto done;
3965 	if (count == auio.uio_resid) {
3966 		if (union_dircheckp) {
3967 			error = union_dircheckp(td, &vp, fp);
3968 			if (error == -1)
3969 				goto unionread;
3970 			if (error)
3971 				goto done;
3972 		}
3973 #if 0
3974 		if ((vp->v_flag & VROOT) &&
3975 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3976 			struct vnode *tvp = vp;
3977 			vp = vp->v_mount->mnt_vnodecovered;
3978 			vref(vp);
3979 			fp->f_data = vp;
3980 			fp->f_offset = 0;
3981 			vrele(tvp);
3982 			goto unionread;
3983 		}
3984 #endif
3985 	}
3986 
3987 	/*
3988 	 * WARNING!  *basep may not be wide enough to accomodate the
3989 	 * seek offset.   XXX should we hack this to return the upper 32 bits
3990 	 * for offsets greater then 4G?
3991 	 */
3992 	if (basep) {
3993 		*basep = (long)loff;
3994 	}
3995 	*res = count - auio.uio_resid;
3996 done:
3997 	fdrop(fp);
3998 	return (error);
3999 }
4000 
4001 /*
4002  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4003  *
4004  * Read a block of directory entries in a file system independent format.
4005  */
4006 int
4007 sys_getdirentries(struct getdirentries_args *uap)
4008 {
4009 	long base;
4010 	int error;
4011 
4012 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4013 				   &uap->sysmsg_result, UIO_USERSPACE);
4014 
4015 	if (error == 0 && uap->basep)
4016 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
4017 	return (error);
4018 }
4019 
4020 /*
4021  * getdents_args(int fd, char *buf, size_t count)
4022  */
4023 int
4024 sys_getdents(struct getdents_args *uap)
4025 {
4026 	int error;
4027 
4028 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4029 				   &uap->sysmsg_result, UIO_USERSPACE);
4030 
4031 	return (error);
4032 }
4033 
4034 /*
4035  * Set the mode mask for creation of filesystem nodes.
4036  *
4037  * umask(int newmask)
4038  */
4039 int
4040 sys_umask(struct umask_args *uap)
4041 {
4042 	struct thread *td = curthread;
4043 	struct proc *p = td->td_proc;
4044 	struct filedesc *fdp;
4045 
4046 	fdp = p->p_fd;
4047 	uap->sysmsg_result = fdp->fd_cmask;
4048 	fdp->fd_cmask = uap->newmask & ALLPERMS;
4049 	return (0);
4050 }
4051 
4052 /*
4053  * revoke(char *path)
4054  *
4055  * Void all references to file by ripping underlying filesystem
4056  * away from vnode.
4057  */
4058 int
4059 sys_revoke(struct revoke_args *uap)
4060 {
4061 	struct nlookupdata nd;
4062 	struct vattr vattr;
4063 	struct vnode *vp;
4064 	struct ucred *cred;
4065 	int error;
4066 
4067 	vp = NULL;
4068 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4069 	if (error == 0)
4070 		error = nlookup(&nd);
4071 	if (error == 0)
4072 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4073 	cred = crhold(nd.nl_cred);
4074 	nlookup_done(&nd);
4075 	if (error == 0) {
4076 		if (error == 0)
4077 			error = VOP_GETATTR(vp, &vattr);
4078 		if (error == 0 && cred->cr_uid != vattr.va_uid)
4079 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4080 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4081 			if (vcount(vp) > 0)
4082 				error = vrevoke(vp, cred);
4083 		} else if (error == 0) {
4084 			error = vrevoke(vp, cred);
4085 		}
4086 		vrele(vp);
4087 	}
4088 	if (cred)
4089 		crfree(cred);
4090 	return (error);
4091 }
4092 
4093 /*
4094  * getfh_args(char *fname, fhandle_t *fhp)
4095  *
4096  * Get (NFS) file handle
4097  *
4098  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4099  * mount.  This allows nullfs mounts to be explicitly exported.
4100  *
4101  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4102  *
4103  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
4104  *	    work, but you do not really have protection against access to
4105  *	    the related parent directories.
4106  */
4107 int
4108 sys_getfh(struct getfh_args *uap)
4109 {
4110 	struct thread *td = curthread;
4111 	struct nlookupdata nd;
4112 	fhandle_t fh;
4113 	struct vnode *vp;
4114 	struct mount *mp;
4115 	int error;
4116 
4117 	/*
4118 	 * Must be super user
4119 	 */
4120 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
4121 		return (error);
4122 
4123 	vp = NULL;
4124 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4125 	if (error == 0)
4126 		error = nlookup(&nd);
4127 	if (error == 0)
4128 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4129 	mp = nd.nl_nch.mount;
4130 	nlookup_done(&nd);
4131 	if (error == 0) {
4132 		bzero(&fh, sizeof(fh));
4133 		fh.fh_fsid = mp->mnt_stat.f_fsid;
4134 		error = VFS_VPTOFH(vp, &fh.fh_fid);
4135 		vput(vp);
4136 		if (error == 0)
4137 			error = copyout(&fh, uap->fhp, sizeof(fh));
4138 	}
4139 	return (error);
4140 }
4141 
4142 /*
4143  * fhopen_args(const struct fhandle *u_fhp, int flags)
4144  *
4145  * syscall for the rpc.lockd to use to translate a NFS file handle into
4146  * an open descriptor.
4147  *
4148  * warning: do not remove the priv_check() call or this becomes one giant
4149  * security hole.
4150  */
4151 int
4152 sys_fhopen(struct fhopen_args *uap)
4153 {
4154 	struct thread *td = curthread;
4155 	struct filedesc *fdp = td->td_proc->p_fd;
4156 	struct mount *mp;
4157 	struct vnode *vp;
4158 	struct fhandle fhp;
4159 	struct vattr vat;
4160 	struct vattr *vap = &vat;
4161 	struct flock lf;
4162 	int fmode, mode, error = 0, type;
4163 	struct file *nfp;
4164 	struct file *fp;
4165 	int indx;
4166 
4167 	/*
4168 	 * Must be super user
4169 	 */
4170 	error = priv_check(td, PRIV_ROOT);
4171 	if (error)
4172 		return (error);
4173 
4174 	fmode = FFLAGS(uap->flags);
4175 
4176 	/*
4177 	 * Why not allow a non-read/write open for our lockd?
4178 	 */
4179 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4180 		return (EINVAL);
4181 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4182 	if (error)
4183 		return(error);
4184 
4185 	/*
4186 	 * Find the mount point
4187 	 */
4188 	mp = vfs_getvfs(&fhp.fh_fsid);
4189 	if (mp == NULL) {
4190 		error = ESTALE;
4191 		goto  done;
4192 	}
4193 	/* now give me my vnode, it gets returned to me locked */
4194 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4195 	if (error)
4196 		goto done;
4197  	/*
4198 	 * from now on we have to make sure not
4199 	 * to forget about the vnode
4200 	 * any error that causes an abort must vput(vp)
4201 	 * just set error = err and 'goto bad;'.
4202 	 */
4203 
4204 	/*
4205 	 * from vn_open
4206 	 */
4207 	if (vp->v_type == VLNK) {
4208 		error = EMLINK;
4209 		goto bad;
4210 	}
4211 	if (vp->v_type == VSOCK) {
4212 		error = EOPNOTSUPP;
4213 		goto bad;
4214 	}
4215 	mode = 0;
4216 	if (fmode & (FWRITE | O_TRUNC)) {
4217 		if (vp->v_type == VDIR) {
4218 			error = EISDIR;
4219 			goto bad;
4220 		}
4221 		error = vn_writechk(vp, NULL);
4222 		if (error)
4223 			goto bad;
4224 		mode |= VWRITE;
4225 	}
4226 	if (fmode & FREAD)
4227 		mode |= VREAD;
4228 	if (mode) {
4229 		error = VOP_ACCESS(vp, mode, td->td_ucred);
4230 		if (error)
4231 			goto bad;
4232 	}
4233 	if (fmode & O_TRUNC) {
4234 		vn_unlock(vp);				/* XXX */
4235 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
4236 		VATTR_NULL(vap);
4237 		vap->va_size = 0;
4238 		error = VOP_SETATTR(vp, vap, td->td_ucred);
4239 		if (error)
4240 			goto bad;
4241 	}
4242 
4243 	/*
4244 	 * VOP_OPEN needs the file pointer so it can potentially override
4245 	 * it.
4246 	 *
4247 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
4248 	 * directory.  XXX
4249 	 */
4250 	if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4251 		goto bad;
4252 	fp = nfp;
4253 
4254 	error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4255 	if (error) {
4256 		/*
4257 		 * setting f_ops this way prevents VOP_CLOSE from being
4258 		 * called or fdrop() releasing the vp from v_data.   Since
4259 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4260 		 */
4261 		fp->f_ops = &badfileops;
4262 		fp->f_data = NULL;
4263 		goto bad_drop;
4264 	}
4265 
4266 	/*
4267 	 * The fp is given its own reference, we still have our ref and lock.
4268 	 *
4269 	 * Assert that all regular files must be created with a VM object.
4270 	 */
4271 	if (vp->v_type == VREG && vp->v_object == NULL) {
4272 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4273 		goto bad_drop;
4274 	}
4275 
4276 	/*
4277 	 * The open was successful.  Handle any locking requirements.
4278 	 */
4279 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
4280 		lf.l_whence = SEEK_SET;
4281 		lf.l_start = 0;
4282 		lf.l_len = 0;
4283 		if (fmode & O_EXLOCK)
4284 			lf.l_type = F_WRLCK;
4285 		else
4286 			lf.l_type = F_RDLCK;
4287 		if (fmode & FNONBLOCK)
4288 			type = 0;
4289 		else
4290 			type = F_WAIT;
4291 		vn_unlock(vp);
4292 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4293 			/*
4294 			 * release our private reference.
4295 			 */
4296 			fsetfd(fdp, NULL, indx);
4297 			fdrop(fp);
4298 			vrele(vp);
4299 			goto done;
4300 		}
4301 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4302 		fp->f_flag |= FHASLOCK;
4303 	}
4304 
4305 	/*
4306 	 * Clean up.  Associate the file pointer with the previously
4307 	 * reserved descriptor and return it.
4308 	 */
4309 	vput(vp);
4310 	fsetfd(fdp, fp, indx);
4311 	fdrop(fp);
4312 	uap->sysmsg_result = indx;
4313 	if (uap->flags & O_CLOEXEC)
4314 		error = fsetfdflags(fdp, indx, UF_EXCLOSE);
4315 	return (error);
4316 
4317 bad_drop:
4318 	fsetfd(fdp, NULL, indx);
4319 	fdrop(fp);
4320 bad:
4321 	vput(vp);
4322 done:
4323 	return (error);
4324 }
4325 
4326 /*
4327  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4328  */
4329 int
4330 sys_fhstat(struct fhstat_args *uap)
4331 {
4332 	struct thread *td = curthread;
4333 	struct stat sb;
4334 	fhandle_t fh;
4335 	struct mount *mp;
4336 	struct vnode *vp;
4337 	int error;
4338 
4339 	/*
4340 	 * Must be super user
4341 	 */
4342 	error = priv_check(td, PRIV_ROOT);
4343 	if (error)
4344 		return (error);
4345 
4346 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4347 	if (error)
4348 		return (error);
4349 
4350 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4351 		error = ESTALE;
4352 	if (error == 0) {
4353 		if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4354 			error = vn_stat(vp, &sb, td->td_ucred);
4355 			vput(vp);
4356 		}
4357 	}
4358 	if (error == 0)
4359 		error = copyout(&sb, uap->sb, sizeof(sb));
4360 	return (error);
4361 }
4362 
4363 /*
4364  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4365  */
4366 int
4367 sys_fhstatfs(struct fhstatfs_args *uap)
4368 {
4369 	struct thread *td = curthread;
4370 	struct proc *p = td->td_proc;
4371 	struct statfs *sp;
4372 	struct mount *mp;
4373 	struct vnode *vp;
4374 	struct statfs sb;
4375 	char *fullpath, *freepath;
4376 	fhandle_t fh;
4377 	int error;
4378 
4379 	/*
4380 	 * Must be super user
4381 	 */
4382 	if ((error = priv_check(td, PRIV_ROOT)))
4383 		return (error);
4384 
4385 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4386 		return (error);
4387 
4388 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4389 		error = ESTALE;
4390 		goto done;
4391 	}
4392 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4393 		error = ESTALE;
4394 		goto done;
4395 	}
4396 
4397 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4398 		goto done;
4399 	mp = vp->v_mount;
4400 	sp = &mp->mnt_stat;
4401 	vput(vp);
4402 	if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4403 		goto done;
4404 
4405 	error = mount_path(p, mp, &fullpath, &freepath);
4406 	if (error)
4407 		goto done;
4408 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4409 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4410 	kfree(freepath, M_TEMP);
4411 
4412 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4413 	if (priv_check(td, PRIV_ROOT)) {
4414 		bcopy(sp, &sb, sizeof(sb));
4415 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4416 		sp = &sb;
4417 	}
4418 	error = copyout(sp, uap->buf, sizeof(*sp));
4419 done:
4420 	return (error);
4421 }
4422 
4423 /*
4424  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4425  */
4426 int
4427 sys_fhstatvfs(struct fhstatvfs_args *uap)
4428 {
4429 	struct thread *td = curthread;
4430 	struct proc *p = td->td_proc;
4431 	struct statvfs *sp;
4432 	struct mount *mp;
4433 	struct vnode *vp;
4434 	fhandle_t fh;
4435 	int error;
4436 
4437 	/*
4438 	 * Must be super user
4439 	 */
4440 	if ((error = priv_check(td, PRIV_ROOT)))
4441 		return (error);
4442 
4443 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4444 		return (error);
4445 
4446 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4447 		error = ESTALE;
4448 		goto done;
4449 	}
4450 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4451 		error = ESTALE;
4452 		goto done;
4453 	}
4454 
4455 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4456 		goto done;
4457 	mp = vp->v_mount;
4458 	sp = &mp->mnt_vstat;
4459 	vput(vp);
4460 	if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4461 		goto done;
4462 
4463 	sp->f_flag = 0;
4464 	if (mp->mnt_flag & MNT_RDONLY)
4465 		sp->f_flag |= ST_RDONLY;
4466 	if (mp->mnt_flag & MNT_NOSUID)
4467 		sp->f_flag |= ST_NOSUID;
4468 	error = copyout(sp, uap->buf, sizeof(*sp));
4469 done:
4470 	return (error);
4471 }
4472 
4473 
4474 /*
4475  * Syscall to push extended attribute configuration information into the
4476  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4477  * a command (int cmd), and attribute name and misc data.  For now, the
4478  * attribute name is left in userspace for consumption by the VFS_op.
4479  * It will probably be changed to be copied into sysspace by the
4480  * syscall in the future, once issues with various consumers of the
4481  * attribute code have raised their hands.
4482  *
4483  * Currently this is used only by UFS Extended Attributes.
4484  */
4485 int
4486 sys_extattrctl(struct extattrctl_args *uap)
4487 {
4488 	struct nlookupdata nd;
4489 	struct vnode *vp;
4490 	char attrname[EXTATTR_MAXNAMELEN];
4491 	int error;
4492 	size_t size;
4493 
4494 	attrname[0] = 0;
4495 	vp = NULL;
4496 	error = 0;
4497 
4498 	if (error == 0 && uap->filename) {
4499 		error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4500 				     NLC_FOLLOW);
4501 		if (error == 0)
4502 			error = nlookup(&nd);
4503 		if (error == 0)
4504 			error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4505 		nlookup_done(&nd);
4506 	}
4507 
4508 	if (error == 0 && uap->attrname) {
4509 		error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4510 				  &size);
4511 	}
4512 
4513 	if (error == 0) {
4514 		error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4515 		if (error == 0)
4516 			error = nlookup(&nd);
4517 		if (error == 0)
4518 			error = ncp_writechk(&nd.nl_nch);
4519 		if (error == 0) {
4520 			error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4521 					       uap->attrnamespace,
4522 					       uap->attrname, nd.nl_cred);
4523 		}
4524 		nlookup_done(&nd);
4525 	}
4526 
4527 	return (error);
4528 }
4529 
4530 /*
4531  * Syscall to get a named extended attribute on a file or directory.
4532  */
4533 int
4534 sys_extattr_set_file(struct extattr_set_file_args *uap)
4535 {
4536 	char attrname[EXTATTR_MAXNAMELEN];
4537 	struct nlookupdata nd;
4538 	struct vnode *vp;
4539 	struct uio auio;
4540 	struct iovec aiov;
4541 	int error;
4542 
4543 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4544 	if (error)
4545 		return (error);
4546 
4547 	vp = NULL;
4548 
4549 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4550 	if (error == 0)
4551 		error = nlookup(&nd);
4552 	if (error == 0)
4553 		error = ncp_writechk(&nd.nl_nch);
4554 	if (error == 0)
4555 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4556 	if (error) {
4557 		nlookup_done(&nd);
4558 		return (error);
4559 	}
4560 
4561 	bzero(&auio, sizeof(auio));
4562 	aiov.iov_base = uap->data;
4563 	aiov.iov_len = uap->nbytes;
4564 	auio.uio_iov = &aiov;
4565 	auio.uio_iovcnt = 1;
4566 	auio.uio_offset = 0;
4567 	auio.uio_resid = uap->nbytes;
4568 	auio.uio_rw = UIO_WRITE;
4569 	auio.uio_td = curthread;
4570 
4571 	error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4572 			       &auio, nd.nl_cred);
4573 
4574 	vput(vp);
4575 	nlookup_done(&nd);
4576 	return (error);
4577 }
4578 
4579 /*
4580  * Syscall to get a named extended attribute on a file or directory.
4581  */
4582 int
4583 sys_extattr_get_file(struct extattr_get_file_args *uap)
4584 {
4585 	char attrname[EXTATTR_MAXNAMELEN];
4586 	struct nlookupdata nd;
4587 	struct uio auio;
4588 	struct iovec aiov;
4589 	struct vnode *vp;
4590 	int error;
4591 
4592 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4593 	if (error)
4594 		return (error);
4595 
4596 	vp = NULL;
4597 
4598 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4599 	if (error == 0)
4600 		error = nlookup(&nd);
4601 	if (error == 0)
4602 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4603 	if (error) {
4604 		nlookup_done(&nd);
4605 		return (error);
4606 	}
4607 
4608 	bzero(&auio, sizeof(auio));
4609 	aiov.iov_base = uap->data;
4610 	aiov.iov_len = uap->nbytes;
4611 	auio.uio_iov = &aiov;
4612 	auio.uio_iovcnt = 1;
4613 	auio.uio_offset = 0;
4614 	auio.uio_resid = uap->nbytes;
4615 	auio.uio_rw = UIO_READ;
4616 	auio.uio_td = curthread;
4617 
4618 	error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4619 				&auio, nd.nl_cred);
4620 	uap->sysmsg_result = uap->nbytes - auio.uio_resid;
4621 
4622 	vput(vp);
4623 	nlookup_done(&nd);
4624 	return(error);
4625 }
4626 
4627 /*
4628  * Syscall to delete a named extended attribute from a file or directory.
4629  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4630  */
4631 int
4632 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4633 {
4634 	char attrname[EXTATTR_MAXNAMELEN];
4635 	struct nlookupdata nd;
4636 	struct vnode *vp;
4637 	int error;
4638 
4639 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4640 	if (error)
4641 		return(error);
4642 
4643 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4644 	if (error == 0)
4645 		error = nlookup(&nd);
4646 	if (error == 0)
4647 		error = ncp_writechk(&nd.nl_nch);
4648 	if (error == 0) {
4649 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4650 		if (error == 0) {
4651 			error = VOP_SETEXTATTR(vp, uap->attrnamespace,
4652 					       attrname, NULL, nd.nl_cred);
4653 			vput(vp);
4654 		}
4655 	}
4656 	nlookup_done(&nd);
4657 	return(error);
4658 }
4659 
4660 /*
4661  * Determine if the mount is visible to the process.
4662  */
4663 static int
4664 chroot_visible_mnt(struct mount *mp, struct proc *p)
4665 {
4666 	struct nchandle nch;
4667 
4668 	/*
4669 	 * Traverse from the mount point upwards.  If we hit the process
4670 	 * root then the mount point is visible to the process.
4671 	 */
4672 	nch = mp->mnt_ncmountpt;
4673 	while (nch.ncp) {
4674 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4675 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4676 			return(1);
4677 		}
4678 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4679 			nch = nch.mount->mnt_ncmounton;
4680 		} else {
4681 			nch.ncp = nch.ncp->nc_parent;
4682 		}
4683 	}
4684 
4685 	/*
4686 	 * If the mount point is not visible to the process, but the
4687 	 * process root is in a subdirectory of the mount, return
4688 	 * TRUE anyway.
4689 	 */
4690 	if (p->p_fd->fd_nrdir.mount == mp)
4691 		return(1);
4692 
4693 	return(0);
4694 }
4695 
4696