xref: /freebsd/sys/kern/vfs_syscalls.c (revision 0afa8e065e14bb8fd338d75690e0238c00167d40)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/bio.h>
48 #include <sys/buf.h>
49 #include <sys/capsicum.h>
50 #include <sys/disk.h>
51 #include <sys/sysent.h>
52 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/namei.h>
57 #include <sys/filedesc.h>
58 #include <sys/kernel.h>
59 #include <sys/fcntl.h>
60 #include <sys/file.h>
61 #include <sys/filio.h>
62 #include <sys/limits.h>
63 #include <sys/linker.h>
64 #include <sys/rwlock.h>
65 #include <sys/sdt.h>
66 #include <sys/stat.h>
67 #include <sys/sx.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
70 #include <sys/priv.h>
71 #include <sys/proc.h>
72 #include <sys/dirent.h>
73 #include <sys/jail.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79 
80 #include <machine/stdarg.h>
81 
82 #include <security/audit/audit.h>
83 #include <security/mac/mac_framework.h>
84 
85 #include <vm/vm.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/uma.h>
89 
90 #include <fs/devfs/devfs.h>
91 
92 MALLOC_DEFINE(M_FADVISE, "fadvise", "posix_fadvise(2) information");
93 
94 static int kern_chflagsat(struct thread *td, int fd, const char *path,
95     enum uio_seg pathseg, u_long flags, int atflag);
96 static int setfflags(struct thread *td, struct vnode *, u_long);
97 static int getutimes(const struct timeval *, enum uio_seg, struct timespec *);
98 static int getutimens(const struct timespec *, enum uio_seg,
99     struct timespec *, int *);
100 static int setutimes(struct thread *td, struct vnode *,
101     const struct timespec *, int, int);
102 static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred,
103     struct thread *td);
104 static int kern_fhlinkat(struct thread *td, int fd, const char *path,
105     enum uio_seg pathseg, fhandle_t *fhp);
106 static int kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg,
107     size_t count, struct thread *td);
108 static int kern_linkat_vp(struct thread *td, struct vnode *vp, int fd,
109     const char *path, enum uio_seg segflag);
110 
111 static uint64_t
112 at2cnpflags(u_int at_flags, u_int mask)
113 {
114 	u_int64_t res;
115 
116 	MPASS((at_flags & (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW)) !=
117 	    (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW));
118 
119 	res = 0;
120 	at_flags &= mask;
121 	if ((at_flags & AT_RESOLVE_BENEATH) != 0)
122 		res |= RBENEATH;
123 	if ((at_flags & AT_SYMLINK_FOLLOW) != 0)
124 		res |= FOLLOW;
125 	/* NOFOLLOW is pseudo flag */
126 	if ((mask & AT_SYMLINK_NOFOLLOW) != 0) {
127 		res |= (at_flags & AT_SYMLINK_NOFOLLOW) != 0 ? NOFOLLOW :
128 		    FOLLOW;
129 	}
130 	if ((mask & AT_EMPTY_PATH) != 0 && (at_flags & AT_EMPTY_PATH) != 0)
131 		res |= EMPTYPATH;
132 	return (res);
133 }
134 
135 int
136 kern_sync(struct thread *td)
137 {
138 	struct mount *mp, *nmp;
139 	int save;
140 
141 	mtx_lock(&mountlist_mtx);
142 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
143 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
144 			nmp = TAILQ_NEXT(mp, mnt_list);
145 			continue;
146 		}
147 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
148 		    vn_start_write(NULL, &mp, V_NOWAIT) == 0) {
149 			save = curthread_pflags_set(TDP_SYNCIO);
150 			vfs_periodic(mp, MNT_NOWAIT);
151 			VFS_SYNC(mp, MNT_NOWAIT);
152 			curthread_pflags_restore(save);
153 			vn_finished_write(mp);
154 		}
155 		mtx_lock(&mountlist_mtx);
156 		nmp = TAILQ_NEXT(mp, mnt_list);
157 		vfs_unbusy(mp);
158 	}
159 	mtx_unlock(&mountlist_mtx);
160 	return (0);
161 }
162 
163 /*
164  * Sync each mounted filesystem.
165  */
166 #ifndef _SYS_SYSPROTO_H_
167 struct sync_args {
168 	int     dummy;
169 };
170 #endif
171 /* ARGSUSED */
172 int
173 sys_sync(struct thread *td, struct sync_args *uap)
174 {
175 
176 	return (kern_sync(td));
177 }
178 
179 /*
180  * Change filesystem quotas.
181  */
182 #ifndef _SYS_SYSPROTO_H_
183 struct quotactl_args {
184 	char *path;
185 	int cmd;
186 	int uid;
187 	caddr_t arg;
188 };
189 #endif
190 int
191 sys_quotactl(struct thread *td, struct quotactl_args *uap)
192 {
193 	struct mount *mp;
194 	struct nameidata nd;
195 	int error;
196 	bool mp_busy;
197 
198 	AUDIT_ARG_CMD(uap->cmd);
199 	AUDIT_ARG_UID(uap->uid);
200 	if (!prison_allow(td->td_ucred, PR_ALLOW_QUOTAS))
201 		return (EPERM);
202 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE,
203 	    uap->path, td);
204 	if ((error = namei(&nd)) != 0)
205 		return (error);
206 	NDFREE(&nd, NDF_ONLY_PNBUF);
207 	mp = nd.ni_vp->v_mount;
208 	vfs_ref(mp);
209 	vput(nd.ni_vp);
210 	error = vfs_busy(mp, 0);
211 	if (error != 0) {
212 		vfs_rel(mp);
213 		return (error);
214 	}
215 	mp_busy = true;
216 	error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, uap->arg, &mp_busy);
217 
218 	/*
219 	 * Since quota on/off operations typically need to open quota
220 	 * files, the implementation may need to unbusy the mount point
221 	 * before calling into namei.  Otherwise, unmount might be
222 	 * started between two vfs_busy() invocations (first is ours,
223 	 * second is from mount point cross-walk code in lookup()),
224 	 * causing deadlock.
225 	 *
226 	 * Avoid unbusying mp if the implementation indicates it has
227 	 * already done so.
228 	 */
229 	if (mp_busy)
230 		vfs_unbusy(mp);
231 	vfs_rel(mp);
232 	return (error);
233 }
234 
235 /*
236  * Used by statfs conversion routines to scale the block size up if
237  * necessary so that all of the block counts are <= 'max_size'.  Note
238  * that 'max_size' should be a bitmask, i.e. 2^n - 1 for some non-zero
239  * value of 'n'.
240  */
241 void
242 statfs_scale_blocks(struct statfs *sf, long max_size)
243 {
244 	uint64_t count;
245 	int shift;
246 
247 	KASSERT(powerof2(max_size + 1), ("%s: invalid max_size", __func__));
248 
249 	/*
250 	 * Attempt to scale the block counts to give a more accurate
251 	 * overview to userland of the ratio of free space to used
252 	 * space.  To do this, find the largest block count and compute
253 	 * a divisor that lets it fit into a signed integer <= max_size.
254 	 */
255 	if (sf->f_bavail < 0)
256 		count = -sf->f_bavail;
257 	else
258 		count = sf->f_bavail;
259 	count = MAX(sf->f_blocks, MAX(sf->f_bfree, count));
260 	if (count <= max_size)
261 		return;
262 
263 	count >>= flsl(max_size);
264 	shift = 0;
265 	while (count > 0) {
266 		shift++;
267 		count >>=1;
268 	}
269 
270 	sf->f_bsize <<= shift;
271 	sf->f_blocks >>= shift;
272 	sf->f_bfree >>= shift;
273 	sf->f_bavail >>= shift;
274 }
275 
276 static int
277 kern_do_statfs(struct thread *td, struct mount *mp, struct statfs *buf)
278 {
279 	int error;
280 
281 	if (mp == NULL)
282 		return (EBADF);
283 	error = vfs_busy(mp, 0);
284 	vfs_rel(mp);
285 	if (error != 0)
286 		return (error);
287 #ifdef MAC
288 	error = mac_mount_check_stat(td->td_ucred, mp);
289 	if (error != 0)
290 		goto out;
291 #endif
292 	error = VFS_STATFS(mp, buf);
293 	if (error != 0)
294 		goto out;
295 	if (priv_check_cred_vfs_generation(td->td_ucred)) {
296 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
297 		prison_enforce_statfs(td->td_ucred, mp, buf);
298 	}
299 out:
300 	vfs_unbusy(mp);
301 	return (error);
302 }
303 
304 /*
305  * Get filesystem statistics.
306  */
307 #ifndef _SYS_SYSPROTO_H_
308 struct statfs_args {
309 	char *path;
310 	struct statfs *buf;
311 };
312 #endif
313 int
314 sys_statfs(struct thread *td, struct statfs_args *uap)
315 {
316 	struct statfs *sfp;
317 	int error;
318 
319 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
320 	error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp);
321 	if (error == 0)
322 		error = copyout(sfp, uap->buf, sizeof(struct statfs));
323 	free(sfp, M_STATFS);
324 	return (error);
325 }
326 
327 int
328 kern_statfs(struct thread *td, const char *path, enum uio_seg pathseg,
329     struct statfs *buf)
330 {
331 	struct mount *mp;
332 	struct nameidata nd;
333 	int error;
334 
335 	NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, td);
336 	error = namei(&nd);
337 	if (error != 0)
338 		return (error);
339 	mp = vfs_ref_from_vp(nd.ni_vp);
340 	NDFREE_NOTHING(&nd);
341 	vrele(nd.ni_vp);
342 	return (kern_do_statfs(td, mp, buf));
343 }
344 
345 /*
346  * Get filesystem statistics.
347  */
348 #ifndef _SYS_SYSPROTO_H_
349 struct fstatfs_args {
350 	int fd;
351 	struct statfs *buf;
352 };
353 #endif
354 int
355 sys_fstatfs(struct thread *td, struct fstatfs_args *uap)
356 {
357 	struct statfs *sfp;
358 	int error;
359 
360 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
361 	error = kern_fstatfs(td, uap->fd, sfp);
362 	if (error == 0)
363 		error = copyout(sfp, uap->buf, sizeof(struct statfs));
364 	free(sfp, M_STATFS);
365 	return (error);
366 }
367 
368 int
369 kern_fstatfs(struct thread *td, int fd, struct statfs *buf)
370 {
371 	struct file *fp;
372 	struct mount *mp;
373 	struct vnode *vp;
374 	int error;
375 
376 	AUDIT_ARG_FD(fd);
377 	error = getvnode_path(td, fd, &cap_fstatfs_rights, &fp);
378 	if (error != 0)
379 		return (error);
380 	vp = fp->f_vnode;
381 #ifdef AUDIT
382 	if (AUDITING_TD(td)) {
383 		vn_lock(vp, LK_SHARED | LK_RETRY);
384 		AUDIT_ARG_VNODE1(vp);
385 		VOP_UNLOCK(vp);
386 	}
387 #endif
388 	mp = vfs_ref_from_vp(vp);
389 	fdrop(fp, td);
390 	return (kern_do_statfs(td, mp, buf));
391 }
392 
393 /*
394  * Get statistics on all filesystems.
395  */
396 #ifndef _SYS_SYSPROTO_H_
397 struct getfsstat_args {
398 	struct statfs *buf;
399 	long bufsize;
400 	int mode;
401 };
402 #endif
403 int
404 sys_getfsstat(struct thread *td, struct getfsstat_args *uap)
405 {
406 	size_t count;
407 	int error;
408 
409 	if (uap->bufsize < 0 || uap->bufsize > SIZE_MAX)
410 		return (EINVAL);
411 	error = kern_getfsstat(td, &uap->buf, uap->bufsize, &count,
412 	    UIO_USERSPACE, uap->mode);
413 	if (error == 0)
414 		td->td_retval[0] = count;
415 	return (error);
416 }
417 
418 /*
419  * If (bufsize > 0 && bufseg == UIO_SYSSPACE)
420  *	The caller is responsible for freeing memory which will be allocated
421  *	in '*buf'.
422  */
423 int
424 kern_getfsstat(struct thread *td, struct statfs **buf, size_t bufsize,
425     size_t *countp, enum uio_seg bufseg, int mode)
426 {
427 	struct mount *mp, *nmp;
428 	struct statfs *sfsp, *sp, *sptmp, *tofree;
429 	size_t count, maxcount;
430 	int error;
431 
432 	switch (mode) {
433 	case MNT_WAIT:
434 	case MNT_NOWAIT:
435 		break;
436 	default:
437 		if (bufseg == UIO_SYSSPACE)
438 			*buf = NULL;
439 		return (EINVAL);
440 	}
441 restart:
442 	maxcount = bufsize / sizeof(struct statfs);
443 	if (bufsize == 0) {
444 		sfsp = NULL;
445 		tofree = NULL;
446 	} else if (bufseg == UIO_USERSPACE) {
447 		sfsp = *buf;
448 		tofree = NULL;
449 	} else /* if (bufseg == UIO_SYSSPACE) */ {
450 		count = 0;
451 		mtx_lock(&mountlist_mtx);
452 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
453 			count++;
454 		}
455 		mtx_unlock(&mountlist_mtx);
456 		if (maxcount > count)
457 			maxcount = count;
458 		tofree = sfsp = *buf = malloc(maxcount * sizeof(struct statfs),
459 		    M_STATFS, M_WAITOK);
460 	}
461 
462 	count = 0;
463 
464 	/*
465 	 * If there is no target buffer they only want the count.
466 	 *
467 	 * This could be TAILQ_FOREACH but it is open-coded to match the original
468 	 * code below.
469 	 */
470 	if (sfsp == NULL) {
471 		mtx_lock(&mountlist_mtx);
472 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
473 			if (prison_canseemount(td->td_ucred, mp) != 0) {
474 				nmp = TAILQ_NEXT(mp, mnt_list);
475 				continue;
476 			}
477 #ifdef MAC
478 			if (mac_mount_check_stat(td->td_ucred, mp) != 0) {
479 				nmp = TAILQ_NEXT(mp, mnt_list);
480 				continue;
481 			}
482 #endif
483 			count++;
484 			nmp = TAILQ_NEXT(mp, mnt_list);
485 		}
486 		mtx_unlock(&mountlist_mtx);
487 		*countp = count;
488 		return (0);
489 	}
490 
491 	/*
492 	 * They want the entire thing.
493 	 *
494 	 * Short-circuit the corner case of no room for anything, avoids
495 	 * relocking below.
496 	 */
497 	if (maxcount < 1) {
498 		goto out;
499 	}
500 
501 	mtx_lock(&mountlist_mtx);
502 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
503 		if (prison_canseemount(td->td_ucred, mp) != 0) {
504 			nmp = TAILQ_NEXT(mp, mnt_list);
505 			continue;
506 		}
507 #ifdef MAC
508 		if (mac_mount_check_stat(td->td_ucred, mp) != 0) {
509 			nmp = TAILQ_NEXT(mp, mnt_list);
510 			continue;
511 		}
512 #endif
513 		if (mode == MNT_WAIT) {
514 			if (vfs_busy(mp, MBF_MNTLSTLOCK) != 0) {
515 				/*
516 				 * If vfs_busy() failed, and MBF_NOWAIT
517 				 * wasn't passed, then the mp is gone.
518 				 * Furthermore, because of MBF_MNTLSTLOCK,
519 				 * the mountlist_mtx was dropped.  We have
520 				 * no other choice than to start over.
521 				 */
522 				mtx_unlock(&mountlist_mtx);
523 				free(tofree, M_STATFS);
524 				goto restart;
525 			}
526 		} else {
527 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
528 				nmp = TAILQ_NEXT(mp, mnt_list);
529 				continue;
530 			}
531 		}
532 		sp = &mp->mnt_stat;
533 		/*
534 		 * If MNT_NOWAIT is specified, do not refresh
535 		 * the fsstat cache.
536 		 */
537 		if (mode != MNT_NOWAIT) {
538 			error = VFS_STATFS(mp, sp);
539 			if (error != 0) {
540 				mtx_lock(&mountlist_mtx);
541 				nmp = TAILQ_NEXT(mp, mnt_list);
542 				vfs_unbusy(mp);
543 				continue;
544 			}
545 		}
546 		if (priv_check_cred_vfs_generation(td->td_ucred)) {
547 			sptmp = malloc(sizeof(struct statfs), M_STATFS,
548 			    M_WAITOK);
549 			*sptmp = *sp;
550 			sptmp->f_fsid.val[0] = sptmp->f_fsid.val[1] = 0;
551 			prison_enforce_statfs(td->td_ucred, mp, sptmp);
552 			sp = sptmp;
553 		} else
554 			sptmp = NULL;
555 		if (bufseg == UIO_SYSSPACE) {
556 			bcopy(sp, sfsp, sizeof(*sp));
557 			free(sptmp, M_STATFS);
558 		} else /* if (bufseg == UIO_USERSPACE) */ {
559 			error = copyout(sp, sfsp, sizeof(*sp));
560 			free(sptmp, M_STATFS);
561 			if (error != 0) {
562 				vfs_unbusy(mp);
563 				return (error);
564 			}
565 		}
566 		sfsp++;
567 		count++;
568 
569 		if (count == maxcount) {
570 			vfs_unbusy(mp);
571 			goto out;
572 		}
573 
574 		mtx_lock(&mountlist_mtx);
575 		nmp = TAILQ_NEXT(mp, mnt_list);
576 		vfs_unbusy(mp);
577 	}
578 	mtx_unlock(&mountlist_mtx);
579 out:
580 	*countp = count;
581 	return (0);
582 }
583 
584 #ifdef COMPAT_FREEBSD4
585 /*
586  * Get old format filesystem statistics.
587  */
588 static void freebsd4_cvtstatfs(struct statfs *, struct ostatfs *);
589 
590 #ifndef _SYS_SYSPROTO_H_
591 struct freebsd4_statfs_args {
592 	char *path;
593 	struct ostatfs *buf;
594 };
595 #endif
596 int
597 freebsd4_statfs(struct thread *td, struct freebsd4_statfs_args *uap)
598 {
599 	struct ostatfs osb;
600 	struct statfs *sfp;
601 	int error;
602 
603 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
604 	error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp);
605 	if (error == 0) {
606 		freebsd4_cvtstatfs(sfp, &osb);
607 		error = copyout(&osb, uap->buf, sizeof(osb));
608 	}
609 	free(sfp, M_STATFS);
610 	return (error);
611 }
612 
613 /*
614  * Get filesystem statistics.
615  */
616 #ifndef _SYS_SYSPROTO_H_
617 struct freebsd4_fstatfs_args {
618 	int fd;
619 	struct ostatfs *buf;
620 };
621 #endif
622 int
623 freebsd4_fstatfs(struct thread *td, struct freebsd4_fstatfs_args *uap)
624 {
625 	struct ostatfs osb;
626 	struct statfs *sfp;
627 	int error;
628 
629 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
630 	error = kern_fstatfs(td, uap->fd, sfp);
631 	if (error == 0) {
632 		freebsd4_cvtstatfs(sfp, &osb);
633 		error = copyout(&osb, uap->buf, sizeof(osb));
634 	}
635 	free(sfp, M_STATFS);
636 	return (error);
637 }
638 
639 /*
640  * Get statistics on all filesystems.
641  */
642 #ifndef _SYS_SYSPROTO_H_
643 struct freebsd4_getfsstat_args {
644 	struct ostatfs *buf;
645 	long bufsize;
646 	int mode;
647 };
648 #endif
649 int
650 freebsd4_getfsstat(struct thread *td, struct freebsd4_getfsstat_args *uap)
651 {
652 	struct statfs *buf, *sp;
653 	struct ostatfs osb;
654 	size_t count, size;
655 	int error;
656 
657 	if (uap->bufsize < 0)
658 		return (EINVAL);
659 	count = uap->bufsize / sizeof(struct ostatfs);
660 	if (count > SIZE_MAX / sizeof(struct statfs))
661 		return (EINVAL);
662 	size = count * sizeof(struct statfs);
663 	error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE,
664 	    uap->mode);
665 	if (error == 0)
666 		td->td_retval[0] = count;
667 	if (size != 0) {
668 		sp = buf;
669 		while (count != 0 && error == 0) {
670 			freebsd4_cvtstatfs(sp, &osb);
671 			error = copyout(&osb, uap->buf, sizeof(osb));
672 			sp++;
673 			uap->buf++;
674 			count--;
675 		}
676 		free(buf, M_STATFS);
677 	}
678 	return (error);
679 }
680 
681 /*
682  * Implement fstatfs() for (NFS) file handles.
683  */
684 #ifndef _SYS_SYSPROTO_H_
685 struct freebsd4_fhstatfs_args {
686 	struct fhandle *u_fhp;
687 	struct ostatfs *buf;
688 };
689 #endif
690 int
691 freebsd4_fhstatfs(struct thread *td, struct freebsd4_fhstatfs_args *uap)
692 {
693 	struct ostatfs osb;
694 	struct statfs *sfp;
695 	fhandle_t fh;
696 	int error;
697 
698 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
699 	if (error != 0)
700 		return (error);
701 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
702 	error = kern_fhstatfs(td, fh, sfp);
703 	if (error == 0) {
704 		freebsd4_cvtstatfs(sfp, &osb);
705 		error = copyout(&osb, uap->buf, sizeof(osb));
706 	}
707 	free(sfp, M_STATFS);
708 	return (error);
709 }
710 
711 /*
712  * Convert a new format statfs structure to an old format statfs structure.
713  */
714 static void
715 freebsd4_cvtstatfs(struct statfs *nsp, struct ostatfs *osp)
716 {
717 
718 	statfs_scale_blocks(nsp, LONG_MAX);
719 	bzero(osp, sizeof(*osp));
720 	osp->f_bsize = nsp->f_bsize;
721 	osp->f_iosize = MIN(nsp->f_iosize, LONG_MAX);
722 	osp->f_blocks = nsp->f_blocks;
723 	osp->f_bfree = nsp->f_bfree;
724 	osp->f_bavail = nsp->f_bavail;
725 	osp->f_files = MIN(nsp->f_files, LONG_MAX);
726 	osp->f_ffree = MIN(nsp->f_ffree, LONG_MAX);
727 	osp->f_owner = nsp->f_owner;
728 	osp->f_type = nsp->f_type;
729 	osp->f_flags = nsp->f_flags;
730 	osp->f_syncwrites = MIN(nsp->f_syncwrites, LONG_MAX);
731 	osp->f_asyncwrites = MIN(nsp->f_asyncwrites, LONG_MAX);
732 	osp->f_syncreads = MIN(nsp->f_syncreads, LONG_MAX);
733 	osp->f_asyncreads = MIN(nsp->f_asyncreads, LONG_MAX);
734 	strlcpy(osp->f_fstypename, nsp->f_fstypename,
735 	    MIN(MFSNAMELEN, OMFSNAMELEN));
736 	strlcpy(osp->f_mntonname, nsp->f_mntonname,
737 	    MIN(MNAMELEN, OMNAMELEN));
738 	strlcpy(osp->f_mntfromname, nsp->f_mntfromname,
739 	    MIN(MNAMELEN, OMNAMELEN));
740 	osp->f_fsid = nsp->f_fsid;
741 }
742 #endif /* COMPAT_FREEBSD4 */
743 
744 #if defined(COMPAT_FREEBSD11)
745 /*
746  * Get old format filesystem statistics.
747  */
748 static void freebsd11_cvtstatfs(struct statfs *, struct freebsd11_statfs *);
749 
750 int
751 freebsd11_statfs(struct thread *td, struct freebsd11_statfs_args *uap)
752 {
753 	struct freebsd11_statfs osb;
754 	struct statfs *sfp;
755 	int error;
756 
757 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
758 	error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp);
759 	if (error == 0) {
760 		freebsd11_cvtstatfs(sfp, &osb);
761 		error = copyout(&osb, uap->buf, sizeof(osb));
762 	}
763 	free(sfp, M_STATFS);
764 	return (error);
765 }
766 
767 /*
768  * Get filesystem statistics.
769  */
770 int
771 freebsd11_fstatfs(struct thread *td, struct freebsd11_fstatfs_args *uap)
772 {
773 	struct freebsd11_statfs osb;
774 	struct statfs *sfp;
775 	int error;
776 
777 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
778 	error = kern_fstatfs(td, uap->fd, sfp);
779 	if (error == 0) {
780 		freebsd11_cvtstatfs(sfp, &osb);
781 		error = copyout(&osb, uap->buf, sizeof(osb));
782 	}
783 	free(sfp, M_STATFS);
784 	return (error);
785 }
786 
787 /*
788  * Get statistics on all filesystems.
789  */
790 int
791 freebsd11_getfsstat(struct thread *td, struct freebsd11_getfsstat_args *uap)
792 {
793 	struct freebsd11_statfs osb;
794 	struct statfs *buf, *sp;
795 	size_t count, size;
796 	int error;
797 
798 	count = uap->bufsize / sizeof(struct ostatfs);
799 	size = count * sizeof(struct statfs);
800 	error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE,
801 	    uap->mode);
802 	if (error == 0)
803 		td->td_retval[0] = count;
804 	if (size > 0) {
805 		sp = buf;
806 		while (count > 0 && error == 0) {
807 			freebsd11_cvtstatfs(sp, &osb);
808 			error = copyout(&osb, uap->buf, sizeof(osb));
809 			sp++;
810 			uap->buf++;
811 			count--;
812 		}
813 		free(buf, M_STATFS);
814 	}
815 	return (error);
816 }
817 
818 /*
819  * Implement fstatfs() for (NFS) file handles.
820  */
821 int
822 freebsd11_fhstatfs(struct thread *td, struct freebsd11_fhstatfs_args *uap)
823 {
824 	struct freebsd11_statfs osb;
825 	struct statfs *sfp;
826 	fhandle_t fh;
827 	int error;
828 
829 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
830 	if (error)
831 		return (error);
832 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
833 	error = kern_fhstatfs(td, fh, sfp);
834 	if (error == 0) {
835 		freebsd11_cvtstatfs(sfp, &osb);
836 		error = copyout(&osb, uap->buf, sizeof(osb));
837 	}
838 	free(sfp, M_STATFS);
839 	return (error);
840 }
841 
842 /*
843  * Convert a new format statfs structure to an old format statfs structure.
844  */
845 static void
846 freebsd11_cvtstatfs(struct statfs *nsp, struct freebsd11_statfs *osp)
847 {
848 
849 	bzero(osp, sizeof(*osp));
850 	osp->f_version = FREEBSD11_STATFS_VERSION;
851 	osp->f_type = nsp->f_type;
852 	osp->f_flags = nsp->f_flags;
853 	osp->f_bsize = nsp->f_bsize;
854 	osp->f_iosize = nsp->f_iosize;
855 	osp->f_blocks = nsp->f_blocks;
856 	osp->f_bfree = nsp->f_bfree;
857 	osp->f_bavail = nsp->f_bavail;
858 	osp->f_files = nsp->f_files;
859 	osp->f_ffree = nsp->f_ffree;
860 	osp->f_syncwrites = nsp->f_syncwrites;
861 	osp->f_asyncwrites = nsp->f_asyncwrites;
862 	osp->f_syncreads = nsp->f_syncreads;
863 	osp->f_asyncreads = nsp->f_asyncreads;
864 	osp->f_namemax = nsp->f_namemax;
865 	osp->f_owner = nsp->f_owner;
866 	osp->f_fsid = nsp->f_fsid;
867 	strlcpy(osp->f_fstypename, nsp->f_fstypename,
868 	    MIN(MFSNAMELEN, sizeof(osp->f_fstypename)));
869 	strlcpy(osp->f_mntonname, nsp->f_mntonname,
870 	    MIN(MNAMELEN, sizeof(osp->f_mntonname)));
871 	strlcpy(osp->f_mntfromname, nsp->f_mntfromname,
872 	    MIN(MNAMELEN, sizeof(osp->f_mntfromname)));
873 }
874 #endif /* COMPAT_FREEBSD11 */
875 
876 /*
877  * Change current working directory to a given file descriptor.
878  */
879 #ifndef _SYS_SYSPROTO_H_
880 struct fchdir_args {
881 	int	fd;
882 };
883 #endif
884 int
885 sys_fchdir(struct thread *td, struct fchdir_args *uap)
886 {
887 	struct vnode *vp, *tdp;
888 	struct mount *mp;
889 	struct file *fp;
890 	int error;
891 
892 	AUDIT_ARG_FD(uap->fd);
893 	error = getvnode_path(td, uap->fd, &cap_fchdir_rights,
894 	    &fp);
895 	if (error != 0)
896 		return (error);
897 	vp = fp->f_vnode;
898 	vref(vp);
899 	fdrop(fp, td);
900 	vn_lock(vp, LK_SHARED | LK_RETRY);
901 	AUDIT_ARG_VNODE1(vp);
902 	error = change_dir(vp, td);
903 	while (!error && (mp = vp->v_mountedhere) != NULL) {
904 		if (vfs_busy(mp, 0))
905 			continue;
906 		error = VFS_ROOT(mp, LK_SHARED, &tdp);
907 		vfs_unbusy(mp);
908 		if (error != 0)
909 			break;
910 		vput(vp);
911 		vp = tdp;
912 	}
913 	if (error != 0) {
914 		vput(vp);
915 		return (error);
916 	}
917 	VOP_UNLOCK(vp);
918 	pwd_chdir(td, vp);
919 	return (0);
920 }
921 
922 /*
923  * Change current working directory (``.'').
924  */
925 #ifndef _SYS_SYSPROTO_H_
926 struct chdir_args {
927 	char	*path;
928 };
929 #endif
930 int
931 sys_chdir(struct thread *td, struct chdir_args *uap)
932 {
933 
934 	return (kern_chdir(td, uap->path, UIO_USERSPACE));
935 }
936 
937 int
938 kern_chdir(struct thread *td, const char *path, enum uio_seg pathseg)
939 {
940 	struct nameidata nd;
941 	int error;
942 
943 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1,
944 	    pathseg, path, td);
945 	if ((error = namei(&nd)) != 0)
946 		return (error);
947 	if ((error = change_dir(nd.ni_vp, td)) != 0) {
948 		vput(nd.ni_vp);
949 		NDFREE_NOTHING(&nd);
950 		return (error);
951 	}
952 	VOP_UNLOCK(nd.ni_vp);
953 	NDFREE_NOTHING(&nd);
954 	pwd_chdir(td, nd.ni_vp);
955 	return (0);
956 }
957 
958 static int unprivileged_chroot = 0;
959 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_chroot, CTLFLAG_RW,
960     &unprivileged_chroot, 0,
961     "Unprivileged processes can use chroot(2)");
962 /*
963  * Change notion of root (``/'') directory.
964  */
965 #ifndef _SYS_SYSPROTO_H_
966 struct chroot_args {
967 	char	*path;
968 };
969 #endif
970 int
971 sys_chroot(struct thread *td, struct chroot_args *uap)
972 {
973 	struct nameidata nd;
974 	struct proc *p;
975 	int error;
976 
977 	error = priv_check(td, PRIV_VFS_CHROOT);
978 	if (error != 0) {
979 		p = td->td_proc;
980 		PROC_LOCK(p);
981 		if (unprivileged_chroot == 0 ||
982 		    (p->p_flag2 & P2_NO_NEW_PRIVS) == 0) {
983 			PROC_UNLOCK(p);
984 			return (error);
985 		}
986 		PROC_UNLOCK(p);
987 	}
988 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1,
989 	    UIO_USERSPACE, uap->path, td);
990 	error = namei(&nd);
991 	if (error != 0)
992 		goto error;
993 	error = change_dir(nd.ni_vp, td);
994 	if (error != 0)
995 		goto e_vunlock;
996 #ifdef MAC
997 	error = mac_vnode_check_chroot(td->td_ucred, nd.ni_vp);
998 	if (error != 0)
999 		goto e_vunlock;
1000 #endif
1001 	VOP_UNLOCK(nd.ni_vp);
1002 	error = pwd_chroot(td, nd.ni_vp);
1003 	vrele(nd.ni_vp);
1004 	NDFREE_NOTHING(&nd);
1005 	return (error);
1006 e_vunlock:
1007 	vput(nd.ni_vp);
1008 error:
1009 	NDFREE_NOTHING(&nd);
1010 	return (error);
1011 }
1012 
1013 /*
1014  * Common routine for chroot and chdir.  Callers must provide a locked vnode
1015  * instance.
1016  */
1017 int
1018 change_dir(struct vnode *vp, struct thread *td)
1019 {
1020 #ifdef MAC
1021 	int error;
1022 #endif
1023 
1024 	ASSERT_VOP_LOCKED(vp, "change_dir(): vp not locked");
1025 	if (vp->v_type != VDIR)
1026 		return (ENOTDIR);
1027 #ifdef MAC
1028 	error = mac_vnode_check_chdir(td->td_ucred, vp);
1029 	if (error != 0)
1030 		return (error);
1031 #endif
1032 	return (VOP_ACCESS(vp, VEXEC, td->td_ucred, td));
1033 }
1034 
1035 static __inline void
1036 flags_to_rights(int flags, cap_rights_t *rightsp)
1037 {
1038 	if (flags & O_EXEC) {
1039 		cap_rights_set_one(rightsp, CAP_FEXECVE);
1040 		if (flags & O_PATH)
1041 			return;
1042 	} else {
1043 		switch ((flags & O_ACCMODE)) {
1044 		case O_RDONLY:
1045 			cap_rights_set_one(rightsp, CAP_READ);
1046 			break;
1047 		case O_RDWR:
1048 			cap_rights_set_one(rightsp, CAP_READ);
1049 			/* FALLTHROUGH */
1050 		case O_WRONLY:
1051 			cap_rights_set_one(rightsp, CAP_WRITE);
1052 			if (!(flags & (O_APPEND | O_TRUNC)))
1053 				cap_rights_set_one(rightsp, CAP_SEEK);
1054 			break;
1055 		}
1056 	}
1057 
1058 	if (flags & O_CREAT)
1059 		cap_rights_set_one(rightsp, CAP_CREATE);
1060 
1061 	if (flags & O_TRUNC)
1062 		cap_rights_set_one(rightsp, CAP_FTRUNCATE);
1063 
1064 	if (flags & (O_SYNC | O_FSYNC))
1065 		cap_rights_set_one(rightsp, CAP_FSYNC);
1066 
1067 	if (flags & (O_EXLOCK | O_SHLOCK))
1068 		cap_rights_set_one(rightsp, CAP_FLOCK);
1069 }
1070 
1071 /*
1072  * Check permissions, allocate an open file structure, and call the device
1073  * open routine if any.
1074  */
1075 #ifndef _SYS_SYSPROTO_H_
1076 struct open_args {
1077 	char	*path;
1078 	int	flags;
1079 	int	mode;
1080 };
1081 #endif
1082 int
1083 sys_open(struct thread *td, struct open_args *uap)
1084 {
1085 
1086 	return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1087 	    uap->flags, uap->mode));
1088 }
1089 
1090 #ifndef _SYS_SYSPROTO_H_
1091 struct openat_args {
1092 	int	fd;
1093 	char	*path;
1094 	int	flag;
1095 	int	mode;
1096 };
1097 #endif
1098 int
1099 sys_openat(struct thread *td, struct openat_args *uap)
1100 {
1101 
1102 	AUDIT_ARG_FD(uap->fd);
1103 	return (kern_openat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag,
1104 	    uap->mode));
1105 }
1106 
1107 int
1108 kern_openat(struct thread *td, int fd, const char *path, enum uio_seg pathseg,
1109     int flags, int mode)
1110 {
1111 	struct proc *p = td->td_proc;
1112 	struct filedesc *fdp;
1113 	struct pwddesc *pdp;
1114 	struct file *fp;
1115 	struct vnode *vp;
1116 	struct nameidata nd;
1117 	cap_rights_t rights;
1118 	int cmode, error, indx;
1119 
1120 	indx = -1;
1121 	fdp = p->p_fd;
1122 	pdp = p->p_pd;
1123 
1124 	AUDIT_ARG_FFLAGS(flags);
1125 	AUDIT_ARG_MODE(mode);
1126 	cap_rights_init_one(&rights, CAP_LOOKUP);
1127 	flags_to_rights(flags, &rights);
1128 
1129 	/*
1130 	 * Only one of the O_EXEC, O_RDONLY, O_WRONLY and O_RDWR flags
1131 	 * may be specified.  On the other hand, for O_PATH any mode
1132 	 * except O_EXEC is ignored.
1133 	 */
1134 	if ((flags & O_PATH) != 0) {
1135 		flags &= ~(O_CREAT | O_ACCMODE);
1136 	} else if ((flags & O_EXEC) != 0) {
1137 		if (flags & O_ACCMODE)
1138 			return (EINVAL);
1139 	} else if ((flags & O_ACCMODE) == O_ACCMODE) {
1140 		return (EINVAL);
1141 	} else {
1142 		flags = FFLAGS(flags);
1143 	}
1144 
1145 	/*
1146 	 * Allocate a file structure. The descriptor to reference it
1147 	 * is allocated and used by finstall_refed() below.
1148 	 */
1149 	error = falloc_noinstall(td, &fp);
1150 	if (error != 0)
1151 		return (error);
1152 	/* Set the flags early so the finit in devfs can pick them up. */
1153 	fp->f_flag = flags & FMASK;
1154 	cmode = ((mode & ~pdp->pd_cmask) & ALLPERMS) & ~S_ISTXT;
1155 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd,
1156 	    &rights, td);
1157 	td->td_dupfd = -1;		/* XXX check for fdopen */
1158 	error = vn_open(&nd, &flags, cmode, fp);
1159 	if (error != 0) {
1160 		/*
1161 		 * If the vn_open replaced the method vector, something
1162 		 * wonderous happened deep below and we just pass it up
1163 		 * pretending we know what we do.
1164 		 */
1165 		if (error == ENXIO && fp->f_ops != &badfileops) {
1166 			MPASS((flags & O_PATH) == 0);
1167 			goto success;
1168 		}
1169 
1170 		/*
1171 		 * Handle special fdopen() case. bleh.
1172 		 *
1173 		 * Don't do this for relative (capability) lookups; we don't
1174 		 * understand exactly what would happen, and we don't think
1175 		 * that it ever should.
1176 		 */
1177 		if ((nd.ni_resflags & NIRES_STRICTREL) == 0 &&
1178 		    (error == ENODEV || error == ENXIO) &&
1179 		    td->td_dupfd >= 0) {
1180 			error = dupfdopen(td, fdp, td->td_dupfd, flags, error,
1181 			    &indx);
1182 			if (error == 0)
1183 				goto success;
1184 		}
1185 
1186 		goto bad;
1187 	}
1188 	td->td_dupfd = 0;
1189 	NDFREE(&nd, NDF_ONLY_PNBUF);
1190 	vp = nd.ni_vp;
1191 
1192 	/*
1193 	 * Store the vnode, for any f_type. Typically, the vnode use
1194 	 * count is decremented by direct call to vn_closefile() for
1195 	 * files that switched type in the cdevsw fdopen() method.
1196 	 */
1197 	fp->f_vnode = vp;
1198 
1199 	/*
1200 	 * If the file wasn't claimed by devfs bind it to the normal
1201 	 * vnode operations here.
1202 	 */
1203 	if (fp->f_ops == &badfileops) {
1204 		KASSERT(vp->v_type != VFIFO || (flags & O_PATH) != 0,
1205 		    ("Unexpected fifo fp %p vp %p", fp, vp));
1206 		if ((flags & O_PATH) != 0) {
1207 			finit(fp, (flags & FMASK) | (fp->f_flag & FKQALLOWED),
1208 			    DTYPE_VNODE, NULL, &path_fileops);
1209 			vhold(vp);
1210 			vunref(vp);
1211 		} else {
1212 			finit_vnode(fp, flags, NULL, &vnops);
1213 		}
1214 	}
1215 
1216 	VOP_UNLOCK(vp);
1217 	if (flags & O_TRUNC) {
1218 		error = fo_truncate(fp, 0, td->td_ucred, td);
1219 		if (error != 0)
1220 			goto bad;
1221 	}
1222 success:
1223 	/*
1224 	 * If we haven't already installed the FD (for dupfdopen), do so now.
1225 	 */
1226 	if (indx == -1) {
1227 		struct filecaps *fcaps;
1228 
1229 #ifdef CAPABILITIES
1230 		if ((nd.ni_resflags & NIRES_STRICTREL) != 0)
1231 			fcaps = &nd.ni_filecaps;
1232 		else
1233 #endif
1234 			fcaps = NULL;
1235 		error = finstall_refed(td, fp, &indx, flags, fcaps);
1236 		/* On success finstall_refed() consumes fcaps. */
1237 		if (error != 0) {
1238 			filecaps_free(&nd.ni_filecaps);
1239 			goto bad;
1240 		}
1241 	} else {
1242 		filecaps_free(&nd.ni_filecaps);
1243 		falloc_abort(td, fp);
1244 	}
1245 
1246 	td->td_retval[0] = indx;
1247 	return (0);
1248 bad:
1249 	KASSERT(indx == -1, ("indx=%d, should be -1", indx));
1250 	falloc_abort(td, fp);
1251 	return (error);
1252 }
1253 
1254 #ifdef COMPAT_43
1255 /*
1256  * Create a file.
1257  */
1258 #ifndef _SYS_SYSPROTO_H_
1259 struct ocreat_args {
1260 	char	*path;
1261 	int	mode;
1262 };
1263 #endif
1264 int
1265 ocreat(struct thread *td, struct ocreat_args *uap)
1266 {
1267 
1268 	return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1269 	    O_WRONLY | O_CREAT | O_TRUNC, uap->mode));
1270 }
1271 #endif /* COMPAT_43 */
1272 
1273 /*
1274  * Create a special file.
1275  */
1276 #ifndef _SYS_SYSPROTO_H_
1277 struct mknodat_args {
1278 	int	fd;
1279 	char	*path;
1280 	mode_t	mode;
1281 	dev_t	dev;
1282 };
1283 #endif
1284 int
1285 sys_mknodat(struct thread *td, struct mknodat_args *uap)
1286 {
1287 
1288 	return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode,
1289 	    uap->dev));
1290 }
1291 
1292 #if defined(COMPAT_FREEBSD11)
1293 int
1294 freebsd11_mknod(struct thread *td,
1295     struct freebsd11_mknod_args *uap)
1296 {
1297 
1298 	return (kern_mknodat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1299 	    uap->mode, uap->dev));
1300 }
1301 
1302 int
1303 freebsd11_mknodat(struct thread *td,
1304     struct freebsd11_mknodat_args *uap)
1305 {
1306 
1307 	return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode,
1308 	    uap->dev));
1309 }
1310 #endif /* COMPAT_FREEBSD11 */
1311 
1312 int
1313 kern_mknodat(struct thread *td, int fd, const char *path, enum uio_seg pathseg,
1314     int mode, dev_t dev)
1315 {
1316 	struct vnode *vp;
1317 	struct mount *mp;
1318 	struct vattr vattr;
1319 	struct nameidata nd;
1320 	int error, whiteout = 0;
1321 
1322 	AUDIT_ARG_MODE(mode);
1323 	AUDIT_ARG_DEV(dev);
1324 	switch (mode & S_IFMT) {
1325 	case S_IFCHR:
1326 	case S_IFBLK:
1327 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1328 		if (error == 0 && dev == VNOVAL)
1329 			error = EINVAL;
1330 		break;
1331 	case S_IFWHT:
1332 		error = priv_check(td, PRIV_VFS_MKNOD_WHT);
1333 		break;
1334 	case S_IFIFO:
1335 		if (dev == 0)
1336 			return (kern_mkfifoat(td, fd, path, pathseg, mode));
1337 		/* FALLTHROUGH */
1338 	default:
1339 		error = EINVAL;
1340 		break;
1341 	}
1342 	if (error != 0)
1343 		return (error);
1344 	NDPREINIT(&nd);
1345 restart:
1346 	bwillwrite();
1347 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
1348 	    NOCACHE, pathseg, path, fd, &cap_mknodat_rights,
1349 	    td);
1350 	if ((error = namei(&nd)) != 0)
1351 		return (error);
1352 	vp = nd.ni_vp;
1353 	if (vp != NULL) {
1354 		NDFREE(&nd, NDF_ONLY_PNBUF);
1355 		if (vp == nd.ni_dvp)
1356 			vrele(nd.ni_dvp);
1357 		else
1358 			vput(nd.ni_dvp);
1359 		vrele(vp);
1360 		return (EEXIST);
1361 	} else {
1362 		VATTR_NULL(&vattr);
1363 		vattr.va_mode = (mode & ALLPERMS) &
1364 		    ~td->td_proc->p_pd->pd_cmask;
1365 		vattr.va_rdev = dev;
1366 		whiteout = 0;
1367 
1368 		switch (mode & S_IFMT) {
1369 		case S_IFCHR:
1370 			vattr.va_type = VCHR;
1371 			break;
1372 		case S_IFBLK:
1373 			vattr.va_type = VBLK;
1374 			break;
1375 		case S_IFWHT:
1376 			whiteout = 1;
1377 			break;
1378 		default:
1379 			panic("kern_mknod: invalid mode");
1380 		}
1381 	}
1382 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1383 		NDFREE(&nd, NDF_ONLY_PNBUF);
1384 		vput(nd.ni_dvp);
1385 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1386 			return (error);
1387 		goto restart;
1388 	}
1389 #ifdef MAC
1390 	if (error == 0 && !whiteout)
1391 		error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp,
1392 		    &nd.ni_cnd, &vattr);
1393 #endif
1394 	if (error == 0) {
1395 		if (whiteout)
1396 			error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, CREATE);
1397 		else {
1398 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp,
1399 						&nd.ni_cnd, &vattr);
1400 		}
1401 	}
1402 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 && !whiteout ? &nd.ni_vp : NULL,
1403 	    true);
1404 	vn_finished_write(mp);
1405 	NDFREE(&nd, NDF_ONLY_PNBUF);
1406 	if (error == ERELOOKUP)
1407 		goto restart;
1408 	return (error);
1409 }
1410 
1411 /*
1412  * Create a named pipe.
1413  */
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct mkfifo_args {
1416 	char	*path;
1417 	int	mode;
1418 };
1419 #endif
1420 int
1421 sys_mkfifo(struct thread *td, struct mkfifo_args *uap)
1422 {
1423 
1424 	return (kern_mkfifoat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1425 	    uap->mode));
1426 }
1427 
1428 #ifndef _SYS_SYSPROTO_H_
1429 struct mkfifoat_args {
1430 	int	fd;
1431 	char	*path;
1432 	mode_t	mode;
1433 };
1434 #endif
1435 int
1436 sys_mkfifoat(struct thread *td, struct mkfifoat_args *uap)
1437 {
1438 
1439 	return (kern_mkfifoat(td, uap->fd, uap->path, UIO_USERSPACE,
1440 	    uap->mode));
1441 }
1442 
1443 int
1444 kern_mkfifoat(struct thread *td, int fd, const char *path,
1445     enum uio_seg pathseg, int mode)
1446 {
1447 	struct mount *mp;
1448 	struct vattr vattr;
1449 	struct nameidata nd;
1450 	int error;
1451 
1452 	AUDIT_ARG_MODE(mode);
1453 	NDPREINIT(&nd);
1454 restart:
1455 	bwillwrite();
1456 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
1457 	    NOCACHE, pathseg, path, fd, &cap_mkfifoat_rights,
1458 	    td);
1459 	if ((error = namei(&nd)) != 0)
1460 		return (error);
1461 	if (nd.ni_vp != NULL) {
1462 		NDFREE(&nd, NDF_ONLY_PNBUF);
1463 		if (nd.ni_vp == nd.ni_dvp)
1464 			vrele(nd.ni_dvp);
1465 		else
1466 			vput(nd.ni_dvp);
1467 		vrele(nd.ni_vp);
1468 		return (EEXIST);
1469 	}
1470 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1471 		NDFREE(&nd, NDF_ONLY_PNBUF);
1472 		vput(nd.ni_dvp);
1473 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1474 			return (error);
1475 		goto restart;
1476 	}
1477 	VATTR_NULL(&vattr);
1478 	vattr.va_type = VFIFO;
1479 	vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_pd->pd_cmask;
1480 #ifdef MAC
1481 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
1482 	    &vattr);
1483 	if (error != 0)
1484 		goto out;
1485 #endif
1486 	error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
1487 #ifdef MAC
1488 out:
1489 #endif
1490 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true);
1491 	vn_finished_write(mp);
1492 	NDFREE(&nd, NDF_ONLY_PNBUF);
1493 	if (error == ERELOOKUP)
1494 		goto restart;
1495 	return (error);
1496 }
1497 
1498 /*
1499  * Make a hard file link.
1500  */
1501 #ifndef _SYS_SYSPROTO_H_
1502 struct link_args {
1503 	char	*path;
1504 	char	*link;
1505 };
1506 #endif
1507 int
1508 sys_link(struct thread *td, struct link_args *uap)
1509 {
1510 
1511 	return (kern_linkat(td, AT_FDCWD, AT_FDCWD, uap->path, uap->link,
1512 	    UIO_USERSPACE, AT_SYMLINK_FOLLOW));
1513 }
1514 
1515 #ifndef _SYS_SYSPROTO_H_
1516 struct linkat_args {
1517 	int	fd1;
1518 	char	*path1;
1519 	int	fd2;
1520 	char	*path2;
1521 	int	flag;
1522 };
1523 #endif
1524 int
1525 sys_linkat(struct thread *td, struct linkat_args *uap)
1526 {
1527 	int flag;
1528 
1529 	flag = uap->flag;
1530 	if ((flag & ~(AT_SYMLINK_FOLLOW | AT_RESOLVE_BENEATH |
1531 	    AT_EMPTY_PATH)) != 0)
1532 		return (EINVAL);
1533 
1534 	return (kern_linkat(td, uap->fd1, uap->fd2, uap->path1, uap->path2,
1535 	    UIO_USERSPACE, flag));
1536 }
1537 
1538 int hardlink_check_uid = 0;
1539 SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1540     &hardlink_check_uid, 0,
1541     "Unprivileged processes cannot create hard links to files owned by other "
1542     "users");
1543 static int hardlink_check_gid = 0;
1544 SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1545     &hardlink_check_gid, 0,
1546     "Unprivileged processes cannot create hard links to files owned by other "
1547     "groups");
1548 
1549 static int
1550 can_hardlink(struct vnode *vp, struct ucred *cred)
1551 {
1552 	struct vattr va;
1553 	int error;
1554 
1555 	if (!hardlink_check_uid && !hardlink_check_gid)
1556 		return (0);
1557 
1558 	error = VOP_GETATTR(vp, &va, cred);
1559 	if (error != 0)
1560 		return (error);
1561 
1562 	if (hardlink_check_uid && cred->cr_uid != va.va_uid) {
1563 		error = priv_check_cred(cred, PRIV_VFS_LINK);
1564 		if (error != 0)
1565 			return (error);
1566 	}
1567 
1568 	if (hardlink_check_gid && !groupmember(va.va_gid, cred)) {
1569 		error = priv_check_cred(cred, PRIV_VFS_LINK);
1570 		if (error != 0)
1571 			return (error);
1572 	}
1573 
1574 	return (0);
1575 }
1576 
1577 int
1578 kern_linkat(struct thread *td, int fd1, int fd2, const char *path1,
1579     const char *path2, enum uio_seg segflag, int flag)
1580 {
1581 	struct nameidata nd;
1582 	int error;
1583 
1584 	NDPREINIT(&nd);
1585 	do {
1586 		bwillwrite();
1587 		NDINIT_ATRIGHTS(&nd, LOOKUP, AUDITVNODE1 | at2cnpflags(flag,
1588 		    AT_SYMLINK_FOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH),
1589 		    segflag, path1, fd1, &cap_linkat_source_rights, td);
1590 		if ((error = namei(&nd)) != 0)
1591 			return (error);
1592 		NDFREE(&nd, NDF_ONLY_PNBUF);
1593 		if ((nd.ni_resflags & NIRES_EMPTYPATH) != 0) {
1594 			error = priv_check(td, PRIV_VFS_FHOPEN);
1595 			if (error != 0) {
1596 				vrele(nd.ni_vp);
1597 				return (error);
1598 			}
1599 		}
1600 		error = kern_linkat_vp(td, nd.ni_vp, fd2, path2, segflag);
1601 	} while (error ==  EAGAIN || error == ERELOOKUP);
1602 	return (error);
1603 }
1604 
1605 static int
1606 kern_linkat_vp(struct thread *td, struct vnode *vp, int fd, const char *path,
1607     enum uio_seg segflag)
1608 {
1609 	struct nameidata nd;
1610 	struct mount *mp;
1611 	int error;
1612 
1613 	if (vp->v_type == VDIR) {
1614 		vrele(vp);
1615 		return (EPERM);		/* POSIX */
1616 	}
1617 	NDINIT_ATRIGHTS(&nd, CREATE,
1618 	    LOCKPARENT | SAVENAME | AUDITVNODE2 | NOCACHE, segflag, path, fd,
1619 	    &cap_linkat_target_rights, td);
1620 	if ((error = namei(&nd)) == 0) {
1621 		if (nd.ni_vp != NULL) {
1622 			NDFREE(&nd, NDF_ONLY_PNBUF);
1623 			if (nd.ni_dvp == nd.ni_vp)
1624 				vrele(nd.ni_dvp);
1625 			else
1626 				vput(nd.ni_dvp);
1627 			vrele(nd.ni_vp);
1628 			vrele(vp);
1629 			return (EEXIST);
1630 		} else if (nd.ni_dvp->v_mount != vp->v_mount) {
1631 			/*
1632 			 * Cross-device link.  No need to recheck
1633 			 * vp->v_type, since it cannot change, except
1634 			 * to VBAD.
1635 			 */
1636 			NDFREE(&nd, NDF_ONLY_PNBUF);
1637 			vput(nd.ni_dvp);
1638 			vrele(vp);
1639 			return (EXDEV);
1640 		} else if ((error = vn_lock(vp, LK_EXCLUSIVE)) == 0) {
1641 			error = can_hardlink(vp, td->td_ucred);
1642 #ifdef MAC
1643 			if (error == 0)
1644 				error = mac_vnode_check_link(td->td_ucred,
1645 				    nd.ni_dvp, vp, &nd.ni_cnd);
1646 #endif
1647 			if (error != 0) {
1648 				vput(vp);
1649 				vput(nd.ni_dvp);
1650 				NDFREE(&nd, NDF_ONLY_PNBUF);
1651 				return (error);
1652 			}
1653 			error = vn_start_write(vp, &mp, V_NOWAIT);
1654 			if (error != 0) {
1655 				vput(vp);
1656 				vput(nd.ni_dvp);
1657 				NDFREE(&nd, NDF_ONLY_PNBUF);
1658 				error = vn_start_write(NULL, &mp,
1659 				    V_XSLEEP | PCATCH);
1660 				if (error != 0)
1661 					return (error);
1662 				return (EAGAIN);
1663 			}
1664 			error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
1665 			VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
1666 			vn_finished_write(mp);
1667 			NDFREE(&nd, NDF_ONLY_PNBUF);
1668 			vp = NULL;
1669 		} else {
1670 			vput(nd.ni_dvp);
1671 			NDFREE(&nd, NDF_ONLY_PNBUF);
1672 			vrele(vp);
1673 			return (EAGAIN);
1674 		}
1675 	}
1676 	if (vp != NULL)
1677 		vrele(vp);
1678 	return (error);
1679 }
1680 
1681 /*
1682  * Make a symbolic link.
1683  */
1684 #ifndef _SYS_SYSPROTO_H_
1685 struct symlink_args {
1686 	char	*path;
1687 	char	*link;
1688 };
1689 #endif
1690 int
1691 sys_symlink(struct thread *td, struct symlink_args *uap)
1692 {
1693 
1694 	return (kern_symlinkat(td, uap->path, AT_FDCWD, uap->link,
1695 	    UIO_USERSPACE));
1696 }
1697 
1698 #ifndef _SYS_SYSPROTO_H_
1699 struct symlinkat_args {
1700 	char	*path;
1701 	int	fd;
1702 	char	*path2;
1703 };
1704 #endif
1705 int
1706 sys_symlinkat(struct thread *td, struct symlinkat_args *uap)
1707 {
1708 
1709 	return (kern_symlinkat(td, uap->path1, uap->fd, uap->path2,
1710 	    UIO_USERSPACE));
1711 }
1712 
1713 int
1714 kern_symlinkat(struct thread *td, const char *path1, int fd, const char *path2,
1715     enum uio_seg segflg)
1716 {
1717 	struct mount *mp;
1718 	struct vattr vattr;
1719 	const char *syspath;
1720 	char *tmppath;
1721 	struct nameidata nd;
1722 	int error;
1723 
1724 	if (segflg == UIO_SYSSPACE) {
1725 		syspath = path1;
1726 	} else {
1727 		tmppath = uma_zalloc(namei_zone, M_WAITOK);
1728 		if ((error = copyinstr(path1, tmppath, MAXPATHLEN, NULL)) != 0)
1729 			goto out;
1730 		syspath = tmppath;
1731 	}
1732 	AUDIT_ARG_TEXT(syspath);
1733 	NDPREINIT(&nd);
1734 restart:
1735 	bwillwrite();
1736 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
1737 	    NOCACHE, segflg, path2, fd, &cap_symlinkat_rights,
1738 	    td);
1739 	if ((error = namei(&nd)) != 0)
1740 		goto out;
1741 	if (nd.ni_vp) {
1742 		NDFREE(&nd, NDF_ONLY_PNBUF);
1743 		if (nd.ni_vp == nd.ni_dvp)
1744 			vrele(nd.ni_dvp);
1745 		else
1746 			vput(nd.ni_dvp);
1747 		vrele(nd.ni_vp);
1748 		nd.ni_vp = NULL;
1749 		error = EEXIST;
1750 		goto out;
1751 	}
1752 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1753 		NDFREE(&nd, NDF_ONLY_PNBUF);
1754 		vput(nd.ni_dvp);
1755 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1756 			goto out;
1757 		goto restart;
1758 	}
1759 	VATTR_NULL(&vattr);
1760 	vattr.va_mode = ACCESSPERMS &~ td->td_proc->p_pd->pd_cmask;
1761 #ifdef MAC
1762 	vattr.va_type = VLNK;
1763 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
1764 	    &vattr);
1765 	if (error != 0)
1766 		goto out2;
1767 #endif
1768 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr, syspath);
1769 #ifdef MAC
1770 out2:
1771 #endif
1772 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true);
1773 	vn_finished_write(mp);
1774 	NDFREE(&nd, NDF_ONLY_PNBUF);
1775 	if (error == ERELOOKUP)
1776 		goto restart;
1777 out:
1778 	if (segflg != UIO_SYSSPACE)
1779 		uma_zfree(namei_zone, tmppath);
1780 	return (error);
1781 }
1782 
1783 /*
1784  * Delete a whiteout from the filesystem.
1785  */
1786 #ifndef _SYS_SYSPROTO_H_
1787 struct undelete_args {
1788 	char *path;
1789 };
1790 #endif
1791 int
1792 sys_undelete(struct thread *td, struct undelete_args *uap)
1793 {
1794 	struct mount *mp;
1795 	struct nameidata nd;
1796 	int error;
1797 
1798 	NDPREINIT(&nd);
1799 restart:
1800 	bwillwrite();
1801 	NDINIT(&nd, DELETE, LOCKPARENT | DOWHITEOUT | AUDITVNODE1,
1802 	    UIO_USERSPACE, uap->path, td);
1803 	error = namei(&nd);
1804 	if (error != 0)
1805 		return (error);
1806 
1807 	if (nd.ni_vp != NULLVP || !(nd.ni_cnd.cn_flags & ISWHITEOUT)) {
1808 		NDFREE(&nd, NDF_ONLY_PNBUF);
1809 		if (nd.ni_vp == nd.ni_dvp)
1810 			vrele(nd.ni_dvp);
1811 		else
1812 			vput(nd.ni_dvp);
1813 		if (nd.ni_vp)
1814 			vrele(nd.ni_vp);
1815 		return (EEXIST);
1816 	}
1817 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1818 		NDFREE(&nd, NDF_ONLY_PNBUF);
1819 		vput(nd.ni_dvp);
1820 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1821 			return (error);
1822 		goto restart;
1823 	}
1824 	error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, DELETE);
1825 	NDFREE(&nd, NDF_ONLY_PNBUF);
1826 	vput(nd.ni_dvp);
1827 	vn_finished_write(mp);
1828 	if (error == ERELOOKUP)
1829 		goto restart;
1830 	return (error);
1831 }
1832 
1833 /*
1834  * Delete a name from the filesystem.
1835  */
1836 #ifndef _SYS_SYSPROTO_H_
1837 struct unlink_args {
1838 	char	*path;
1839 };
1840 #endif
1841 int
1842 sys_unlink(struct thread *td, struct unlink_args *uap)
1843 {
1844 
1845 	return (kern_funlinkat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE,
1846 	    0, 0));
1847 }
1848 
1849 static int
1850 kern_funlinkat_ex(struct thread *td, int dfd, const char *path, int fd,
1851     int flag, enum uio_seg pathseg, ino_t oldinum)
1852 {
1853 
1854 	if ((flag & ~(AT_REMOVEDIR | AT_RESOLVE_BENEATH)) != 0)
1855 		return (EINVAL);
1856 
1857 	if ((flag & AT_REMOVEDIR) != 0)
1858 		return (kern_frmdirat(td, dfd, path, fd, UIO_USERSPACE, 0));
1859 
1860 	return (kern_funlinkat(td, dfd, path, fd, UIO_USERSPACE, 0, 0));
1861 }
1862 
1863 #ifndef _SYS_SYSPROTO_H_
1864 struct unlinkat_args {
1865 	int	fd;
1866 	char	*path;
1867 	int	flag;
1868 };
1869 #endif
1870 int
1871 sys_unlinkat(struct thread *td, struct unlinkat_args *uap)
1872 {
1873 
1874 	return (kern_funlinkat_ex(td, uap->fd, uap->path, FD_NONE, uap->flag,
1875 	    UIO_USERSPACE, 0));
1876 }
1877 
1878 #ifndef _SYS_SYSPROTO_H_
1879 struct funlinkat_args {
1880 	int		dfd;
1881 	const char	*path;
1882 	int		fd;
1883 	int		flag;
1884 };
1885 #endif
1886 int
1887 sys_funlinkat(struct thread *td, struct funlinkat_args *uap)
1888 {
1889 
1890 	return (kern_funlinkat_ex(td, uap->dfd, uap->path, uap->fd, uap->flag,
1891 	    UIO_USERSPACE, 0));
1892 }
1893 
1894 int
1895 kern_funlinkat(struct thread *td, int dfd, const char *path, int fd,
1896     enum uio_seg pathseg, int flag, ino_t oldinum)
1897 {
1898 	struct mount *mp;
1899 	struct file *fp;
1900 	struct vnode *vp;
1901 	struct nameidata nd;
1902 	struct stat sb;
1903 	int error;
1904 
1905 	fp = NULL;
1906 	if (fd != FD_NONE) {
1907 		error = getvnode_path(td, fd, &cap_no_rights, &fp);
1908 		if (error != 0)
1909 			return (error);
1910 	}
1911 
1912 	NDPREINIT(&nd);
1913 restart:
1914 	bwillwrite();
1915 	NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 |
1916 	    at2cnpflags(flag, AT_RESOLVE_BENEATH),
1917 	    pathseg, path, dfd, &cap_unlinkat_rights, td);
1918 	if ((error = namei(&nd)) != 0) {
1919 		if (error == EINVAL)
1920 			error = EPERM;
1921 		goto fdout;
1922 	}
1923 	vp = nd.ni_vp;
1924 	if (vp->v_type == VDIR && oldinum == 0) {
1925 		error = EPERM;		/* POSIX */
1926 	} else if (oldinum != 0 &&
1927 	    ((error = VOP_STAT(vp, &sb, td->td_ucred, NOCRED, td)) == 0) &&
1928 	    sb.st_ino != oldinum) {
1929 		error = EIDRM;	/* Identifier removed */
1930 	} else if (fp != NULL && fp->f_vnode != vp) {
1931 		if (VN_IS_DOOMED(fp->f_vnode))
1932 			error = EBADF;
1933 		else
1934 			error = EDEADLK;
1935 	} else {
1936 		/*
1937 		 * The root of a mounted filesystem cannot be deleted.
1938 		 *
1939 		 * XXX: can this only be a VDIR case?
1940 		 */
1941 		if (vp->v_vflag & VV_ROOT)
1942 			error = EBUSY;
1943 	}
1944 	if (error == 0) {
1945 		if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1946 			NDFREE(&nd, NDF_ONLY_PNBUF);
1947 			vput(nd.ni_dvp);
1948 			if (vp == nd.ni_dvp)
1949 				vrele(vp);
1950 			else
1951 				vput(vp);
1952 			if ((error = vn_start_write(NULL, &mp,
1953 			    V_XSLEEP | PCATCH)) != 0) {
1954 				goto fdout;
1955 			}
1956 			goto restart;
1957 		}
1958 #ifdef MAC
1959 		error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp,
1960 		    &nd.ni_cnd);
1961 		if (error != 0)
1962 			goto out;
1963 #endif
1964 		vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK);
1965 		error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
1966 #ifdef MAC
1967 out:
1968 #endif
1969 		vn_finished_write(mp);
1970 	}
1971 	NDFREE(&nd, NDF_ONLY_PNBUF);
1972 	vput(nd.ni_dvp);
1973 	if (vp == nd.ni_dvp)
1974 		vrele(vp);
1975 	else
1976 		vput(vp);
1977 	if (error == ERELOOKUP)
1978 		goto restart;
1979 fdout:
1980 	if (fp != NULL)
1981 		fdrop(fp, td);
1982 	return (error);
1983 }
1984 
1985 /*
1986  * Reposition read/write file offset.
1987  */
1988 #ifndef _SYS_SYSPROTO_H_
1989 struct lseek_args {
1990 	int	fd;
1991 	int	pad;
1992 	off_t	offset;
1993 	int	whence;
1994 };
1995 #endif
1996 int
1997 sys_lseek(struct thread *td, struct lseek_args *uap)
1998 {
1999 
2000 	return (kern_lseek(td, uap->fd, uap->offset, uap->whence));
2001 }
2002 
2003 int
2004 kern_lseek(struct thread *td, int fd, off_t offset, int whence)
2005 {
2006 	struct file *fp;
2007 	int error;
2008 
2009 	AUDIT_ARG_FD(fd);
2010 	error = fget(td, fd, &cap_seek_rights, &fp);
2011 	if (error != 0)
2012 		return (error);
2013 	error = (fp->f_ops->fo_flags & DFLAG_SEEKABLE) != 0 ?
2014 	    fo_seek(fp, offset, whence, td) : ESPIPE;
2015 	fdrop(fp, td);
2016 	return (error);
2017 }
2018 
2019 #if defined(COMPAT_43)
2020 /*
2021  * Reposition read/write file offset.
2022  */
2023 #ifndef _SYS_SYSPROTO_H_
2024 struct olseek_args {
2025 	int	fd;
2026 	long	offset;
2027 	int	whence;
2028 };
2029 #endif
2030 int
2031 olseek(struct thread *td, struct olseek_args *uap)
2032 {
2033 
2034 	return (kern_lseek(td, uap->fd, uap->offset, uap->whence));
2035 }
2036 #endif /* COMPAT_43 */
2037 
2038 #if defined(COMPAT_FREEBSD6)
2039 /* Version with the 'pad' argument */
2040 int
2041 freebsd6_lseek(struct thread *td, struct freebsd6_lseek_args *uap)
2042 {
2043 
2044 	return (kern_lseek(td, uap->fd, uap->offset, uap->whence));
2045 }
2046 #endif
2047 
2048 /*
2049  * Check access permissions using passed credentials.
2050  */
2051 static int
2052 vn_access(struct vnode *vp, int user_flags, struct ucred *cred,
2053      struct thread *td)
2054 {
2055 	accmode_t accmode;
2056 	int error;
2057 
2058 	/* Flags == 0 means only check for existence. */
2059 	if (user_flags == 0)
2060 		return (0);
2061 
2062 	accmode = 0;
2063 	if (user_flags & R_OK)
2064 		accmode |= VREAD;
2065 	if (user_flags & W_OK)
2066 		accmode |= VWRITE;
2067 	if (user_flags & X_OK)
2068 		accmode |= VEXEC;
2069 #ifdef MAC
2070 	error = mac_vnode_check_access(cred, vp, accmode);
2071 	if (error != 0)
2072 		return (error);
2073 #endif
2074 	if ((accmode & VWRITE) == 0 || (error = vn_writechk(vp)) == 0)
2075 		error = VOP_ACCESS(vp, accmode, cred, td);
2076 	return (error);
2077 }
2078 
2079 /*
2080  * Check access permissions using "real" credentials.
2081  */
2082 #ifndef _SYS_SYSPROTO_H_
2083 struct access_args {
2084 	char	*path;
2085 	int	amode;
2086 };
2087 #endif
2088 int
2089 sys_access(struct thread *td, struct access_args *uap)
2090 {
2091 
2092 	return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2093 	    0, uap->amode));
2094 }
2095 
2096 #ifndef _SYS_SYSPROTO_H_
2097 struct faccessat_args {
2098 	int	dirfd;
2099 	char	*path;
2100 	int	amode;
2101 	int	flag;
2102 }
2103 #endif
2104 int
2105 sys_faccessat(struct thread *td, struct faccessat_args *uap)
2106 {
2107 
2108 	return (kern_accessat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag,
2109 	    uap->amode));
2110 }
2111 
2112 int
2113 kern_accessat(struct thread *td, int fd, const char *path,
2114     enum uio_seg pathseg, int flag, int amode)
2115 {
2116 	struct ucred *cred, *usecred;
2117 	struct vnode *vp;
2118 	struct nameidata nd;
2119 	int error;
2120 
2121 	if ((flag & ~(AT_EACCESS | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0)
2122 		return (EINVAL);
2123 	if (amode != F_OK && (amode & ~(R_OK | W_OK | X_OK)) != 0)
2124 		return (EINVAL);
2125 
2126 	/*
2127 	 * Create and modify a temporary credential instead of one that
2128 	 * is potentially shared (if we need one).
2129 	 */
2130 	cred = td->td_ucred;
2131 	if ((flag & AT_EACCESS) == 0 &&
2132 	    ((cred->cr_uid != cred->cr_ruid ||
2133 	    cred->cr_rgid != cred->cr_groups[0]))) {
2134 		usecred = crdup(cred);
2135 		usecred->cr_uid = cred->cr_ruid;
2136 		usecred->cr_groups[0] = cred->cr_rgid;
2137 		td->td_ucred = usecred;
2138 	} else
2139 		usecred = cred;
2140 	AUDIT_ARG_VALUE(amode);
2141 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF |
2142 	    AUDITVNODE1 | at2cnpflags(flag, AT_RESOLVE_BENEATH |
2143 	    AT_EMPTY_PATH), pathseg, path, fd, &cap_fstat_rights, td);
2144 	if ((error = namei(&nd)) != 0)
2145 		goto out;
2146 	vp = nd.ni_vp;
2147 
2148 	error = vn_access(vp, amode, usecred, td);
2149 	NDFREE_NOTHING(&nd);
2150 	vput(vp);
2151 out:
2152 	if (usecred != cred) {
2153 		td->td_ucred = cred;
2154 		crfree(usecred);
2155 	}
2156 	return (error);
2157 }
2158 
2159 /*
2160  * Check access permissions using "effective" credentials.
2161  */
2162 #ifndef _SYS_SYSPROTO_H_
2163 struct eaccess_args {
2164 	char	*path;
2165 	int	amode;
2166 };
2167 #endif
2168 int
2169 sys_eaccess(struct thread *td, struct eaccess_args *uap)
2170 {
2171 
2172 	return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2173 	    AT_EACCESS, uap->amode));
2174 }
2175 
2176 #if defined(COMPAT_43)
2177 /*
2178  * Get file status; this version follows links.
2179  */
2180 #ifndef _SYS_SYSPROTO_H_
2181 struct ostat_args {
2182 	char	*path;
2183 	struct ostat *ub;
2184 };
2185 #endif
2186 int
2187 ostat(struct thread *td, struct ostat_args *uap)
2188 {
2189 	struct stat sb;
2190 	struct ostat osb;
2191 	int error;
2192 
2193 	error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE,
2194 	    &sb, NULL);
2195 	if (error != 0)
2196 		return (error);
2197 	cvtstat(&sb, &osb);
2198 	return (copyout(&osb, uap->ub, sizeof (osb)));
2199 }
2200 
2201 /*
2202  * Get file status; this version does not follow links.
2203  */
2204 #ifndef _SYS_SYSPROTO_H_
2205 struct olstat_args {
2206 	char	*path;
2207 	struct ostat *ub;
2208 };
2209 #endif
2210 int
2211 olstat(struct thread *td, struct olstat_args *uap)
2212 {
2213 	struct stat sb;
2214 	struct ostat osb;
2215 	int error;
2216 
2217 	error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path,
2218 	    UIO_USERSPACE, &sb, NULL);
2219 	if (error != 0)
2220 		return (error);
2221 	cvtstat(&sb, &osb);
2222 	return (copyout(&osb, uap->ub, sizeof (osb)));
2223 }
2224 
2225 /*
2226  * Convert from an old to a new stat structure.
2227  * XXX: many values are blindly truncated.
2228  */
2229 void
2230 cvtstat(struct stat *st, struct ostat *ost)
2231 {
2232 
2233 	bzero(ost, sizeof(*ost));
2234 	ost->st_dev = st->st_dev;
2235 	ost->st_ino = st->st_ino;
2236 	ost->st_mode = st->st_mode;
2237 	ost->st_nlink = st->st_nlink;
2238 	ost->st_uid = st->st_uid;
2239 	ost->st_gid = st->st_gid;
2240 	ost->st_rdev = st->st_rdev;
2241 	ost->st_size = MIN(st->st_size, INT32_MAX);
2242 	ost->st_atim = st->st_atim;
2243 	ost->st_mtim = st->st_mtim;
2244 	ost->st_ctim = st->st_ctim;
2245 	ost->st_blksize = st->st_blksize;
2246 	ost->st_blocks = st->st_blocks;
2247 	ost->st_flags = st->st_flags;
2248 	ost->st_gen = st->st_gen;
2249 }
2250 #endif /* COMPAT_43 */
2251 
2252 #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11)
2253 int ino64_trunc_error;
2254 SYSCTL_INT(_vfs, OID_AUTO, ino64_trunc_error, CTLFLAG_RW,
2255     &ino64_trunc_error, 0,
2256     "Error on truncation of device, file or inode number, or link count");
2257 
2258 int
2259 freebsd11_cvtstat(struct stat *st, struct freebsd11_stat *ost)
2260 {
2261 
2262 	ost->st_dev = st->st_dev;
2263 	if (ost->st_dev != st->st_dev) {
2264 		switch (ino64_trunc_error) {
2265 		default:
2266 			/*
2267 			 * Since dev_t is almost raw, don't clamp to the
2268 			 * maximum for case 2, but ignore the error.
2269 			 */
2270 			break;
2271 		case 1:
2272 			return (EOVERFLOW);
2273 		}
2274 	}
2275 	ost->st_ino = st->st_ino;
2276 	if (ost->st_ino != st->st_ino) {
2277 		switch (ino64_trunc_error) {
2278 		default:
2279 		case 0:
2280 			break;
2281 		case 1:
2282 			return (EOVERFLOW);
2283 		case 2:
2284 			ost->st_ino = UINT32_MAX;
2285 			break;
2286 		}
2287 	}
2288 	ost->st_mode = st->st_mode;
2289 	ost->st_nlink = st->st_nlink;
2290 	if (ost->st_nlink != st->st_nlink) {
2291 		switch (ino64_trunc_error) {
2292 		default:
2293 		case 0:
2294 			break;
2295 		case 1:
2296 			return (EOVERFLOW);
2297 		case 2:
2298 			ost->st_nlink = UINT16_MAX;
2299 			break;
2300 		}
2301 	}
2302 	ost->st_uid = st->st_uid;
2303 	ost->st_gid = st->st_gid;
2304 	ost->st_rdev = st->st_rdev;
2305 	if (ost->st_rdev != st->st_rdev) {
2306 		switch (ino64_trunc_error) {
2307 		default:
2308 			break;
2309 		case 1:
2310 			return (EOVERFLOW);
2311 		}
2312 	}
2313 	ost->st_atim = st->st_atim;
2314 	ost->st_mtim = st->st_mtim;
2315 	ost->st_ctim = st->st_ctim;
2316 	ost->st_size = st->st_size;
2317 	ost->st_blocks = st->st_blocks;
2318 	ost->st_blksize = st->st_blksize;
2319 	ost->st_flags = st->st_flags;
2320 	ost->st_gen = st->st_gen;
2321 	ost->st_lspare = 0;
2322 	ost->st_birthtim = st->st_birthtim;
2323 	bzero((char *)&ost->st_birthtim + sizeof(ost->st_birthtim),
2324 	    sizeof(*ost) - offsetof(struct freebsd11_stat,
2325 	    st_birthtim) - sizeof(ost->st_birthtim));
2326 	return (0);
2327 }
2328 
2329 int
2330 freebsd11_stat(struct thread *td, struct freebsd11_stat_args* uap)
2331 {
2332 	struct stat sb;
2333 	struct freebsd11_stat osb;
2334 	int error;
2335 
2336 	error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE,
2337 	    &sb, NULL);
2338 	if (error != 0)
2339 		return (error);
2340 	error = freebsd11_cvtstat(&sb, &osb);
2341 	if (error == 0)
2342 		error = copyout(&osb, uap->ub, sizeof(osb));
2343 	return (error);
2344 }
2345 
2346 int
2347 freebsd11_lstat(struct thread *td, struct freebsd11_lstat_args* uap)
2348 {
2349 	struct stat sb;
2350 	struct freebsd11_stat osb;
2351 	int error;
2352 
2353 	error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path,
2354 	    UIO_USERSPACE, &sb, NULL);
2355 	if (error != 0)
2356 		return (error);
2357 	error = freebsd11_cvtstat(&sb, &osb);
2358 	if (error == 0)
2359 		error = copyout(&osb, uap->ub, sizeof(osb));
2360 	return (error);
2361 }
2362 
2363 int
2364 freebsd11_fhstat(struct thread *td, struct freebsd11_fhstat_args* uap)
2365 {
2366 	struct fhandle fh;
2367 	struct stat sb;
2368 	struct freebsd11_stat osb;
2369 	int error;
2370 
2371 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
2372 	if (error != 0)
2373 		return (error);
2374 	error = kern_fhstat(td, fh, &sb);
2375 	if (error != 0)
2376 		return (error);
2377 	error = freebsd11_cvtstat(&sb, &osb);
2378 	if (error == 0)
2379 		error = copyout(&osb, uap->sb, sizeof(osb));
2380 	return (error);
2381 }
2382 
2383 int
2384 freebsd11_fstatat(struct thread *td, struct freebsd11_fstatat_args* uap)
2385 {
2386 	struct stat sb;
2387 	struct freebsd11_stat osb;
2388 	int error;
2389 
2390 	error = kern_statat(td, uap->flag, uap->fd, uap->path,
2391 	    UIO_USERSPACE, &sb, NULL);
2392 	if (error != 0)
2393 		return (error);
2394 	error = freebsd11_cvtstat(&sb, &osb);
2395 	if (error == 0)
2396 		error = copyout(&osb, uap->buf, sizeof(osb));
2397 	return (error);
2398 }
2399 #endif	/* COMPAT_FREEBSD11 */
2400 
2401 /*
2402  * Get file status
2403  */
2404 #ifndef _SYS_SYSPROTO_H_
2405 struct fstatat_args {
2406 	int	fd;
2407 	char	*path;
2408 	struct stat	*buf;
2409 	int	flag;
2410 }
2411 #endif
2412 int
2413 sys_fstatat(struct thread *td, struct fstatat_args *uap)
2414 {
2415 	struct stat sb;
2416 	int error;
2417 
2418 	error = kern_statat(td, uap->flag, uap->fd, uap->path,
2419 	    UIO_USERSPACE, &sb, NULL);
2420 	if (error == 0)
2421 		error = copyout(&sb, uap->buf, sizeof (sb));
2422 	return (error);
2423 }
2424 
2425 int
2426 kern_statat(struct thread *td, int flag, int fd, const char *path,
2427     enum uio_seg pathseg, struct stat *sbp,
2428     void (*hook)(struct vnode *vp, struct stat *sbp))
2429 {
2430 	struct nameidata nd;
2431 	int error;
2432 
2433 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
2434 	    AT_EMPTY_PATH)) != 0)
2435 		return (EINVAL);
2436 
2437 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_RESOLVE_BENEATH |
2438 	    AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH) | LOCKSHARED | LOCKLEAF |
2439 	    AUDITVNODE1, pathseg, path, fd, &cap_fstat_rights, td);
2440 
2441 	if ((error = namei(&nd)) != 0) {
2442 		if (error == ENOTDIR &&
2443 		    (nd.ni_resflags & NIRES_EMPTYPATH) != 0)
2444 			error = kern_fstat(td, fd, sbp);
2445 		return (error);
2446 	}
2447 	error = VOP_STAT(nd.ni_vp, sbp, td->td_ucred, NOCRED, td);
2448 	if (error == 0) {
2449 		if (__predict_false(hook != NULL))
2450 			hook(nd.ni_vp, sbp);
2451 	}
2452 	NDFREE_NOTHING(&nd);
2453 	vput(nd.ni_vp);
2454 #ifdef __STAT_TIME_T_EXT
2455 	sbp->st_atim_ext = 0;
2456 	sbp->st_mtim_ext = 0;
2457 	sbp->st_ctim_ext = 0;
2458 	sbp->st_btim_ext = 0;
2459 #endif
2460 #ifdef KTRACE
2461 	if (KTRPOINT(td, KTR_STRUCT))
2462 		ktrstat_error(sbp, error);
2463 #endif
2464 	return (error);
2465 }
2466 
2467 #if defined(COMPAT_FREEBSD11)
2468 /*
2469  * Implementation of the NetBSD [l]stat() functions.
2470  */
2471 void
2472 freebsd11_cvtnstat(struct stat *sb, struct nstat *nsb)
2473 {
2474 
2475 	bzero(nsb, sizeof(*nsb));
2476 	nsb->st_dev = sb->st_dev;
2477 	nsb->st_ino = sb->st_ino;
2478 	nsb->st_mode = sb->st_mode;
2479 	nsb->st_nlink = sb->st_nlink;
2480 	nsb->st_uid = sb->st_uid;
2481 	nsb->st_gid = sb->st_gid;
2482 	nsb->st_rdev = sb->st_rdev;
2483 	nsb->st_atim = sb->st_atim;
2484 	nsb->st_mtim = sb->st_mtim;
2485 	nsb->st_ctim = sb->st_ctim;
2486 	nsb->st_size = sb->st_size;
2487 	nsb->st_blocks = sb->st_blocks;
2488 	nsb->st_blksize = sb->st_blksize;
2489 	nsb->st_flags = sb->st_flags;
2490 	nsb->st_gen = sb->st_gen;
2491 	nsb->st_birthtim = sb->st_birthtim;
2492 }
2493 
2494 #ifndef _SYS_SYSPROTO_H_
2495 struct freebsd11_nstat_args {
2496 	char	*path;
2497 	struct nstat *ub;
2498 };
2499 #endif
2500 int
2501 freebsd11_nstat(struct thread *td, struct freebsd11_nstat_args *uap)
2502 {
2503 	struct stat sb;
2504 	struct nstat nsb;
2505 	int error;
2506 
2507 	error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE,
2508 	    &sb, NULL);
2509 	if (error != 0)
2510 		return (error);
2511 	freebsd11_cvtnstat(&sb, &nsb);
2512 	return (copyout(&nsb, uap->ub, sizeof (nsb)));
2513 }
2514 
2515 /*
2516  * NetBSD lstat.  Get file status; this version does not follow links.
2517  */
2518 #ifndef _SYS_SYSPROTO_H_
2519 struct freebsd11_nlstat_args {
2520 	char	*path;
2521 	struct nstat *ub;
2522 };
2523 #endif
2524 int
2525 freebsd11_nlstat(struct thread *td, struct freebsd11_nlstat_args *uap)
2526 {
2527 	struct stat sb;
2528 	struct nstat nsb;
2529 	int error;
2530 
2531 	error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path,
2532 	    UIO_USERSPACE, &sb, NULL);
2533 	if (error != 0)
2534 		return (error);
2535 	freebsd11_cvtnstat(&sb, &nsb);
2536 	return (copyout(&nsb, uap->ub, sizeof (nsb)));
2537 }
2538 #endif /* COMPAT_FREEBSD11 */
2539 
2540 /*
2541  * Get configurable pathname variables.
2542  */
2543 #ifndef _SYS_SYSPROTO_H_
2544 struct pathconf_args {
2545 	char	*path;
2546 	int	name;
2547 };
2548 #endif
2549 int
2550 sys_pathconf(struct thread *td, struct pathconf_args *uap)
2551 {
2552 	long value;
2553 	int error;
2554 
2555 	error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, FOLLOW,
2556 	    &value);
2557 	if (error == 0)
2558 		td->td_retval[0] = value;
2559 	return (error);
2560 }
2561 
2562 #ifndef _SYS_SYSPROTO_H_
2563 struct lpathconf_args {
2564 	char	*path;
2565 	int	name;
2566 };
2567 #endif
2568 int
2569 sys_lpathconf(struct thread *td, struct lpathconf_args *uap)
2570 {
2571 	long value;
2572 	int error;
2573 
2574 	error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name,
2575 	    NOFOLLOW, &value);
2576 	if (error == 0)
2577 		td->td_retval[0] = value;
2578 	return (error);
2579 }
2580 
2581 int
2582 kern_pathconf(struct thread *td, const char *path, enum uio_seg pathseg,
2583     int name, u_long flags, long *valuep)
2584 {
2585 	struct nameidata nd;
2586 	int error;
2587 
2588 	NDINIT(&nd, LOOKUP, LOCKSHARED | LOCKLEAF | AUDITVNODE1 | flags,
2589 	    pathseg, path, td);
2590 	if ((error = namei(&nd)) != 0)
2591 		return (error);
2592 	NDFREE_NOTHING(&nd);
2593 
2594 	error = VOP_PATHCONF(nd.ni_vp, name, valuep);
2595 	vput(nd.ni_vp);
2596 	return (error);
2597 }
2598 
2599 /*
2600  * Return target name of a symbolic link.
2601  */
2602 #ifndef _SYS_SYSPROTO_H_
2603 struct readlink_args {
2604 	char	*path;
2605 	char	*buf;
2606 	size_t	count;
2607 };
2608 #endif
2609 int
2610 sys_readlink(struct thread *td, struct readlink_args *uap)
2611 {
2612 
2613 	return (kern_readlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2614 	    uap->buf, UIO_USERSPACE, uap->count));
2615 }
2616 #ifndef _SYS_SYSPROTO_H_
2617 struct readlinkat_args {
2618 	int	fd;
2619 	char	*path;
2620 	char	*buf;
2621 	size_t	bufsize;
2622 };
2623 #endif
2624 int
2625 sys_readlinkat(struct thread *td, struct readlinkat_args *uap)
2626 {
2627 
2628 	return (kern_readlinkat(td, uap->fd, uap->path, UIO_USERSPACE,
2629 	    uap->buf, UIO_USERSPACE, uap->bufsize));
2630 }
2631 
2632 int
2633 kern_readlinkat(struct thread *td, int fd, const char *path,
2634     enum uio_seg pathseg, char *buf, enum uio_seg bufseg, size_t count)
2635 {
2636 	struct vnode *vp;
2637 	struct nameidata nd;
2638 	int error;
2639 
2640 	if (count > IOSIZE_MAX)
2641 		return (EINVAL);
2642 
2643 	NDINIT_AT(&nd, LOOKUP, NOFOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1,
2644 	    pathseg, path, fd, td);
2645 
2646 	if ((error = namei(&nd)) != 0)
2647 		return (error);
2648 	NDFREE_NOTHING(&nd);
2649 	vp = nd.ni_vp;
2650 
2651 	error = kern_readlink_vp(vp, buf, bufseg, count, td);
2652 	vput(vp);
2653 
2654 	return (error);
2655 }
2656 
2657 /*
2658  * Helper function to readlink from a vnode
2659  */
2660 static int
2661 kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg, size_t count,
2662     struct thread *td)
2663 {
2664 	struct iovec aiov;
2665 	struct uio auio;
2666 	int error;
2667 
2668 	ASSERT_VOP_LOCKED(vp, "kern_readlink_vp(): vp not locked");
2669 #ifdef MAC
2670 	error = mac_vnode_check_readlink(td->td_ucred, vp);
2671 	if (error != 0)
2672 		return (error);
2673 #endif
2674 	if (vp->v_type != VLNK && (vp->v_vflag & VV_READLINK) == 0)
2675 		return (EINVAL);
2676 
2677 	aiov.iov_base = buf;
2678 	aiov.iov_len = count;
2679 	auio.uio_iov = &aiov;
2680 	auio.uio_iovcnt = 1;
2681 	auio.uio_offset = 0;
2682 	auio.uio_rw = UIO_READ;
2683 	auio.uio_segflg = bufseg;
2684 	auio.uio_td = td;
2685 	auio.uio_resid = count;
2686 	error = VOP_READLINK(vp, &auio, td->td_ucred);
2687 	td->td_retval[0] = count - auio.uio_resid;
2688 	return (error);
2689 }
2690 
2691 /*
2692  * Common implementation code for chflags() and fchflags().
2693  */
2694 static int
2695 setfflags(struct thread *td, struct vnode *vp, u_long flags)
2696 {
2697 	struct mount *mp;
2698 	struct vattr vattr;
2699 	int error;
2700 
2701 	/* We can't support the value matching VNOVAL. */
2702 	if (flags == VNOVAL)
2703 		return (EOPNOTSUPP);
2704 
2705 	/*
2706 	 * Prevent non-root users from setting flags on devices.  When
2707 	 * a device is reused, users can retain ownership of the device
2708 	 * if they are allowed to set flags and programs assume that
2709 	 * chown can't fail when done as root.
2710 	 */
2711 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
2712 		error = priv_check(td, PRIV_VFS_CHFLAGS_DEV);
2713 		if (error != 0)
2714 			return (error);
2715 	}
2716 
2717 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
2718 		return (error);
2719 	VATTR_NULL(&vattr);
2720 	vattr.va_flags = flags;
2721 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2722 #ifdef MAC
2723 	error = mac_vnode_check_setflags(td->td_ucred, vp, vattr.va_flags);
2724 	if (error == 0)
2725 #endif
2726 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2727 	VOP_UNLOCK(vp);
2728 	vn_finished_write(mp);
2729 	return (error);
2730 }
2731 
2732 /*
2733  * Change flags of a file given a path name.
2734  */
2735 #ifndef _SYS_SYSPROTO_H_
2736 struct chflags_args {
2737 	const char *path;
2738 	u_long	flags;
2739 };
2740 #endif
2741 int
2742 sys_chflags(struct thread *td, struct chflags_args *uap)
2743 {
2744 
2745 	return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2746 	    uap->flags, 0));
2747 }
2748 
2749 #ifndef _SYS_SYSPROTO_H_
2750 struct chflagsat_args {
2751 	int	fd;
2752 	const char *path;
2753 	u_long	flags;
2754 	int	atflag;
2755 }
2756 #endif
2757 int
2758 sys_chflagsat(struct thread *td, struct chflagsat_args *uap)
2759 {
2760 
2761 	if ((uap->atflag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
2762 	    AT_EMPTY_PATH)) != 0)
2763 		return (EINVAL);
2764 
2765 	return (kern_chflagsat(td, uap->fd, uap->path, UIO_USERSPACE,
2766 	    uap->flags, uap->atflag));
2767 }
2768 
2769 /*
2770  * Same as chflags() but doesn't follow symlinks.
2771  */
2772 #ifndef _SYS_SYSPROTO_H_
2773 struct lchflags_args {
2774 	const char *path;
2775 	u_long flags;
2776 };
2777 #endif
2778 int
2779 sys_lchflags(struct thread *td, struct lchflags_args *uap)
2780 {
2781 
2782 	return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2783 	    uap->flags, AT_SYMLINK_NOFOLLOW));
2784 }
2785 
2786 static int
2787 kern_chflagsat(struct thread *td, int fd, const char *path,
2788     enum uio_seg pathseg, u_long flags, int atflag)
2789 {
2790 	struct nameidata nd;
2791 	int error;
2792 
2793 	AUDIT_ARG_FFLAGS(flags);
2794 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(atflag, AT_SYMLINK_NOFOLLOW |
2795 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path,
2796 	    fd, &cap_fchflags_rights, td);
2797 	if ((error = namei(&nd)) != 0)
2798 		return (error);
2799 	NDFREE_NOTHING(&nd);
2800 	error = setfflags(td, nd.ni_vp, flags);
2801 	vrele(nd.ni_vp);
2802 	return (error);
2803 }
2804 
2805 /*
2806  * Change flags of a file given a file descriptor.
2807  */
2808 #ifndef _SYS_SYSPROTO_H_
2809 struct fchflags_args {
2810 	int	fd;
2811 	u_long	flags;
2812 };
2813 #endif
2814 int
2815 sys_fchflags(struct thread *td, struct fchflags_args *uap)
2816 {
2817 	struct file *fp;
2818 	int error;
2819 
2820 	AUDIT_ARG_FD(uap->fd);
2821 	AUDIT_ARG_FFLAGS(uap->flags);
2822 	error = getvnode(td, uap->fd, &cap_fchflags_rights,
2823 	    &fp);
2824 	if (error != 0)
2825 		return (error);
2826 #ifdef AUDIT
2827 	if (AUDITING_TD(td)) {
2828 		vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY);
2829 		AUDIT_ARG_VNODE1(fp->f_vnode);
2830 		VOP_UNLOCK(fp->f_vnode);
2831 	}
2832 #endif
2833 	error = setfflags(td, fp->f_vnode, uap->flags);
2834 	fdrop(fp, td);
2835 	return (error);
2836 }
2837 
2838 /*
2839  * Common implementation code for chmod(), lchmod() and fchmod().
2840  */
2841 int
2842 setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode)
2843 {
2844 	struct mount *mp;
2845 	struct vattr vattr;
2846 	int error;
2847 
2848 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
2849 		return (error);
2850 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2851 	VATTR_NULL(&vattr);
2852 	vattr.va_mode = mode & ALLPERMS;
2853 #ifdef MAC
2854 	error = mac_vnode_check_setmode(cred, vp, vattr.va_mode);
2855 	if (error == 0)
2856 #endif
2857 		error = VOP_SETATTR(vp, &vattr, cred);
2858 	VOP_UNLOCK(vp);
2859 	vn_finished_write(mp);
2860 	return (error);
2861 }
2862 
2863 /*
2864  * Change mode of a file given path name.
2865  */
2866 #ifndef _SYS_SYSPROTO_H_
2867 struct chmod_args {
2868 	char	*path;
2869 	int	mode;
2870 };
2871 #endif
2872 int
2873 sys_chmod(struct thread *td, struct chmod_args *uap)
2874 {
2875 
2876 	return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2877 	    uap->mode, 0));
2878 }
2879 
2880 #ifndef _SYS_SYSPROTO_H_
2881 struct fchmodat_args {
2882 	int	dirfd;
2883 	char	*path;
2884 	mode_t	mode;
2885 	int	flag;
2886 }
2887 #endif
2888 int
2889 sys_fchmodat(struct thread *td, struct fchmodat_args *uap)
2890 {
2891 
2892 	if ((uap->flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
2893 	    AT_EMPTY_PATH)) != 0)
2894 		return (EINVAL);
2895 
2896 	return (kern_fchmodat(td, uap->fd, uap->path, UIO_USERSPACE,
2897 	    uap->mode, uap->flag));
2898 }
2899 
2900 /*
2901  * Change mode of a file given path name (don't follow links.)
2902  */
2903 #ifndef _SYS_SYSPROTO_H_
2904 struct lchmod_args {
2905 	char	*path;
2906 	int	mode;
2907 };
2908 #endif
2909 int
2910 sys_lchmod(struct thread *td, struct lchmod_args *uap)
2911 {
2912 
2913 	return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2914 	    uap->mode, AT_SYMLINK_NOFOLLOW));
2915 }
2916 
2917 int
2918 kern_fchmodat(struct thread *td, int fd, const char *path,
2919     enum uio_seg pathseg, mode_t mode, int flag)
2920 {
2921 	struct nameidata nd;
2922 	int error;
2923 
2924 	AUDIT_ARG_MODE(mode);
2925 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW |
2926 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path,
2927 	    fd, &cap_fchmod_rights, td);
2928 	if ((error = namei(&nd)) != 0)
2929 		return (error);
2930 	NDFREE_NOTHING(&nd);
2931 	error = setfmode(td, td->td_ucred, nd.ni_vp, mode);
2932 	vrele(nd.ni_vp);
2933 	return (error);
2934 }
2935 
2936 /*
2937  * Change mode of a file given a file descriptor.
2938  */
2939 #ifndef _SYS_SYSPROTO_H_
2940 struct fchmod_args {
2941 	int	fd;
2942 	int	mode;
2943 };
2944 #endif
2945 int
2946 sys_fchmod(struct thread *td, struct fchmod_args *uap)
2947 {
2948 	struct file *fp;
2949 	int error;
2950 
2951 	AUDIT_ARG_FD(uap->fd);
2952 	AUDIT_ARG_MODE(uap->mode);
2953 
2954 	error = fget(td, uap->fd, &cap_fchmod_rights, &fp);
2955 	if (error != 0)
2956 		return (error);
2957 	error = fo_chmod(fp, uap->mode, td->td_ucred, td);
2958 	fdrop(fp, td);
2959 	return (error);
2960 }
2961 
2962 /*
2963  * Common implementation for chown(), lchown(), and fchown()
2964  */
2965 int
2966 setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid,
2967     gid_t gid)
2968 {
2969 	struct mount *mp;
2970 	struct vattr vattr;
2971 	int error;
2972 
2973 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
2974 		return (error);
2975 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2976 	VATTR_NULL(&vattr);
2977 	vattr.va_uid = uid;
2978 	vattr.va_gid = gid;
2979 #ifdef MAC
2980 	error = mac_vnode_check_setowner(cred, vp, vattr.va_uid,
2981 	    vattr.va_gid);
2982 	if (error == 0)
2983 #endif
2984 		error = VOP_SETATTR(vp, &vattr, cred);
2985 	VOP_UNLOCK(vp);
2986 	vn_finished_write(mp);
2987 	return (error);
2988 }
2989 
2990 /*
2991  * Set ownership given a path name.
2992  */
2993 #ifndef _SYS_SYSPROTO_H_
2994 struct chown_args {
2995 	char	*path;
2996 	int	uid;
2997 	int	gid;
2998 };
2999 #endif
3000 int
3001 sys_chown(struct thread *td, struct chown_args *uap)
3002 {
3003 
3004 	return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid,
3005 	    uap->gid, 0));
3006 }
3007 
3008 #ifndef _SYS_SYSPROTO_H_
3009 struct fchownat_args {
3010 	int fd;
3011 	const char * path;
3012 	uid_t uid;
3013 	gid_t gid;
3014 	int flag;
3015 };
3016 #endif
3017 int
3018 sys_fchownat(struct thread *td, struct fchownat_args *uap)
3019 {
3020 
3021 	if ((uap->flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
3022 	    AT_EMPTY_PATH)) != 0)
3023 		return (EINVAL);
3024 
3025 	return (kern_fchownat(td, uap->fd, uap->path, UIO_USERSPACE, uap->uid,
3026 	    uap->gid, uap->flag));
3027 }
3028 
3029 int
3030 kern_fchownat(struct thread *td, int fd, const char *path,
3031     enum uio_seg pathseg, int uid, int gid, int flag)
3032 {
3033 	struct nameidata nd;
3034 	int error;
3035 
3036 	AUDIT_ARG_OWNER(uid, gid);
3037 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW |
3038 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path,
3039 	    fd, &cap_fchown_rights, td);
3040 
3041 	if ((error = namei(&nd)) != 0)
3042 		return (error);
3043 	NDFREE_NOTHING(&nd);
3044 	error = setfown(td, td->td_ucred, nd.ni_vp, uid, gid);
3045 	vrele(nd.ni_vp);
3046 	return (error);
3047 }
3048 
3049 /*
3050  * Set ownership given a path name, do not cross symlinks.
3051  */
3052 #ifndef _SYS_SYSPROTO_H_
3053 struct lchown_args {
3054 	char	*path;
3055 	int	uid;
3056 	int	gid;
3057 };
3058 #endif
3059 int
3060 sys_lchown(struct thread *td, struct lchown_args *uap)
3061 {
3062 
3063 	return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
3064 	    uap->uid, uap->gid, AT_SYMLINK_NOFOLLOW));
3065 }
3066 
3067 /*
3068  * Set ownership given a file descriptor.
3069  */
3070 #ifndef _SYS_SYSPROTO_H_
3071 struct fchown_args {
3072 	int	fd;
3073 	int	uid;
3074 	int	gid;
3075 };
3076 #endif
3077 int
3078 sys_fchown(struct thread *td, struct fchown_args *uap)
3079 {
3080 	struct file *fp;
3081 	int error;
3082 
3083 	AUDIT_ARG_FD(uap->fd);
3084 	AUDIT_ARG_OWNER(uap->uid, uap->gid);
3085 	error = fget(td, uap->fd, &cap_fchown_rights, &fp);
3086 	if (error != 0)
3087 		return (error);
3088 	error = fo_chown(fp, uap->uid, uap->gid, td->td_ucred, td);
3089 	fdrop(fp, td);
3090 	return (error);
3091 }
3092 
3093 /*
3094  * Common implementation code for utimes(), lutimes(), and futimes().
3095  */
3096 static int
3097 getutimes(const struct timeval *usrtvp, enum uio_seg tvpseg,
3098     struct timespec *tsp)
3099 {
3100 	struct timeval tv[2];
3101 	const struct timeval *tvp;
3102 	int error;
3103 
3104 	if (usrtvp == NULL) {
3105 		vfs_timestamp(&tsp[0]);
3106 		tsp[1] = tsp[0];
3107 	} else {
3108 		if (tvpseg == UIO_SYSSPACE) {
3109 			tvp = usrtvp;
3110 		} else {
3111 			if ((error = copyin(usrtvp, tv, sizeof(tv))) != 0)
3112 				return (error);
3113 			tvp = tv;
3114 		}
3115 
3116 		if (tvp[0].tv_usec < 0 || tvp[0].tv_usec >= 1000000 ||
3117 		    tvp[1].tv_usec < 0 || tvp[1].tv_usec >= 1000000)
3118 			return (EINVAL);
3119 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3120 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3121 	}
3122 	return (0);
3123 }
3124 
3125 /*
3126  * Common implementation code for futimens(), utimensat().
3127  */
3128 #define	UTIMENS_NULL	0x1
3129 #define	UTIMENS_EXIT	0x2
3130 static int
3131 getutimens(const struct timespec *usrtsp, enum uio_seg tspseg,
3132     struct timespec *tsp, int *retflags)
3133 {
3134 	struct timespec tsnow;
3135 	int error;
3136 
3137 	vfs_timestamp(&tsnow);
3138 	*retflags = 0;
3139 	if (usrtsp == NULL) {
3140 		tsp[0] = tsnow;
3141 		tsp[1] = tsnow;
3142 		*retflags |= UTIMENS_NULL;
3143 		return (0);
3144 	}
3145 	if (tspseg == UIO_SYSSPACE) {
3146 		tsp[0] = usrtsp[0];
3147 		tsp[1] = usrtsp[1];
3148 	} else if ((error = copyin(usrtsp, tsp, sizeof(*tsp) * 2)) != 0)
3149 		return (error);
3150 	if (tsp[0].tv_nsec == UTIME_OMIT && tsp[1].tv_nsec == UTIME_OMIT)
3151 		*retflags |= UTIMENS_EXIT;
3152 	if (tsp[0].tv_nsec == UTIME_NOW && tsp[1].tv_nsec == UTIME_NOW)
3153 		*retflags |= UTIMENS_NULL;
3154 	if (tsp[0].tv_nsec == UTIME_OMIT)
3155 		tsp[0].tv_sec = VNOVAL;
3156 	else if (tsp[0].tv_nsec == UTIME_NOW)
3157 		tsp[0] = tsnow;
3158 	else if (tsp[0].tv_nsec < 0 || tsp[0].tv_nsec >= 1000000000L)
3159 		return (EINVAL);
3160 	if (tsp[1].tv_nsec == UTIME_OMIT)
3161 		tsp[1].tv_sec = VNOVAL;
3162 	else if (tsp[1].tv_nsec == UTIME_NOW)
3163 		tsp[1] = tsnow;
3164 	else if (tsp[1].tv_nsec < 0 || tsp[1].tv_nsec >= 1000000000L)
3165 		return (EINVAL);
3166 
3167 	return (0);
3168 }
3169 
3170 /*
3171  * Common implementation code for utimes(), lutimes(), futimes(), futimens(),
3172  * and utimensat().
3173  */
3174 static int
3175 setutimes(struct thread *td, struct vnode *vp, const struct timespec *ts,
3176     int numtimes, int nullflag)
3177 {
3178 	struct mount *mp;
3179 	struct vattr vattr;
3180 	int error;
3181 	bool setbirthtime;
3182 
3183 	setbirthtime = false;
3184 	vattr.va_birthtime.tv_sec = VNOVAL;
3185 	vattr.va_birthtime.tv_nsec = 0;
3186 
3187 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
3188 		return (error);
3189 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3190 	if (numtimes < 3 && VOP_GETATTR(vp, &vattr, td->td_ucred) == 0 &&
3191 	    timespeccmp(&ts[1], &vattr.va_birthtime, < ))
3192 		setbirthtime = true;
3193 	VATTR_NULL(&vattr);
3194 	vattr.va_atime = ts[0];
3195 	vattr.va_mtime = ts[1];
3196 	if (setbirthtime)
3197 		vattr.va_birthtime = ts[1];
3198 	if (numtimes > 2)
3199 		vattr.va_birthtime = ts[2];
3200 	if (nullflag)
3201 		vattr.va_vaflags |= VA_UTIMES_NULL;
3202 #ifdef MAC
3203 	error = mac_vnode_check_setutimes(td->td_ucred, vp, vattr.va_atime,
3204 	    vattr.va_mtime);
3205 #endif
3206 	if (error == 0)
3207 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3208 	VOP_UNLOCK(vp);
3209 	vn_finished_write(mp);
3210 	return (error);
3211 }
3212 
3213 /*
3214  * Set the access and modification times of a file.
3215  */
3216 #ifndef _SYS_SYSPROTO_H_
3217 struct utimes_args {
3218 	char	*path;
3219 	struct	timeval *tptr;
3220 };
3221 #endif
3222 int
3223 sys_utimes(struct thread *td, struct utimes_args *uap)
3224 {
3225 
3226 	return (kern_utimesat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
3227 	    uap->tptr, UIO_USERSPACE));
3228 }
3229 
3230 #ifndef _SYS_SYSPROTO_H_
3231 struct futimesat_args {
3232 	int fd;
3233 	const char * path;
3234 	const struct timeval * times;
3235 };
3236 #endif
3237 int
3238 sys_futimesat(struct thread *td, struct futimesat_args *uap)
3239 {
3240 
3241 	return (kern_utimesat(td, uap->fd, uap->path, UIO_USERSPACE,
3242 	    uap->times, UIO_USERSPACE));
3243 }
3244 
3245 int
3246 kern_utimesat(struct thread *td, int fd, const char *path,
3247     enum uio_seg pathseg, struct timeval *tptr, enum uio_seg tptrseg)
3248 {
3249 	struct nameidata nd;
3250 	struct timespec ts[2];
3251 	int error;
3252 
3253 	if ((error = getutimes(tptr, tptrseg, ts)) != 0)
3254 		return (error);
3255 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd,
3256 	    &cap_futimes_rights, td);
3257 
3258 	if ((error = namei(&nd)) != 0)
3259 		return (error);
3260 	NDFREE_NOTHING(&nd);
3261 	error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL);
3262 	vrele(nd.ni_vp);
3263 	return (error);
3264 }
3265 
3266 /*
3267  * Set the access and modification times of a file.
3268  */
3269 #ifndef _SYS_SYSPROTO_H_
3270 struct lutimes_args {
3271 	char	*path;
3272 	struct	timeval *tptr;
3273 };
3274 #endif
3275 int
3276 sys_lutimes(struct thread *td, struct lutimes_args *uap)
3277 {
3278 
3279 	return (kern_lutimes(td, uap->path, UIO_USERSPACE, uap->tptr,
3280 	    UIO_USERSPACE));
3281 }
3282 
3283 int
3284 kern_lutimes(struct thread *td, const char *path, enum uio_seg pathseg,
3285     struct timeval *tptr, enum uio_seg tptrseg)
3286 {
3287 	struct timespec ts[2];
3288 	struct nameidata nd;
3289 	int error;
3290 
3291 	if ((error = getutimes(tptr, tptrseg, ts)) != 0)
3292 		return (error);
3293 	NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, pathseg, path, td);
3294 	if ((error = namei(&nd)) != 0)
3295 		return (error);
3296 	NDFREE_NOTHING(&nd);
3297 	error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL);
3298 	vrele(nd.ni_vp);
3299 	return (error);
3300 }
3301 
3302 /*
3303  * Set the access and modification times of a file.
3304  */
3305 #ifndef _SYS_SYSPROTO_H_
3306 struct futimes_args {
3307 	int	fd;
3308 	struct	timeval *tptr;
3309 };
3310 #endif
3311 int
3312 sys_futimes(struct thread *td, struct futimes_args *uap)
3313 {
3314 
3315 	return (kern_futimes(td, uap->fd, uap->tptr, UIO_USERSPACE));
3316 }
3317 
3318 int
3319 kern_futimes(struct thread *td, int fd, struct timeval *tptr,
3320     enum uio_seg tptrseg)
3321 {
3322 	struct timespec ts[2];
3323 	struct file *fp;
3324 	int error;
3325 
3326 	AUDIT_ARG_FD(fd);
3327 	error = getutimes(tptr, tptrseg, ts);
3328 	if (error != 0)
3329 		return (error);
3330 	error = getvnode(td, fd, &cap_futimes_rights, &fp);
3331 	if (error != 0)
3332 		return (error);
3333 #ifdef AUDIT
3334 	if (AUDITING_TD(td)) {
3335 		vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY);
3336 		AUDIT_ARG_VNODE1(fp->f_vnode);
3337 		VOP_UNLOCK(fp->f_vnode);
3338 	}
3339 #endif
3340 	error = setutimes(td, fp->f_vnode, ts, 2, tptr == NULL);
3341 	fdrop(fp, td);
3342 	return (error);
3343 }
3344 
3345 int
3346 sys_futimens(struct thread *td, struct futimens_args *uap)
3347 {
3348 
3349 	return (kern_futimens(td, uap->fd, uap->times, UIO_USERSPACE));
3350 }
3351 
3352 int
3353 kern_futimens(struct thread *td, int fd, struct timespec *tptr,
3354     enum uio_seg tptrseg)
3355 {
3356 	struct timespec ts[2];
3357 	struct file *fp;
3358 	int error, flags;
3359 
3360 	AUDIT_ARG_FD(fd);
3361 	error = getutimens(tptr, tptrseg, ts, &flags);
3362 	if (error != 0)
3363 		return (error);
3364 	if (flags & UTIMENS_EXIT)
3365 		return (0);
3366 	error = getvnode(td, fd, &cap_futimes_rights, &fp);
3367 	if (error != 0)
3368 		return (error);
3369 #ifdef AUDIT
3370 	if (AUDITING_TD(td)) {
3371 		vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY);
3372 		AUDIT_ARG_VNODE1(fp->f_vnode);
3373 		VOP_UNLOCK(fp->f_vnode);
3374 	}
3375 #endif
3376 	error = setutimes(td, fp->f_vnode, ts, 2, flags & UTIMENS_NULL);
3377 	fdrop(fp, td);
3378 	return (error);
3379 }
3380 
3381 int
3382 sys_utimensat(struct thread *td, struct utimensat_args *uap)
3383 {
3384 
3385 	return (kern_utimensat(td, uap->fd, uap->path, UIO_USERSPACE,
3386 	    uap->times, UIO_USERSPACE, uap->flag));
3387 }
3388 
3389 int
3390 kern_utimensat(struct thread *td, int fd, const char *path,
3391     enum uio_seg pathseg, struct timespec *tptr, enum uio_seg tptrseg,
3392     int flag)
3393 {
3394 	struct nameidata nd;
3395 	struct timespec ts[2];
3396 	int error, flags;
3397 
3398 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
3399 	    AT_EMPTY_PATH)) != 0)
3400 		return (EINVAL);
3401 
3402 	if ((error = getutimens(tptr, tptrseg, ts, &flags)) != 0)
3403 		return (error);
3404 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW |
3405 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1,
3406 	    pathseg, path, fd, &cap_futimes_rights, td);
3407 	if ((error = namei(&nd)) != 0)
3408 		return (error);
3409 	/*
3410 	 * We are allowed to call namei() regardless of 2xUTIME_OMIT.
3411 	 * POSIX states:
3412 	 * "If both tv_nsec fields are UTIME_OMIT... EACCESS may be detected."
3413 	 * "Search permission is denied by a component of the path prefix."
3414 	 */
3415 	NDFREE_NOTHING(&nd);
3416 	if ((flags & UTIMENS_EXIT) == 0)
3417 		error = setutimes(td, nd.ni_vp, ts, 2, flags & UTIMENS_NULL);
3418 	vrele(nd.ni_vp);
3419 	return (error);
3420 }
3421 
3422 /*
3423  * Truncate a file given its path name.
3424  */
3425 #ifndef _SYS_SYSPROTO_H_
3426 struct truncate_args {
3427 	char	*path;
3428 	int	pad;
3429 	off_t	length;
3430 };
3431 #endif
3432 int
3433 sys_truncate(struct thread *td, struct truncate_args *uap)
3434 {
3435 
3436 	return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length));
3437 }
3438 
3439 int
3440 kern_truncate(struct thread *td, const char *path, enum uio_seg pathseg,
3441     off_t length)
3442 {
3443 	struct mount *mp;
3444 	struct vnode *vp;
3445 	void *rl_cookie;
3446 	struct vattr vattr;
3447 	struct nameidata nd;
3448 	int error;
3449 
3450 	if (length < 0)
3451 		return (EINVAL);
3452 	NDPREINIT(&nd);
3453 retry:
3454 	NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, td);
3455 	if ((error = namei(&nd)) != 0)
3456 		return (error);
3457 	vp = nd.ni_vp;
3458 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3459 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) {
3460 		vn_rangelock_unlock(vp, rl_cookie);
3461 		vrele(vp);
3462 		return (error);
3463 	}
3464 	NDFREE(&nd, NDF_ONLY_PNBUF);
3465 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3466 	if (vp->v_type == VDIR)
3467 		error = EISDIR;
3468 #ifdef MAC
3469 	else if ((error = mac_vnode_check_write(td->td_ucred, NOCRED, vp))) {
3470 	}
3471 #endif
3472 	else if ((error = vn_writechk(vp)) == 0 &&
3473 	    (error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td)) == 0) {
3474 		VATTR_NULL(&vattr);
3475 		vattr.va_size = length;
3476 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3477 	}
3478 	VOP_UNLOCK(vp);
3479 	vn_finished_write(mp);
3480 	vn_rangelock_unlock(vp, rl_cookie);
3481 	vrele(vp);
3482 	if (error == ERELOOKUP)
3483 		goto retry;
3484 	return (error);
3485 }
3486 
3487 #if defined(COMPAT_43)
3488 /*
3489  * Truncate a file given its path name.
3490  */
3491 #ifndef _SYS_SYSPROTO_H_
3492 struct otruncate_args {
3493 	char	*path;
3494 	long	length;
3495 };
3496 #endif
3497 int
3498 otruncate(struct thread *td, struct otruncate_args *uap)
3499 {
3500 
3501 	return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length));
3502 }
3503 #endif /* COMPAT_43 */
3504 
3505 #if defined(COMPAT_FREEBSD6)
3506 /* Versions with the pad argument */
3507 int
3508 freebsd6_truncate(struct thread *td, struct freebsd6_truncate_args *uap)
3509 {
3510 
3511 	return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length));
3512 }
3513 
3514 int
3515 freebsd6_ftruncate(struct thread *td, struct freebsd6_ftruncate_args *uap)
3516 {
3517 
3518 	return (kern_ftruncate(td, uap->fd, uap->length));
3519 }
3520 #endif
3521 
3522 int
3523 kern_fsync(struct thread *td, int fd, bool fullsync)
3524 {
3525 	struct vnode *vp;
3526 	struct mount *mp;
3527 	struct file *fp;
3528 	int error;
3529 
3530 	AUDIT_ARG_FD(fd);
3531 	error = getvnode(td, fd, &cap_fsync_rights, &fp);
3532 	if (error != 0)
3533 		return (error);
3534 	vp = fp->f_vnode;
3535 #if 0
3536 	if (!fullsync)
3537 		/* XXXKIB: compete outstanding aio writes */;
3538 #endif
3539 retry:
3540 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
3541 	if (error != 0)
3542 		goto drop;
3543 	vn_lock(vp, vn_lktype_write(mp, vp) | LK_RETRY);
3544 	AUDIT_ARG_VNODE1(vp);
3545 	if (vp->v_object != NULL) {
3546 		VM_OBJECT_WLOCK(vp->v_object);
3547 		vm_object_page_clean(vp->v_object, 0, 0, 0);
3548 		VM_OBJECT_WUNLOCK(vp->v_object);
3549 	}
3550 	error = fullsync ? VOP_FSYNC(vp, MNT_WAIT, td) : VOP_FDATASYNC(vp, td);
3551 	VOP_UNLOCK(vp);
3552 	vn_finished_write(mp);
3553 	if (error == ERELOOKUP)
3554 		goto retry;
3555 drop:
3556 	fdrop(fp, td);
3557 	return (error);
3558 }
3559 
3560 /*
3561  * Sync an open file.
3562  */
3563 #ifndef _SYS_SYSPROTO_H_
3564 struct fsync_args {
3565 	int	fd;
3566 };
3567 #endif
3568 int
3569 sys_fsync(struct thread *td, struct fsync_args *uap)
3570 {
3571 
3572 	return (kern_fsync(td, uap->fd, true));
3573 }
3574 
3575 int
3576 sys_fdatasync(struct thread *td, struct fdatasync_args *uap)
3577 {
3578 
3579 	return (kern_fsync(td, uap->fd, false));
3580 }
3581 
3582 /*
3583  * Rename files.  Source and destination must either both be directories, or
3584  * both not be directories.  If target is a directory, it must be empty.
3585  */
3586 #ifndef _SYS_SYSPROTO_H_
3587 struct rename_args {
3588 	char	*from;
3589 	char	*to;
3590 };
3591 #endif
3592 int
3593 sys_rename(struct thread *td, struct rename_args *uap)
3594 {
3595 
3596 	return (kern_renameat(td, AT_FDCWD, uap->from, AT_FDCWD,
3597 	    uap->to, UIO_USERSPACE));
3598 }
3599 
3600 #ifndef _SYS_SYSPROTO_H_
3601 struct renameat_args {
3602 	int	oldfd;
3603 	char	*old;
3604 	int	newfd;
3605 	char	*new;
3606 };
3607 #endif
3608 int
3609 sys_renameat(struct thread *td, struct renameat_args *uap)
3610 {
3611 
3612 	return (kern_renameat(td, uap->oldfd, uap->old, uap->newfd, uap->new,
3613 	    UIO_USERSPACE));
3614 }
3615 
3616 #ifdef MAC
3617 static int
3618 kern_renameat_mac(struct thread *td, int oldfd, const char *old, int newfd,
3619     const char *new, enum uio_seg pathseg, struct nameidata *fromnd)
3620 {
3621 	int error;
3622 
3623 	NDINIT_ATRIGHTS(fromnd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART |
3624 	    AUDITVNODE1, pathseg, old, oldfd, &cap_renameat_source_rights, td);
3625 	if ((error = namei(fromnd)) != 0)
3626 		return (error);
3627 	error = mac_vnode_check_rename_from(td->td_ucred, fromnd->ni_dvp,
3628 	    fromnd->ni_vp, &fromnd->ni_cnd);
3629 	VOP_UNLOCK(fromnd->ni_dvp);
3630 	if (fromnd->ni_dvp != fromnd->ni_vp)
3631 		VOP_UNLOCK(fromnd->ni_vp);
3632 	if (error != 0) {
3633 		NDFREE(fromnd, NDF_ONLY_PNBUF);
3634 		vrele(fromnd->ni_dvp);
3635 		vrele(fromnd->ni_vp);
3636 		if (fromnd->ni_startdir)
3637 			vrele(fromnd->ni_startdir);
3638 	}
3639 	return (error);
3640 }
3641 #endif
3642 
3643 int
3644 kern_renameat(struct thread *td, int oldfd, const char *old, int newfd,
3645     const char *new, enum uio_seg pathseg)
3646 {
3647 	struct mount *mp = NULL;
3648 	struct vnode *tvp, *fvp, *tdvp;
3649 	struct nameidata fromnd, tond;
3650 	u_int64_t tondflags;
3651 	int error;
3652 
3653 again:
3654 	bwillwrite();
3655 #ifdef MAC
3656 	if (mac_vnode_check_rename_from_enabled()) {
3657 		error = kern_renameat_mac(td, oldfd, old, newfd, new, pathseg,
3658 		    &fromnd);
3659 		if (error != 0)
3660 			return (error);
3661 	} else {
3662 #endif
3663 	NDINIT_ATRIGHTS(&fromnd, DELETE, WANTPARENT | SAVESTART | AUDITVNODE1,
3664 	    pathseg, old, oldfd, &cap_renameat_source_rights, td);
3665 	if ((error = namei(&fromnd)) != 0)
3666 		return (error);
3667 #ifdef MAC
3668 	}
3669 #endif
3670 	fvp = fromnd.ni_vp;
3671 	tondflags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNODE2;
3672 	if (fromnd.ni_vp->v_type == VDIR)
3673 		tondflags |= WILLBEDIR;
3674 	NDINIT_ATRIGHTS(&tond, RENAME, tondflags, pathseg, new, newfd,
3675 	    &cap_renameat_target_rights, td);
3676 	if ((error = namei(&tond)) != 0) {
3677 		/* Translate error code for rename("dir1", "dir2/."). */
3678 		if (error == EISDIR && fvp->v_type == VDIR)
3679 			error = EINVAL;
3680 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3681 		vrele(fromnd.ni_dvp);
3682 		vrele(fvp);
3683 		goto out1;
3684 	}
3685 	tdvp = tond.ni_dvp;
3686 	tvp = tond.ni_vp;
3687 	error = vn_start_write(fvp, &mp, V_NOWAIT);
3688 	if (error != 0) {
3689 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3690 		NDFREE(&tond, NDF_ONLY_PNBUF);
3691 		if (tvp != NULL)
3692 			vput(tvp);
3693 		if (tdvp == tvp)
3694 			vrele(tdvp);
3695 		else
3696 			vput(tdvp);
3697 		vrele(fromnd.ni_dvp);
3698 		vrele(fvp);
3699 		vrele(tond.ni_startdir);
3700 		if (fromnd.ni_startdir != NULL)
3701 			vrele(fromnd.ni_startdir);
3702 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
3703 		if (error != 0)
3704 			return (error);
3705 		goto again;
3706 	}
3707 	if (tvp != NULL) {
3708 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
3709 			error = ENOTDIR;
3710 			goto out;
3711 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
3712 			error = EISDIR;
3713 			goto out;
3714 		}
3715 #ifdef CAPABILITIES
3716 		if (newfd != AT_FDCWD && (tond.ni_resflags & NIRES_ABS) == 0) {
3717 			/*
3718 			 * If the target already exists we require CAP_UNLINKAT
3719 			 * from 'newfd', when newfd was used for the lookup.
3720 			 */
3721 			error = cap_check(&tond.ni_filecaps.fc_rights,
3722 			    &cap_unlinkat_rights);
3723 			if (error != 0)
3724 				goto out;
3725 		}
3726 #endif
3727 	}
3728 	if (fvp == tdvp) {
3729 		error = EINVAL;
3730 		goto out;
3731 	}
3732 	/*
3733 	 * If the source is the same as the destination (that is, if they
3734 	 * are links to the same vnode), then there is nothing to do.
3735 	 */
3736 	if (fvp == tvp)
3737 		error = ERESTART;
3738 #ifdef MAC
3739 	else
3740 		error = mac_vnode_check_rename_to(td->td_ucred, tdvp,
3741 		    tond.ni_vp, fromnd.ni_dvp == tdvp, &tond.ni_cnd);
3742 #endif
3743 out:
3744 	if (error == 0) {
3745 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
3746 		    tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
3747 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3748 		NDFREE(&tond, NDF_ONLY_PNBUF);
3749 	} else {
3750 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3751 		NDFREE(&tond, NDF_ONLY_PNBUF);
3752 		if (tvp != NULL)
3753 			vput(tvp);
3754 		if (tdvp == tvp)
3755 			vrele(tdvp);
3756 		else
3757 			vput(tdvp);
3758 		vrele(fromnd.ni_dvp);
3759 		vrele(fvp);
3760 	}
3761 	vrele(tond.ni_startdir);
3762 	vn_finished_write(mp);
3763 out1:
3764 	if (fromnd.ni_startdir)
3765 		vrele(fromnd.ni_startdir);
3766 	if (error == ERESTART)
3767 		return (0);
3768 	if (error == ERELOOKUP)
3769 		goto again;
3770 	return (error);
3771 }
3772 
3773 /*
3774  * Make a directory file.
3775  */
3776 #ifndef _SYS_SYSPROTO_H_
3777 struct mkdir_args {
3778 	char	*path;
3779 	int	mode;
3780 };
3781 #endif
3782 int
3783 sys_mkdir(struct thread *td, struct mkdir_args *uap)
3784 {
3785 
3786 	return (kern_mkdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
3787 	    uap->mode));
3788 }
3789 
3790 #ifndef _SYS_SYSPROTO_H_
3791 struct mkdirat_args {
3792 	int	fd;
3793 	char	*path;
3794 	mode_t	mode;
3795 };
3796 #endif
3797 int
3798 sys_mkdirat(struct thread *td, struct mkdirat_args *uap)
3799 {
3800 
3801 	return (kern_mkdirat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode));
3802 }
3803 
3804 int
3805 kern_mkdirat(struct thread *td, int fd, const char *path, enum uio_seg segflg,
3806     int mode)
3807 {
3808 	struct mount *mp;
3809 	struct vattr vattr;
3810 	struct nameidata nd;
3811 	int error;
3812 
3813 	AUDIT_ARG_MODE(mode);
3814 	NDPREINIT(&nd);
3815 restart:
3816 	bwillwrite();
3817 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
3818 	    NC_NOMAKEENTRY | NC_KEEPPOSENTRY | FAILIFEXISTS | WILLBEDIR,
3819 	    segflg, path, fd, &cap_mkdirat_rights, td);
3820 	if ((error = namei(&nd)) != 0)
3821 		return (error);
3822 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
3823 		NDFREE(&nd, NDF_ONLY_PNBUF);
3824 		vput(nd.ni_dvp);
3825 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
3826 			return (error);
3827 		goto restart;
3828 	}
3829 	VATTR_NULL(&vattr);
3830 	vattr.va_type = VDIR;
3831 	vattr.va_mode = (mode & ACCESSPERMS) &~ td->td_proc->p_pd->pd_cmask;
3832 #ifdef MAC
3833 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
3834 	    &vattr);
3835 	if (error != 0)
3836 		goto out;
3837 #endif
3838 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
3839 #ifdef MAC
3840 out:
3841 #endif
3842 	NDFREE(&nd, NDF_ONLY_PNBUF);
3843 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true);
3844 	vn_finished_write(mp);
3845 	if (error == ERELOOKUP)
3846 		goto restart;
3847 	return (error);
3848 }
3849 
3850 /*
3851  * Remove a directory file.
3852  */
3853 #ifndef _SYS_SYSPROTO_H_
3854 struct rmdir_args {
3855 	char	*path;
3856 };
3857 #endif
3858 int
3859 sys_rmdir(struct thread *td, struct rmdir_args *uap)
3860 {
3861 
3862 	return (kern_frmdirat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE,
3863 	    0));
3864 }
3865 
3866 int
3867 kern_frmdirat(struct thread *td, int dfd, const char *path, int fd,
3868     enum uio_seg pathseg, int flag)
3869 {
3870 	struct mount *mp;
3871 	struct vnode *vp;
3872 	struct file *fp;
3873 	struct nameidata nd;
3874 	cap_rights_t rights;
3875 	int error;
3876 
3877 	fp = NULL;
3878 	if (fd != FD_NONE) {
3879 		error = getvnode(td, fd, cap_rights_init_one(&rights,
3880 		    CAP_LOOKUP), &fp);
3881 		if (error != 0)
3882 			return (error);
3883 	}
3884 
3885 	NDPREINIT(&nd);
3886 restart:
3887 	bwillwrite();
3888 	NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 |
3889 	    at2cnpflags(flag, AT_RESOLVE_BENEATH),
3890 	    pathseg, path, dfd, &cap_unlinkat_rights, td);
3891 	if ((error = namei(&nd)) != 0)
3892 		goto fdout;
3893 	vp = nd.ni_vp;
3894 	if (vp->v_type != VDIR) {
3895 		error = ENOTDIR;
3896 		goto out;
3897 	}
3898 	/*
3899 	 * No rmdir "." please.
3900 	 */
3901 	if (nd.ni_dvp == vp) {
3902 		error = EINVAL;
3903 		goto out;
3904 	}
3905 	/*
3906 	 * The root of a mounted filesystem cannot be deleted.
3907 	 */
3908 	if (vp->v_vflag & VV_ROOT) {
3909 		error = EBUSY;
3910 		goto out;
3911 	}
3912 
3913 	if (fp != NULL && fp->f_vnode != vp) {
3914 		if (VN_IS_DOOMED(fp->f_vnode))
3915 			error = EBADF;
3916 		else
3917 			error = EDEADLK;
3918 		goto out;
3919 	}
3920 
3921 #ifdef MAC
3922 	error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp,
3923 	    &nd.ni_cnd);
3924 	if (error != 0)
3925 		goto out;
3926 #endif
3927 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
3928 		NDFREE(&nd, NDF_ONLY_PNBUF);
3929 		vput(vp);
3930 		if (nd.ni_dvp == vp)
3931 			vrele(nd.ni_dvp);
3932 		else
3933 			vput(nd.ni_dvp);
3934 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
3935 			goto fdout;
3936 		goto restart;
3937 	}
3938 	vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK);
3939 	error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
3940 	vn_finished_write(mp);
3941 out:
3942 	NDFREE(&nd, NDF_ONLY_PNBUF);
3943 	vput(vp);
3944 	if (nd.ni_dvp == vp)
3945 		vrele(nd.ni_dvp);
3946 	else
3947 		vput(nd.ni_dvp);
3948 	if (error == ERELOOKUP)
3949 		goto restart;
3950 fdout:
3951 	if (fp != NULL)
3952 		fdrop(fp, td);
3953 	return (error);
3954 }
3955 
3956 #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11)
3957 int
3958 freebsd11_kern_getdirentries(struct thread *td, int fd, char *ubuf, u_int count,
3959     long *basep, void (*func)(struct freebsd11_dirent *))
3960 {
3961 	struct freebsd11_dirent dstdp;
3962 	struct dirent *dp, *edp;
3963 	char *dirbuf;
3964 	off_t base;
3965 	ssize_t resid, ucount;
3966 	int error;
3967 
3968 	/* XXX arbitrary sanity limit on `count'. */
3969 	count = min(count, 64 * 1024);
3970 
3971 	dirbuf = malloc(count, M_TEMP, M_WAITOK);
3972 
3973 	error = kern_getdirentries(td, fd, dirbuf, count, &base, &resid,
3974 	    UIO_SYSSPACE);
3975 	if (error != 0)
3976 		goto done;
3977 	if (basep != NULL)
3978 		*basep = base;
3979 
3980 	ucount = 0;
3981 	for (dp = (struct dirent *)dirbuf,
3982 	    edp = (struct dirent *)&dirbuf[count - resid];
3983 	    ucount < count && dp < edp; ) {
3984 		if (dp->d_reclen == 0)
3985 			break;
3986 		MPASS(dp->d_reclen >= _GENERIC_DIRLEN(0));
3987 		if (dp->d_namlen >= sizeof(dstdp.d_name))
3988 			continue;
3989 		dstdp.d_type = dp->d_type;
3990 		dstdp.d_namlen = dp->d_namlen;
3991 		dstdp.d_fileno = dp->d_fileno;		/* truncate */
3992 		if (dstdp.d_fileno != dp->d_fileno) {
3993 			switch (ino64_trunc_error) {
3994 			default:
3995 			case 0:
3996 				break;
3997 			case 1:
3998 				error = EOVERFLOW;
3999 				goto done;
4000 			case 2:
4001 				dstdp.d_fileno = UINT32_MAX;
4002 				break;
4003 			}
4004 		}
4005 		dstdp.d_reclen = sizeof(dstdp) - sizeof(dstdp.d_name) +
4006 		    ((dp->d_namlen + 1 + 3) &~ 3);
4007 		bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen);
4008 		bzero(dstdp.d_name + dstdp.d_namlen,
4009 		    dstdp.d_reclen - offsetof(struct freebsd11_dirent, d_name) -
4010 		    dstdp.d_namlen);
4011 		MPASS(dstdp.d_reclen <= dp->d_reclen);
4012 		MPASS(ucount + dstdp.d_reclen <= count);
4013 		if (func != NULL)
4014 			func(&dstdp);
4015 		error = copyout(&dstdp, ubuf + ucount, dstdp.d_reclen);
4016 		if (error != 0)
4017 			break;
4018 		dp = (struct dirent *)((char *)dp + dp->d_reclen);
4019 		ucount += dstdp.d_reclen;
4020 	}
4021 
4022 done:
4023 	free(dirbuf, M_TEMP);
4024 	if (error == 0)
4025 		td->td_retval[0] = ucount;
4026 	return (error);
4027 }
4028 #endif /* COMPAT */
4029 
4030 #ifdef COMPAT_43
4031 static void
4032 ogetdirentries_cvt(struct freebsd11_dirent *dp)
4033 {
4034 #if (BYTE_ORDER == LITTLE_ENDIAN)
4035 	/*
4036 	 * The expected low byte of dp->d_namlen is our dp->d_type.
4037 	 * The high MBZ byte of dp->d_namlen is our dp->d_namlen.
4038 	 */
4039 	dp->d_type = dp->d_namlen;
4040 	dp->d_namlen = 0;
4041 #else
4042 	/*
4043 	 * The dp->d_type is the high byte of the expected dp->d_namlen,
4044 	 * so must be zero'ed.
4045 	 */
4046 	dp->d_type = 0;
4047 #endif
4048 }
4049 
4050 /*
4051  * Read a block of directory entries in a filesystem independent format.
4052  */
4053 #ifndef _SYS_SYSPROTO_H_
4054 struct ogetdirentries_args {
4055 	int	fd;
4056 	char	*buf;
4057 	u_int	count;
4058 	long	*basep;
4059 };
4060 #endif
4061 int
4062 ogetdirentries(struct thread *td, struct ogetdirentries_args *uap)
4063 {
4064 	long loff;
4065 	int error;
4066 
4067 	error = kern_ogetdirentries(td, uap, &loff);
4068 	if (error == 0)
4069 		error = copyout(&loff, uap->basep, sizeof(long));
4070 	return (error);
4071 }
4072 
4073 int
4074 kern_ogetdirentries(struct thread *td, struct ogetdirentries_args *uap,
4075     long *ploff)
4076 {
4077 	long base;
4078 	int error;
4079 
4080 	/* XXX arbitrary sanity limit on `count'. */
4081 	if (uap->count > 64 * 1024)
4082 		return (EINVAL);
4083 
4084 	error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count,
4085 	    &base, ogetdirentries_cvt);
4086 
4087 	if (error == 0 && uap->basep != NULL)
4088 		error = copyout(&base, uap->basep, sizeof(long));
4089 
4090 	return (error);
4091 }
4092 #endif /* COMPAT_43 */
4093 
4094 #if defined(COMPAT_FREEBSD11)
4095 #ifndef _SYS_SYSPROTO_H_
4096 struct freebsd11_getdirentries_args {
4097 	int	fd;
4098 	char	*buf;
4099 	u_int	count;
4100 	long	*basep;
4101 };
4102 #endif
4103 int
4104 freebsd11_getdirentries(struct thread *td,
4105     struct freebsd11_getdirentries_args *uap)
4106 {
4107 	long base;
4108 	int error;
4109 
4110 	error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count,
4111 	    &base, NULL);
4112 
4113 	if (error == 0 && uap->basep != NULL)
4114 		error = copyout(&base, uap->basep, sizeof(long));
4115 	return (error);
4116 }
4117 
4118 int
4119 freebsd11_getdents(struct thread *td, struct freebsd11_getdents_args *uap)
4120 {
4121 	struct freebsd11_getdirentries_args ap;
4122 
4123 	ap.fd = uap->fd;
4124 	ap.buf = uap->buf;
4125 	ap.count = uap->count;
4126 	ap.basep = NULL;
4127 	return (freebsd11_getdirentries(td, &ap));
4128 }
4129 #endif /* COMPAT_FREEBSD11 */
4130 
4131 /*
4132  * Read a block of directory entries in a filesystem independent format.
4133  */
4134 int
4135 sys_getdirentries(struct thread *td, struct getdirentries_args *uap)
4136 {
4137 	off_t base;
4138 	int error;
4139 
4140 	error = kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base,
4141 	    NULL, UIO_USERSPACE);
4142 	if (error != 0)
4143 		return (error);
4144 	if (uap->basep != NULL)
4145 		error = copyout(&base, uap->basep, sizeof(off_t));
4146 	return (error);
4147 }
4148 
4149 int
4150 kern_getdirentries(struct thread *td, int fd, char *buf, size_t count,
4151     off_t *basep, ssize_t *residp, enum uio_seg bufseg)
4152 {
4153 	struct vnode *vp;
4154 	struct file *fp;
4155 	struct uio auio;
4156 	struct iovec aiov;
4157 	off_t loff;
4158 	int error, eofflag;
4159 	off_t foffset;
4160 
4161 	AUDIT_ARG_FD(fd);
4162 	if (count > IOSIZE_MAX)
4163 		return (EINVAL);
4164 	auio.uio_resid = count;
4165 	error = getvnode(td, fd, &cap_read_rights, &fp);
4166 	if (error != 0)
4167 		return (error);
4168 	if ((fp->f_flag & FREAD) == 0) {
4169 		fdrop(fp, td);
4170 		return (EBADF);
4171 	}
4172 	vp = fp->f_vnode;
4173 	foffset = foffset_lock(fp, 0);
4174 unionread:
4175 	if (vp->v_type != VDIR) {
4176 		error = EINVAL;
4177 		goto fail;
4178 	}
4179 	aiov.iov_base = buf;
4180 	aiov.iov_len = count;
4181 	auio.uio_iov = &aiov;
4182 	auio.uio_iovcnt = 1;
4183 	auio.uio_rw = UIO_READ;
4184 	auio.uio_segflg = bufseg;
4185 	auio.uio_td = td;
4186 	vn_lock(vp, LK_SHARED | LK_RETRY);
4187 	AUDIT_ARG_VNODE1(vp);
4188 	loff = auio.uio_offset = foffset;
4189 #ifdef MAC
4190 	error = mac_vnode_check_readdir(td->td_ucred, vp);
4191 	if (error == 0)
4192 #endif
4193 		error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL,
4194 		    NULL);
4195 	foffset = auio.uio_offset;
4196 	if (error != 0) {
4197 		VOP_UNLOCK(vp);
4198 		goto fail;
4199 	}
4200 	if (count == auio.uio_resid &&
4201 	    (vp->v_vflag & VV_ROOT) &&
4202 	    (vp->v_mount->mnt_flag & MNT_UNION)) {
4203 		struct vnode *tvp = vp;
4204 
4205 		vp = vp->v_mount->mnt_vnodecovered;
4206 		VREF(vp);
4207 		fp->f_vnode = vp;
4208 		foffset = 0;
4209 		vput(tvp);
4210 		goto unionread;
4211 	}
4212 	VOP_UNLOCK(vp);
4213 	*basep = loff;
4214 	if (residp != NULL)
4215 		*residp = auio.uio_resid;
4216 	td->td_retval[0] = count - auio.uio_resid;
4217 fail:
4218 	foffset_unlock(fp, foffset, 0);
4219 	fdrop(fp, td);
4220 	return (error);
4221 }
4222 
4223 /*
4224  * Set the mode mask for creation of filesystem nodes.
4225  */
4226 #ifndef _SYS_SYSPROTO_H_
4227 struct umask_args {
4228 	int	newmask;
4229 };
4230 #endif
4231 int
4232 sys_umask(struct thread *td, struct umask_args *uap)
4233 {
4234 	struct pwddesc *pdp;
4235 
4236 	pdp = td->td_proc->p_pd;
4237 	PWDDESC_XLOCK(pdp);
4238 	td->td_retval[0] = pdp->pd_cmask;
4239 	pdp->pd_cmask = uap->newmask & ALLPERMS;
4240 	PWDDESC_XUNLOCK(pdp);
4241 	return (0);
4242 }
4243 
4244 /*
4245  * Void all references to file by ripping underlying filesystem away from
4246  * vnode.
4247  */
4248 #ifndef _SYS_SYSPROTO_H_
4249 struct revoke_args {
4250 	char	*path;
4251 };
4252 #endif
4253 int
4254 sys_revoke(struct thread *td, struct revoke_args *uap)
4255 {
4256 	struct vnode *vp;
4257 	struct vattr vattr;
4258 	struct nameidata nd;
4259 	int error;
4260 
4261 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE,
4262 	    uap->path, td);
4263 	if ((error = namei(&nd)) != 0)
4264 		return (error);
4265 	vp = nd.ni_vp;
4266 	NDFREE_NOTHING(&nd);
4267 	if (vp->v_type != VCHR || vp->v_rdev == NULL) {
4268 		error = EINVAL;
4269 		goto out;
4270 	}
4271 #ifdef MAC
4272 	error = mac_vnode_check_revoke(td->td_ucred, vp);
4273 	if (error != 0)
4274 		goto out;
4275 #endif
4276 	error = VOP_GETATTR(vp, &vattr, td->td_ucred);
4277 	if (error != 0)
4278 		goto out;
4279 	if (td->td_ucred->cr_uid != vattr.va_uid) {
4280 		error = priv_check(td, PRIV_VFS_ADMIN);
4281 		if (error != 0)
4282 			goto out;
4283 	}
4284 	if (devfs_usecount(vp) > 0)
4285 		VOP_REVOKE(vp, REVOKEALL);
4286 out:
4287 	vput(vp);
4288 	return (error);
4289 }
4290 
4291 /*
4292  * This variant of getvnode() allows O_PATH files.  Caller should
4293  * ensure that returned file and vnode are only used for compatible
4294  * semantics.
4295  */
4296 int
4297 getvnode_path(struct thread *td, int fd, cap_rights_t *rightsp,
4298     struct file **fpp)
4299 {
4300 	struct file *fp;
4301 	int error;
4302 
4303 	error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, &fp);
4304 	if (error != 0)
4305 		return (error);
4306 
4307 	/*
4308 	 * The file could be not of the vnode type, or it may be not
4309 	 * yet fully initialized, in which case the f_vnode pointer
4310 	 * may be set, but f_ops is still badfileops.  E.g.,
4311 	 * devfs_open() transiently create such situation to
4312 	 * facilitate csw d_fdopen().
4313 	 *
4314 	 * Dupfdopen() handling in kern_openat() installs the
4315 	 * half-baked file into the process descriptor table, allowing
4316 	 * other thread to dereference it. Guard against the race by
4317 	 * checking f_ops.
4318 	 */
4319 	if (fp->f_vnode == NULL || fp->f_ops == &badfileops) {
4320 		fdrop(fp, td);
4321 		return (EINVAL);
4322 	}
4323 
4324 	*fpp = fp;
4325 	return (0);
4326 }
4327 
4328 /*
4329  * Convert a user file descriptor to a kernel file entry and check
4330  * that, if it is a capability, the correct rights are present.
4331  * A reference on the file entry is held upon returning.
4332  */
4333 int
4334 getvnode(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
4335 {
4336 	int error;
4337 
4338 	error = getvnode_path(td, fd, rightsp, fpp);
4339 
4340 	/*
4341 	 * Filter out O_PATH file descriptors, most getvnode() callers
4342 	 * do not call fo_ methods.
4343 	 */
4344 	if (error == 0 && (*fpp)->f_ops == &path_fileops) {
4345 		fdrop(*fpp, td);
4346 		error = EBADF;
4347 	}
4348 
4349 	return (error);
4350 }
4351 
4352 /*
4353  * Get an (NFS) file handle.
4354  */
4355 #ifndef _SYS_SYSPROTO_H_
4356 struct lgetfh_args {
4357 	char *fname;
4358 	fhandle_t *fhp;
4359 };
4360 #endif
4361 int
4362 sys_lgetfh(struct thread *td, struct lgetfh_args *uap)
4363 {
4364 
4365 	return (kern_getfhat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->fname,
4366 	    UIO_USERSPACE, uap->fhp, UIO_USERSPACE));
4367 }
4368 
4369 #ifndef _SYS_SYSPROTO_H_
4370 struct getfh_args {
4371 	char *fname;
4372 	fhandle_t *fhp;
4373 };
4374 #endif
4375 int
4376 sys_getfh(struct thread *td, struct getfh_args *uap)
4377 {
4378 
4379 	return (kern_getfhat(td, 0, AT_FDCWD, uap->fname, UIO_USERSPACE,
4380 	    uap->fhp, UIO_USERSPACE));
4381 }
4382 
4383 /*
4384  * syscall for the rpc.lockd to use to translate an open descriptor into
4385  * a NFS file handle.
4386  *
4387  * warning: do not remove the priv_check() call or this becomes one giant
4388  * security hole.
4389  */
4390 #ifndef _SYS_SYSPROTO_H_
4391 struct getfhat_args {
4392 	int fd;
4393 	char *path;
4394 	fhandle_t *fhp;
4395 	int flags;
4396 };
4397 #endif
4398 int
4399 sys_getfhat(struct thread *td, struct getfhat_args *uap)
4400 {
4401 
4402 	if ((uap->flags & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH)) != 0)
4403 		return (EINVAL);
4404 	return (kern_getfhat(td, uap->flags, uap->fd, uap->path, UIO_USERSPACE,
4405 	    uap->fhp, UIO_USERSPACE));
4406 }
4407 
4408 int
4409 kern_getfhat(struct thread *td, int flags, int fd, const char *path,
4410     enum uio_seg pathseg, fhandle_t *fhp, enum uio_seg fhseg)
4411 {
4412 	struct nameidata nd;
4413 	fhandle_t fh;
4414 	struct vnode *vp;
4415 	int error;
4416 
4417 	error = priv_check(td, PRIV_VFS_GETFH);
4418 	if (error != 0)
4419 		return (error);
4420 	NDINIT_AT(&nd, LOOKUP, at2cnpflags(flags, AT_SYMLINK_NOFOLLOW |
4421 	    AT_RESOLVE_BENEATH) | LOCKLEAF | AUDITVNODE1, pathseg, path,
4422 	    fd, td);
4423 	error = namei(&nd);
4424 	if (error != 0)
4425 		return (error);
4426 	NDFREE_NOTHING(&nd);
4427 	vp = nd.ni_vp;
4428 	bzero(&fh, sizeof(fh));
4429 	fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
4430 	error = VOP_VPTOFH(vp, &fh.fh_fid);
4431 	vput(vp);
4432 	if (error == 0) {
4433 		if (fhseg == UIO_USERSPACE)
4434 			error = copyout(&fh, fhp, sizeof (fh));
4435 		else
4436 			memcpy(fhp, &fh, sizeof(fh));
4437 	}
4438 	return (error);
4439 }
4440 
4441 #ifndef _SYS_SYSPROTO_H_
4442 struct fhlink_args {
4443 	fhandle_t *fhp;
4444 	const char *to;
4445 };
4446 #endif
4447 int
4448 sys_fhlink(struct thread *td, struct fhlink_args *uap)
4449 {
4450 
4451 	return (kern_fhlinkat(td, AT_FDCWD, uap->to, UIO_USERSPACE, uap->fhp));
4452 }
4453 
4454 #ifndef _SYS_SYSPROTO_H_
4455 struct fhlinkat_args {
4456 	fhandle_t *fhp;
4457 	int tofd;
4458 	const char *to;
4459 };
4460 #endif
4461 int
4462 sys_fhlinkat(struct thread *td, struct fhlinkat_args *uap)
4463 {
4464 
4465 	return (kern_fhlinkat(td, uap->tofd, uap->to, UIO_USERSPACE, uap->fhp));
4466 }
4467 
4468 static int
4469 kern_fhlinkat(struct thread *td, int fd, const char *path,
4470     enum uio_seg pathseg, fhandle_t *fhp)
4471 {
4472 	fhandle_t fh;
4473 	struct mount *mp;
4474 	struct vnode *vp;
4475 	int error;
4476 
4477 	error = priv_check(td, PRIV_VFS_GETFH);
4478 	if (error != 0)
4479 		return (error);
4480 	error = copyin(fhp, &fh, sizeof(fh));
4481 	if (error != 0)
4482 		return (error);
4483 	do {
4484 		bwillwrite();
4485 		if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4486 			return (ESTALE);
4487 		error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp);
4488 		vfs_unbusy(mp);
4489 		if (error != 0)
4490 			return (error);
4491 		VOP_UNLOCK(vp);
4492 		error = kern_linkat_vp(td, vp, fd, path, pathseg);
4493 	} while (error == EAGAIN || error == ERELOOKUP);
4494 	return (error);
4495 }
4496 
4497 #ifndef _SYS_SYSPROTO_H_
4498 struct fhreadlink_args {
4499 	fhandle_t *fhp;
4500 	char *buf;
4501 	size_t bufsize;
4502 };
4503 #endif
4504 int
4505 sys_fhreadlink(struct thread *td, struct fhreadlink_args *uap)
4506 {
4507 	fhandle_t fh;
4508 	struct mount *mp;
4509 	struct vnode *vp;
4510 	int error;
4511 
4512 	error = priv_check(td, PRIV_VFS_GETFH);
4513 	if (error != 0)
4514 		return (error);
4515 	if (uap->bufsize > IOSIZE_MAX)
4516 		return (EINVAL);
4517 	error = copyin(uap->fhp, &fh, sizeof(fh));
4518 	if (error != 0)
4519 		return (error);
4520 	if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4521 		return (ESTALE);
4522 	error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp);
4523 	vfs_unbusy(mp);
4524 	if (error != 0)
4525 		return (error);
4526 	error = kern_readlink_vp(vp, uap->buf, UIO_USERSPACE, uap->bufsize, td);
4527 	vput(vp);
4528 	return (error);
4529 }
4530 
4531 /*
4532  * syscall for the rpc.lockd to use to translate a NFS file handle into an
4533  * open descriptor.
4534  *
4535  * warning: do not remove the priv_check() call or this becomes one giant
4536  * security hole.
4537  */
4538 #ifndef _SYS_SYSPROTO_H_
4539 struct fhopen_args {
4540 	const struct fhandle *u_fhp;
4541 	int flags;
4542 };
4543 #endif
4544 int
4545 sys_fhopen(struct thread *td, struct fhopen_args *uap)
4546 {
4547 	return (kern_fhopen(td, uap->u_fhp, uap->flags));
4548 }
4549 
4550 int
4551 kern_fhopen(struct thread *td, const struct fhandle *u_fhp, int flags)
4552 {
4553 	struct mount *mp;
4554 	struct vnode *vp;
4555 	struct fhandle fhp;
4556 	struct file *fp;
4557 	int fmode, error;
4558 	int indx;
4559 
4560 	error = priv_check(td, PRIV_VFS_FHOPEN);
4561 	if (error != 0)
4562 		return (error);
4563 	indx = -1;
4564 	fmode = FFLAGS(flags);
4565 	/* why not allow a non-read/write open for our lockd? */
4566 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4567 		return (EINVAL);
4568 	error = copyin(u_fhp, &fhp, sizeof(fhp));
4569 	if (error != 0)
4570 		return(error);
4571 	/* find the mount point */
4572 	mp = vfs_busyfs(&fhp.fh_fsid);
4573 	if (mp == NULL)
4574 		return (ESTALE);
4575 	/* now give me my vnode, it gets returned to me locked */
4576 	error = VFS_FHTOVP(mp, &fhp.fh_fid, LK_EXCLUSIVE, &vp);
4577 	vfs_unbusy(mp);
4578 	if (error != 0)
4579 		return (error);
4580 
4581 	error = falloc_noinstall(td, &fp);
4582 	if (error != 0) {
4583 		vput(vp);
4584 		return (error);
4585 	}
4586 	/*
4587 	 * An extra reference on `fp' has been held for us by
4588 	 * falloc_noinstall().
4589 	 */
4590 
4591 #ifdef INVARIANTS
4592 	td->td_dupfd = -1;
4593 #endif
4594 	error = vn_open_vnode(vp, fmode, td->td_ucred, td, fp);
4595 	if (error != 0) {
4596 		KASSERT(fp->f_ops == &badfileops,
4597 		    ("VOP_OPEN in fhopen() set f_ops"));
4598 		KASSERT(td->td_dupfd < 0,
4599 		    ("fhopen() encountered fdopen()"));
4600 
4601 		vput(vp);
4602 		goto bad;
4603 	}
4604 #ifdef INVARIANTS
4605 	td->td_dupfd = 0;
4606 #endif
4607 	fp->f_vnode = vp;
4608 	finit_vnode(fp, fmode, NULL, &vnops);
4609 	VOP_UNLOCK(vp);
4610 	if ((fmode & O_TRUNC) != 0) {
4611 		error = fo_truncate(fp, 0, td->td_ucred, td);
4612 		if (error != 0)
4613 			goto bad;
4614 	}
4615 
4616 	error = finstall(td, fp, &indx, fmode, NULL);
4617 bad:
4618 	fdrop(fp, td);
4619 	td->td_retval[0] = indx;
4620 	return (error);
4621 }
4622 
4623 /*
4624  * Stat an (NFS) file handle.
4625  */
4626 #ifndef _SYS_SYSPROTO_H_
4627 struct fhstat_args {
4628 	struct fhandle *u_fhp;
4629 	struct stat *sb;
4630 };
4631 #endif
4632 int
4633 sys_fhstat(struct thread *td, struct fhstat_args *uap)
4634 {
4635 	struct stat sb;
4636 	struct fhandle fh;
4637 	int error;
4638 
4639 	error = copyin(uap->u_fhp, &fh, sizeof(fh));
4640 	if (error != 0)
4641 		return (error);
4642 	error = kern_fhstat(td, fh, &sb);
4643 	if (error == 0)
4644 		error = copyout(&sb, uap->sb, sizeof(sb));
4645 	return (error);
4646 }
4647 
4648 int
4649 kern_fhstat(struct thread *td, struct fhandle fh, struct stat *sb)
4650 {
4651 	struct mount *mp;
4652 	struct vnode *vp;
4653 	int error;
4654 
4655 	error = priv_check(td, PRIV_VFS_FHSTAT);
4656 	if (error != 0)
4657 		return (error);
4658 	if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4659 		return (ESTALE);
4660 	error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp);
4661 	vfs_unbusy(mp);
4662 	if (error != 0)
4663 		return (error);
4664 	error = VOP_STAT(vp, sb, td->td_ucred, NOCRED, td);
4665 	vput(vp);
4666 	return (error);
4667 }
4668 
4669 /*
4670  * Implement fstatfs() for (NFS) file handles.
4671  */
4672 #ifndef _SYS_SYSPROTO_H_
4673 struct fhstatfs_args {
4674 	struct fhandle *u_fhp;
4675 	struct statfs *buf;
4676 };
4677 #endif
4678 int
4679 sys_fhstatfs(struct thread *td, struct fhstatfs_args *uap)
4680 {
4681 	struct statfs *sfp;
4682 	fhandle_t fh;
4683 	int error;
4684 
4685 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4686 	if (error != 0)
4687 		return (error);
4688 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
4689 	error = kern_fhstatfs(td, fh, sfp);
4690 	if (error == 0)
4691 		error = copyout(sfp, uap->buf, sizeof(*sfp));
4692 	free(sfp, M_STATFS);
4693 	return (error);
4694 }
4695 
4696 int
4697 kern_fhstatfs(struct thread *td, fhandle_t fh, struct statfs *buf)
4698 {
4699 	struct mount *mp;
4700 	struct vnode *vp;
4701 	int error;
4702 
4703 	error = priv_check(td, PRIV_VFS_FHSTATFS);
4704 	if (error != 0)
4705 		return (error);
4706 	if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4707 		return (ESTALE);
4708 	error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp);
4709 	if (error != 0) {
4710 		vfs_unbusy(mp);
4711 		return (error);
4712 	}
4713 	vput(vp);
4714 	error = prison_canseemount(td->td_ucred, mp);
4715 	if (error != 0)
4716 		goto out;
4717 #ifdef MAC
4718 	error = mac_mount_check_stat(td->td_ucred, mp);
4719 	if (error != 0)
4720 		goto out;
4721 #endif
4722 	error = VFS_STATFS(mp, buf);
4723 out:
4724 	vfs_unbusy(mp);
4725 	return (error);
4726 }
4727 
4728 /*
4729  * Unlike madvise(2), we do not make a best effort to remember every
4730  * possible caching hint.  Instead, we remember the last setting with
4731  * the exception that we will allow POSIX_FADV_NORMAL to adjust the
4732  * region of any current setting.
4733  */
4734 int
4735 kern_posix_fadvise(struct thread *td, int fd, off_t offset, off_t len,
4736     int advice)
4737 {
4738 	struct fadvise_info *fa, *new;
4739 	struct file *fp;
4740 	struct vnode *vp;
4741 	off_t end;
4742 	int error;
4743 
4744 	if (offset < 0 || len < 0 || offset > OFF_MAX - len)
4745 		return (EINVAL);
4746 	AUDIT_ARG_VALUE(advice);
4747 	switch (advice) {
4748 	case POSIX_FADV_SEQUENTIAL:
4749 	case POSIX_FADV_RANDOM:
4750 	case POSIX_FADV_NOREUSE:
4751 		new = malloc(sizeof(*fa), M_FADVISE, M_WAITOK);
4752 		break;
4753 	case POSIX_FADV_NORMAL:
4754 	case POSIX_FADV_WILLNEED:
4755 	case POSIX_FADV_DONTNEED:
4756 		new = NULL;
4757 		break;
4758 	default:
4759 		return (EINVAL);
4760 	}
4761 	/* XXX: CAP_POSIX_FADVISE? */
4762 	AUDIT_ARG_FD(fd);
4763 	error = fget(td, fd, &cap_no_rights, &fp);
4764 	if (error != 0)
4765 		goto out;
4766 	AUDIT_ARG_FILE(td->td_proc, fp);
4767 	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
4768 		error = ESPIPE;
4769 		goto out;
4770 	}
4771 	if (fp->f_type != DTYPE_VNODE) {
4772 		error = ENODEV;
4773 		goto out;
4774 	}
4775 	vp = fp->f_vnode;
4776 	if (vp->v_type != VREG) {
4777 		error = ENODEV;
4778 		goto out;
4779 	}
4780 	if (len == 0)
4781 		end = OFF_MAX;
4782 	else
4783 		end = offset + len - 1;
4784 	switch (advice) {
4785 	case POSIX_FADV_SEQUENTIAL:
4786 	case POSIX_FADV_RANDOM:
4787 	case POSIX_FADV_NOREUSE:
4788 		/*
4789 		 * Try to merge any existing non-standard region with
4790 		 * this new region if possible, otherwise create a new
4791 		 * non-standard region for this request.
4792 		 */
4793 		mtx_pool_lock(mtxpool_sleep, fp);
4794 		fa = fp->f_advice;
4795 		if (fa != NULL && fa->fa_advice == advice &&
4796 		    ((fa->fa_start <= end && fa->fa_end >= offset) ||
4797 		    (end != OFF_MAX && fa->fa_start == end + 1) ||
4798 		    (fa->fa_end != OFF_MAX && fa->fa_end + 1 == offset))) {
4799 			if (offset < fa->fa_start)
4800 				fa->fa_start = offset;
4801 			if (end > fa->fa_end)
4802 				fa->fa_end = end;
4803 		} else {
4804 			new->fa_advice = advice;
4805 			new->fa_start = offset;
4806 			new->fa_end = end;
4807 			fp->f_advice = new;
4808 			new = fa;
4809 		}
4810 		mtx_pool_unlock(mtxpool_sleep, fp);
4811 		break;
4812 	case POSIX_FADV_NORMAL:
4813 		/*
4814 		 * If a the "normal" region overlaps with an existing
4815 		 * non-standard region, trim or remove the
4816 		 * non-standard region.
4817 		 */
4818 		mtx_pool_lock(mtxpool_sleep, fp);
4819 		fa = fp->f_advice;
4820 		if (fa != NULL) {
4821 			if (offset <= fa->fa_start && end >= fa->fa_end) {
4822 				new = fa;
4823 				fp->f_advice = NULL;
4824 			} else if (offset <= fa->fa_start &&
4825 			    end >= fa->fa_start)
4826 				fa->fa_start = end + 1;
4827 			else if (offset <= fa->fa_end && end >= fa->fa_end)
4828 				fa->fa_end = offset - 1;
4829 			else if (offset >= fa->fa_start && end <= fa->fa_end) {
4830 				/*
4831 				 * If the "normal" region is a middle
4832 				 * portion of the existing
4833 				 * non-standard region, just remove
4834 				 * the whole thing rather than picking
4835 				 * one side or the other to
4836 				 * preserve.
4837 				 */
4838 				new = fa;
4839 				fp->f_advice = NULL;
4840 			}
4841 		}
4842 		mtx_pool_unlock(mtxpool_sleep, fp);
4843 		break;
4844 	case POSIX_FADV_WILLNEED:
4845 	case POSIX_FADV_DONTNEED:
4846 		error = VOP_ADVISE(vp, offset, end, advice);
4847 		break;
4848 	}
4849 out:
4850 	if (fp != NULL)
4851 		fdrop(fp, td);
4852 	free(new, M_FADVISE);
4853 	return (error);
4854 }
4855 
4856 int
4857 sys_posix_fadvise(struct thread *td, struct posix_fadvise_args *uap)
4858 {
4859 	int error;
4860 
4861 	error = kern_posix_fadvise(td, uap->fd, uap->offset, uap->len,
4862 	    uap->advice);
4863 	return (kern_posix_error(td, error));
4864 }
4865 
4866 int
4867 kern_copy_file_range(struct thread *td, int infd, off_t *inoffp, int outfd,
4868     off_t *outoffp, size_t len, unsigned int flags)
4869 {
4870 	struct file *infp, *outfp;
4871 	struct vnode *invp, *outvp;
4872 	int error;
4873 	size_t retlen;
4874 	void *rl_rcookie, *rl_wcookie;
4875 	off_t savinoff, savoutoff;
4876 
4877 	infp = outfp = NULL;
4878 	rl_rcookie = rl_wcookie = NULL;
4879 	savinoff = -1;
4880 	error = 0;
4881 	retlen = 0;
4882 
4883 	if (flags != 0) {
4884 		error = EINVAL;
4885 		goto out;
4886 	}
4887 	if (len > SSIZE_MAX)
4888 		/*
4889 		 * Although the len argument is size_t, the return argument
4890 		 * is ssize_t (which is signed).  Therefore a size that won't
4891 		 * fit in ssize_t can't be returned.
4892 		 */
4893 		len = SSIZE_MAX;
4894 
4895 	/* Get the file structures for the file descriptors. */
4896 	error = fget_read(td, infd, &cap_read_rights, &infp);
4897 	if (error != 0)
4898 		goto out;
4899 	if (infp->f_ops == &badfileops) {
4900 		error = EBADF;
4901 		goto out;
4902 	}
4903 	if (infp->f_vnode == NULL) {
4904 		error = EINVAL;
4905 		goto out;
4906 	}
4907 	error = fget_write(td, outfd, &cap_write_rights, &outfp);
4908 	if (error != 0)
4909 		goto out;
4910 	if (outfp->f_ops == &badfileops) {
4911 		error = EBADF;
4912 		goto out;
4913 	}
4914 	if (outfp->f_vnode == NULL) {
4915 		error = EINVAL;
4916 		goto out;
4917 	}
4918 
4919 	/* Set the offset pointers to the correct place. */
4920 	if (inoffp == NULL)
4921 		inoffp = &infp->f_offset;
4922 	if (outoffp == NULL)
4923 		outoffp = &outfp->f_offset;
4924 	savinoff = *inoffp;
4925 	savoutoff = *outoffp;
4926 
4927 	invp = infp->f_vnode;
4928 	outvp = outfp->f_vnode;
4929 	/* Sanity check the f_flag bits. */
4930 	if ((outfp->f_flag & (FWRITE | FAPPEND)) != FWRITE ||
4931 	    (infp->f_flag & FREAD) == 0) {
4932 		error = EBADF;
4933 		goto out;
4934 	}
4935 
4936 	/* If len == 0, just return 0. */
4937 	if (len == 0)
4938 		goto out;
4939 
4940 	/*
4941 	 * If infp and outfp refer to the same file, the byte ranges cannot
4942 	 * overlap.
4943 	 */
4944 	if (invp == outvp && ((savinoff <= savoutoff && savinoff + len >
4945 	    savoutoff) || (savinoff > savoutoff && savoutoff + len >
4946 	    savinoff))) {
4947 		error = EINVAL;
4948 		goto out;
4949 	}
4950 
4951 	/* Range lock the byte ranges for both invp and outvp. */
4952 	for (;;) {
4953 		rl_wcookie = vn_rangelock_wlock(outvp, *outoffp, *outoffp +
4954 		    len);
4955 		rl_rcookie = vn_rangelock_tryrlock(invp, *inoffp, *inoffp +
4956 		    len);
4957 		if (rl_rcookie != NULL)
4958 			break;
4959 		vn_rangelock_unlock(outvp, rl_wcookie);
4960 		rl_rcookie = vn_rangelock_rlock(invp, *inoffp, *inoffp + len);
4961 		vn_rangelock_unlock(invp, rl_rcookie);
4962 	}
4963 
4964 	retlen = len;
4965 	error = vn_copy_file_range(invp, inoffp, outvp, outoffp, &retlen,
4966 	    flags, infp->f_cred, outfp->f_cred, td);
4967 out:
4968 	if (rl_rcookie != NULL)
4969 		vn_rangelock_unlock(invp, rl_rcookie);
4970 	if (rl_wcookie != NULL)
4971 		vn_rangelock_unlock(outvp, rl_wcookie);
4972 	if (savinoff != -1 && (error == EINTR || error == ERESTART)) {
4973 		*inoffp = savinoff;
4974 		*outoffp = savoutoff;
4975 	}
4976 	if (outfp != NULL)
4977 		fdrop(outfp, td);
4978 	if (infp != NULL)
4979 		fdrop(infp, td);
4980 	td->td_retval[0] = retlen;
4981 	return (error);
4982 }
4983 
4984 int
4985 sys_copy_file_range(struct thread *td, struct copy_file_range_args *uap)
4986 {
4987 	off_t inoff, outoff, *inoffp, *outoffp;
4988 	int error;
4989 
4990 	inoffp = outoffp = NULL;
4991 	if (uap->inoffp != NULL) {
4992 		error = copyin(uap->inoffp, &inoff, sizeof(off_t));
4993 		if (error != 0)
4994 			return (error);
4995 		inoffp = &inoff;
4996 	}
4997 	if (uap->outoffp != NULL) {
4998 		error = copyin(uap->outoffp, &outoff, sizeof(off_t));
4999 		if (error != 0)
5000 			return (error);
5001 		outoffp = &outoff;
5002 	}
5003 	error = kern_copy_file_range(td, uap->infd, inoffp, uap->outfd,
5004 	    outoffp, uap->len, uap->flags);
5005 	if (error == 0 && uap->inoffp != NULL)
5006 		error = copyout(inoffp, uap->inoffp, sizeof(off_t));
5007 	if (error == 0 && uap->outoffp != NULL)
5008 		error = copyout(outoffp, uap->outoffp, sizeof(off_t));
5009 	return (error);
5010 }
5011