xref: /freebsd/sys/kern/vfs_syscalls.c (revision 833a452e9f082a7982a31c21f0da437dbbe0a39d)
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 #ifdef COMPAT_FREEBSD11
48 #include <sys/abi_compat.h>
49 #endif
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/capsicum.h>
53 #include <sys/disk.h>
54 #include <sys/sysent.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/sysproto.h>
59 #include <sys/namei.h>
60 #include <sys/filedesc.h>
61 #include <sys/kernel.h>
62 #include <sys/fcntl.h>
63 #include <sys/file.h>
64 #include <sys/filio.h>
65 #include <sys/limits.h>
66 #include <sys/linker.h>
67 #include <sys/rwlock.h>
68 #include <sys/sdt.h>
69 #include <sys/stat.h>
70 #include <sys/sx.h>
71 #include <sys/unistd.h>
72 #include <sys/vnode.h>
73 #include <sys/priv.h>
74 #include <sys/proc.h>
75 #include <sys/dirent.h>
76 #include <sys/jail.h>
77 #include <sys/syscallsubr.h>
78 #include <sys/sysctl.h>
79 #ifdef KTRACE
80 #include <sys/ktrace.h>
81 #endif
82 
83 #include <machine/stdarg.h>
84 
85 #include <security/audit/audit.h>
86 #include <security/mac/mac_framework.h>
87 
88 #include <vm/vm.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/uma.h>
92 
93 #include <fs/devfs/devfs.h>
94 
95 MALLOC_DEFINE(M_FADVISE, "fadvise", "posix_fadvise(2) information");
96 
97 static int kern_chflagsat(struct thread *td, int fd, const char *path,
98     enum uio_seg pathseg, u_long flags, int atflag);
99 static int setfflags(struct thread *td, struct vnode *, u_long);
100 static int getutimes(const struct timeval *, enum uio_seg, struct timespec *);
101 static int getutimens(const struct timespec *, enum uio_seg,
102     struct timespec *, int *);
103 static int setutimes(struct thread *td, struct vnode *,
104     const struct timespec *, int, int);
105 static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred,
106     struct thread *td);
107 static int kern_fhlinkat(struct thread *td, int fd, const char *path,
108     enum uio_seg pathseg, fhandle_t *fhp);
109 static int kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg,
110     size_t count, struct thread *td);
111 static int kern_linkat_vp(struct thread *td, struct vnode *vp, int fd,
112     const char *path, enum uio_seg segflag);
113 
114 static uint64_t
115 at2cnpflags(u_int at_flags, u_int mask)
116 {
117 	uint64_t res;
118 
119 	MPASS((at_flags & (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW)) !=
120 	    (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW));
121 
122 	res = 0;
123 	at_flags &= mask;
124 	if ((at_flags & AT_RESOLVE_BENEATH) != 0)
125 		res |= RBENEATH;
126 	if ((at_flags & AT_SYMLINK_FOLLOW) != 0)
127 		res |= FOLLOW;
128 	/* NOFOLLOW is pseudo flag */
129 	if ((mask & AT_SYMLINK_NOFOLLOW) != 0) {
130 		res |= (at_flags & AT_SYMLINK_NOFOLLOW) != 0 ? NOFOLLOW :
131 		    FOLLOW;
132 	}
133 	if ((mask & AT_EMPTY_PATH) != 0 && (at_flags & AT_EMPTY_PATH) != 0)
134 		res |= EMPTYPATH;
135 	return (res);
136 }
137 
138 int
139 kern_sync(struct thread *td)
140 {
141 	struct mount *mp, *nmp;
142 	int save;
143 
144 	mtx_lock(&mountlist_mtx);
145 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
146 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
147 			nmp = TAILQ_NEXT(mp, mnt_list);
148 			continue;
149 		}
150 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
151 		    vn_start_write(NULL, &mp, V_NOWAIT) == 0) {
152 			save = curthread_pflags_set(TDP_SYNCIO);
153 			vfs_periodic(mp, MNT_NOWAIT);
154 			VFS_SYNC(mp, MNT_NOWAIT);
155 			curthread_pflags_restore(save);
156 			vn_finished_write(mp);
157 		}
158 		mtx_lock(&mountlist_mtx);
159 		nmp = TAILQ_NEXT(mp, mnt_list);
160 		vfs_unbusy(mp);
161 	}
162 	mtx_unlock(&mountlist_mtx);
163 	return (0);
164 }
165 
166 /*
167  * Sync each mounted filesystem.
168  */
169 #ifndef _SYS_SYSPROTO_H_
170 struct sync_args {
171 	int     dummy;
172 };
173 #endif
174 /* ARGSUSED */
175 int
176 sys_sync(struct thread *td, struct sync_args *uap)
177 {
178 
179 	return (kern_sync(td));
180 }
181 
182 /*
183  * Change filesystem quotas.
184  */
185 #ifndef _SYS_SYSPROTO_H_
186 struct quotactl_args {
187 	char *path;
188 	int cmd;
189 	int uid;
190 	caddr_t arg;
191 };
192 #endif
193 int
194 sys_quotactl(struct thread *td, struct quotactl_args *uap)
195 {
196 	struct mount *mp;
197 	struct nameidata nd;
198 	int error;
199 	bool mp_busy;
200 
201 	AUDIT_ARG_CMD(uap->cmd);
202 	AUDIT_ARG_UID(uap->uid);
203 	if (!prison_allow(td->td_ucred, PR_ALLOW_QUOTAS))
204 		return (EPERM);
205 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE,
206 	    uap->path);
207 	if ((error = namei(&nd)) != 0)
208 		return (error);
209 	NDFREE(&nd, NDF_ONLY_PNBUF);
210 	mp = nd.ni_vp->v_mount;
211 	vfs_ref(mp);
212 	vput(nd.ni_vp);
213 	error = vfs_busy(mp, 0);
214 	if (error != 0) {
215 		vfs_rel(mp);
216 		return (error);
217 	}
218 	mp_busy = true;
219 	error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, uap->arg, &mp_busy);
220 
221 	/*
222 	 * Since quota on/off operations typically need to open quota
223 	 * files, the implementation may need to unbusy the mount point
224 	 * before calling into namei.  Otherwise, unmount might be
225 	 * started between two vfs_busy() invocations (first is ours,
226 	 * second is from mount point cross-walk code in lookup()),
227 	 * causing deadlock.
228 	 *
229 	 * Avoid unbusying mp if the implementation indicates it has
230 	 * already done so.
231 	 */
232 	if (mp_busy)
233 		vfs_unbusy(mp);
234 	vfs_rel(mp);
235 	return (error);
236 }
237 
238 /*
239  * Used by statfs conversion routines to scale the block size up if
240  * necessary so that all of the block counts are <= 'max_size'.  Note
241  * that 'max_size' should be a bitmask, i.e. 2^n - 1 for some non-zero
242  * value of 'n'.
243  */
244 void
245 statfs_scale_blocks(struct statfs *sf, long max_size)
246 {
247 	uint64_t count;
248 	int shift;
249 
250 	KASSERT(powerof2(max_size + 1), ("%s: invalid max_size", __func__));
251 
252 	/*
253 	 * Attempt to scale the block counts to give a more accurate
254 	 * overview to userland of the ratio of free space to used
255 	 * space.  To do this, find the largest block count and compute
256 	 * a divisor that lets it fit into a signed integer <= max_size.
257 	 */
258 	if (sf->f_bavail < 0)
259 		count = -sf->f_bavail;
260 	else
261 		count = sf->f_bavail;
262 	count = MAX(sf->f_blocks, MAX(sf->f_bfree, count));
263 	if (count <= max_size)
264 		return;
265 
266 	count >>= flsl(max_size);
267 	shift = 0;
268 	while (count > 0) {
269 		shift++;
270 		count >>=1;
271 	}
272 
273 	sf->f_bsize <<= shift;
274 	sf->f_blocks >>= shift;
275 	sf->f_bfree >>= shift;
276 	sf->f_bavail >>= shift;
277 }
278 
279 static int
280 kern_do_statfs(struct thread *td, struct mount *mp, struct statfs *buf)
281 {
282 	int error;
283 
284 	if (mp == NULL)
285 		return (EBADF);
286 	error = vfs_busy(mp, 0);
287 	vfs_rel(mp);
288 	if (error != 0)
289 		return (error);
290 #ifdef MAC
291 	error = mac_mount_check_stat(td->td_ucred, mp);
292 	if (error != 0)
293 		goto out;
294 #endif
295 	error = VFS_STATFS(mp, buf);
296 	if (error != 0)
297 		goto out;
298 	if (priv_check_cred_vfs_generation(td->td_ucred)) {
299 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
300 		prison_enforce_statfs(td->td_ucred, mp, buf);
301 	}
302 out:
303 	vfs_unbusy(mp);
304 	return (error);
305 }
306 
307 /*
308  * Get filesystem statistics.
309  */
310 #ifndef _SYS_SYSPROTO_H_
311 struct statfs_args {
312 	char *path;
313 	struct statfs *buf;
314 };
315 #endif
316 int
317 sys_statfs(struct thread *td, struct statfs_args *uap)
318 {
319 	struct statfs *sfp;
320 	int error;
321 
322 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
323 	error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp);
324 	if (error == 0)
325 		error = copyout(sfp, uap->buf, sizeof(struct statfs));
326 	free(sfp, M_STATFS);
327 	return (error);
328 }
329 
330 int
331 kern_statfs(struct thread *td, const char *path, enum uio_seg pathseg,
332     struct statfs *buf)
333 {
334 	struct mount *mp;
335 	struct nameidata nd;
336 	int error;
337 
338 	NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path);
339 	error = namei(&nd);
340 	if (error != 0)
341 		return (error);
342 	mp = vfs_ref_from_vp(nd.ni_vp);
343 	NDFREE_NOTHING(&nd);
344 	vrele(nd.ni_vp);
345 	return (kern_do_statfs(td, mp, buf));
346 }
347 
348 /*
349  * Get filesystem statistics.
350  */
351 #ifndef _SYS_SYSPROTO_H_
352 struct fstatfs_args {
353 	int fd;
354 	struct statfs *buf;
355 };
356 #endif
357 int
358 sys_fstatfs(struct thread *td, struct fstatfs_args *uap)
359 {
360 	struct statfs *sfp;
361 	int error;
362 
363 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
364 	error = kern_fstatfs(td, uap->fd, sfp);
365 	if (error == 0)
366 		error = copyout(sfp, uap->buf, sizeof(struct statfs));
367 	free(sfp, M_STATFS);
368 	return (error);
369 }
370 
371 int
372 kern_fstatfs(struct thread *td, int fd, struct statfs *buf)
373 {
374 	struct file *fp;
375 	struct mount *mp;
376 	struct vnode *vp;
377 	int error;
378 
379 	AUDIT_ARG_FD(fd);
380 	error = getvnode_path(td, fd, &cap_fstatfs_rights, &fp);
381 	if (error != 0)
382 		return (error);
383 	vp = fp->f_vnode;
384 #ifdef AUDIT
385 	if (AUDITING_TD(td)) {
386 		vn_lock(vp, LK_SHARED | LK_RETRY);
387 		AUDIT_ARG_VNODE1(vp);
388 		VOP_UNLOCK(vp);
389 	}
390 #endif
391 	mp = vfs_ref_from_vp(vp);
392 	fdrop(fp, td);
393 	return (kern_do_statfs(td, mp, buf));
394 }
395 
396 /*
397  * Get statistics on all filesystems.
398  */
399 #ifndef _SYS_SYSPROTO_H_
400 struct getfsstat_args {
401 	struct statfs *buf;
402 	long bufsize;
403 	int mode;
404 };
405 #endif
406 int
407 sys_getfsstat(struct thread *td, struct getfsstat_args *uap)
408 {
409 	size_t count;
410 	int error;
411 
412 	if (uap->bufsize < 0 || uap->bufsize > SIZE_MAX)
413 		return (EINVAL);
414 	error = kern_getfsstat(td, &uap->buf, uap->bufsize, &count,
415 	    UIO_USERSPACE, uap->mode);
416 	if (error == 0)
417 		td->td_retval[0] = count;
418 	return (error);
419 }
420 
421 /*
422  * If (bufsize > 0 && bufseg == UIO_SYSSPACE)
423  *	The caller is responsible for freeing memory which will be allocated
424  *	in '*buf'.
425  */
426 int
427 kern_getfsstat(struct thread *td, struct statfs **buf, size_t bufsize,
428     size_t *countp, enum uio_seg bufseg, int mode)
429 {
430 	struct mount *mp, *nmp;
431 	struct statfs *sfsp, *sp, *sptmp, *tofree;
432 	size_t count, maxcount;
433 	int error;
434 
435 	switch (mode) {
436 	case MNT_WAIT:
437 	case MNT_NOWAIT:
438 		break;
439 	default:
440 		if (bufseg == UIO_SYSSPACE)
441 			*buf = NULL;
442 		return (EINVAL);
443 	}
444 restart:
445 	maxcount = bufsize / sizeof(struct statfs);
446 	if (bufsize == 0) {
447 		sfsp = NULL;
448 		tofree = NULL;
449 	} else if (bufseg == UIO_USERSPACE) {
450 		sfsp = *buf;
451 		tofree = NULL;
452 	} else /* if (bufseg == UIO_SYSSPACE) */ {
453 		count = 0;
454 		mtx_lock(&mountlist_mtx);
455 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
456 			count++;
457 		}
458 		mtx_unlock(&mountlist_mtx);
459 		if (maxcount > count)
460 			maxcount = count;
461 		tofree = sfsp = *buf = malloc(maxcount * sizeof(struct statfs),
462 		    M_STATFS, M_WAITOK);
463 	}
464 
465 	count = 0;
466 
467 	/*
468 	 * If there is no target buffer they only want the count.
469 	 *
470 	 * This could be TAILQ_FOREACH but it is open-coded to match the original
471 	 * code below.
472 	 */
473 	if (sfsp == NULL) {
474 		mtx_lock(&mountlist_mtx);
475 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
476 			if (prison_canseemount(td->td_ucred, mp) != 0) {
477 				nmp = TAILQ_NEXT(mp, mnt_list);
478 				continue;
479 			}
480 #ifdef MAC
481 			if (mac_mount_check_stat(td->td_ucred, mp) != 0) {
482 				nmp = TAILQ_NEXT(mp, mnt_list);
483 				continue;
484 			}
485 #endif
486 			count++;
487 			nmp = TAILQ_NEXT(mp, mnt_list);
488 		}
489 		mtx_unlock(&mountlist_mtx);
490 		*countp = count;
491 		return (0);
492 	}
493 
494 	/*
495 	 * They want the entire thing.
496 	 *
497 	 * Short-circuit the corner case of no room for anything, avoids
498 	 * relocking below.
499 	 */
500 	if (maxcount < 1) {
501 		goto out;
502 	}
503 
504 	mtx_lock(&mountlist_mtx);
505 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
506 		if (prison_canseemount(td->td_ucred, mp) != 0) {
507 			nmp = TAILQ_NEXT(mp, mnt_list);
508 			continue;
509 		}
510 #ifdef MAC
511 		if (mac_mount_check_stat(td->td_ucred, mp) != 0) {
512 			nmp = TAILQ_NEXT(mp, mnt_list);
513 			continue;
514 		}
515 #endif
516 		if (mode == MNT_WAIT) {
517 			if (vfs_busy(mp, MBF_MNTLSTLOCK) != 0) {
518 				/*
519 				 * If vfs_busy() failed, and MBF_NOWAIT
520 				 * wasn't passed, then the mp is gone.
521 				 * Furthermore, because of MBF_MNTLSTLOCK,
522 				 * the mountlist_mtx was dropped.  We have
523 				 * no other choice than to start over.
524 				 */
525 				mtx_unlock(&mountlist_mtx);
526 				free(tofree, M_STATFS);
527 				goto restart;
528 			}
529 		} else {
530 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
531 				nmp = TAILQ_NEXT(mp, mnt_list);
532 				continue;
533 			}
534 		}
535 		sp = &mp->mnt_stat;
536 		/*
537 		 * If MNT_NOWAIT is specified, do not refresh
538 		 * the fsstat cache.
539 		 */
540 		if (mode != MNT_NOWAIT) {
541 			error = VFS_STATFS(mp, sp);
542 			if (error != 0) {
543 				mtx_lock(&mountlist_mtx);
544 				nmp = TAILQ_NEXT(mp, mnt_list);
545 				vfs_unbusy(mp);
546 				continue;
547 			}
548 		}
549 		if (priv_check_cred_vfs_generation(td->td_ucred)) {
550 			sptmp = malloc(sizeof(struct statfs), M_STATFS,
551 			    M_WAITOK);
552 			*sptmp = *sp;
553 			sptmp->f_fsid.val[0] = sptmp->f_fsid.val[1] = 0;
554 			prison_enforce_statfs(td->td_ucred, mp, sptmp);
555 			sp = sptmp;
556 		} else
557 			sptmp = NULL;
558 		if (bufseg == UIO_SYSSPACE) {
559 			bcopy(sp, sfsp, sizeof(*sp));
560 			free(sptmp, M_STATFS);
561 		} else /* if (bufseg == UIO_USERSPACE) */ {
562 			error = copyout(sp, sfsp, sizeof(*sp));
563 			free(sptmp, M_STATFS);
564 			if (error != 0) {
565 				vfs_unbusy(mp);
566 				return (error);
567 			}
568 		}
569 		sfsp++;
570 		count++;
571 
572 		if (count == maxcount) {
573 			vfs_unbusy(mp);
574 			goto out;
575 		}
576 
577 		mtx_lock(&mountlist_mtx);
578 		nmp = TAILQ_NEXT(mp, mnt_list);
579 		vfs_unbusy(mp);
580 	}
581 	mtx_unlock(&mountlist_mtx);
582 out:
583 	*countp = count;
584 	return (0);
585 }
586 
587 #ifdef COMPAT_FREEBSD4
588 /*
589  * Get old format filesystem statistics.
590  */
591 static void freebsd4_cvtstatfs(struct statfs *, struct ostatfs *);
592 
593 #ifndef _SYS_SYSPROTO_H_
594 struct freebsd4_statfs_args {
595 	char *path;
596 	struct ostatfs *buf;
597 };
598 #endif
599 int
600 freebsd4_statfs(struct thread *td, struct freebsd4_statfs_args *uap)
601 {
602 	struct ostatfs osb;
603 	struct statfs *sfp;
604 	int error;
605 
606 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
607 	error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp);
608 	if (error == 0) {
609 		freebsd4_cvtstatfs(sfp, &osb);
610 		error = copyout(&osb, uap->buf, sizeof(osb));
611 	}
612 	free(sfp, M_STATFS);
613 	return (error);
614 }
615 
616 /*
617  * Get filesystem statistics.
618  */
619 #ifndef _SYS_SYSPROTO_H_
620 struct freebsd4_fstatfs_args {
621 	int fd;
622 	struct ostatfs *buf;
623 };
624 #endif
625 int
626 freebsd4_fstatfs(struct thread *td, struct freebsd4_fstatfs_args *uap)
627 {
628 	struct ostatfs osb;
629 	struct statfs *sfp;
630 	int error;
631 
632 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
633 	error = kern_fstatfs(td, uap->fd, sfp);
634 	if (error == 0) {
635 		freebsd4_cvtstatfs(sfp, &osb);
636 		error = copyout(&osb, uap->buf, sizeof(osb));
637 	}
638 	free(sfp, M_STATFS);
639 	return (error);
640 }
641 
642 /*
643  * Get statistics on all filesystems.
644  */
645 #ifndef _SYS_SYSPROTO_H_
646 struct freebsd4_getfsstat_args {
647 	struct ostatfs *buf;
648 	long bufsize;
649 	int mode;
650 };
651 #endif
652 int
653 freebsd4_getfsstat(struct thread *td, struct freebsd4_getfsstat_args *uap)
654 {
655 	struct statfs *buf, *sp;
656 	struct ostatfs osb;
657 	size_t count, size;
658 	int error;
659 
660 	if (uap->bufsize < 0)
661 		return (EINVAL);
662 	count = uap->bufsize / sizeof(struct ostatfs);
663 	if (count > SIZE_MAX / sizeof(struct statfs))
664 		return (EINVAL);
665 	size = count * sizeof(struct statfs);
666 	error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE,
667 	    uap->mode);
668 	if (error == 0)
669 		td->td_retval[0] = count;
670 	if (size != 0) {
671 		sp = buf;
672 		while (count != 0 && error == 0) {
673 			freebsd4_cvtstatfs(sp, &osb);
674 			error = copyout(&osb, uap->buf, sizeof(osb));
675 			sp++;
676 			uap->buf++;
677 			count--;
678 		}
679 		free(buf, M_STATFS);
680 	}
681 	return (error);
682 }
683 
684 /*
685  * Implement fstatfs() for (NFS) file handles.
686  */
687 #ifndef _SYS_SYSPROTO_H_
688 struct freebsd4_fhstatfs_args {
689 	struct fhandle *u_fhp;
690 	struct ostatfs *buf;
691 };
692 #endif
693 int
694 freebsd4_fhstatfs(struct thread *td, struct freebsd4_fhstatfs_args *uap)
695 {
696 	struct ostatfs osb;
697 	struct statfs *sfp;
698 	fhandle_t fh;
699 	int error;
700 
701 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
702 	if (error != 0)
703 		return (error);
704 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
705 	error = kern_fhstatfs(td, fh, sfp);
706 	if (error == 0) {
707 		freebsd4_cvtstatfs(sfp, &osb);
708 		error = copyout(&osb, uap->buf, sizeof(osb));
709 	}
710 	free(sfp, M_STATFS);
711 	return (error);
712 }
713 
714 /*
715  * Convert a new format statfs structure to an old format statfs structure.
716  */
717 static void
718 freebsd4_cvtstatfs(struct statfs *nsp, struct ostatfs *osp)
719 {
720 
721 	statfs_scale_blocks(nsp, LONG_MAX);
722 	bzero(osp, sizeof(*osp));
723 	osp->f_bsize = nsp->f_bsize;
724 	osp->f_iosize = MIN(nsp->f_iosize, LONG_MAX);
725 	osp->f_blocks = nsp->f_blocks;
726 	osp->f_bfree = nsp->f_bfree;
727 	osp->f_bavail = nsp->f_bavail;
728 	osp->f_files = MIN(nsp->f_files, LONG_MAX);
729 	osp->f_ffree = MIN(nsp->f_ffree, LONG_MAX);
730 	osp->f_owner = nsp->f_owner;
731 	osp->f_type = nsp->f_type;
732 	osp->f_flags = nsp->f_flags;
733 	osp->f_syncwrites = MIN(nsp->f_syncwrites, LONG_MAX);
734 	osp->f_asyncwrites = MIN(nsp->f_asyncwrites, LONG_MAX);
735 	osp->f_syncreads = MIN(nsp->f_syncreads, LONG_MAX);
736 	osp->f_asyncreads = MIN(nsp->f_asyncreads, LONG_MAX);
737 	strlcpy(osp->f_fstypename, nsp->f_fstypename,
738 	    MIN(MFSNAMELEN, OMFSNAMELEN));
739 	strlcpy(osp->f_mntonname, nsp->f_mntonname,
740 	    MIN(MNAMELEN, OMNAMELEN));
741 	strlcpy(osp->f_mntfromname, nsp->f_mntfromname,
742 	    MIN(MNAMELEN, OMNAMELEN));
743 	osp->f_fsid = nsp->f_fsid;
744 }
745 #endif /* COMPAT_FREEBSD4 */
746 
747 #if defined(COMPAT_FREEBSD11)
748 /*
749  * Get old format filesystem statistics.
750  */
751 static void freebsd11_cvtstatfs(struct statfs *, struct freebsd11_statfs *);
752 
753 int
754 freebsd11_statfs(struct thread *td, struct freebsd11_statfs_args *uap)
755 {
756 	struct freebsd11_statfs osb;
757 	struct statfs *sfp;
758 	int error;
759 
760 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
761 	error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp);
762 	if (error == 0) {
763 		freebsd11_cvtstatfs(sfp, &osb);
764 		error = copyout(&osb, uap->buf, sizeof(osb));
765 	}
766 	free(sfp, M_STATFS);
767 	return (error);
768 }
769 
770 /*
771  * Get filesystem statistics.
772  */
773 int
774 freebsd11_fstatfs(struct thread *td, struct freebsd11_fstatfs_args *uap)
775 {
776 	struct freebsd11_statfs osb;
777 	struct statfs *sfp;
778 	int error;
779 
780 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
781 	error = kern_fstatfs(td, uap->fd, sfp);
782 	if (error == 0) {
783 		freebsd11_cvtstatfs(sfp, &osb);
784 		error = copyout(&osb, uap->buf, sizeof(osb));
785 	}
786 	free(sfp, M_STATFS);
787 	return (error);
788 }
789 
790 /*
791  * Get statistics on all filesystems.
792  */
793 int
794 freebsd11_getfsstat(struct thread *td, struct freebsd11_getfsstat_args *uap)
795 {
796 	return (kern_freebsd11_getfsstat(td, uap->buf, uap->bufsize, uap->mode));
797 }
798 
799 int
800 kern_freebsd11_getfsstat(struct thread *td, struct freebsd11_statfs * ubuf,
801     long bufsize, int mode)
802 {
803 	struct freebsd11_statfs osb;
804 	struct statfs *buf, *sp;
805 	size_t count, size;
806 	int error;
807 
808 	if (bufsize < 0)
809 		return (EINVAL);
810 
811 	count = bufsize / sizeof(struct ostatfs);
812 	size = count * sizeof(struct statfs);
813 	error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, mode);
814 	if (error == 0)
815 		td->td_retval[0] = count;
816 	if (size > 0) {
817 		sp = buf;
818 		while (count > 0 && error == 0) {
819 			freebsd11_cvtstatfs(sp, &osb);
820 			error = copyout(&osb, ubuf, sizeof(osb));
821 			sp++;
822 			ubuf++;
823 			count--;
824 		}
825 		free(buf, M_STATFS);
826 	}
827 	return (error);
828 }
829 
830 /*
831  * Implement fstatfs() for (NFS) file handles.
832  */
833 int
834 freebsd11_fhstatfs(struct thread *td, struct freebsd11_fhstatfs_args *uap)
835 {
836 	struct freebsd11_statfs osb;
837 	struct statfs *sfp;
838 	fhandle_t fh;
839 	int error;
840 
841 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
842 	if (error)
843 		return (error);
844 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
845 	error = kern_fhstatfs(td, fh, sfp);
846 	if (error == 0) {
847 		freebsd11_cvtstatfs(sfp, &osb);
848 		error = copyout(&osb, uap->buf, sizeof(osb));
849 	}
850 	free(sfp, M_STATFS);
851 	return (error);
852 }
853 
854 /*
855  * Convert a new format statfs structure to an old format statfs structure.
856  */
857 static void
858 freebsd11_cvtstatfs(struct statfs *nsp, struct freebsd11_statfs *osp)
859 {
860 
861 	bzero(osp, sizeof(*osp));
862 	osp->f_version = FREEBSD11_STATFS_VERSION;
863 	osp->f_type = nsp->f_type;
864 	osp->f_flags = nsp->f_flags;
865 	osp->f_bsize = nsp->f_bsize;
866 	osp->f_iosize = nsp->f_iosize;
867 	osp->f_blocks = nsp->f_blocks;
868 	osp->f_bfree = nsp->f_bfree;
869 	osp->f_bavail = nsp->f_bavail;
870 	osp->f_files = nsp->f_files;
871 	osp->f_ffree = nsp->f_ffree;
872 	osp->f_syncwrites = nsp->f_syncwrites;
873 	osp->f_asyncwrites = nsp->f_asyncwrites;
874 	osp->f_syncreads = nsp->f_syncreads;
875 	osp->f_asyncreads = nsp->f_asyncreads;
876 	osp->f_namemax = nsp->f_namemax;
877 	osp->f_owner = nsp->f_owner;
878 	osp->f_fsid = nsp->f_fsid;
879 	strlcpy(osp->f_fstypename, nsp->f_fstypename,
880 	    MIN(MFSNAMELEN, sizeof(osp->f_fstypename)));
881 	strlcpy(osp->f_mntonname, nsp->f_mntonname,
882 	    MIN(MNAMELEN, sizeof(osp->f_mntonname)));
883 	strlcpy(osp->f_mntfromname, nsp->f_mntfromname,
884 	    MIN(MNAMELEN, sizeof(osp->f_mntfromname)));
885 }
886 #endif /* COMPAT_FREEBSD11 */
887 
888 /*
889  * Change current working directory to a given file descriptor.
890  */
891 #ifndef _SYS_SYSPROTO_H_
892 struct fchdir_args {
893 	int	fd;
894 };
895 #endif
896 int
897 sys_fchdir(struct thread *td, struct fchdir_args *uap)
898 {
899 	struct vnode *vp, *tdp;
900 	struct mount *mp;
901 	struct file *fp;
902 	int error;
903 
904 	AUDIT_ARG_FD(uap->fd);
905 	error = getvnode_path(td, uap->fd, &cap_fchdir_rights,
906 	    &fp);
907 	if (error != 0)
908 		return (error);
909 	vp = fp->f_vnode;
910 	vref(vp);
911 	fdrop(fp, td);
912 	vn_lock(vp, LK_SHARED | LK_RETRY);
913 	AUDIT_ARG_VNODE1(vp);
914 	error = change_dir(vp, td);
915 	while (!error && (mp = vp->v_mountedhere) != NULL) {
916 		if (vfs_busy(mp, 0))
917 			continue;
918 		error = VFS_ROOT(mp, LK_SHARED, &tdp);
919 		vfs_unbusy(mp);
920 		if (error != 0)
921 			break;
922 		vput(vp);
923 		vp = tdp;
924 	}
925 	if (error != 0) {
926 		vput(vp);
927 		return (error);
928 	}
929 	VOP_UNLOCK(vp);
930 	pwd_chdir(td, vp);
931 	return (0);
932 }
933 
934 /*
935  * Change current working directory (``.'').
936  */
937 #ifndef _SYS_SYSPROTO_H_
938 struct chdir_args {
939 	char	*path;
940 };
941 #endif
942 int
943 sys_chdir(struct thread *td, struct chdir_args *uap)
944 {
945 
946 	return (kern_chdir(td, uap->path, UIO_USERSPACE));
947 }
948 
949 int
950 kern_chdir(struct thread *td, const char *path, enum uio_seg pathseg)
951 {
952 	struct nameidata nd;
953 	int error;
954 
955 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1,
956 	    pathseg, path);
957 	if ((error = namei(&nd)) != 0)
958 		return (error);
959 	if ((error = change_dir(nd.ni_vp, td)) != 0) {
960 		vput(nd.ni_vp);
961 		NDFREE_NOTHING(&nd);
962 		return (error);
963 	}
964 	VOP_UNLOCK(nd.ni_vp);
965 	NDFREE_NOTHING(&nd);
966 	pwd_chdir(td, nd.ni_vp);
967 	return (0);
968 }
969 
970 static int unprivileged_chroot = 0;
971 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_chroot, CTLFLAG_RW,
972     &unprivileged_chroot, 0,
973     "Unprivileged processes can use chroot(2)");
974 /*
975  * Change notion of root (``/'') directory.
976  */
977 #ifndef _SYS_SYSPROTO_H_
978 struct chroot_args {
979 	char	*path;
980 };
981 #endif
982 int
983 sys_chroot(struct thread *td, struct chroot_args *uap)
984 {
985 	struct nameidata nd;
986 	struct proc *p;
987 	int error;
988 
989 	error = priv_check(td, PRIV_VFS_CHROOT);
990 	if (error != 0) {
991 		p = td->td_proc;
992 		PROC_LOCK(p);
993 		if (unprivileged_chroot == 0 ||
994 		    (p->p_flag2 & P2_NO_NEW_PRIVS) == 0) {
995 			PROC_UNLOCK(p);
996 			return (error);
997 		}
998 		PROC_UNLOCK(p);
999 	}
1000 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1,
1001 	    UIO_USERSPACE, uap->path);
1002 	error = namei(&nd);
1003 	if (error != 0)
1004 		goto error;
1005 	error = change_dir(nd.ni_vp, td);
1006 	if (error != 0)
1007 		goto e_vunlock;
1008 #ifdef MAC
1009 	error = mac_vnode_check_chroot(td->td_ucred, nd.ni_vp);
1010 	if (error != 0)
1011 		goto e_vunlock;
1012 #endif
1013 	VOP_UNLOCK(nd.ni_vp);
1014 	error = pwd_chroot(td, nd.ni_vp);
1015 	vrele(nd.ni_vp);
1016 	NDFREE_NOTHING(&nd);
1017 	return (error);
1018 e_vunlock:
1019 	vput(nd.ni_vp);
1020 error:
1021 	NDFREE_NOTHING(&nd);
1022 	return (error);
1023 }
1024 
1025 /*
1026  * Common routine for chroot and chdir.  Callers must provide a locked vnode
1027  * instance.
1028  */
1029 int
1030 change_dir(struct vnode *vp, struct thread *td)
1031 {
1032 #ifdef MAC
1033 	int error;
1034 #endif
1035 
1036 	ASSERT_VOP_LOCKED(vp, "change_dir(): vp not locked");
1037 	if (vp->v_type != VDIR)
1038 		return (ENOTDIR);
1039 #ifdef MAC
1040 	error = mac_vnode_check_chdir(td->td_ucred, vp);
1041 	if (error != 0)
1042 		return (error);
1043 #endif
1044 	return (VOP_ACCESS(vp, VEXEC, td->td_ucred, td));
1045 }
1046 
1047 static __inline void
1048 flags_to_rights(int flags, cap_rights_t *rightsp)
1049 {
1050 	if (flags & O_EXEC) {
1051 		cap_rights_set_one(rightsp, CAP_FEXECVE);
1052 		if (flags & O_PATH)
1053 			return;
1054 	} else {
1055 		switch ((flags & O_ACCMODE)) {
1056 		case O_RDONLY:
1057 			cap_rights_set_one(rightsp, CAP_READ);
1058 			break;
1059 		case O_RDWR:
1060 			cap_rights_set_one(rightsp, CAP_READ);
1061 			/* FALLTHROUGH */
1062 		case O_WRONLY:
1063 			cap_rights_set_one(rightsp, CAP_WRITE);
1064 			if (!(flags & (O_APPEND | O_TRUNC)))
1065 				cap_rights_set_one(rightsp, CAP_SEEK);
1066 			break;
1067 		}
1068 	}
1069 
1070 	if (flags & O_CREAT)
1071 		cap_rights_set_one(rightsp, CAP_CREATE);
1072 
1073 	if (flags & O_TRUNC)
1074 		cap_rights_set_one(rightsp, CAP_FTRUNCATE);
1075 
1076 	if (flags & (O_SYNC | O_FSYNC))
1077 		cap_rights_set_one(rightsp, CAP_FSYNC);
1078 
1079 	if (flags & (O_EXLOCK | O_SHLOCK))
1080 		cap_rights_set_one(rightsp, CAP_FLOCK);
1081 }
1082 
1083 /*
1084  * Check permissions, allocate an open file structure, and call the device
1085  * open routine if any.
1086  */
1087 #ifndef _SYS_SYSPROTO_H_
1088 struct open_args {
1089 	char	*path;
1090 	int	flags;
1091 	int	mode;
1092 };
1093 #endif
1094 int
1095 sys_open(struct thread *td, struct open_args *uap)
1096 {
1097 
1098 	return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1099 	    uap->flags, uap->mode));
1100 }
1101 
1102 #ifndef _SYS_SYSPROTO_H_
1103 struct openat_args {
1104 	int	fd;
1105 	char	*path;
1106 	int	flag;
1107 	int	mode;
1108 };
1109 #endif
1110 int
1111 sys_openat(struct thread *td, struct openat_args *uap)
1112 {
1113 
1114 	AUDIT_ARG_FD(uap->fd);
1115 	return (kern_openat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag,
1116 	    uap->mode));
1117 }
1118 
1119 int
1120 kern_openat(struct thread *td, int fd, const char *path, enum uio_seg pathseg,
1121     int flags, int mode)
1122 {
1123 	struct proc *p = td->td_proc;
1124 	struct filedesc *fdp;
1125 	struct pwddesc *pdp;
1126 	struct file *fp;
1127 	struct vnode *vp;
1128 	struct nameidata nd;
1129 	cap_rights_t rights;
1130 	int cmode, error, indx;
1131 
1132 	indx = -1;
1133 	fdp = p->p_fd;
1134 	pdp = p->p_pd;
1135 
1136 	AUDIT_ARG_FFLAGS(flags);
1137 	AUDIT_ARG_MODE(mode);
1138 	cap_rights_init_one(&rights, CAP_LOOKUP);
1139 	flags_to_rights(flags, &rights);
1140 
1141 	/*
1142 	 * Only one of the O_EXEC, O_RDONLY, O_WRONLY and O_RDWR flags
1143 	 * may be specified.  On the other hand, for O_PATH any mode
1144 	 * except O_EXEC is ignored.
1145 	 */
1146 	if ((flags & O_PATH) != 0) {
1147 		flags &= ~(O_CREAT | O_ACCMODE);
1148 	} else if ((flags & O_EXEC) != 0) {
1149 		if (flags & O_ACCMODE)
1150 			return (EINVAL);
1151 	} else if ((flags & O_ACCMODE) == O_ACCMODE) {
1152 		return (EINVAL);
1153 	} else {
1154 		flags = FFLAGS(flags);
1155 	}
1156 
1157 	/*
1158 	 * Allocate a file structure. The descriptor to reference it
1159 	 * is allocated and used by finstall_refed() below.
1160 	 */
1161 	error = falloc_noinstall(td, &fp);
1162 	if (error != 0)
1163 		return (error);
1164 	/* Set the flags early so the finit in devfs can pick them up. */
1165 	fp->f_flag = flags & FMASK;
1166 	cmode = ((mode & ~pdp->pd_cmask) & ALLPERMS) & ~S_ISTXT;
1167 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd,
1168 	    &rights);
1169 	td->td_dupfd = -1;		/* XXX check for fdopen */
1170 	error = vn_open(&nd, &flags, cmode, fp);
1171 	if (error != 0) {
1172 		/*
1173 		 * If the vn_open replaced the method vector, something
1174 		 * wonderous happened deep below and we just pass it up
1175 		 * pretending we know what we do.
1176 		 */
1177 		if (error == ENXIO && fp->f_ops != &badfileops) {
1178 			MPASS((flags & O_PATH) == 0);
1179 			goto success;
1180 		}
1181 
1182 		/*
1183 		 * Handle special fdopen() case. bleh.
1184 		 *
1185 		 * Don't do this for relative (capability) lookups; we don't
1186 		 * understand exactly what would happen, and we don't think
1187 		 * that it ever should.
1188 		 */
1189 		if ((nd.ni_resflags & NIRES_STRICTREL) == 0 &&
1190 		    (error == ENODEV || error == ENXIO) &&
1191 		    td->td_dupfd >= 0) {
1192 			error = dupfdopen(td, fdp, td->td_dupfd, flags, error,
1193 			    &indx);
1194 			if (error == 0)
1195 				goto success;
1196 		}
1197 
1198 		goto bad;
1199 	}
1200 	td->td_dupfd = 0;
1201 	NDFREE(&nd, NDF_ONLY_PNBUF);
1202 	vp = nd.ni_vp;
1203 
1204 	/*
1205 	 * Store the vnode, for any f_type. Typically, the vnode use
1206 	 * count is decremented by direct call to vn_closefile() for
1207 	 * files that switched type in the cdevsw fdopen() method.
1208 	 */
1209 	fp->f_vnode = vp;
1210 
1211 	/*
1212 	 * If the file wasn't claimed by devfs bind it to the normal
1213 	 * vnode operations here.
1214 	 */
1215 	if (fp->f_ops == &badfileops) {
1216 		KASSERT(vp->v_type != VFIFO || (flags & O_PATH) != 0,
1217 		    ("Unexpected fifo fp %p vp %p", fp, vp));
1218 		if ((flags & O_PATH) != 0) {
1219 			finit(fp, (flags & FMASK) | (fp->f_flag & FKQALLOWED),
1220 			    DTYPE_VNODE, NULL, &path_fileops);
1221 			vhold(vp);
1222 			vunref(vp);
1223 		} else {
1224 			finit_vnode(fp, flags, NULL, &vnops);
1225 		}
1226 	}
1227 
1228 	VOP_UNLOCK(vp);
1229 	if (flags & O_TRUNC) {
1230 		error = fo_truncate(fp, 0, td->td_ucred, td);
1231 		if (error != 0)
1232 			goto bad;
1233 	}
1234 success:
1235 	/*
1236 	 * If we haven't already installed the FD (for dupfdopen), do so now.
1237 	 */
1238 	if (indx == -1) {
1239 		struct filecaps *fcaps;
1240 
1241 #ifdef CAPABILITIES
1242 		if ((nd.ni_resflags & NIRES_STRICTREL) != 0)
1243 			fcaps = &nd.ni_filecaps;
1244 		else
1245 #endif
1246 			fcaps = NULL;
1247 		error = finstall_refed(td, fp, &indx, flags, fcaps);
1248 		/* On success finstall_refed() consumes fcaps. */
1249 		if (error != 0) {
1250 			filecaps_free(&nd.ni_filecaps);
1251 			goto bad;
1252 		}
1253 	} else {
1254 		filecaps_free(&nd.ni_filecaps);
1255 		falloc_abort(td, fp);
1256 	}
1257 
1258 	td->td_retval[0] = indx;
1259 	return (0);
1260 bad:
1261 	KASSERT(indx == -1, ("indx=%d, should be -1", indx));
1262 	falloc_abort(td, fp);
1263 	return (error);
1264 }
1265 
1266 #ifdef COMPAT_43
1267 /*
1268  * Create a file.
1269  */
1270 #ifndef _SYS_SYSPROTO_H_
1271 struct ocreat_args {
1272 	char	*path;
1273 	int	mode;
1274 };
1275 #endif
1276 int
1277 ocreat(struct thread *td, struct ocreat_args *uap)
1278 {
1279 
1280 	return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1281 	    O_WRONLY | O_CREAT | O_TRUNC, uap->mode));
1282 }
1283 #endif /* COMPAT_43 */
1284 
1285 /*
1286  * Create a special file.
1287  */
1288 #ifndef _SYS_SYSPROTO_H_
1289 struct mknodat_args {
1290 	int	fd;
1291 	char	*path;
1292 	mode_t	mode;
1293 	dev_t	dev;
1294 };
1295 #endif
1296 int
1297 sys_mknodat(struct thread *td, struct mknodat_args *uap)
1298 {
1299 
1300 	return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode,
1301 	    uap->dev));
1302 }
1303 
1304 #if defined(COMPAT_FREEBSD11)
1305 int
1306 freebsd11_mknod(struct thread *td,
1307     struct freebsd11_mknod_args *uap)
1308 {
1309 
1310 	return (kern_mknodat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1311 	    uap->mode, uap->dev));
1312 }
1313 
1314 int
1315 freebsd11_mknodat(struct thread *td,
1316     struct freebsd11_mknodat_args *uap)
1317 {
1318 
1319 	return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode,
1320 	    uap->dev));
1321 }
1322 #endif /* COMPAT_FREEBSD11 */
1323 
1324 int
1325 kern_mknodat(struct thread *td, int fd, const char *path, enum uio_seg pathseg,
1326     int mode, dev_t dev)
1327 {
1328 	struct vnode *vp;
1329 	struct mount *mp;
1330 	struct vattr vattr;
1331 	struct nameidata nd;
1332 	int error, whiteout = 0;
1333 
1334 	AUDIT_ARG_MODE(mode);
1335 	AUDIT_ARG_DEV(dev);
1336 	switch (mode & S_IFMT) {
1337 	case S_IFCHR:
1338 	case S_IFBLK:
1339 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1340 		if (error == 0 && dev == VNOVAL)
1341 			error = EINVAL;
1342 		break;
1343 	case S_IFWHT:
1344 		error = priv_check(td, PRIV_VFS_MKNOD_WHT);
1345 		break;
1346 	case S_IFIFO:
1347 		if (dev == 0)
1348 			return (kern_mkfifoat(td, fd, path, pathseg, mode));
1349 		/* FALLTHROUGH */
1350 	default:
1351 		error = EINVAL;
1352 		break;
1353 	}
1354 	if (error != 0)
1355 		return (error);
1356 	NDPREINIT(&nd);
1357 restart:
1358 	bwillwrite();
1359 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
1360 	    NOCACHE, pathseg, path, fd, &cap_mknodat_rights);
1361 	if ((error = namei(&nd)) != 0)
1362 		return (error);
1363 	vp = nd.ni_vp;
1364 	if (vp != NULL) {
1365 		NDFREE(&nd, NDF_ONLY_PNBUF);
1366 		if (vp == nd.ni_dvp)
1367 			vrele(nd.ni_dvp);
1368 		else
1369 			vput(nd.ni_dvp);
1370 		vrele(vp);
1371 		return (EEXIST);
1372 	} else {
1373 		VATTR_NULL(&vattr);
1374 		vattr.va_mode = (mode & ALLPERMS) &
1375 		    ~td->td_proc->p_pd->pd_cmask;
1376 		vattr.va_rdev = dev;
1377 		whiteout = 0;
1378 
1379 		switch (mode & S_IFMT) {
1380 		case S_IFCHR:
1381 			vattr.va_type = VCHR;
1382 			break;
1383 		case S_IFBLK:
1384 			vattr.va_type = VBLK;
1385 			break;
1386 		case S_IFWHT:
1387 			whiteout = 1;
1388 			break;
1389 		default:
1390 			panic("kern_mknod: invalid mode");
1391 		}
1392 	}
1393 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1394 		NDFREE(&nd, NDF_ONLY_PNBUF);
1395 		vput(nd.ni_dvp);
1396 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1397 			return (error);
1398 		goto restart;
1399 	}
1400 #ifdef MAC
1401 	if (error == 0 && !whiteout)
1402 		error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp,
1403 		    &nd.ni_cnd, &vattr);
1404 #endif
1405 	if (error == 0) {
1406 		if (whiteout)
1407 			error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, CREATE);
1408 		else {
1409 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp,
1410 						&nd.ni_cnd, &vattr);
1411 		}
1412 	}
1413 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 && !whiteout ? &nd.ni_vp : NULL,
1414 	    true);
1415 	vn_finished_write(mp);
1416 	NDFREE(&nd, NDF_ONLY_PNBUF);
1417 	if (error == ERELOOKUP)
1418 		goto restart;
1419 	return (error);
1420 }
1421 
1422 /*
1423  * Create a named pipe.
1424  */
1425 #ifndef _SYS_SYSPROTO_H_
1426 struct mkfifo_args {
1427 	char	*path;
1428 	int	mode;
1429 };
1430 #endif
1431 int
1432 sys_mkfifo(struct thread *td, struct mkfifo_args *uap)
1433 {
1434 
1435 	return (kern_mkfifoat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
1436 	    uap->mode));
1437 }
1438 
1439 #ifndef _SYS_SYSPROTO_H_
1440 struct mkfifoat_args {
1441 	int	fd;
1442 	char	*path;
1443 	mode_t	mode;
1444 };
1445 #endif
1446 int
1447 sys_mkfifoat(struct thread *td, struct mkfifoat_args *uap)
1448 {
1449 
1450 	return (kern_mkfifoat(td, uap->fd, uap->path, UIO_USERSPACE,
1451 	    uap->mode));
1452 }
1453 
1454 int
1455 kern_mkfifoat(struct thread *td, int fd, const char *path,
1456     enum uio_seg pathseg, int mode)
1457 {
1458 	struct mount *mp;
1459 	struct vattr vattr;
1460 	struct nameidata nd;
1461 	int error;
1462 
1463 	AUDIT_ARG_MODE(mode);
1464 	NDPREINIT(&nd);
1465 restart:
1466 	bwillwrite();
1467 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
1468 	    NOCACHE, pathseg, path, fd, &cap_mkfifoat_rights);
1469 	if ((error = namei(&nd)) != 0)
1470 		return (error);
1471 	if (nd.ni_vp != NULL) {
1472 		NDFREE(&nd, NDF_ONLY_PNBUF);
1473 		if (nd.ni_vp == nd.ni_dvp)
1474 			vrele(nd.ni_dvp);
1475 		else
1476 			vput(nd.ni_dvp);
1477 		vrele(nd.ni_vp);
1478 		return (EEXIST);
1479 	}
1480 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1481 		NDFREE(&nd, NDF_ONLY_PNBUF);
1482 		vput(nd.ni_dvp);
1483 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1484 			return (error);
1485 		goto restart;
1486 	}
1487 	VATTR_NULL(&vattr);
1488 	vattr.va_type = VFIFO;
1489 	vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_pd->pd_cmask;
1490 #ifdef MAC
1491 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
1492 	    &vattr);
1493 	if (error != 0)
1494 		goto out;
1495 #endif
1496 	error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
1497 #ifdef MAC
1498 out:
1499 #endif
1500 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true);
1501 	vn_finished_write(mp);
1502 	NDFREE(&nd, NDF_ONLY_PNBUF);
1503 	if (error == ERELOOKUP)
1504 		goto restart;
1505 	return (error);
1506 }
1507 
1508 /*
1509  * Make a hard file link.
1510  */
1511 #ifndef _SYS_SYSPROTO_H_
1512 struct link_args {
1513 	char	*path;
1514 	char	*link;
1515 };
1516 #endif
1517 int
1518 sys_link(struct thread *td, struct link_args *uap)
1519 {
1520 
1521 	return (kern_linkat(td, AT_FDCWD, AT_FDCWD, uap->path, uap->link,
1522 	    UIO_USERSPACE, AT_SYMLINK_FOLLOW));
1523 }
1524 
1525 #ifndef _SYS_SYSPROTO_H_
1526 struct linkat_args {
1527 	int	fd1;
1528 	char	*path1;
1529 	int	fd2;
1530 	char	*path2;
1531 	int	flag;
1532 };
1533 #endif
1534 int
1535 sys_linkat(struct thread *td, struct linkat_args *uap)
1536 {
1537 
1538 	return (kern_linkat(td, uap->fd1, uap->fd2, uap->path1, uap->path2,
1539 	    UIO_USERSPACE, uap->flag));
1540 }
1541 
1542 int hardlink_check_uid = 0;
1543 SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1544     &hardlink_check_uid, 0,
1545     "Unprivileged processes cannot create hard links to files owned by other "
1546     "users");
1547 static int hardlink_check_gid = 0;
1548 SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1549     &hardlink_check_gid, 0,
1550     "Unprivileged processes cannot create hard links to files owned by other "
1551     "groups");
1552 
1553 static int
1554 can_hardlink(struct vnode *vp, struct ucred *cred)
1555 {
1556 	struct vattr va;
1557 	int error;
1558 
1559 	if (!hardlink_check_uid && !hardlink_check_gid)
1560 		return (0);
1561 
1562 	error = VOP_GETATTR(vp, &va, cred);
1563 	if (error != 0)
1564 		return (error);
1565 
1566 	if (hardlink_check_uid && cred->cr_uid != va.va_uid) {
1567 		error = priv_check_cred(cred, PRIV_VFS_LINK);
1568 		if (error != 0)
1569 			return (error);
1570 	}
1571 
1572 	if (hardlink_check_gid && !groupmember(va.va_gid, cred)) {
1573 		error = priv_check_cred(cred, PRIV_VFS_LINK);
1574 		if (error != 0)
1575 			return (error);
1576 	}
1577 
1578 	return (0);
1579 }
1580 
1581 int
1582 kern_linkat(struct thread *td, int fd1, int fd2, const char *path1,
1583     const char *path2, enum uio_seg segflag, int flag)
1584 {
1585 	struct nameidata nd;
1586 	int error;
1587 
1588 	if ((flag & ~(AT_SYMLINK_FOLLOW | AT_RESOLVE_BENEATH |
1589 	    AT_EMPTY_PATH)) != 0)
1590 		return (EINVAL);
1591 
1592 	NDPREINIT(&nd);
1593 	do {
1594 		bwillwrite();
1595 		NDINIT_ATRIGHTS(&nd, LOOKUP, AUDITVNODE1 | at2cnpflags(flag,
1596 		    AT_SYMLINK_FOLLOW | AT_RESOLVE_BENEATH | AT_EMPTY_PATH),
1597 		    segflag, path1, fd1, &cap_linkat_source_rights);
1598 		if ((error = namei(&nd)) != 0)
1599 			return (error);
1600 		NDFREE(&nd, NDF_ONLY_PNBUF);
1601 		if ((nd.ni_resflags & NIRES_EMPTYPATH) != 0) {
1602 			error = priv_check(td, PRIV_VFS_FHOPEN);
1603 			if (error != 0) {
1604 				vrele(nd.ni_vp);
1605 				return (error);
1606 			}
1607 		}
1608 		error = kern_linkat_vp(td, nd.ni_vp, fd2, path2, segflag);
1609 	} while (error ==  EAGAIN || error == ERELOOKUP);
1610 	return (error);
1611 }
1612 
1613 static int
1614 kern_linkat_vp(struct thread *td, struct vnode *vp, int fd, const char *path,
1615     enum uio_seg segflag)
1616 {
1617 	struct nameidata nd;
1618 	struct mount *mp;
1619 	int error;
1620 
1621 	if (vp->v_type == VDIR) {
1622 		vrele(vp);
1623 		return (EPERM);		/* POSIX */
1624 	}
1625 	NDINIT_ATRIGHTS(&nd, CREATE,
1626 	    LOCKPARENT | SAVENAME | AUDITVNODE2 | NOCACHE, segflag, path, fd,
1627 	    &cap_linkat_target_rights);
1628 	if ((error = namei(&nd)) == 0) {
1629 		if (nd.ni_vp != NULL) {
1630 			NDFREE(&nd, NDF_ONLY_PNBUF);
1631 			if (nd.ni_dvp == nd.ni_vp)
1632 				vrele(nd.ni_dvp);
1633 			else
1634 				vput(nd.ni_dvp);
1635 			vrele(nd.ni_vp);
1636 			vrele(vp);
1637 			return (EEXIST);
1638 		} else if (nd.ni_dvp->v_mount != vp->v_mount) {
1639 			/*
1640 			 * Cross-device link.  No need to recheck
1641 			 * vp->v_type, since it cannot change, except
1642 			 * to VBAD.
1643 			 */
1644 			NDFREE(&nd, NDF_ONLY_PNBUF);
1645 			vput(nd.ni_dvp);
1646 			vrele(vp);
1647 			return (EXDEV);
1648 		} else if ((error = vn_lock(vp, LK_EXCLUSIVE)) == 0) {
1649 			error = can_hardlink(vp, td->td_ucred);
1650 #ifdef MAC
1651 			if (error == 0)
1652 				error = mac_vnode_check_link(td->td_ucred,
1653 				    nd.ni_dvp, vp, &nd.ni_cnd);
1654 #endif
1655 			if (error != 0) {
1656 				vput(vp);
1657 				vput(nd.ni_dvp);
1658 				NDFREE(&nd, NDF_ONLY_PNBUF);
1659 				return (error);
1660 			}
1661 			error = vn_start_write(vp, &mp, V_NOWAIT);
1662 			if (error != 0) {
1663 				vput(vp);
1664 				vput(nd.ni_dvp);
1665 				NDFREE(&nd, NDF_ONLY_PNBUF);
1666 				error = vn_start_write(NULL, &mp,
1667 				    V_XSLEEP | PCATCH);
1668 				if (error != 0)
1669 					return (error);
1670 				return (EAGAIN);
1671 			}
1672 			error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
1673 			VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
1674 			vn_finished_write(mp);
1675 			NDFREE(&nd, NDF_ONLY_PNBUF);
1676 			vp = NULL;
1677 		} else {
1678 			vput(nd.ni_dvp);
1679 			NDFREE(&nd, NDF_ONLY_PNBUF);
1680 			vrele(vp);
1681 			return (EAGAIN);
1682 		}
1683 	}
1684 	if (vp != NULL)
1685 		vrele(vp);
1686 	return (error);
1687 }
1688 
1689 /*
1690  * Make a symbolic link.
1691  */
1692 #ifndef _SYS_SYSPROTO_H_
1693 struct symlink_args {
1694 	char	*path;
1695 	char	*link;
1696 };
1697 #endif
1698 int
1699 sys_symlink(struct thread *td, struct symlink_args *uap)
1700 {
1701 
1702 	return (kern_symlinkat(td, uap->path, AT_FDCWD, uap->link,
1703 	    UIO_USERSPACE));
1704 }
1705 
1706 #ifndef _SYS_SYSPROTO_H_
1707 struct symlinkat_args {
1708 	char	*path;
1709 	int	fd;
1710 	char	*path2;
1711 };
1712 #endif
1713 int
1714 sys_symlinkat(struct thread *td, struct symlinkat_args *uap)
1715 {
1716 
1717 	return (kern_symlinkat(td, uap->path1, uap->fd, uap->path2,
1718 	    UIO_USERSPACE));
1719 }
1720 
1721 int
1722 kern_symlinkat(struct thread *td, const char *path1, int fd, const char *path2,
1723     enum uio_seg segflg)
1724 {
1725 	struct mount *mp;
1726 	struct vattr vattr;
1727 	const char *syspath;
1728 	char *tmppath;
1729 	struct nameidata nd;
1730 	int error;
1731 
1732 	if (segflg == UIO_SYSSPACE) {
1733 		syspath = path1;
1734 	} else {
1735 		tmppath = uma_zalloc(namei_zone, M_WAITOK);
1736 		if ((error = copyinstr(path1, tmppath, MAXPATHLEN, NULL)) != 0)
1737 			goto out;
1738 		syspath = tmppath;
1739 	}
1740 	AUDIT_ARG_TEXT(syspath);
1741 	NDPREINIT(&nd);
1742 restart:
1743 	bwillwrite();
1744 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
1745 	    NOCACHE, segflg, path2, fd, &cap_symlinkat_rights);
1746 	if ((error = namei(&nd)) != 0)
1747 		goto out;
1748 	if (nd.ni_vp) {
1749 		NDFREE(&nd, NDF_ONLY_PNBUF);
1750 		if (nd.ni_vp == nd.ni_dvp)
1751 			vrele(nd.ni_dvp);
1752 		else
1753 			vput(nd.ni_dvp);
1754 		vrele(nd.ni_vp);
1755 		nd.ni_vp = NULL;
1756 		error = EEXIST;
1757 		goto out;
1758 	}
1759 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1760 		NDFREE(&nd, NDF_ONLY_PNBUF);
1761 		vput(nd.ni_dvp);
1762 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1763 			goto out;
1764 		goto restart;
1765 	}
1766 	VATTR_NULL(&vattr);
1767 	vattr.va_mode = ACCESSPERMS &~ td->td_proc->p_pd->pd_cmask;
1768 #ifdef MAC
1769 	vattr.va_type = VLNK;
1770 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
1771 	    &vattr);
1772 	if (error != 0)
1773 		goto out2;
1774 #endif
1775 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr, syspath);
1776 #ifdef MAC
1777 out2:
1778 #endif
1779 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true);
1780 	vn_finished_write(mp);
1781 	NDFREE(&nd, NDF_ONLY_PNBUF);
1782 	if (error == ERELOOKUP)
1783 		goto restart;
1784 out:
1785 	if (segflg != UIO_SYSSPACE)
1786 		uma_zfree(namei_zone, tmppath);
1787 	return (error);
1788 }
1789 
1790 /*
1791  * Delete a whiteout from the filesystem.
1792  */
1793 #ifndef _SYS_SYSPROTO_H_
1794 struct undelete_args {
1795 	char *path;
1796 };
1797 #endif
1798 int
1799 sys_undelete(struct thread *td, struct undelete_args *uap)
1800 {
1801 	struct mount *mp;
1802 	struct nameidata nd;
1803 	int error;
1804 
1805 	NDPREINIT(&nd);
1806 restart:
1807 	bwillwrite();
1808 	NDINIT(&nd, DELETE, LOCKPARENT | DOWHITEOUT | AUDITVNODE1,
1809 	    UIO_USERSPACE, uap->path);
1810 	error = namei(&nd);
1811 	if (error != 0)
1812 		return (error);
1813 
1814 	if (nd.ni_vp != NULLVP || !(nd.ni_cnd.cn_flags & ISWHITEOUT)) {
1815 		NDFREE(&nd, NDF_ONLY_PNBUF);
1816 		if (nd.ni_vp == nd.ni_dvp)
1817 			vrele(nd.ni_dvp);
1818 		else
1819 			vput(nd.ni_dvp);
1820 		if (nd.ni_vp)
1821 			vrele(nd.ni_vp);
1822 		return (EEXIST);
1823 	}
1824 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1825 		NDFREE(&nd, NDF_ONLY_PNBUF);
1826 		vput(nd.ni_dvp);
1827 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
1828 			return (error);
1829 		goto restart;
1830 	}
1831 	error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, DELETE);
1832 	NDFREE(&nd, NDF_ONLY_PNBUF);
1833 	vput(nd.ni_dvp);
1834 	vn_finished_write(mp);
1835 	if (error == ERELOOKUP)
1836 		goto restart;
1837 	return (error);
1838 }
1839 
1840 /*
1841  * Delete a name from the filesystem.
1842  */
1843 #ifndef _SYS_SYSPROTO_H_
1844 struct unlink_args {
1845 	char	*path;
1846 };
1847 #endif
1848 int
1849 sys_unlink(struct thread *td, struct unlink_args *uap)
1850 {
1851 
1852 	return (kern_funlinkat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE,
1853 	    0, 0));
1854 }
1855 
1856 static int
1857 kern_funlinkat_ex(struct thread *td, int dfd, const char *path, int fd,
1858     int flag, enum uio_seg pathseg, ino_t oldinum)
1859 {
1860 
1861 	if ((flag & ~(AT_REMOVEDIR | AT_RESOLVE_BENEATH)) != 0)
1862 		return (EINVAL);
1863 
1864 	if ((flag & AT_REMOVEDIR) != 0)
1865 		return (kern_frmdirat(td, dfd, path, fd, UIO_USERSPACE, 0));
1866 
1867 	return (kern_funlinkat(td, dfd, path, fd, UIO_USERSPACE, 0, 0));
1868 }
1869 
1870 #ifndef _SYS_SYSPROTO_H_
1871 struct unlinkat_args {
1872 	int	fd;
1873 	char	*path;
1874 	int	flag;
1875 };
1876 #endif
1877 int
1878 sys_unlinkat(struct thread *td, struct unlinkat_args *uap)
1879 {
1880 
1881 	return (kern_funlinkat_ex(td, uap->fd, uap->path, FD_NONE, uap->flag,
1882 	    UIO_USERSPACE, 0));
1883 }
1884 
1885 #ifndef _SYS_SYSPROTO_H_
1886 struct funlinkat_args {
1887 	int		dfd;
1888 	const char	*path;
1889 	int		fd;
1890 	int		flag;
1891 };
1892 #endif
1893 int
1894 sys_funlinkat(struct thread *td, struct funlinkat_args *uap)
1895 {
1896 
1897 	return (kern_funlinkat_ex(td, uap->dfd, uap->path, uap->fd, uap->flag,
1898 	    UIO_USERSPACE, 0));
1899 }
1900 
1901 int
1902 kern_funlinkat(struct thread *td, int dfd, const char *path, int fd,
1903     enum uio_seg pathseg, int flag, ino_t oldinum)
1904 {
1905 	struct mount *mp;
1906 	struct file *fp;
1907 	struct vnode *vp;
1908 	struct nameidata nd;
1909 	struct stat sb;
1910 	int error;
1911 
1912 	fp = NULL;
1913 	if (fd != FD_NONE) {
1914 		error = getvnode_path(td, fd, &cap_no_rights, &fp);
1915 		if (error != 0)
1916 			return (error);
1917 	}
1918 
1919 	NDPREINIT(&nd);
1920 restart:
1921 	bwillwrite();
1922 	NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 |
1923 	    at2cnpflags(flag, AT_RESOLVE_BENEATH),
1924 	    pathseg, path, dfd, &cap_unlinkat_rights);
1925 	if ((error = namei(&nd)) != 0) {
1926 		if (error == EINVAL)
1927 			error = EPERM;
1928 		goto fdout;
1929 	}
1930 	vp = nd.ni_vp;
1931 	if (vp->v_type == VDIR && oldinum == 0) {
1932 		error = EPERM;		/* POSIX */
1933 	} else if (oldinum != 0 &&
1934 	    ((error = VOP_STAT(vp, &sb, td->td_ucred, NOCRED)) == 0) &&
1935 	    sb.st_ino != oldinum) {
1936 		error = EIDRM;	/* Identifier removed */
1937 	} else if (fp != NULL && fp->f_vnode != vp) {
1938 		if (VN_IS_DOOMED(fp->f_vnode))
1939 			error = EBADF;
1940 		else
1941 			error = EDEADLK;
1942 	} else {
1943 		/*
1944 		 * The root of a mounted filesystem cannot be deleted.
1945 		 *
1946 		 * XXX: can this only be a VDIR case?
1947 		 */
1948 		if (vp->v_vflag & VV_ROOT)
1949 			error = EBUSY;
1950 	}
1951 	if (error == 0) {
1952 		if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
1953 			NDFREE(&nd, NDF_ONLY_PNBUF);
1954 			vput(nd.ni_dvp);
1955 			if (vp == nd.ni_dvp)
1956 				vrele(vp);
1957 			else
1958 				vput(vp);
1959 			if ((error = vn_start_write(NULL, &mp,
1960 			    V_XSLEEP | PCATCH)) != 0) {
1961 				goto fdout;
1962 			}
1963 			goto restart;
1964 		}
1965 #ifdef MAC
1966 		error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp,
1967 		    &nd.ni_cnd);
1968 		if (error != 0)
1969 			goto out;
1970 #endif
1971 		vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK);
1972 		error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
1973 #ifdef MAC
1974 out:
1975 #endif
1976 		vn_finished_write(mp);
1977 	}
1978 	NDFREE(&nd, NDF_ONLY_PNBUF);
1979 	vput(nd.ni_dvp);
1980 	if (vp == nd.ni_dvp)
1981 		vrele(vp);
1982 	else
1983 		vput(vp);
1984 	if (error == ERELOOKUP)
1985 		goto restart;
1986 fdout:
1987 	if (fp != NULL)
1988 		fdrop(fp, td);
1989 	return (error);
1990 }
1991 
1992 /*
1993  * Reposition read/write file offset.
1994  */
1995 #ifndef _SYS_SYSPROTO_H_
1996 struct lseek_args {
1997 	int	fd;
1998 	int	pad;
1999 	off_t	offset;
2000 	int	whence;
2001 };
2002 #endif
2003 int
2004 sys_lseek(struct thread *td, struct lseek_args *uap)
2005 {
2006 
2007 	return (kern_lseek(td, uap->fd, uap->offset, uap->whence));
2008 }
2009 
2010 int
2011 kern_lseek(struct thread *td, int fd, off_t offset, int whence)
2012 {
2013 	struct file *fp;
2014 	int error;
2015 
2016 	AUDIT_ARG_FD(fd);
2017 	error = fget(td, fd, &cap_seek_rights, &fp);
2018 	if (error != 0)
2019 		return (error);
2020 	error = (fp->f_ops->fo_flags & DFLAG_SEEKABLE) != 0 ?
2021 	    fo_seek(fp, offset, whence, td) : ESPIPE;
2022 	fdrop(fp, td);
2023 	return (error);
2024 }
2025 
2026 #if defined(COMPAT_43)
2027 /*
2028  * Reposition read/write file offset.
2029  */
2030 #ifndef _SYS_SYSPROTO_H_
2031 struct olseek_args {
2032 	int	fd;
2033 	long	offset;
2034 	int	whence;
2035 };
2036 #endif
2037 int
2038 olseek(struct thread *td, struct olseek_args *uap)
2039 {
2040 
2041 	return (kern_lseek(td, uap->fd, uap->offset, uap->whence));
2042 }
2043 #endif /* COMPAT_43 */
2044 
2045 #if defined(COMPAT_FREEBSD6)
2046 /* Version with the 'pad' argument */
2047 int
2048 freebsd6_lseek(struct thread *td, struct freebsd6_lseek_args *uap)
2049 {
2050 
2051 	return (kern_lseek(td, uap->fd, uap->offset, uap->whence));
2052 }
2053 #endif
2054 
2055 /*
2056  * Check access permissions using passed credentials.
2057  */
2058 static int
2059 vn_access(struct vnode *vp, int user_flags, struct ucred *cred,
2060      struct thread *td)
2061 {
2062 	accmode_t accmode;
2063 	int error;
2064 
2065 	/* Flags == 0 means only check for existence. */
2066 	if (user_flags == 0)
2067 		return (0);
2068 
2069 	accmode = 0;
2070 	if (user_flags & R_OK)
2071 		accmode |= VREAD;
2072 	if (user_flags & W_OK)
2073 		accmode |= VWRITE;
2074 	if (user_flags & X_OK)
2075 		accmode |= VEXEC;
2076 #ifdef MAC
2077 	error = mac_vnode_check_access(cred, vp, accmode);
2078 	if (error != 0)
2079 		return (error);
2080 #endif
2081 	if ((accmode & VWRITE) == 0 || (error = vn_writechk(vp)) == 0)
2082 		error = VOP_ACCESS(vp, accmode, cred, td);
2083 	return (error);
2084 }
2085 
2086 /*
2087  * Check access permissions using "real" credentials.
2088  */
2089 #ifndef _SYS_SYSPROTO_H_
2090 struct access_args {
2091 	char	*path;
2092 	int	amode;
2093 };
2094 #endif
2095 int
2096 sys_access(struct thread *td, struct access_args *uap)
2097 {
2098 
2099 	return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2100 	    0, uap->amode));
2101 }
2102 
2103 #ifndef _SYS_SYSPROTO_H_
2104 struct faccessat_args {
2105 	int	dirfd;
2106 	char	*path;
2107 	int	amode;
2108 	int	flag;
2109 }
2110 #endif
2111 int
2112 sys_faccessat(struct thread *td, struct faccessat_args *uap)
2113 {
2114 
2115 	return (kern_accessat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag,
2116 	    uap->amode));
2117 }
2118 
2119 int
2120 kern_accessat(struct thread *td, int fd, const char *path,
2121     enum uio_seg pathseg, int flag, int amode)
2122 {
2123 	struct ucred *cred, *usecred;
2124 	struct vnode *vp;
2125 	struct nameidata nd;
2126 	int error;
2127 
2128 	if ((flag & ~(AT_EACCESS | AT_RESOLVE_BENEATH | AT_EMPTY_PATH)) != 0)
2129 		return (EINVAL);
2130 	if (amode != F_OK && (amode & ~(R_OK | W_OK | X_OK)) != 0)
2131 		return (EINVAL);
2132 
2133 	/*
2134 	 * Create and modify a temporary credential instead of one that
2135 	 * is potentially shared (if we need one).
2136 	 */
2137 	cred = td->td_ucred;
2138 	if ((flag & AT_EACCESS) == 0 &&
2139 	    ((cred->cr_uid != cred->cr_ruid ||
2140 	    cred->cr_rgid != cred->cr_groups[0]))) {
2141 		usecred = crdup(cred);
2142 		usecred->cr_uid = cred->cr_ruid;
2143 		usecred->cr_groups[0] = cred->cr_rgid;
2144 		td->td_ucred = usecred;
2145 	} else
2146 		usecred = cred;
2147 	AUDIT_ARG_VALUE(amode);
2148 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF |
2149 	    AUDITVNODE1 | at2cnpflags(flag, AT_RESOLVE_BENEATH |
2150 	    AT_EMPTY_PATH), pathseg, path, fd, &cap_fstat_rights);
2151 	if ((error = namei(&nd)) != 0)
2152 		goto out;
2153 	vp = nd.ni_vp;
2154 
2155 	error = vn_access(vp, amode, usecred, td);
2156 	NDFREE_NOTHING(&nd);
2157 	vput(vp);
2158 out:
2159 	if (usecred != cred) {
2160 		td->td_ucred = cred;
2161 		crfree(usecred);
2162 	}
2163 	return (error);
2164 }
2165 
2166 /*
2167  * Check access permissions using "effective" credentials.
2168  */
2169 #ifndef _SYS_SYSPROTO_H_
2170 struct eaccess_args {
2171 	char	*path;
2172 	int	amode;
2173 };
2174 #endif
2175 int
2176 sys_eaccess(struct thread *td, struct eaccess_args *uap)
2177 {
2178 
2179 	return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2180 	    AT_EACCESS, uap->amode));
2181 }
2182 
2183 #if defined(COMPAT_43)
2184 /*
2185  * Get file status; this version follows links.
2186  */
2187 #ifndef _SYS_SYSPROTO_H_
2188 struct ostat_args {
2189 	char	*path;
2190 	struct ostat *ub;
2191 };
2192 #endif
2193 int
2194 ostat(struct thread *td, struct ostat_args *uap)
2195 {
2196 	struct stat sb;
2197 	struct ostat osb;
2198 	int error;
2199 
2200 	error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE,
2201 	    &sb, NULL);
2202 	if (error != 0)
2203 		return (error);
2204 	cvtstat(&sb, &osb);
2205 	return (copyout(&osb, uap->ub, sizeof (osb)));
2206 }
2207 
2208 /*
2209  * Get file status; this version does not follow links.
2210  */
2211 #ifndef _SYS_SYSPROTO_H_
2212 struct olstat_args {
2213 	char	*path;
2214 	struct ostat *ub;
2215 };
2216 #endif
2217 int
2218 olstat(struct thread *td, struct olstat_args *uap)
2219 {
2220 	struct stat sb;
2221 	struct ostat osb;
2222 	int error;
2223 
2224 	error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path,
2225 	    UIO_USERSPACE, &sb, NULL);
2226 	if (error != 0)
2227 		return (error);
2228 	cvtstat(&sb, &osb);
2229 	return (copyout(&osb, uap->ub, sizeof (osb)));
2230 }
2231 
2232 /*
2233  * Convert from an old to a new stat structure.
2234  * XXX: many values are blindly truncated.
2235  */
2236 void
2237 cvtstat(struct stat *st, struct ostat *ost)
2238 {
2239 
2240 	bzero(ost, sizeof(*ost));
2241 	ost->st_dev = st->st_dev;
2242 	ost->st_ino = st->st_ino;
2243 	ost->st_mode = st->st_mode;
2244 	ost->st_nlink = st->st_nlink;
2245 	ost->st_uid = st->st_uid;
2246 	ost->st_gid = st->st_gid;
2247 	ost->st_rdev = st->st_rdev;
2248 	ost->st_size = MIN(st->st_size, INT32_MAX);
2249 	ost->st_atim = st->st_atim;
2250 	ost->st_mtim = st->st_mtim;
2251 	ost->st_ctim = st->st_ctim;
2252 	ost->st_blksize = st->st_blksize;
2253 	ost->st_blocks = st->st_blocks;
2254 	ost->st_flags = st->st_flags;
2255 	ost->st_gen = st->st_gen;
2256 }
2257 #endif /* COMPAT_43 */
2258 
2259 #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11)
2260 int ino64_trunc_error;
2261 SYSCTL_INT(_vfs, OID_AUTO, ino64_trunc_error, CTLFLAG_RW,
2262     &ino64_trunc_error, 0,
2263     "Error on truncation of device, file or inode number, or link count");
2264 
2265 int
2266 freebsd11_cvtstat(struct stat *st, struct freebsd11_stat *ost)
2267 {
2268 
2269 	ost->st_dev = st->st_dev;
2270 	if (ost->st_dev != st->st_dev) {
2271 		switch (ino64_trunc_error) {
2272 		default:
2273 			/*
2274 			 * Since dev_t is almost raw, don't clamp to the
2275 			 * maximum for case 2, but ignore the error.
2276 			 */
2277 			break;
2278 		case 1:
2279 			return (EOVERFLOW);
2280 		}
2281 	}
2282 	ost->st_ino = st->st_ino;
2283 	if (ost->st_ino != st->st_ino) {
2284 		switch (ino64_trunc_error) {
2285 		default:
2286 		case 0:
2287 			break;
2288 		case 1:
2289 			return (EOVERFLOW);
2290 		case 2:
2291 			ost->st_ino = UINT32_MAX;
2292 			break;
2293 		}
2294 	}
2295 	ost->st_mode = st->st_mode;
2296 	ost->st_nlink = st->st_nlink;
2297 	if (ost->st_nlink != st->st_nlink) {
2298 		switch (ino64_trunc_error) {
2299 		default:
2300 		case 0:
2301 			break;
2302 		case 1:
2303 			return (EOVERFLOW);
2304 		case 2:
2305 			ost->st_nlink = UINT16_MAX;
2306 			break;
2307 		}
2308 	}
2309 	ost->st_uid = st->st_uid;
2310 	ost->st_gid = st->st_gid;
2311 	ost->st_rdev = st->st_rdev;
2312 	if (ost->st_rdev != st->st_rdev) {
2313 		switch (ino64_trunc_error) {
2314 		default:
2315 			break;
2316 		case 1:
2317 			return (EOVERFLOW);
2318 		}
2319 	}
2320 	ost->st_atim = st->st_atim;
2321 	ost->st_mtim = st->st_mtim;
2322 	ost->st_ctim = st->st_ctim;
2323 	ost->st_size = st->st_size;
2324 	ost->st_blocks = st->st_blocks;
2325 	ost->st_blksize = st->st_blksize;
2326 	ost->st_flags = st->st_flags;
2327 	ost->st_gen = st->st_gen;
2328 	ost->st_lspare = 0;
2329 	ost->st_birthtim = st->st_birthtim;
2330 	bzero((char *)&ost->st_birthtim + sizeof(ost->st_birthtim),
2331 	    sizeof(*ost) - offsetof(struct freebsd11_stat,
2332 	    st_birthtim) - sizeof(ost->st_birthtim));
2333 	return (0);
2334 }
2335 
2336 int
2337 freebsd11_stat(struct thread *td, struct freebsd11_stat_args* uap)
2338 {
2339 	struct stat sb;
2340 	struct freebsd11_stat osb;
2341 	int error;
2342 
2343 	error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE,
2344 	    &sb, NULL);
2345 	if (error != 0)
2346 		return (error);
2347 	error = freebsd11_cvtstat(&sb, &osb);
2348 	if (error == 0)
2349 		error = copyout(&osb, uap->ub, sizeof(osb));
2350 	return (error);
2351 }
2352 
2353 int
2354 freebsd11_lstat(struct thread *td, struct freebsd11_lstat_args* uap)
2355 {
2356 	struct stat sb;
2357 	struct freebsd11_stat osb;
2358 	int error;
2359 
2360 	error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path,
2361 	    UIO_USERSPACE, &sb, NULL);
2362 	if (error != 0)
2363 		return (error);
2364 	error = freebsd11_cvtstat(&sb, &osb);
2365 	if (error == 0)
2366 		error = copyout(&osb, uap->ub, sizeof(osb));
2367 	return (error);
2368 }
2369 
2370 int
2371 freebsd11_fhstat(struct thread *td, struct freebsd11_fhstat_args* uap)
2372 {
2373 	struct fhandle fh;
2374 	struct stat sb;
2375 	struct freebsd11_stat osb;
2376 	int error;
2377 
2378 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
2379 	if (error != 0)
2380 		return (error);
2381 	error = kern_fhstat(td, fh, &sb);
2382 	if (error != 0)
2383 		return (error);
2384 	error = freebsd11_cvtstat(&sb, &osb);
2385 	if (error == 0)
2386 		error = copyout(&osb, uap->sb, sizeof(osb));
2387 	return (error);
2388 }
2389 
2390 int
2391 freebsd11_fstatat(struct thread *td, struct freebsd11_fstatat_args* uap)
2392 {
2393 	struct stat sb;
2394 	struct freebsd11_stat osb;
2395 	int error;
2396 
2397 	error = kern_statat(td, uap->flag, uap->fd, uap->path,
2398 	    UIO_USERSPACE, &sb, NULL);
2399 	if (error != 0)
2400 		return (error);
2401 	error = freebsd11_cvtstat(&sb, &osb);
2402 	if (error == 0)
2403 		error = copyout(&osb, uap->buf, sizeof(osb));
2404 	return (error);
2405 }
2406 #endif	/* COMPAT_FREEBSD11 */
2407 
2408 /*
2409  * Get file status
2410  */
2411 #ifndef _SYS_SYSPROTO_H_
2412 struct fstatat_args {
2413 	int	fd;
2414 	char	*path;
2415 	struct stat	*buf;
2416 	int	flag;
2417 }
2418 #endif
2419 int
2420 sys_fstatat(struct thread *td, struct fstatat_args *uap)
2421 {
2422 	struct stat sb;
2423 	int error;
2424 
2425 	error = kern_statat(td, uap->flag, uap->fd, uap->path,
2426 	    UIO_USERSPACE, &sb, NULL);
2427 	if (error == 0)
2428 		error = copyout(&sb, uap->buf, sizeof (sb));
2429 	return (error);
2430 }
2431 
2432 int
2433 kern_statat(struct thread *td, int flag, int fd, const char *path,
2434     enum uio_seg pathseg, struct stat *sbp,
2435     void (*hook)(struct vnode *vp, struct stat *sbp))
2436 {
2437 	struct nameidata nd;
2438 	int error;
2439 
2440 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
2441 	    AT_EMPTY_PATH)) != 0)
2442 		return (EINVAL);
2443 
2444 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_RESOLVE_BENEATH |
2445 	    AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH) | LOCKSHARED | LOCKLEAF |
2446 	    AUDITVNODE1, pathseg, path, fd, &cap_fstat_rights);
2447 
2448 	if ((error = namei(&nd)) != 0) {
2449 		if (error == ENOTDIR &&
2450 		    (nd.ni_resflags & NIRES_EMPTYPATH) != 0)
2451 			error = kern_fstat(td, fd, sbp);
2452 		return (error);
2453 	}
2454 	error = VOP_STAT(nd.ni_vp, sbp, td->td_ucred, NOCRED);
2455 	if (error == 0) {
2456 		if (__predict_false(hook != NULL))
2457 			hook(nd.ni_vp, sbp);
2458 	}
2459 	NDFREE_NOTHING(&nd);
2460 	vput(nd.ni_vp);
2461 #ifdef __STAT_TIME_T_EXT
2462 	sbp->st_atim_ext = 0;
2463 	sbp->st_mtim_ext = 0;
2464 	sbp->st_ctim_ext = 0;
2465 	sbp->st_btim_ext = 0;
2466 #endif
2467 #ifdef KTRACE
2468 	if (KTRPOINT(td, KTR_STRUCT))
2469 		ktrstat_error(sbp, error);
2470 #endif
2471 	return (error);
2472 }
2473 
2474 #if defined(COMPAT_FREEBSD11)
2475 /*
2476  * Implementation of the NetBSD [l]stat() functions.
2477  */
2478 int
2479 freebsd11_cvtnstat(struct stat *sb, struct nstat *nsb)
2480 {
2481 	struct freebsd11_stat sb11;
2482 	int error;
2483 
2484 	error = freebsd11_cvtstat(sb, &sb11);
2485 	if (error != 0)
2486 		return (error);
2487 
2488 	bzero(nsb, sizeof(*nsb));
2489 	CP(sb11, *nsb, st_dev);
2490 	CP(sb11, *nsb, st_ino);
2491 	CP(sb11, *nsb, st_mode);
2492 	CP(sb11, *nsb, st_nlink);
2493 	CP(sb11, *nsb, st_uid);
2494 	CP(sb11, *nsb, st_gid);
2495 	CP(sb11, *nsb, st_rdev);
2496 	CP(sb11, *nsb, st_atim);
2497 	CP(sb11, *nsb, st_mtim);
2498 	CP(sb11, *nsb, st_ctim);
2499 	CP(sb11, *nsb, st_size);
2500 	CP(sb11, *nsb, st_blocks);
2501 	CP(sb11, *nsb, st_blksize);
2502 	CP(sb11, *nsb, st_flags);
2503 	CP(sb11, *nsb, st_gen);
2504 	CP(sb11, *nsb, st_birthtim);
2505 	return (0);
2506 }
2507 
2508 #ifndef _SYS_SYSPROTO_H_
2509 struct freebsd11_nstat_args {
2510 	char	*path;
2511 	struct nstat *ub;
2512 };
2513 #endif
2514 int
2515 freebsd11_nstat(struct thread *td, struct freebsd11_nstat_args *uap)
2516 {
2517 	struct stat sb;
2518 	struct nstat nsb;
2519 	int error;
2520 
2521 	error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE,
2522 	    &sb, NULL);
2523 	if (error != 0)
2524 		return (error);
2525 	error = freebsd11_cvtnstat(&sb, &nsb);
2526 	if (error == 0)
2527 		error = copyout(&nsb, uap->ub, sizeof (nsb));
2528 	return (error);
2529 }
2530 
2531 /*
2532  * NetBSD lstat.  Get file status; this version does not follow links.
2533  */
2534 #ifndef _SYS_SYSPROTO_H_
2535 struct freebsd11_nlstat_args {
2536 	char	*path;
2537 	struct nstat *ub;
2538 };
2539 #endif
2540 int
2541 freebsd11_nlstat(struct thread *td, struct freebsd11_nlstat_args *uap)
2542 {
2543 	struct stat sb;
2544 	struct nstat nsb;
2545 	int error;
2546 
2547 	error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path,
2548 	    UIO_USERSPACE, &sb, NULL);
2549 	if (error != 0)
2550 		return (error);
2551 	error = freebsd11_cvtnstat(&sb, &nsb);
2552 	if (error == 0)
2553 		error = copyout(&nsb, uap->ub, sizeof (nsb));
2554 	return (error);
2555 }
2556 #endif /* COMPAT_FREEBSD11 */
2557 
2558 /*
2559  * Get configurable pathname variables.
2560  */
2561 #ifndef _SYS_SYSPROTO_H_
2562 struct pathconf_args {
2563 	char	*path;
2564 	int	name;
2565 };
2566 #endif
2567 int
2568 sys_pathconf(struct thread *td, struct pathconf_args *uap)
2569 {
2570 	long value;
2571 	int error;
2572 
2573 	error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, FOLLOW,
2574 	    &value);
2575 	if (error == 0)
2576 		td->td_retval[0] = value;
2577 	return (error);
2578 }
2579 
2580 #ifndef _SYS_SYSPROTO_H_
2581 struct lpathconf_args {
2582 	char	*path;
2583 	int	name;
2584 };
2585 #endif
2586 int
2587 sys_lpathconf(struct thread *td, struct lpathconf_args *uap)
2588 {
2589 	long value;
2590 	int error;
2591 
2592 	error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name,
2593 	    NOFOLLOW, &value);
2594 	if (error == 0)
2595 		td->td_retval[0] = value;
2596 	return (error);
2597 }
2598 
2599 int
2600 kern_pathconf(struct thread *td, const char *path, enum uio_seg pathseg,
2601     int name, u_long flags, long *valuep)
2602 {
2603 	struct nameidata nd;
2604 	int error;
2605 
2606 	NDINIT(&nd, LOOKUP, LOCKSHARED | LOCKLEAF | AUDITVNODE1 | flags,
2607 	       pathseg, path);
2608 	if ((error = namei(&nd)) != 0)
2609 		return (error);
2610 	NDFREE_NOTHING(&nd);
2611 
2612 	error = VOP_PATHCONF(nd.ni_vp, name, valuep);
2613 	vput(nd.ni_vp);
2614 	return (error);
2615 }
2616 
2617 /*
2618  * Return target name of a symbolic link.
2619  */
2620 #ifndef _SYS_SYSPROTO_H_
2621 struct readlink_args {
2622 	char	*path;
2623 	char	*buf;
2624 	size_t	count;
2625 };
2626 #endif
2627 int
2628 sys_readlink(struct thread *td, struct readlink_args *uap)
2629 {
2630 
2631 	return (kern_readlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2632 	    uap->buf, UIO_USERSPACE, uap->count));
2633 }
2634 #ifndef _SYS_SYSPROTO_H_
2635 struct readlinkat_args {
2636 	int	fd;
2637 	char	*path;
2638 	char	*buf;
2639 	size_t	bufsize;
2640 };
2641 #endif
2642 int
2643 sys_readlinkat(struct thread *td, struct readlinkat_args *uap)
2644 {
2645 
2646 	return (kern_readlinkat(td, uap->fd, uap->path, UIO_USERSPACE,
2647 	    uap->buf, UIO_USERSPACE, uap->bufsize));
2648 }
2649 
2650 int
2651 kern_readlinkat(struct thread *td, int fd, const char *path,
2652     enum uio_seg pathseg, char *buf, enum uio_seg bufseg, size_t count)
2653 {
2654 	struct vnode *vp;
2655 	struct nameidata nd;
2656 	int error;
2657 
2658 	if (count > IOSIZE_MAX)
2659 		return (EINVAL);
2660 
2661 	NDINIT_AT(&nd, LOOKUP, NOFOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1 |
2662 	    EMPTYPATH, pathseg, path, fd);
2663 
2664 	if ((error = namei(&nd)) != 0)
2665 		return (error);
2666 	NDFREE_NOTHING(&nd);
2667 	vp = nd.ni_vp;
2668 
2669 	error = kern_readlink_vp(vp, buf, bufseg, count, td);
2670 	vput(vp);
2671 
2672 	return (error);
2673 }
2674 
2675 /*
2676  * Helper function to readlink from a vnode
2677  */
2678 static int
2679 kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg, size_t count,
2680     struct thread *td)
2681 {
2682 	struct iovec aiov;
2683 	struct uio auio;
2684 	int error;
2685 
2686 	ASSERT_VOP_LOCKED(vp, "kern_readlink_vp(): vp not locked");
2687 #ifdef MAC
2688 	error = mac_vnode_check_readlink(td->td_ucred, vp);
2689 	if (error != 0)
2690 		return (error);
2691 #endif
2692 	if (vp->v_type != VLNK && (vp->v_vflag & VV_READLINK) == 0)
2693 		return (EINVAL);
2694 
2695 	aiov.iov_base = buf;
2696 	aiov.iov_len = count;
2697 	auio.uio_iov = &aiov;
2698 	auio.uio_iovcnt = 1;
2699 	auio.uio_offset = 0;
2700 	auio.uio_rw = UIO_READ;
2701 	auio.uio_segflg = bufseg;
2702 	auio.uio_td = td;
2703 	auio.uio_resid = count;
2704 	error = VOP_READLINK(vp, &auio, td->td_ucred);
2705 	td->td_retval[0] = count - auio.uio_resid;
2706 	return (error);
2707 }
2708 
2709 /*
2710  * Common implementation code for chflags() and fchflags().
2711  */
2712 static int
2713 setfflags(struct thread *td, struct vnode *vp, u_long flags)
2714 {
2715 	struct mount *mp;
2716 	struct vattr vattr;
2717 	int error;
2718 
2719 	/* We can't support the value matching VNOVAL. */
2720 	if (flags == VNOVAL)
2721 		return (EOPNOTSUPP);
2722 
2723 	/*
2724 	 * Prevent non-root users from setting flags on devices.  When
2725 	 * a device is reused, users can retain ownership of the device
2726 	 * if they are allowed to set flags and programs assume that
2727 	 * chown can't fail when done as root.
2728 	 */
2729 	if (vp->v_type == VCHR || vp->v_type == VBLK) {
2730 		error = priv_check(td, PRIV_VFS_CHFLAGS_DEV);
2731 		if (error != 0)
2732 			return (error);
2733 	}
2734 
2735 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
2736 		return (error);
2737 	VATTR_NULL(&vattr);
2738 	vattr.va_flags = flags;
2739 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2740 #ifdef MAC
2741 	error = mac_vnode_check_setflags(td->td_ucred, vp, vattr.va_flags);
2742 	if (error == 0)
2743 #endif
2744 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2745 	VOP_UNLOCK(vp);
2746 	vn_finished_write(mp);
2747 	return (error);
2748 }
2749 
2750 /*
2751  * Change flags of a file given a path name.
2752  */
2753 #ifndef _SYS_SYSPROTO_H_
2754 struct chflags_args {
2755 	const char *path;
2756 	u_long	flags;
2757 };
2758 #endif
2759 int
2760 sys_chflags(struct thread *td, struct chflags_args *uap)
2761 {
2762 
2763 	return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2764 	    uap->flags, 0));
2765 }
2766 
2767 #ifndef _SYS_SYSPROTO_H_
2768 struct chflagsat_args {
2769 	int	fd;
2770 	const char *path;
2771 	u_long	flags;
2772 	int	atflag;
2773 }
2774 #endif
2775 int
2776 sys_chflagsat(struct thread *td, struct chflagsat_args *uap)
2777 {
2778 
2779 	return (kern_chflagsat(td, uap->fd, uap->path, UIO_USERSPACE,
2780 	    uap->flags, uap->atflag));
2781 }
2782 
2783 /*
2784  * Same as chflags() but doesn't follow symlinks.
2785  */
2786 #ifndef _SYS_SYSPROTO_H_
2787 struct lchflags_args {
2788 	const char *path;
2789 	u_long flags;
2790 };
2791 #endif
2792 int
2793 sys_lchflags(struct thread *td, struct lchflags_args *uap)
2794 {
2795 
2796 	return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2797 	    uap->flags, AT_SYMLINK_NOFOLLOW));
2798 }
2799 
2800 static int
2801 kern_chflagsat(struct thread *td, int fd, const char *path,
2802     enum uio_seg pathseg, u_long flags, int atflag)
2803 {
2804 	struct nameidata nd;
2805 	int error;
2806 
2807 	if ((atflag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
2808 	    AT_EMPTY_PATH)) != 0)
2809 		return (EINVAL);
2810 
2811 	AUDIT_ARG_FFLAGS(flags);
2812 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(atflag, AT_SYMLINK_NOFOLLOW |
2813 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path,
2814 	    fd, &cap_fchflags_rights);
2815 	if ((error = namei(&nd)) != 0)
2816 		return (error);
2817 	NDFREE_NOTHING(&nd);
2818 	error = setfflags(td, nd.ni_vp, flags);
2819 	vrele(nd.ni_vp);
2820 	return (error);
2821 }
2822 
2823 /*
2824  * Change flags of a file given a file descriptor.
2825  */
2826 #ifndef _SYS_SYSPROTO_H_
2827 struct fchflags_args {
2828 	int	fd;
2829 	u_long	flags;
2830 };
2831 #endif
2832 int
2833 sys_fchflags(struct thread *td, struct fchflags_args *uap)
2834 {
2835 	struct file *fp;
2836 	int error;
2837 
2838 	AUDIT_ARG_FD(uap->fd);
2839 	AUDIT_ARG_FFLAGS(uap->flags);
2840 	error = getvnode(td, uap->fd, &cap_fchflags_rights,
2841 	    &fp);
2842 	if (error != 0)
2843 		return (error);
2844 #ifdef AUDIT
2845 	if (AUDITING_TD(td)) {
2846 		vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY);
2847 		AUDIT_ARG_VNODE1(fp->f_vnode);
2848 		VOP_UNLOCK(fp->f_vnode);
2849 	}
2850 #endif
2851 	error = setfflags(td, fp->f_vnode, uap->flags);
2852 	fdrop(fp, td);
2853 	return (error);
2854 }
2855 
2856 /*
2857  * Common implementation code for chmod(), lchmod() and fchmod().
2858  */
2859 int
2860 setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode)
2861 {
2862 	struct mount *mp;
2863 	struct vattr vattr;
2864 	int error;
2865 
2866 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
2867 		return (error);
2868 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2869 	VATTR_NULL(&vattr);
2870 	vattr.va_mode = mode & ALLPERMS;
2871 #ifdef MAC
2872 	error = mac_vnode_check_setmode(cred, vp, vattr.va_mode);
2873 	if (error == 0)
2874 #endif
2875 		error = VOP_SETATTR(vp, &vattr, cred);
2876 	VOP_UNLOCK(vp);
2877 	vn_finished_write(mp);
2878 	return (error);
2879 }
2880 
2881 /*
2882  * Change mode of a file given path name.
2883  */
2884 #ifndef _SYS_SYSPROTO_H_
2885 struct chmod_args {
2886 	char	*path;
2887 	int	mode;
2888 };
2889 #endif
2890 int
2891 sys_chmod(struct thread *td, struct chmod_args *uap)
2892 {
2893 
2894 	return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2895 	    uap->mode, 0));
2896 }
2897 
2898 #ifndef _SYS_SYSPROTO_H_
2899 struct fchmodat_args {
2900 	int	dirfd;
2901 	char	*path;
2902 	mode_t	mode;
2903 	int	flag;
2904 }
2905 #endif
2906 int
2907 sys_fchmodat(struct thread *td, struct fchmodat_args *uap)
2908 {
2909 
2910 	return (kern_fchmodat(td, uap->fd, uap->path, UIO_USERSPACE,
2911 	    uap->mode, uap->flag));
2912 }
2913 
2914 /*
2915  * Change mode of a file given path name (don't follow links.)
2916  */
2917 #ifndef _SYS_SYSPROTO_H_
2918 struct lchmod_args {
2919 	char	*path;
2920 	int	mode;
2921 };
2922 #endif
2923 int
2924 sys_lchmod(struct thread *td, struct lchmod_args *uap)
2925 {
2926 
2927 	return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
2928 	    uap->mode, AT_SYMLINK_NOFOLLOW));
2929 }
2930 
2931 int
2932 kern_fchmodat(struct thread *td, int fd, const char *path,
2933     enum uio_seg pathseg, mode_t mode, int flag)
2934 {
2935 	struct nameidata nd;
2936 	int error;
2937 
2938 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
2939 	    AT_EMPTY_PATH)) != 0)
2940 		return (EINVAL);
2941 
2942 	AUDIT_ARG_MODE(mode);
2943 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW |
2944 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path,
2945 	    fd, &cap_fchmod_rights);
2946 	if ((error = namei(&nd)) != 0)
2947 		return (error);
2948 	NDFREE_NOTHING(&nd);
2949 	error = setfmode(td, td->td_ucred, nd.ni_vp, mode);
2950 	vrele(nd.ni_vp);
2951 	return (error);
2952 }
2953 
2954 /*
2955  * Change mode of a file given a file descriptor.
2956  */
2957 #ifndef _SYS_SYSPROTO_H_
2958 struct fchmod_args {
2959 	int	fd;
2960 	int	mode;
2961 };
2962 #endif
2963 int
2964 sys_fchmod(struct thread *td, struct fchmod_args *uap)
2965 {
2966 	struct file *fp;
2967 	int error;
2968 
2969 	AUDIT_ARG_FD(uap->fd);
2970 	AUDIT_ARG_MODE(uap->mode);
2971 
2972 	error = fget(td, uap->fd, &cap_fchmod_rights, &fp);
2973 	if (error != 0)
2974 		return (error);
2975 	error = fo_chmod(fp, uap->mode, td->td_ucred, td);
2976 	fdrop(fp, td);
2977 	return (error);
2978 }
2979 
2980 /*
2981  * Common implementation for chown(), lchown(), and fchown()
2982  */
2983 int
2984 setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid,
2985     gid_t gid)
2986 {
2987 	struct mount *mp;
2988 	struct vattr vattr;
2989 	int error;
2990 
2991 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
2992 		return (error);
2993 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2994 	VATTR_NULL(&vattr);
2995 	vattr.va_uid = uid;
2996 	vattr.va_gid = gid;
2997 #ifdef MAC
2998 	error = mac_vnode_check_setowner(cred, vp, vattr.va_uid,
2999 	    vattr.va_gid);
3000 	if (error == 0)
3001 #endif
3002 		error = VOP_SETATTR(vp, &vattr, cred);
3003 	VOP_UNLOCK(vp);
3004 	vn_finished_write(mp);
3005 	return (error);
3006 }
3007 
3008 /*
3009  * Set ownership given a path name.
3010  */
3011 #ifndef _SYS_SYSPROTO_H_
3012 struct chown_args {
3013 	char	*path;
3014 	int	uid;
3015 	int	gid;
3016 };
3017 #endif
3018 int
3019 sys_chown(struct thread *td, struct chown_args *uap)
3020 {
3021 
3022 	return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid,
3023 	    uap->gid, 0));
3024 }
3025 
3026 #ifndef _SYS_SYSPROTO_H_
3027 struct fchownat_args {
3028 	int fd;
3029 	const char * path;
3030 	uid_t uid;
3031 	gid_t gid;
3032 	int flag;
3033 };
3034 #endif
3035 int
3036 sys_fchownat(struct thread *td, struct fchownat_args *uap)
3037 {
3038 
3039 	return (kern_fchownat(td, uap->fd, uap->path, UIO_USERSPACE, uap->uid,
3040 	    uap->gid, uap->flag));
3041 }
3042 
3043 int
3044 kern_fchownat(struct thread *td, int fd, const char *path,
3045     enum uio_seg pathseg, int uid, int gid, int flag)
3046 {
3047 	struct nameidata nd;
3048 	int error;
3049 
3050 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
3051 	    AT_EMPTY_PATH)) != 0)
3052 		return (EINVAL);
3053 
3054 	AUDIT_ARG_OWNER(uid, gid);
3055 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW |
3056 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1, pathseg, path,
3057 	    fd, &cap_fchown_rights);
3058 
3059 	if ((error = namei(&nd)) != 0)
3060 		return (error);
3061 	NDFREE_NOTHING(&nd);
3062 	error = setfown(td, td->td_ucred, nd.ni_vp, uid, gid);
3063 	vrele(nd.ni_vp);
3064 	return (error);
3065 }
3066 
3067 /*
3068  * Set ownership given a path name, do not cross symlinks.
3069  */
3070 #ifndef _SYS_SYSPROTO_H_
3071 struct lchown_args {
3072 	char	*path;
3073 	int	uid;
3074 	int	gid;
3075 };
3076 #endif
3077 int
3078 sys_lchown(struct thread *td, struct lchown_args *uap)
3079 {
3080 
3081 	return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
3082 	    uap->uid, uap->gid, AT_SYMLINK_NOFOLLOW));
3083 }
3084 
3085 /*
3086  * Set ownership given a file descriptor.
3087  */
3088 #ifndef _SYS_SYSPROTO_H_
3089 struct fchown_args {
3090 	int	fd;
3091 	int	uid;
3092 	int	gid;
3093 };
3094 #endif
3095 int
3096 sys_fchown(struct thread *td, struct fchown_args *uap)
3097 {
3098 	struct file *fp;
3099 	int error;
3100 
3101 	AUDIT_ARG_FD(uap->fd);
3102 	AUDIT_ARG_OWNER(uap->uid, uap->gid);
3103 	error = fget(td, uap->fd, &cap_fchown_rights, &fp);
3104 	if (error != 0)
3105 		return (error);
3106 	error = fo_chown(fp, uap->uid, uap->gid, td->td_ucred, td);
3107 	fdrop(fp, td);
3108 	return (error);
3109 }
3110 
3111 /*
3112  * Common implementation code for utimes(), lutimes(), and futimes().
3113  */
3114 static int
3115 getutimes(const struct timeval *usrtvp, enum uio_seg tvpseg,
3116     struct timespec *tsp)
3117 {
3118 	struct timeval tv[2];
3119 	const struct timeval *tvp;
3120 	int error;
3121 
3122 	if (usrtvp == NULL) {
3123 		vfs_timestamp(&tsp[0]);
3124 		tsp[1] = tsp[0];
3125 	} else {
3126 		if (tvpseg == UIO_SYSSPACE) {
3127 			tvp = usrtvp;
3128 		} else {
3129 			if ((error = copyin(usrtvp, tv, sizeof(tv))) != 0)
3130 				return (error);
3131 			tvp = tv;
3132 		}
3133 
3134 		if (tvp[0].tv_usec < 0 || tvp[0].tv_usec >= 1000000 ||
3135 		    tvp[1].tv_usec < 0 || tvp[1].tv_usec >= 1000000)
3136 			return (EINVAL);
3137 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3138 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3139 	}
3140 	return (0);
3141 }
3142 
3143 /*
3144  * Common implementation code for futimens(), utimensat().
3145  */
3146 #define	UTIMENS_NULL	0x1
3147 #define	UTIMENS_EXIT	0x2
3148 static int
3149 getutimens(const struct timespec *usrtsp, enum uio_seg tspseg,
3150     struct timespec *tsp, int *retflags)
3151 {
3152 	struct timespec tsnow;
3153 	int error;
3154 
3155 	vfs_timestamp(&tsnow);
3156 	*retflags = 0;
3157 	if (usrtsp == NULL) {
3158 		tsp[0] = tsnow;
3159 		tsp[1] = tsnow;
3160 		*retflags |= UTIMENS_NULL;
3161 		return (0);
3162 	}
3163 	if (tspseg == UIO_SYSSPACE) {
3164 		tsp[0] = usrtsp[0];
3165 		tsp[1] = usrtsp[1];
3166 	} else if ((error = copyin(usrtsp, tsp, sizeof(*tsp) * 2)) != 0)
3167 		return (error);
3168 	if (tsp[0].tv_nsec == UTIME_OMIT && tsp[1].tv_nsec == UTIME_OMIT)
3169 		*retflags |= UTIMENS_EXIT;
3170 	if (tsp[0].tv_nsec == UTIME_NOW && tsp[1].tv_nsec == UTIME_NOW)
3171 		*retflags |= UTIMENS_NULL;
3172 	if (tsp[0].tv_nsec == UTIME_OMIT)
3173 		tsp[0].tv_sec = VNOVAL;
3174 	else if (tsp[0].tv_nsec == UTIME_NOW)
3175 		tsp[0] = tsnow;
3176 	else if (tsp[0].tv_nsec < 0 || tsp[0].tv_nsec >= 1000000000L)
3177 		return (EINVAL);
3178 	if (tsp[1].tv_nsec == UTIME_OMIT)
3179 		tsp[1].tv_sec = VNOVAL;
3180 	else if (tsp[1].tv_nsec == UTIME_NOW)
3181 		tsp[1] = tsnow;
3182 	else if (tsp[1].tv_nsec < 0 || tsp[1].tv_nsec >= 1000000000L)
3183 		return (EINVAL);
3184 
3185 	return (0);
3186 }
3187 
3188 /*
3189  * Common implementation code for utimes(), lutimes(), futimes(), futimens(),
3190  * and utimensat().
3191  */
3192 static int
3193 setutimes(struct thread *td, struct vnode *vp, const struct timespec *ts,
3194     int numtimes, int nullflag)
3195 {
3196 	struct mount *mp;
3197 	struct vattr vattr;
3198 	int error;
3199 	bool setbirthtime;
3200 
3201 	setbirthtime = false;
3202 	vattr.va_birthtime.tv_sec = VNOVAL;
3203 	vattr.va_birthtime.tv_nsec = 0;
3204 
3205 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
3206 		return (error);
3207 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3208 	if (numtimes < 3 && VOP_GETATTR(vp, &vattr, td->td_ucred) == 0 &&
3209 	    timespeccmp(&ts[1], &vattr.va_birthtime, < ))
3210 		setbirthtime = true;
3211 	VATTR_NULL(&vattr);
3212 	vattr.va_atime = ts[0];
3213 	vattr.va_mtime = ts[1];
3214 	if (setbirthtime)
3215 		vattr.va_birthtime = ts[1];
3216 	if (numtimes > 2)
3217 		vattr.va_birthtime = ts[2];
3218 	if (nullflag)
3219 		vattr.va_vaflags |= VA_UTIMES_NULL;
3220 #ifdef MAC
3221 	error = mac_vnode_check_setutimes(td->td_ucred, vp, vattr.va_atime,
3222 	    vattr.va_mtime);
3223 #endif
3224 	if (error == 0)
3225 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3226 	VOP_UNLOCK(vp);
3227 	vn_finished_write(mp);
3228 	return (error);
3229 }
3230 
3231 /*
3232  * Set the access and modification times of a file.
3233  */
3234 #ifndef _SYS_SYSPROTO_H_
3235 struct utimes_args {
3236 	char	*path;
3237 	struct	timeval *tptr;
3238 };
3239 #endif
3240 int
3241 sys_utimes(struct thread *td, struct utimes_args *uap)
3242 {
3243 
3244 	return (kern_utimesat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
3245 	    uap->tptr, UIO_USERSPACE));
3246 }
3247 
3248 #ifndef _SYS_SYSPROTO_H_
3249 struct futimesat_args {
3250 	int fd;
3251 	const char * path;
3252 	const struct timeval * times;
3253 };
3254 #endif
3255 int
3256 sys_futimesat(struct thread *td, struct futimesat_args *uap)
3257 {
3258 
3259 	return (kern_utimesat(td, uap->fd, uap->path, UIO_USERSPACE,
3260 	    uap->times, UIO_USERSPACE));
3261 }
3262 
3263 int
3264 kern_utimesat(struct thread *td, int fd, const char *path,
3265     enum uio_seg pathseg, const struct timeval *tptr, enum uio_seg tptrseg)
3266 {
3267 	struct nameidata nd;
3268 	struct timespec ts[2];
3269 	int error;
3270 
3271 	if ((error = getutimes(tptr, tptrseg, ts)) != 0)
3272 		return (error);
3273 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd,
3274 	    &cap_futimes_rights);
3275 
3276 	if ((error = namei(&nd)) != 0)
3277 		return (error);
3278 	NDFREE_NOTHING(&nd);
3279 	error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL);
3280 	vrele(nd.ni_vp);
3281 	return (error);
3282 }
3283 
3284 /*
3285  * Set the access and modification times of a file.
3286  */
3287 #ifndef _SYS_SYSPROTO_H_
3288 struct lutimes_args {
3289 	char	*path;
3290 	struct	timeval *tptr;
3291 };
3292 #endif
3293 int
3294 sys_lutimes(struct thread *td, struct lutimes_args *uap)
3295 {
3296 
3297 	return (kern_lutimes(td, uap->path, UIO_USERSPACE, uap->tptr,
3298 	    UIO_USERSPACE));
3299 }
3300 
3301 int
3302 kern_lutimes(struct thread *td, const char *path, enum uio_seg pathseg,
3303     const struct timeval *tptr, enum uio_seg tptrseg)
3304 {
3305 	struct timespec ts[2];
3306 	struct nameidata nd;
3307 	int error;
3308 
3309 	if ((error = getutimes(tptr, tptrseg, ts)) != 0)
3310 		return (error);
3311 	NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, pathseg, path);
3312 	if ((error = namei(&nd)) != 0)
3313 		return (error);
3314 	NDFREE_NOTHING(&nd);
3315 	error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL);
3316 	vrele(nd.ni_vp);
3317 	return (error);
3318 }
3319 
3320 /*
3321  * Set the access and modification times of a file.
3322  */
3323 #ifndef _SYS_SYSPROTO_H_
3324 struct futimes_args {
3325 	int	fd;
3326 	struct	timeval *tptr;
3327 };
3328 #endif
3329 int
3330 sys_futimes(struct thread *td, struct futimes_args *uap)
3331 {
3332 
3333 	return (kern_futimes(td, uap->fd, uap->tptr, UIO_USERSPACE));
3334 }
3335 
3336 int
3337 kern_futimes(struct thread *td, int fd, const struct timeval *tptr,
3338     enum uio_seg tptrseg)
3339 {
3340 	struct timespec ts[2];
3341 	struct file *fp;
3342 	int error;
3343 
3344 	AUDIT_ARG_FD(fd);
3345 	error = getutimes(tptr, tptrseg, ts);
3346 	if (error != 0)
3347 		return (error);
3348 	error = getvnode(td, fd, &cap_futimes_rights, &fp);
3349 	if (error != 0)
3350 		return (error);
3351 #ifdef AUDIT
3352 	if (AUDITING_TD(td)) {
3353 		vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY);
3354 		AUDIT_ARG_VNODE1(fp->f_vnode);
3355 		VOP_UNLOCK(fp->f_vnode);
3356 	}
3357 #endif
3358 	error = setutimes(td, fp->f_vnode, ts, 2, tptr == NULL);
3359 	fdrop(fp, td);
3360 	return (error);
3361 }
3362 
3363 int
3364 sys_futimens(struct thread *td, struct futimens_args *uap)
3365 {
3366 
3367 	return (kern_futimens(td, uap->fd, uap->times, UIO_USERSPACE));
3368 }
3369 
3370 int
3371 kern_futimens(struct thread *td, int fd, const struct timespec *tptr,
3372     enum uio_seg tptrseg)
3373 {
3374 	struct timespec ts[2];
3375 	struct file *fp;
3376 	int error, flags;
3377 
3378 	AUDIT_ARG_FD(fd);
3379 	error = getutimens(tptr, tptrseg, ts, &flags);
3380 	if (error != 0)
3381 		return (error);
3382 	if (flags & UTIMENS_EXIT)
3383 		return (0);
3384 	error = getvnode(td, fd, &cap_futimes_rights, &fp);
3385 	if (error != 0)
3386 		return (error);
3387 #ifdef AUDIT
3388 	if (AUDITING_TD(td)) {
3389 		vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY);
3390 		AUDIT_ARG_VNODE1(fp->f_vnode);
3391 		VOP_UNLOCK(fp->f_vnode);
3392 	}
3393 #endif
3394 	error = setutimes(td, fp->f_vnode, ts, 2, flags & UTIMENS_NULL);
3395 	fdrop(fp, td);
3396 	return (error);
3397 }
3398 
3399 int
3400 sys_utimensat(struct thread *td, struct utimensat_args *uap)
3401 {
3402 
3403 	return (kern_utimensat(td, uap->fd, uap->path, UIO_USERSPACE,
3404 	    uap->times, UIO_USERSPACE, uap->flag));
3405 }
3406 
3407 int
3408 kern_utimensat(struct thread *td, int fd, const char *path,
3409     enum uio_seg pathseg, const struct timespec *tptr, enum uio_seg tptrseg,
3410     int flag)
3411 {
3412 	struct nameidata nd;
3413 	struct timespec ts[2];
3414 	int error, flags;
3415 
3416 	if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH |
3417 	    AT_EMPTY_PATH)) != 0)
3418 		return (EINVAL);
3419 
3420 	if ((error = getutimens(tptr, tptrseg, ts, &flags)) != 0)
3421 		return (error);
3422 	NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW |
3423 	    AT_RESOLVE_BENEATH | AT_EMPTY_PATH) | AUDITVNODE1,
3424 	    pathseg, path, fd, &cap_futimes_rights);
3425 	if ((error = namei(&nd)) != 0)
3426 		return (error);
3427 	/*
3428 	 * We are allowed to call namei() regardless of 2xUTIME_OMIT.
3429 	 * POSIX states:
3430 	 * "If both tv_nsec fields are UTIME_OMIT... EACCESS may be detected."
3431 	 * "Search permission is denied by a component of the path prefix."
3432 	 */
3433 	NDFREE_NOTHING(&nd);
3434 	if ((flags & UTIMENS_EXIT) == 0)
3435 		error = setutimes(td, nd.ni_vp, ts, 2, flags & UTIMENS_NULL);
3436 	vrele(nd.ni_vp);
3437 	return (error);
3438 }
3439 
3440 /*
3441  * Truncate a file given its path name.
3442  */
3443 #ifndef _SYS_SYSPROTO_H_
3444 struct truncate_args {
3445 	char	*path;
3446 	int	pad;
3447 	off_t	length;
3448 };
3449 #endif
3450 int
3451 sys_truncate(struct thread *td, struct truncate_args *uap)
3452 {
3453 
3454 	return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length));
3455 }
3456 
3457 int
3458 kern_truncate(struct thread *td, const char *path, enum uio_seg pathseg,
3459     off_t length)
3460 {
3461 	struct mount *mp;
3462 	struct vnode *vp;
3463 	void *rl_cookie;
3464 	struct vattr vattr;
3465 	struct nameidata nd;
3466 	int error;
3467 
3468 	if (length < 0)
3469 		return (EINVAL);
3470 	NDPREINIT(&nd);
3471 retry:
3472 	NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path);
3473 	if ((error = namei(&nd)) != 0)
3474 		return (error);
3475 	vp = nd.ni_vp;
3476 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3477 	if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) {
3478 		vn_rangelock_unlock(vp, rl_cookie);
3479 		vrele(vp);
3480 		return (error);
3481 	}
3482 	NDFREE(&nd, NDF_ONLY_PNBUF);
3483 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3484 	if (vp->v_type == VDIR)
3485 		error = EISDIR;
3486 #ifdef MAC
3487 	else if ((error = mac_vnode_check_write(td->td_ucred, NOCRED, vp))) {
3488 	}
3489 #endif
3490 	else if ((error = vn_writechk(vp)) == 0 &&
3491 	    (error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td)) == 0) {
3492 		VATTR_NULL(&vattr);
3493 		vattr.va_size = length;
3494 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3495 	}
3496 	VOP_UNLOCK(vp);
3497 	vn_finished_write(mp);
3498 	vn_rangelock_unlock(vp, rl_cookie);
3499 	vrele(vp);
3500 	if (error == ERELOOKUP)
3501 		goto retry;
3502 	return (error);
3503 }
3504 
3505 #if defined(COMPAT_43)
3506 /*
3507  * Truncate a file given its path name.
3508  */
3509 #ifndef _SYS_SYSPROTO_H_
3510 struct otruncate_args {
3511 	char	*path;
3512 	long	length;
3513 };
3514 #endif
3515 int
3516 otruncate(struct thread *td, struct otruncate_args *uap)
3517 {
3518 
3519 	return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length));
3520 }
3521 #endif /* COMPAT_43 */
3522 
3523 #if defined(COMPAT_FREEBSD6)
3524 /* Versions with the pad argument */
3525 int
3526 freebsd6_truncate(struct thread *td, struct freebsd6_truncate_args *uap)
3527 {
3528 
3529 	return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length));
3530 }
3531 
3532 int
3533 freebsd6_ftruncate(struct thread *td, struct freebsd6_ftruncate_args *uap)
3534 {
3535 
3536 	return (kern_ftruncate(td, uap->fd, uap->length));
3537 }
3538 #endif
3539 
3540 int
3541 kern_fsync(struct thread *td, int fd, bool fullsync)
3542 {
3543 	struct vnode *vp;
3544 	struct mount *mp;
3545 	struct file *fp;
3546 	int error;
3547 
3548 	AUDIT_ARG_FD(fd);
3549 	error = getvnode(td, fd, &cap_fsync_rights, &fp);
3550 	if (error != 0)
3551 		return (error);
3552 	vp = fp->f_vnode;
3553 #if 0
3554 	if (!fullsync)
3555 		/* XXXKIB: compete outstanding aio writes */;
3556 #endif
3557 retry:
3558 	error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
3559 	if (error != 0)
3560 		goto drop;
3561 	vn_lock(vp, vn_lktype_write(mp, vp) | LK_RETRY);
3562 	AUDIT_ARG_VNODE1(vp);
3563 	if (vp->v_object != NULL) {
3564 		VM_OBJECT_WLOCK(vp->v_object);
3565 		vm_object_page_clean(vp->v_object, 0, 0, 0);
3566 		VM_OBJECT_WUNLOCK(vp->v_object);
3567 	}
3568 	error = fullsync ? VOP_FSYNC(vp, MNT_WAIT, td) : VOP_FDATASYNC(vp, td);
3569 	VOP_UNLOCK(vp);
3570 	vn_finished_write(mp);
3571 	if (error == ERELOOKUP)
3572 		goto retry;
3573 drop:
3574 	fdrop(fp, td);
3575 	return (error);
3576 }
3577 
3578 /*
3579  * Sync an open file.
3580  */
3581 #ifndef _SYS_SYSPROTO_H_
3582 struct fsync_args {
3583 	int	fd;
3584 };
3585 #endif
3586 int
3587 sys_fsync(struct thread *td, struct fsync_args *uap)
3588 {
3589 
3590 	return (kern_fsync(td, uap->fd, true));
3591 }
3592 
3593 int
3594 sys_fdatasync(struct thread *td, struct fdatasync_args *uap)
3595 {
3596 
3597 	return (kern_fsync(td, uap->fd, false));
3598 }
3599 
3600 /*
3601  * Rename files.  Source and destination must either both be directories, or
3602  * both not be directories.  If target is a directory, it must be empty.
3603  */
3604 #ifndef _SYS_SYSPROTO_H_
3605 struct rename_args {
3606 	char	*from;
3607 	char	*to;
3608 };
3609 #endif
3610 int
3611 sys_rename(struct thread *td, struct rename_args *uap)
3612 {
3613 
3614 	return (kern_renameat(td, AT_FDCWD, uap->from, AT_FDCWD,
3615 	    uap->to, UIO_USERSPACE));
3616 }
3617 
3618 #ifndef _SYS_SYSPROTO_H_
3619 struct renameat_args {
3620 	int	oldfd;
3621 	char	*old;
3622 	int	newfd;
3623 	char	*new;
3624 };
3625 #endif
3626 int
3627 sys_renameat(struct thread *td, struct renameat_args *uap)
3628 {
3629 
3630 	return (kern_renameat(td, uap->oldfd, uap->old, uap->newfd, uap->new,
3631 	    UIO_USERSPACE));
3632 }
3633 
3634 #ifdef MAC
3635 static int
3636 kern_renameat_mac(struct thread *td, int oldfd, const char *old, int newfd,
3637     const char *new, enum uio_seg pathseg, struct nameidata *fromnd)
3638 {
3639 	int error;
3640 
3641 	NDINIT_ATRIGHTS(fromnd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART |
3642 	    AUDITVNODE1, pathseg, old, oldfd, &cap_renameat_source_rights);
3643 	if ((error = namei(fromnd)) != 0)
3644 		return (error);
3645 	error = mac_vnode_check_rename_from(td->td_ucred, fromnd->ni_dvp,
3646 	    fromnd->ni_vp, &fromnd->ni_cnd);
3647 	VOP_UNLOCK(fromnd->ni_dvp);
3648 	if (fromnd->ni_dvp != fromnd->ni_vp)
3649 		VOP_UNLOCK(fromnd->ni_vp);
3650 	if (error != 0) {
3651 		NDFREE(fromnd, NDF_ONLY_PNBUF);
3652 		vrele(fromnd->ni_dvp);
3653 		vrele(fromnd->ni_vp);
3654 		if (fromnd->ni_startdir)
3655 			vrele(fromnd->ni_startdir);
3656 	}
3657 	return (error);
3658 }
3659 #endif
3660 
3661 int
3662 kern_renameat(struct thread *td, int oldfd, const char *old, int newfd,
3663     const char *new, enum uio_seg pathseg)
3664 {
3665 	struct mount *mp = NULL;
3666 	struct vnode *tvp, *fvp, *tdvp;
3667 	struct nameidata fromnd, tond;
3668 	uint64_t tondflags;
3669 	int error;
3670 
3671 again:
3672 	bwillwrite();
3673 #ifdef MAC
3674 	if (mac_vnode_check_rename_from_enabled()) {
3675 		error = kern_renameat_mac(td, oldfd, old, newfd, new, pathseg,
3676 		    &fromnd);
3677 		if (error != 0)
3678 			return (error);
3679 	} else {
3680 #endif
3681 	NDINIT_ATRIGHTS(&fromnd, DELETE, WANTPARENT | SAVESTART | AUDITVNODE1,
3682 	    pathseg, old, oldfd, &cap_renameat_source_rights);
3683 	if ((error = namei(&fromnd)) != 0)
3684 		return (error);
3685 #ifdef MAC
3686 	}
3687 #endif
3688 	fvp = fromnd.ni_vp;
3689 	tondflags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNODE2;
3690 	if (fromnd.ni_vp->v_type == VDIR)
3691 		tondflags |= WILLBEDIR;
3692 	NDINIT_ATRIGHTS(&tond, RENAME, tondflags, pathseg, new, newfd,
3693 	    &cap_renameat_target_rights);
3694 	if ((error = namei(&tond)) != 0) {
3695 		/* Translate error code for rename("dir1", "dir2/."). */
3696 		if (error == EISDIR && fvp->v_type == VDIR)
3697 			error = EINVAL;
3698 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3699 		vrele(fromnd.ni_dvp);
3700 		vrele(fvp);
3701 		goto out1;
3702 	}
3703 	tdvp = tond.ni_dvp;
3704 	tvp = tond.ni_vp;
3705 	error = vn_start_write(fvp, &mp, V_NOWAIT);
3706 	if (error != 0) {
3707 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3708 		NDFREE(&tond, NDF_ONLY_PNBUF);
3709 		if (tvp != NULL)
3710 			vput(tvp);
3711 		if (tdvp == tvp)
3712 			vrele(tdvp);
3713 		else
3714 			vput(tdvp);
3715 		vrele(fromnd.ni_dvp);
3716 		vrele(fvp);
3717 		vrele(tond.ni_startdir);
3718 		if (fromnd.ni_startdir != NULL)
3719 			vrele(fromnd.ni_startdir);
3720 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
3721 		if (error != 0)
3722 			return (error);
3723 		goto again;
3724 	}
3725 	if (tvp != NULL) {
3726 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
3727 			error = ENOTDIR;
3728 			goto out;
3729 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
3730 			error = EISDIR;
3731 			goto out;
3732 		}
3733 #ifdef CAPABILITIES
3734 		if (newfd != AT_FDCWD && (tond.ni_resflags & NIRES_ABS) == 0) {
3735 			/*
3736 			 * If the target already exists we require CAP_UNLINKAT
3737 			 * from 'newfd', when newfd was used for the lookup.
3738 			 */
3739 			error = cap_check(&tond.ni_filecaps.fc_rights,
3740 			    &cap_unlinkat_rights);
3741 			if (error != 0)
3742 				goto out;
3743 		}
3744 #endif
3745 	}
3746 	if (fvp == tdvp) {
3747 		error = EINVAL;
3748 		goto out;
3749 	}
3750 	/*
3751 	 * If the source is the same as the destination (that is, if they
3752 	 * are links to the same vnode), then there is nothing to do.
3753 	 */
3754 	if (fvp == tvp)
3755 		error = ERESTART;
3756 #ifdef MAC
3757 	else
3758 		error = mac_vnode_check_rename_to(td->td_ucred, tdvp,
3759 		    tond.ni_vp, fromnd.ni_dvp == tdvp, &tond.ni_cnd);
3760 #endif
3761 out:
3762 	if (error == 0) {
3763 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
3764 		    tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
3765 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3766 		NDFREE(&tond, NDF_ONLY_PNBUF);
3767 	} else {
3768 		NDFREE(&fromnd, NDF_ONLY_PNBUF);
3769 		NDFREE(&tond, NDF_ONLY_PNBUF);
3770 		if (tvp != NULL)
3771 			vput(tvp);
3772 		if (tdvp == tvp)
3773 			vrele(tdvp);
3774 		else
3775 			vput(tdvp);
3776 		vrele(fromnd.ni_dvp);
3777 		vrele(fvp);
3778 	}
3779 	vrele(tond.ni_startdir);
3780 	vn_finished_write(mp);
3781 out1:
3782 	if (fromnd.ni_startdir)
3783 		vrele(fromnd.ni_startdir);
3784 	if (error == ERESTART)
3785 		return (0);
3786 	if (error == ERELOOKUP)
3787 		goto again;
3788 	return (error);
3789 }
3790 
3791 /*
3792  * Make a directory file.
3793  */
3794 #ifndef _SYS_SYSPROTO_H_
3795 struct mkdir_args {
3796 	char	*path;
3797 	int	mode;
3798 };
3799 #endif
3800 int
3801 sys_mkdir(struct thread *td, struct mkdir_args *uap)
3802 {
3803 
3804 	return (kern_mkdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE,
3805 	    uap->mode));
3806 }
3807 
3808 #ifndef _SYS_SYSPROTO_H_
3809 struct mkdirat_args {
3810 	int	fd;
3811 	char	*path;
3812 	mode_t	mode;
3813 };
3814 #endif
3815 int
3816 sys_mkdirat(struct thread *td, struct mkdirat_args *uap)
3817 {
3818 
3819 	return (kern_mkdirat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode));
3820 }
3821 
3822 int
3823 kern_mkdirat(struct thread *td, int fd, const char *path, enum uio_seg segflg,
3824     int mode)
3825 {
3826 	struct mount *mp;
3827 	struct vattr vattr;
3828 	struct nameidata nd;
3829 	int error;
3830 
3831 	AUDIT_ARG_MODE(mode);
3832 	NDPREINIT(&nd);
3833 restart:
3834 	bwillwrite();
3835 	NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 |
3836 	    NC_NOMAKEENTRY | NC_KEEPPOSENTRY | FAILIFEXISTS | WILLBEDIR,
3837 	    segflg, path, fd, &cap_mkdirat_rights);
3838 	if ((error = namei(&nd)) != 0)
3839 		return (error);
3840 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
3841 		NDFREE(&nd, NDF_ONLY_PNBUF);
3842 		vput(nd.ni_dvp);
3843 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
3844 			return (error);
3845 		goto restart;
3846 	}
3847 	VATTR_NULL(&vattr);
3848 	vattr.va_type = VDIR;
3849 	vattr.va_mode = (mode & ACCESSPERMS) &~ td->td_proc->p_pd->pd_cmask;
3850 #ifdef MAC
3851 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
3852 	    &vattr);
3853 	if (error != 0)
3854 		goto out;
3855 #endif
3856 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
3857 #ifdef MAC
3858 out:
3859 #endif
3860 	NDFREE(&nd, NDF_ONLY_PNBUF);
3861 	VOP_VPUT_PAIR(nd.ni_dvp, error == 0 ? &nd.ni_vp : NULL, true);
3862 	vn_finished_write(mp);
3863 	if (error == ERELOOKUP)
3864 		goto restart;
3865 	return (error);
3866 }
3867 
3868 /*
3869  * Remove a directory file.
3870  */
3871 #ifndef _SYS_SYSPROTO_H_
3872 struct rmdir_args {
3873 	char	*path;
3874 };
3875 #endif
3876 int
3877 sys_rmdir(struct thread *td, struct rmdir_args *uap)
3878 {
3879 
3880 	return (kern_frmdirat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE,
3881 	    0));
3882 }
3883 
3884 int
3885 kern_frmdirat(struct thread *td, int dfd, const char *path, int fd,
3886     enum uio_seg pathseg, int flag)
3887 {
3888 	struct mount *mp;
3889 	struct vnode *vp;
3890 	struct file *fp;
3891 	struct nameidata nd;
3892 	cap_rights_t rights;
3893 	int error;
3894 
3895 	fp = NULL;
3896 	if (fd != FD_NONE) {
3897 		error = getvnode(td, fd, cap_rights_init_one(&rights,
3898 		    CAP_LOOKUP), &fp);
3899 		if (error != 0)
3900 			return (error);
3901 	}
3902 
3903 	NDPREINIT(&nd);
3904 restart:
3905 	bwillwrite();
3906 	NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 |
3907 	    at2cnpflags(flag, AT_RESOLVE_BENEATH),
3908 	    pathseg, path, dfd, &cap_unlinkat_rights);
3909 	if ((error = namei(&nd)) != 0)
3910 		goto fdout;
3911 	vp = nd.ni_vp;
3912 	if (vp->v_type != VDIR) {
3913 		error = ENOTDIR;
3914 		goto out;
3915 	}
3916 	/*
3917 	 * No rmdir "." please.
3918 	 */
3919 	if (nd.ni_dvp == vp) {
3920 		error = EINVAL;
3921 		goto out;
3922 	}
3923 	/*
3924 	 * The root of a mounted filesystem cannot be deleted.
3925 	 */
3926 	if (vp->v_vflag & VV_ROOT) {
3927 		error = EBUSY;
3928 		goto out;
3929 	}
3930 
3931 	if (fp != NULL && fp->f_vnode != vp) {
3932 		if (VN_IS_DOOMED(fp->f_vnode))
3933 			error = EBADF;
3934 		else
3935 			error = EDEADLK;
3936 		goto out;
3937 	}
3938 
3939 #ifdef MAC
3940 	error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp,
3941 	    &nd.ni_cnd);
3942 	if (error != 0)
3943 		goto out;
3944 #endif
3945 	if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
3946 		NDFREE(&nd, NDF_ONLY_PNBUF);
3947 		vput(vp);
3948 		if (nd.ni_dvp == vp)
3949 			vrele(nd.ni_dvp);
3950 		else
3951 			vput(nd.ni_dvp);
3952 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
3953 			goto fdout;
3954 		goto restart;
3955 	}
3956 	vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK);
3957 	error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
3958 	vn_finished_write(mp);
3959 out:
3960 	NDFREE(&nd, NDF_ONLY_PNBUF);
3961 	vput(vp);
3962 	if (nd.ni_dvp == vp)
3963 		vrele(nd.ni_dvp);
3964 	else
3965 		vput(nd.ni_dvp);
3966 	if (error == ERELOOKUP)
3967 		goto restart;
3968 fdout:
3969 	if (fp != NULL)
3970 		fdrop(fp, td);
3971 	return (error);
3972 }
3973 
3974 #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11)
3975 int
3976 freebsd11_kern_getdirentries(struct thread *td, int fd, char *ubuf, u_int count,
3977     long *basep, void (*func)(struct freebsd11_dirent *))
3978 {
3979 	struct freebsd11_dirent dstdp;
3980 	struct dirent *dp, *edp;
3981 	char *dirbuf;
3982 	off_t base;
3983 	ssize_t resid, ucount;
3984 	int error;
3985 
3986 	/* XXX arbitrary sanity limit on `count'. */
3987 	count = min(count, 64 * 1024);
3988 
3989 	dirbuf = malloc(count, M_TEMP, M_WAITOK);
3990 
3991 	error = kern_getdirentries(td, fd, dirbuf, count, &base, &resid,
3992 	    UIO_SYSSPACE);
3993 	if (error != 0)
3994 		goto done;
3995 	if (basep != NULL)
3996 		*basep = base;
3997 
3998 	ucount = 0;
3999 	for (dp = (struct dirent *)dirbuf,
4000 	    edp = (struct dirent *)&dirbuf[count - resid];
4001 	    ucount < count && dp < edp; ) {
4002 		if (dp->d_reclen == 0)
4003 			break;
4004 		MPASS(dp->d_reclen >= _GENERIC_DIRLEN(0));
4005 		if (dp->d_namlen >= sizeof(dstdp.d_name))
4006 			continue;
4007 		dstdp.d_type = dp->d_type;
4008 		dstdp.d_namlen = dp->d_namlen;
4009 		dstdp.d_fileno = dp->d_fileno;		/* truncate */
4010 		if (dstdp.d_fileno != dp->d_fileno) {
4011 			switch (ino64_trunc_error) {
4012 			default:
4013 			case 0:
4014 				break;
4015 			case 1:
4016 				error = EOVERFLOW;
4017 				goto done;
4018 			case 2:
4019 				dstdp.d_fileno = UINT32_MAX;
4020 				break;
4021 			}
4022 		}
4023 		dstdp.d_reclen = sizeof(dstdp) - sizeof(dstdp.d_name) +
4024 		    ((dp->d_namlen + 1 + 3) &~ 3);
4025 		bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen);
4026 		bzero(dstdp.d_name + dstdp.d_namlen,
4027 		    dstdp.d_reclen - offsetof(struct freebsd11_dirent, d_name) -
4028 		    dstdp.d_namlen);
4029 		MPASS(dstdp.d_reclen <= dp->d_reclen);
4030 		MPASS(ucount + dstdp.d_reclen <= count);
4031 		if (func != NULL)
4032 			func(&dstdp);
4033 		error = copyout(&dstdp, ubuf + ucount, dstdp.d_reclen);
4034 		if (error != 0)
4035 			break;
4036 		dp = (struct dirent *)((char *)dp + dp->d_reclen);
4037 		ucount += dstdp.d_reclen;
4038 	}
4039 
4040 done:
4041 	free(dirbuf, M_TEMP);
4042 	if (error == 0)
4043 		td->td_retval[0] = ucount;
4044 	return (error);
4045 }
4046 #endif /* COMPAT */
4047 
4048 #ifdef COMPAT_43
4049 static void
4050 ogetdirentries_cvt(struct freebsd11_dirent *dp)
4051 {
4052 #if (BYTE_ORDER == LITTLE_ENDIAN)
4053 	/*
4054 	 * The expected low byte of dp->d_namlen is our dp->d_type.
4055 	 * The high MBZ byte of dp->d_namlen is our dp->d_namlen.
4056 	 */
4057 	dp->d_type = dp->d_namlen;
4058 	dp->d_namlen = 0;
4059 #else
4060 	/*
4061 	 * The dp->d_type is the high byte of the expected dp->d_namlen,
4062 	 * so must be zero'ed.
4063 	 */
4064 	dp->d_type = 0;
4065 #endif
4066 }
4067 
4068 /*
4069  * Read a block of directory entries in a filesystem independent format.
4070  */
4071 #ifndef _SYS_SYSPROTO_H_
4072 struct ogetdirentries_args {
4073 	int	fd;
4074 	char	*buf;
4075 	u_int	count;
4076 	long	*basep;
4077 };
4078 #endif
4079 int
4080 ogetdirentries(struct thread *td, struct ogetdirentries_args *uap)
4081 {
4082 	long loff;
4083 	int error;
4084 
4085 	error = kern_ogetdirentries(td, uap, &loff);
4086 	if (error == 0)
4087 		error = copyout(&loff, uap->basep, sizeof(long));
4088 	return (error);
4089 }
4090 
4091 int
4092 kern_ogetdirentries(struct thread *td, struct ogetdirentries_args *uap,
4093     long *ploff)
4094 {
4095 	long base;
4096 	int error;
4097 
4098 	/* XXX arbitrary sanity limit on `count'. */
4099 	if (uap->count > 64 * 1024)
4100 		return (EINVAL);
4101 
4102 	error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count,
4103 	    &base, ogetdirentries_cvt);
4104 
4105 	if (error == 0 && uap->basep != NULL)
4106 		error = copyout(&base, uap->basep, sizeof(long));
4107 
4108 	return (error);
4109 }
4110 #endif /* COMPAT_43 */
4111 
4112 #if defined(COMPAT_FREEBSD11)
4113 #ifndef _SYS_SYSPROTO_H_
4114 struct freebsd11_getdirentries_args {
4115 	int	fd;
4116 	char	*buf;
4117 	u_int	count;
4118 	long	*basep;
4119 };
4120 #endif
4121 int
4122 freebsd11_getdirentries(struct thread *td,
4123     struct freebsd11_getdirentries_args *uap)
4124 {
4125 	long base;
4126 	int error;
4127 
4128 	error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count,
4129 	    &base, NULL);
4130 
4131 	if (error == 0 && uap->basep != NULL)
4132 		error = copyout(&base, uap->basep, sizeof(long));
4133 	return (error);
4134 }
4135 
4136 int
4137 freebsd11_getdents(struct thread *td, struct freebsd11_getdents_args *uap)
4138 {
4139 	struct freebsd11_getdirentries_args ap;
4140 
4141 	ap.fd = uap->fd;
4142 	ap.buf = uap->buf;
4143 	ap.count = uap->count;
4144 	ap.basep = NULL;
4145 	return (freebsd11_getdirentries(td, &ap));
4146 }
4147 #endif /* COMPAT_FREEBSD11 */
4148 
4149 /*
4150  * Read a block of directory entries in a filesystem independent format.
4151  */
4152 int
4153 sys_getdirentries(struct thread *td, struct getdirentries_args *uap)
4154 {
4155 	off_t base;
4156 	int error;
4157 
4158 	error = kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base,
4159 	    NULL, UIO_USERSPACE);
4160 	if (error != 0)
4161 		return (error);
4162 	if (uap->basep != NULL)
4163 		error = copyout(&base, uap->basep, sizeof(off_t));
4164 	return (error);
4165 }
4166 
4167 int
4168 kern_getdirentries(struct thread *td, int fd, char *buf, size_t count,
4169     off_t *basep, ssize_t *residp, enum uio_seg bufseg)
4170 {
4171 	struct vnode *vp;
4172 	struct file *fp;
4173 	struct uio auio;
4174 	struct iovec aiov;
4175 	off_t loff;
4176 	int error, eofflag;
4177 	off_t foffset;
4178 
4179 	AUDIT_ARG_FD(fd);
4180 	if (count > IOSIZE_MAX)
4181 		return (EINVAL);
4182 	auio.uio_resid = count;
4183 	error = getvnode(td, fd, &cap_read_rights, &fp);
4184 	if (error != 0)
4185 		return (error);
4186 	if ((fp->f_flag & FREAD) == 0) {
4187 		fdrop(fp, td);
4188 		return (EBADF);
4189 	}
4190 	vp = fp->f_vnode;
4191 	foffset = foffset_lock(fp, 0);
4192 unionread:
4193 	if (vp->v_type != VDIR) {
4194 		error = EINVAL;
4195 		goto fail;
4196 	}
4197 	aiov.iov_base = buf;
4198 	aiov.iov_len = count;
4199 	auio.uio_iov = &aiov;
4200 	auio.uio_iovcnt = 1;
4201 	auio.uio_rw = UIO_READ;
4202 	auio.uio_segflg = bufseg;
4203 	auio.uio_td = td;
4204 	vn_lock(vp, LK_SHARED | LK_RETRY);
4205 	AUDIT_ARG_VNODE1(vp);
4206 	loff = auio.uio_offset = foffset;
4207 #ifdef MAC
4208 	error = mac_vnode_check_readdir(td->td_ucred, vp);
4209 	if (error == 0)
4210 #endif
4211 		error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL,
4212 		    NULL);
4213 	foffset = auio.uio_offset;
4214 	if (error != 0) {
4215 		VOP_UNLOCK(vp);
4216 		goto fail;
4217 	}
4218 	if (count == auio.uio_resid &&
4219 	    (vp->v_vflag & VV_ROOT) &&
4220 	    (vp->v_mount->mnt_flag & MNT_UNION)) {
4221 		struct vnode *tvp = vp;
4222 
4223 		vp = vp->v_mount->mnt_vnodecovered;
4224 		VREF(vp);
4225 		fp->f_vnode = vp;
4226 		foffset = 0;
4227 		vput(tvp);
4228 		goto unionread;
4229 	}
4230 	VOP_UNLOCK(vp);
4231 	*basep = loff;
4232 	if (residp != NULL)
4233 		*residp = auio.uio_resid;
4234 	td->td_retval[0] = count - auio.uio_resid;
4235 fail:
4236 	foffset_unlock(fp, foffset, 0);
4237 	fdrop(fp, td);
4238 	return (error);
4239 }
4240 
4241 /*
4242  * Set the mode mask for creation of filesystem nodes.
4243  */
4244 #ifndef _SYS_SYSPROTO_H_
4245 struct umask_args {
4246 	int	newmask;
4247 };
4248 #endif
4249 int
4250 sys_umask(struct thread *td, struct umask_args *uap)
4251 {
4252 	struct pwddesc *pdp;
4253 
4254 	pdp = td->td_proc->p_pd;
4255 	PWDDESC_XLOCK(pdp);
4256 	td->td_retval[0] = pdp->pd_cmask;
4257 	pdp->pd_cmask = uap->newmask & ALLPERMS;
4258 	PWDDESC_XUNLOCK(pdp);
4259 	return (0);
4260 }
4261 
4262 /*
4263  * Void all references to file by ripping underlying filesystem away from
4264  * vnode.
4265  */
4266 #ifndef _SYS_SYSPROTO_H_
4267 struct revoke_args {
4268 	char	*path;
4269 };
4270 #endif
4271 int
4272 sys_revoke(struct thread *td, struct revoke_args *uap)
4273 {
4274 	struct vnode *vp;
4275 	struct vattr vattr;
4276 	struct nameidata nd;
4277 	int error;
4278 
4279 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE,
4280 	    uap->path);
4281 	if ((error = namei(&nd)) != 0)
4282 		return (error);
4283 	vp = nd.ni_vp;
4284 	NDFREE_NOTHING(&nd);
4285 	if (vp->v_type != VCHR || vp->v_rdev == NULL) {
4286 		error = EINVAL;
4287 		goto out;
4288 	}
4289 #ifdef MAC
4290 	error = mac_vnode_check_revoke(td->td_ucred, vp);
4291 	if (error != 0)
4292 		goto out;
4293 #endif
4294 	error = VOP_GETATTR(vp, &vattr, td->td_ucred);
4295 	if (error != 0)
4296 		goto out;
4297 	if (td->td_ucred->cr_uid != vattr.va_uid) {
4298 		error = priv_check(td, PRIV_VFS_ADMIN);
4299 		if (error != 0)
4300 			goto out;
4301 	}
4302 	if (devfs_usecount(vp) > 0)
4303 		VOP_REVOKE(vp, REVOKEALL);
4304 out:
4305 	vput(vp);
4306 	return (error);
4307 }
4308 
4309 /*
4310  * This variant of getvnode() allows O_PATH files.  Caller should
4311  * ensure that returned file and vnode are only used for compatible
4312  * semantics.
4313  */
4314 int
4315 getvnode_path(struct thread *td, int fd, cap_rights_t *rightsp,
4316     struct file **fpp)
4317 {
4318 	struct file *fp;
4319 	int error;
4320 
4321 	error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, &fp);
4322 	if (error != 0)
4323 		return (error);
4324 
4325 	/*
4326 	 * The file could be not of the vnode type, or it may be not
4327 	 * yet fully initialized, in which case the f_vnode pointer
4328 	 * may be set, but f_ops is still badfileops.  E.g.,
4329 	 * devfs_open() transiently create such situation to
4330 	 * facilitate csw d_fdopen().
4331 	 *
4332 	 * Dupfdopen() handling in kern_openat() installs the
4333 	 * half-baked file into the process descriptor table, allowing
4334 	 * other thread to dereference it. Guard against the race by
4335 	 * checking f_ops.
4336 	 */
4337 	if (__predict_false(fp->f_vnode == NULL || fp->f_ops == &badfileops)) {
4338 		fdrop(fp, td);
4339 		*fpp = NULL;
4340 		return (EINVAL);
4341 	}
4342 
4343 	*fpp = fp;
4344 	return (0);
4345 }
4346 
4347 /*
4348  * Convert a user file descriptor to a kernel file entry and check
4349  * that, if it is a capability, the correct rights are present.
4350  * A reference on the file entry is held upon returning.
4351  */
4352 int
4353 getvnode(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
4354 {
4355 	int error;
4356 
4357 	error = getvnode_path(td, fd, rightsp, fpp);
4358 	if (__predict_false(error != 0))
4359 		return (error);
4360 
4361 	/*
4362 	 * Filter out O_PATH file descriptors, most getvnode() callers
4363 	 * do not call fo_ methods.
4364 	 */
4365 	if (__predict_false((*fpp)->f_ops == &path_fileops)) {
4366 		fdrop(*fpp, td);
4367 		*fpp = NULL;
4368 		error = EBADF;
4369 	}
4370 
4371 	return (error);
4372 }
4373 
4374 /*
4375  * Get an (NFS) file handle.
4376  */
4377 #ifndef _SYS_SYSPROTO_H_
4378 struct lgetfh_args {
4379 	char *fname;
4380 	fhandle_t *fhp;
4381 };
4382 #endif
4383 int
4384 sys_lgetfh(struct thread *td, struct lgetfh_args *uap)
4385 {
4386 
4387 	return (kern_getfhat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->fname,
4388 	    UIO_USERSPACE, uap->fhp, UIO_USERSPACE));
4389 }
4390 
4391 #ifndef _SYS_SYSPROTO_H_
4392 struct getfh_args {
4393 	char *fname;
4394 	fhandle_t *fhp;
4395 };
4396 #endif
4397 int
4398 sys_getfh(struct thread *td, struct getfh_args *uap)
4399 {
4400 
4401 	return (kern_getfhat(td, 0, AT_FDCWD, uap->fname, UIO_USERSPACE,
4402 	    uap->fhp, UIO_USERSPACE));
4403 }
4404 
4405 /*
4406  * syscall for the rpc.lockd to use to translate an open descriptor into
4407  * a NFS file handle.
4408  *
4409  * warning: do not remove the priv_check() call or this becomes one giant
4410  * security hole.
4411  */
4412 #ifndef _SYS_SYSPROTO_H_
4413 struct getfhat_args {
4414 	int fd;
4415 	char *path;
4416 	fhandle_t *fhp;
4417 	int flags;
4418 };
4419 #endif
4420 int
4421 sys_getfhat(struct thread *td, struct getfhat_args *uap)
4422 {
4423 
4424 	return (kern_getfhat(td, uap->flags, uap->fd, uap->path, UIO_USERSPACE,
4425 	    uap->fhp, UIO_USERSPACE));
4426 }
4427 
4428 int
4429 kern_getfhat(struct thread *td, int flags, int fd, const char *path,
4430     enum uio_seg pathseg, fhandle_t *fhp, enum uio_seg fhseg)
4431 {
4432 	struct nameidata nd;
4433 	fhandle_t fh;
4434 	struct vnode *vp;
4435 	int error;
4436 
4437 	if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_RESOLVE_BENEATH)) != 0)
4438 		return (EINVAL);
4439 	error = priv_check(td, PRIV_VFS_GETFH);
4440 	if (error != 0)
4441 		return (error);
4442 	NDINIT_AT(&nd, LOOKUP, at2cnpflags(flags, AT_SYMLINK_NOFOLLOW |
4443 	    AT_RESOLVE_BENEATH) | LOCKLEAF | AUDITVNODE1, pathseg, path,
4444 	    fd);
4445 	error = namei(&nd);
4446 	if (error != 0)
4447 		return (error);
4448 	NDFREE_NOTHING(&nd);
4449 	vp = nd.ni_vp;
4450 	bzero(&fh, sizeof(fh));
4451 	fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
4452 	error = VOP_VPTOFH(vp, &fh.fh_fid);
4453 	vput(vp);
4454 	if (error == 0) {
4455 		if (fhseg == UIO_USERSPACE)
4456 			error = copyout(&fh, fhp, sizeof (fh));
4457 		else
4458 			memcpy(fhp, &fh, sizeof(fh));
4459 	}
4460 	return (error);
4461 }
4462 
4463 #ifndef _SYS_SYSPROTO_H_
4464 struct fhlink_args {
4465 	fhandle_t *fhp;
4466 	const char *to;
4467 };
4468 #endif
4469 int
4470 sys_fhlink(struct thread *td, struct fhlink_args *uap)
4471 {
4472 
4473 	return (kern_fhlinkat(td, AT_FDCWD, uap->to, UIO_USERSPACE, uap->fhp));
4474 }
4475 
4476 #ifndef _SYS_SYSPROTO_H_
4477 struct fhlinkat_args {
4478 	fhandle_t *fhp;
4479 	int tofd;
4480 	const char *to;
4481 };
4482 #endif
4483 int
4484 sys_fhlinkat(struct thread *td, struct fhlinkat_args *uap)
4485 {
4486 
4487 	return (kern_fhlinkat(td, uap->tofd, uap->to, UIO_USERSPACE, uap->fhp));
4488 }
4489 
4490 static int
4491 kern_fhlinkat(struct thread *td, int fd, const char *path,
4492     enum uio_seg pathseg, fhandle_t *fhp)
4493 {
4494 	fhandle_t fh;
4495 	struct mount *mp;
4496 	struct vnode *vp;
4497 	int error;
4498 
4499 	error = priv_check(td, PRIV_VFS_GETFH);
4500 	if (error != 0)
4501 		return (error);
4502 	error = copyin(fhp, &fh, sizeof(fh));
4503 	if (error != 0)
4504 		return (error);
4505 	do {
4506 		bwillwrite();
4507 		if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4508 			return (ESTALE);
4509 		error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp);
4510 		vfs_unbusy(mp);
4511 		if (error != 0)
4512 			return (error);
4513 		VOP_UNLOCK(vp);
4514 		error = kern_linkat_vp(td, vp, fd, path, pathseg);
4515 	} while (error == EAGAIN || error == ERELOOKUP);
4516 	return (error);
4517 }
4518 
4519 #ifndef _SYS_SYSPROTO_H_
4520 struct fhreadlink_args {
4521 	fhandle_t *fhp;
4522 	char *buf;
4523 	size_t bufsize;
4524 };
4525 #endif
4526 int
4527 sys_fhreadlink(struct thread *td, struct fhreadlink_args *uap)
4528 {
4529 	fhandle_t fh;
4530 	struct mount *mp;
4531 	struct vnode *vp;
4532 	int error;
4533 
4534 	error = priv_check(td, PRIV_VFS_GETFH);
4535 	if (error != 0)
4536 		return (error);
4537 	if (uap->bufsize > IOSIZE_MAX)
4538 		return (EINVAL);
4539 	error = copyin(uap->fhp, &fh, sizeof(fh));
4540 	if (error != 0)
4541 		return (error);
4542 	if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4543 		return (ESTALE);
4544 	error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp);
4545 	vfs_unbusy(mp);
4546 	if (error != 0)
4547 		return (error);
4548 	error = kern_readlink_vp(vp, uap->buf, UIO_USERSPACE, uap->bufsize, td);
4549 	vput(vp);
4550 	return (error);
4551 }
4552 
4553 /*
4554  * syscall for the rpc.lockd to use to translate a NFS file handle into an
4555  * open descriptor.
4556  *
4557  * warning: do not remove the priv_check() call or this becomes one giant
4558  * security hole.
4559  */
4560 #ifndef _SYS_SYSPROTO_H_
4561 struct fhopen_args {
4562 	const struct fhandle *u_fhp;
4563 	int flags;
4564 };
4565 #endif
4566 int
4567 sys_fhopen(struct thread *td, struct fhopen_args *uap)
4568 {
4569 	return (kern_fhopen(td, uap->u_fhp, uap->flags));
4570 }
4571 
4572 int
4573 kern_fhopen(struct thread *td, const struct fhandle *u_fhp, int flags)
4574 {
4575 	struct mount *mp;
4576 	struct vnode *vp;
4577 	struct fhandle fhp;
4578 	struct file *fp;
4579 	int fmode, error;
4580 	int indx;
4581 
4582 	error = priv_check(td, PRIV_VFS_FHOPEN);
4583 	if (error != 0)
4584 		return (error);
4585 	indx = -1;
4586 	fmode = FFLAGS(flags);
4587 	/* why not allow a non-read/write open for our lockd? */
4588 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4589 		return (EINVAL);
4590 	error = copyin(u_fhp, &fhp, sizeof(fhp));
4591 	if (error != 0)
4592 		return(error);
4593 	/* find the mount point */
4594 	mp = vfs_busyfs(&fhp.fh_fsid);
4595 	if (mp == NULL)
4596 		return (ESTALE);
4597 	/* now give me my vnode, it gets returned to me locked */
4598 	error = VFS_FHTOVP(mp, &fhp.fh_fid, LK_EXCLUSIVE, &vp);
4599 	vfs_unbusy(mp);
4600 	if (error != 0)
4601 		return (error);
4602 
4603 	error = falloc_noinstall(td, &fp);
4604 	if (error != 0) {
4605 		vput(vp);
4606 		return (error);
4607 	}
4608 	/*
4609 	 * An extra reference on `fp' has been held for us by
4610 	 * falloc_noinstall().
4611 	 */
4612 
4613 #ifdef INVARIANTS
4614 	td->td_dupfd = -1;
4615 #endif
4616 	error = vn_open_vnode(vp, fmode, td->td_ucred, td, fp);
4617 	if (error != 0) {
4618 		KASSERT(fp->f_ops == &badfileops,
4619 		    ("VOP_OPEN in fhopen() set f_ops"));
4620 		KASSERT(td->td_dupfd < 0,
4621 		    ("fhopen() encountered fdopen()"));
4622 
4623 		vput(vp);
4624 		goto bad;
4625 	}
4626 #ifdef INVARIANTS
4627 	td->td_dupfd = 0;
4628 #endif
4629 	fp->f_vnode = vp;
4630 	finit_vnode(fp, fmode, NULL, &vnops);
4631 	VOP_UNLOCK(vp);
4632 	if ((fmode & O_TRUNC) != 0) {
4633 		error = fo_truncate(fp, 0, td->td_ucred, td);
4634 		if (error != 0)
4635 			goto bad;
4636 	}
4637 
4638 	error = finstall(td, fp, &indx, fmode, NULL);
4639 bad:
4640 	fdrop(fp, td);
4641 	td->td_retval[0] = indx;
4642 	return (error);
4643 }
4644 
4645 /*
4646  * Stat an (NFS) file handle.
4647  */
4648 #ifndef _SYS_SYSPROTO_H_
4649 struct fhstat_args {
4650 	struct fhandle *u_fhp;
4651 	struct stat *sb;
4652 };
4653 #endif
4654 int
4655 sys_fhstat(struct thread *td, struct fhstat_args *uap)
4656 {
4657 	struct stat sb;
4658 	struct fhandle fh;
4659 	int error;
4660 
4661 	error = copyin(uap->u_fhp, &fh, sizeof(fh));
4662 	if (error != 0)
4663 		return (error);
4664 	error = kern_fhstat(td, fh, &sb);
4665 	if (error == 0)
4666 		error = copyout(&sb, uap->sb, sizeof(sb));
4667 	return (error);
4668 }
4669 
4670 int
4671 kern_fhstat(struct thread *td, struct fhandle fh, struct stat *sb)
4672 {
4673 	struct mount *mp;
4674 	struct vnode *vp;
4675 	int error;
4676 
4677 	error = priv_check(td, PRIV_VFS_FHSTAT);
4678 	if (error != 0)
4679 		return (error);
4680 	if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4681 		return (ESTALE);
4682 	error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp);
4683 	vfs_unbusy(mp);
4684 	if (error != 0)
4685 		return (error);
4686 	error = VOP_STAT(vp, sb, td->td_ucred, NOCRED);
4687 	vput(vp);
4688 	return (error);
4689 }
4690 
4691 /*
4692  * Implement fstatfs() for (NFS) file handles.
4693  */
4694 #ifndef _SYS_SYSPROTO_H_
4695 struct fhstatfs_args {
4696 	struct fhandle *u_fhp;
4697 	struct statfs *buf;
4698 };
4699 #endif
4700 int
4701 sys_fhstatfs(struct thread *td, struct fhstatfs_args *uap)
4702 {
4703 	struct statfs *sfp;
4704 	fhandle_t fh;
4705 	int error;
4706 
4707 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4708 	if (error != 0)
4709 		return (error);
4710 	sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
4711 	error = kern_fhstatfs(td, fh, sfp);
4712 	if (error == 0)
4713 		error = copyout(sfp, uap->buf, sizeof(*sfp));
4714 	free(sfp, M_STATFS);
4715 	return (error);
4716 }
4717 
4718 int
4719 kern_fhstatfs(struct thread *td, fhandle_t fh, struct statfs *buf)
4720 {
4721 	struct mount *mp;
4722 	struct vnode *vp;
4723 	int error;
4724 
4725 	error = priv_check(td, PRIV_VFS_FHSTATFS);
4726 	if (error != 0)
4727 		return (error);
4728 	if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL)
4729 		return (ESTALE);
4730 	error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp);
4731 	if (error != 0) {
4732 		vfs_unbusy(mp);
4733 		return (error);
4734 	}
4735 	vput(vp);
4736 	error = prison_canseemount(td->td_ucred, mp);
4737 	if (error != 0)
4738 		goto out;
4739 #ifdef MAC
4740 	error = mac_mount_check_stat(td->td_ucred, mp);
4741 	if (error != 0)
4742 		goto out;
4743 #endif
4744 	error = VFS_STATFS(mp, buf);
4745 out:
4746 	vfs_unbusy(mp);
4747 	return (error);
4748 }
4749 
4750 /*
4751  * Unlike madvise(2), we do not make a best effort to remember every
4752  * possible caching hint.  Instead, we remember the last setting with
4753  * the exception that we will allow POSIX_FADV_NORMAL to adjust the
4754  * region of any current setting.
4755  */
4756 int
4757 kern_posix_fadvise(struct thread *td, int fd, off_t offset, off_t len,
4758     int advice)
4759 {
4760 	struct fadvise_info *fa, *new;
4761 	struct file *fp;
4762 	struct vnode *vp;
4763 	off_t end;
4764 	int error;
4765 
4766 	if (offset < 0 || len < 0 || offset > OFF_MAX - len)
4767 		return (EINVAL);
4768 	AUDIT_ARG_VALUE(advice);
4769 	switch (advice) {
4770 	case POSIX_FADV_SEQUENTIAL:
4771 	case POSIX_FADV_RANDOM:
4772 	case POSIX_FADV_NOREUSE:
4773 		new = malloc(sizeof(*fa), M_FADVISE, M_WAITOK);
4774 		break;
4775 	case POSIX_FADV_NORMAL:
4776 	case POSIX_FADV_WILLNEED:
4777 	case POSIX_FADV_DONTNEED:
4778 		new = NULL;
4779 		break;
4780 	default:
4781 		return (EINVAL);
4782 	}
4783 	/* XXX: CAP_POSIX_FADVISE? */
4784 	AUDIT_ARG_FD(fd);
4785 	error = fget(td, fd, &cap_no_rights, &fp);
4786 	if (error != 0)
4787 		goto out;
4788 	AUDIT_ARG_FILE(td->td_proc, fp);
4789 	if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) {
4790 		error = ESPIPE;
4791 		goto out;
4792 	}
4793 	if (fp->f_type != DTYPE_VNODE) {
4794 		error = ENODEV;
4795 		goto out;
4796 	}
4797 	vp = fp->f_vnode;
4798 	if (vp->v_type != VREG) {
4799 		error = ENODEV;
4800 		goto out;
4801 	}
4802 	if (len == 0)
4803 		end = OFF_MAX;
4804 	else
4805 		end = offset + len - 1;
4806 	switch (advice) {
4807 	case POSIX_FADV_SEQUENTIAL:
4808 	case POSIX_FADV_RANDOM:
4809 	case POSIX_FADV_NOREUSE:
4810 		/*
4811 		 * Try to merge any existing non-standard region with
4812 		 * this new region if possible, otherwise create a new
4813 		 * non-standard region for this request.
4814 		 */
4815 		mtx_pool_lock(mtxpool_sleep, fp);
4816 		fa = fp->f_advice;
4817 		if (fa != NULL && fa->fa_advice == advice &&
4818 		    ((fa->fa_start <= end && fa->fa_end >= offset) ||
4819 		    (end != OFF_MAX && fa->fa_start == end + 1) ||
4820 		    (fa->fa_end != OFF_MAX && fa->fa_end + 1 == offset))) {
4821 			if (offset < fa->fa_start)
4822 				fa->fa_start = offset;
4823 			if (end > fa->fa_end)
4824 				fa->fa_end = end;
4825 		} else {
4826 			new->fa_advice = advice;
4827 			new->fa_start = offset;
4828 			new->fa_end = end;
4829 			fp->f_advice = new;
4830 			new = fa;
4831 		}
4832 		mtx_pool_unlock(mtxpool_sleep, fp);
4833 		break;
4834 	case POSIX_FADV_NORMAL:
4835 		/*
4836 		 * If a the "normal" region overlaps with an existing
4837 		 * non-standard region, trim or remove the
4838 		 * non-standard region.
4839 		 */
4840 		mtx_pool_lock(mtxpool_sleep, fp);
4841 		fa = fp->f_advice;
4842 		if (fa != NULL) {
4843 			if (offset <= fa->fa_start && end >= fa->fa_end) {
4844 				new = fa;
4845 				fp->f_advice = NULL;
4846 			} else if (offset <= fa->fa_start &&
4847 			    end >= fa->fa_start)
4848 				fa->fa_start = end + 1;
4849 			else if (offset <= fa->fa_end && end >= fa->fa_end)
4850 				fa->fa_end = offset - 1;
4851 			else if (offset >= fa->fa_start && end <= fa->fa_end) {
4852 				/*
4853 				 * If the "normal" region is a middle
4854 				 * portion of the existing
4855 				 * non-standard region, just remove
4856 				 * the whole thing rather than picking
4857 				 * one side or the other to
4858 				 * preserve.
4859 				 */
4860 				new = fa;
4861 				fp->f_advice = NULL;
4862 			}
4863 		}
4864 		mtx_pool_unlock(mtxpool_sleep, fp);
4865 		break;
4866 	case POSIX_FADV_WILLNEED:
4867 	case POSIX_FADV_DONTNEED:
4868 		error = VOP_ADVISE(vp, offset, end, advice);
4869 		break;
4870 	}
4871 out:
4872 	if (fp != NULL)
4873 		fdrop(fp, td);
4874 	free(new, M_FADVISE);
4875 	return (error);
4876 }
4877 
4878 int
4879 sys_posix_fadvise(struct thread *td, struct posix_fadvise_args *uap)
4880 {
4881 	int error;
4882 
4883 	error = kern_posix_fadvise(td, uap->fd, uap->offset, uap->len,
4884 	    uap->advice);
4885 	return (kern_posix_error(td, error));
4886 }
4887 
4888 int
4889 kern_copy_file_range(struct thread *td, int infd, off_t *inoffp, int outfd,
4890     off_t *outoffp, size_t len, unsigned int flags)
4891 {
4892 	struct file *infp, *outfp;
4893 	struct vnode *invp, *outvp;
4894 	int error;
4895 	size_t retlen;
4896 	void *rl_rcookie, *rl_wcookie;
4897 	off_t savinoff, savoutoff;
4898 
4899 	infp = outfp = NULL;
4900 	rl_rcookie = rl_wcookie = NULL;
4901 	savinoff = -1;
4902 	error = 0;
4903 	retlen = 0;
4904 
4905 	if (flags != 0) {
4906 		error = EINVAL;
4907 		goto out;
4908 	}
4909 	if (len > SSIZE_MAX)
4910 		/*
4911 		 * Although the len argument is size_t, the return argument
4912 		 * is ssize_t (which is signed).  Therefore a size that won't
4913 		 * fit in ssize_t can't be returned.
4914 		 */
4915 		len = SSIZE_MAX;
4916 
4917 	/* Get the file structures for the file descriptors. */
4918 	error = fget_read(td, infd, &cap_read_rights, &infp);
4919 	if (error != 0)
4920 		goto out;
4921 	if (infp->f_ops == &badfileops) {
4922 		error = EBADF;
4923 		goto out;
4924 	}
4925 	if (infp->f_vnode == NULL) {
4926 		error = EINVAL;
4927 		goto out;
4928 	}
4929 	error = fget_write(td, outfd, &cap_write_rights, &outfp);
4930 	if (error != 0)
4931 		goto out;
4932 	if (outfp->f_ops == &badfileops) {
4933 		error = EBADF;
4934 		goto out;
4935 	}
4936 	if (outfp->f_vnode == NULL) {
4937 		error = EINVAL;
4938 		goto out;
4939 	}
4940 
4941 	/* Set the offset pointers to the correct place. */
4942 	if (inoffp == NULL)
4943 		inoffp = &infp->f_offset;
4944 	if (outoffp == NULL)
4945 		outoffp = &outfp->f_offset;
4946 	savinoff = *inoffp;
4947 	savoutoff = *outoffp;
4948 
4949 	invp = infp->f_vnode;
4950 	outvp = outfp->f_vnode;
4951 	/* Sanity check the f_flag bits. */
4952 	if ((outfp->f_flag & (FWRITE | FAPPEND)) != FWRITE ||
4953 	    (infp->f_flag & FREAD) == 0) {
4954 		error = EBADF;
4955 		goto out;
4956 	}
4957 
4958 	/* If len == 0, just return 0. */
4959 	if (len == 0)
4960 		goto out;
4961 
4962 	/*
4963 	 * If infp and outfp refer to the same file, the byte ranges cannot
4964 	 * overlap.
4965 	 */
4966 	if (invp == outvp && ((savinoff <= savoutoff && savinoff + len >
4967 	    savoutoff) || (savinoff > savoutoff && savoutoff + len >
4968 	    savinoff))) {
4969 		error = EINVAL;
4970 		goto out;
4971 	}
4972 
4973 	/* Range lock the byte ranges for both invp and outvp. */
4974 	for (;;) {
4975 		rl_wcookie = vn_rangelock_wlock(outvp, *outoffp, *outoffp +
4976 		    len);
4977 		rl_rcookie = vn_rangelock_tryrlock(invp, *inoffp, *inoffp +
4978 		    len);
4979 		if (rl_rcookie != NULL)
4980 			break;
4981 		vn_rangelock_unlock(outvp, rl_wcookie);
4982 		rl_rcookie = vn_rangelock_rlock(invp, *inoffp, *inoffp + len);
4983 		vn_rangelock_unlock(invp, rl_rcookie);
4984 	}
4985 
4986 	retlen = len;
4987 	error = vn_copy_file_range(invp, inoffp, outvp, outoffp, &retlen,
4988 	    flags, infp->f_cred, outfp->f_cred, td);
4989 out:
4990 	if (rl_rcookie != NULL)
4991 		vn_rangelock_unlock(invp, rl_rcookie);
4992 	if (rl_wcookie != NULL)
4993 		vn_rangelock_unlock(outvp, rl_wcookie);
4994 	if (savinoff != -1 && (error == EINTR || error == ERESTART)) {
4995 		*inoffp = savinoff;
4996 		*outoffp = savoutoff;
4997 	}
4998 	if (outfp != NULL)
4999 		fdrop(outfp, td);
5000 	if (infp != NULL)
5001 		fdrop(infp, td);
5002 	td->td_retval[0] = retlen;
5003 	return (error);
5004 }
5005 
5006 int
5007 sys_copy_file_range(struct thread *td, struct copy_file_range_args *uap)
5008 {
5009 	off_t inoff, outoff, *inoffp, *outoffp;
5010 	int error;
5011 
5012 	inoffp = outoffp = NULL;
5013 	if (uap->inoffp != NULL) {
5014 		error = copyin(uap->inoffp, &inoff, sizeof(off_t));
5015 		if (error != 0)
5016 			return (error);
5017 		inoffp = &inoff;
5018 	}
5019 	if (uap->outoffp != NULL) {
5020 		error = copyin(uap->outoffp, &outoff, sizeof(off_t));
5021 		if (error != 0)
5022 			return (error);
5023 		outoffp = &outoff;
5024 	}
5025 	error = kern_copy_file_range(td, uap->infd, inoffp, uap->outfd,
5026 	    outoffp, uap->len, uap->flags);
5027 	if (error == 0 && uap->inoffp != NULL)
5028 		error = copyout(inoffp, uap->inoffp, sizeof(off_t));
5029 	if (error == 0 && uap->outoffp != NULL)
5030 		error = copyout(outoffp, uap->outoffp, sizeof(off_t));
5031 	return (error);
5032 }
5033