xref: /freebsd/sys/kern/vfs_mount.c (revision 422d05da14fe063e5d187d81a328fa7b362d069f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1999-2004 Poul-Henning Kamp
5  * Copyright (c) 1999 Michael Smith
6  * Copyright (c) 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/conf.h>
44 #include <sys/smp.h>
45 #include <sys/eventhandler.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/libkern.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/filedesc.h>
58 #include <sys/reboot.h>
59 #include <sys/sbuf.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/systm.h>
66 #include <sys/vnode.h>
67 #include <vm/uma.h>
68 
69 #include <geom/geom.h>
70 
71 #include <machine/stdarg.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
77 
78 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
79 		    uint64_t fsflags, struct vfsoptlist **optlist);
80 static void	free_mntarg(struct mntarg *ma);
81 
82 static int	usermount = 0;
83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
84     "Unprivileged users may mount and unmount file systems");
85 
86 static bool	default_autoro = false;
87 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
88     "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
89 
90 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
91 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
92 static uma_zone_t mount_zone;
93 
94 /* List of mounted filesystems. */
95 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
96 
97 /* For any iteration/modification of mountlist */
98 struct mtx mountlist_mtx;
99 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
100 
101 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
102 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
103 
104 /*
105  * Global opts, taken by all filesystems
106  */
107 static const char *global_opts[] = {
108 	"errmsg",
109 	"fstype",
110 	"fspath",
111 	"ro",
112 	"rw",
113 	"nosuid",
114 	"noexec",
115 	NULL
116 };
117 
118 static int
119 mount_init(void *mem, int size, int flags)
120 {
121 	struct mount *mp;
122 
123 	mp = (struct mount *)mem;
124 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
125 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
126 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
127 	mp->mnt_thread_in_ops_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
128 	    M_WAITOK | M_ZERO);
129 	mp->mnt_ref_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
130 	    M_WAITOK | M_ZERO);
131 	mp->mnt_lockref_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
132 	    M_WAITOK | M_ZERO);
133 	mp->mnt_writeopcount_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
134 	    M_WAITOK | M_ZERO);
135 	mp->mnt_ref = 0;
136 	mp->mnt_vfs_ops = 1;
137 	mp->mnt_rootvnode = NULL;
138 	return (0);
139 }
140 
141 static void
142 mount_fini(void *mem, int size)
143 {
144 	struct mount *mp;
145 
146 	mp = (struct mount *)mem;
147 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_writeopcount_pcpu);
148 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_lockref_pcpu);
149 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_ref_pcpu);
150 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_thread_in_ops_pcpu);
151 	lockdestroy(&mp->mnt_explock);
152 	mtx_destroy(&mp->mnt_listmtx);
153 	mtx_destroy(&mp->mnt_mtx);
154 }
155 
156 static void
157 vfs_mount_init(void *dummy __unused)
158 {
159 
160 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
161 	    NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
162 }
163 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
164 
165 /*
166  * ---------------------------------------------------------------------
167  * Functions for building and sanitizing the mount options
168  */
169 
170 /* Remove one mount option. */
171 static void
172 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
173 {
174 
175 	TAILQ_REMOVE(opts, opt, link);
176 	free(opt->name, M_MOUNT);
177 	if (opt->value != NULL)
178 		free(opt->value, M_MOUNT);
179 	free(opt, M_MOUNT);
180 }
181 
182 /* Release all resources related to the mount options. */
183 void
184 vfs_freeopts(struct vfsoptlist *opts)
185 {
186 	struct vfsopt *opt;
187 
188 	while (!TAILQ_EMPTY(opts)) {
189 		opt = TAILQ_FIRST(opts);
190 		vfs_freeopt(opts, opt);
191 	}
192 	free(opts, M_MOUNT);
193 }
194 
195 void
196 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
197 {
198 	struct vfsopt *opt, *temp;
199 
200 	if (opts == NULL)
201 		return;
202 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
203 		if (strcmp(opt->name, name) == 0)
204 			vfs_freeopt(opts, opt);
205 	}
206 }
207 
208 static int
209 vfs_isopt_ro(const char *opt)
210 {
211 
212 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
213 	    strcmp(opt, "norw") == 0)
214 		return (1);
215 	return (0);
216 }
217 
218 static int
219 vfs_isopt_rw(const char *opt)
220 {
221 
222 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
223 		return (1);
224 	return (0);
225 }
226 
227 /*
228  * Check if options are equal (with or without the "no" prefix).
229  */
230 static int
231 vfs_equalopts(const char *opt1, const char *opt2)
232 {
233 	char *p;
234 
235 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
236 	if (strcmp(opt1, opt2) == 0)
237 		return (1);
238 	/* "noopt" vs. "opt" */
239 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
240 		return (1);
241 	/* "opt" vs. "noopt" */
242 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
243 		return (1);
244 	while ((p = strchr(opt1, '.')) != NULL &&
245 	    !strncmp(opt1, opt2, ++p - opt1)) {
246 		opt2 += p - opt1;
247 		opt1 = p;
248 		/* "foo.noopt" vs. "foo.opt" */
249 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
250 			return (1);
251 		/* "foo.opt" vs. "foo.noopt" */
252 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
253 			return (1);
254 	}
255 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
256 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
257 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
258 		return (1);
259 	return (0);
260 }
261 
262 /*
263  * If a mount option is specified several times,
264  * (with or without the "no" prefix) only keep
265  * the last occurrence of it.
266  */
267 static void
268 vfs_sanitizeopts(struct vfsoptlist *opts)
269 {
270 	struct vfsopt *opt, *opt2, *tmp;
271 
272 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
273 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
274 		while (opt2 != NULL) {
275 			if (vfs_equalopts(opt->name, opt2->name)) {
276 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
277 				vfs_freeopt(opts, opt2);
278 				opt2 = tmp;
279 			} else {
280 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
281 			}
282 		}
283 	}
284 }
285 
286 /*
287  * Build a linked list of mount options from a struct uio.
288  */
289 int
290 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
291 {
292 	struct vfsoptlist *opts;
293 	struct vfsopt *opt;
294 	size_t memused, namelen, optlen;
295 	unsigned int i, iovcnt;
296 	int error;
297 
298 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
299 	TAILQ_INIT(opts);
300 	memused = 0;
301 	iovcnt = auio->uio_iovcnt;
302 	for (i = 0; i < iovcnt; i += 2) {
303 		namelen = auio->uio_iov[i].iov_len;
304 		optlen = auio->uio_iov[i + 1].iov_len;
305 		memused += sizeof(struct vfsopt) + optlen + namelen;
306 		/*
307 		 * Avoid consuming too much memory, and attempts to overflow
308 		 * memused.
309 		 */
310 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
311 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
312 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
313 			error = EINVAL;
314 			goto bad;
315 		}
316 
317 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
318 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
319 		opt->value = NULL;
320 		opt->len = 0;
321 		opt->pos = i / 2;
322 		opt->seen = 0;
323 
324 		/*
325 		 * Do this early, so jumps to "bad" will free the current
326 		 * option.
327 		 */
328 		TAILQ_INSERT_TAIL(opts, opt, link);
329 
330 		if (auio->uio_segflg == UIO_SYSSPACE) {
331 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
332 		} else {
333 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
334 			    namelen);
335 			if (error)
336 				goto bad;
337 		}
338 		/* Ensure names are null-terminated strings. */
339 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
340 			error = EINVAL;
341 			goto bad;
342 		}
343 		if (optlen != 0) {
344 			opt->len = optlen;
345 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
346 			if (auio->uio_segflg == UIO_SYSSPACE) {
347 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
348 				    optlen);
349 			} else {
350 				error = copyin(auio->uio_iov[i + 1].iov_base,
351 				    opt->value, optlen);
352 				if (error)
353 					goto bad;
354 			}
355 		}
356 	}
357 	vfs_sanitizeopts(opts);
358 	*options = opts;
359 	return (0);
360 bad:
361 	vfs_freeopts(opts);
362 	return (error);
363 }
364 
365 /*
366  * Merge the old mount options with the new ones passed
367  * in the MNT_UPDATE case.
368  *
369  * XXX: This function will keep a "nofoo" option in the new
370  * options.  E.g, if the option's canonical name is "foo",
371  * "nofoo" ends up in the mount point's active options.
372  */
373 static void
374 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
375 {
376 	struct vfsopt *opt, *new;
377 
378 	TAILQ_FOREACH(opt, oldopts, link) {
379 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
380 		new->name = strdup(opt->name, M_MOUNT);
381 		if (opt->len != 0) {
382 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
383 			bcopy(opt->value, new->value, opt->len);
384 		} else
385 			new->value = NULL;
386 		new->len = opt->len;
387 		new->seen = opt->seen;
388 		TAILQ_INSERT_HEAD(toopts, new, link);
389 	}
390 	vfs_sanitizeopts(toopts);
391 }
392 
393 /*
394  * Mount a filesystem.
395  */
396 #ifndef _SYS_SYSPROTO_H_
397 struct nmount_args {
398 	struct iovec *iovp;
399 	unsigned int iovcnt;
400 	int flags;
401 };
402 #endif
403 int
404 sys_nmount(struct thread *td, struct nmount_args *uap)
405 {
406 	struct uio *auio;
407 	int error;
408 	u_int iovcnt;
409 	uint64_t flags;
410 
411 	/*
412 	 * Mount flags are now 64-bits. On 32-bit archtectures only
413 	 * 32-bits are passed in, but from here on everything handles
414 	 * 64-bit flags correctly.
415 	 */
416 	flags = uap->flags;
417 
418 	AUDIT_ARG_FFLAGS(flags);
419 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
420 	    uap->iovp, uap->iovcnt, flags);
421 
422 	/*
423 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
424 	 * userspace to set this flag, but we must filter it out if we want
425 	 * MNT_UPDATE on the root file system to work.
426 	 * MNT_ROOTFS should only be set by the kernel when mounting its
427 	 * root file system.
428 	 */
429 	flags &= ~MNT_ROOTFS;
430 
431 	iovcnt = uap->iovcnt;
432 	/*
433 	 * Check that we have an even number of iovec's
434 	 * and that we have at least two options.
435 	 */
436 	if ((iovcnt & 1) || (iovcnt < 4)) {
437 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
438 		    uap->iovcnt);
439 		return (EINVAL);
440 	}
441 
442 	error = copyinuio(uap->iovp, iovcnt, &auio);
443 	if (error) {
444 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
445 		    __func__, error);
446 		return (error);
447 	}
448 	error = vfs_donmount(td, flags, auio);
449 
450 	free(auio, M_IOV);
451 	return (error);
452 }
453 
454 /*
455  * ---------------------------------------------------------------------
456  * Various utility functions
457  */
458 
459 void
460 vfs_ref(struct mount *mp)
461 {
462 
463 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
464 	if (vfs_op_thread_enter(mp)) {
465 		vfs_mp_count_add_pcpu(mp, ref, 1);
466 		vfs_op_thread_exit(mp);
467 		return;
468 	}
469 
470 	MNT_ILOCK(mp);
471 	MNT_REF(mp);
472 	MNT_IUNLOCK(mp);
473 }
474 
475 void
476 vfs_rel(struct mount *mp)
477 {
478 
479 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
480 	if (vfs_op_thread_enter(mp)) {
481 		vfs_mp_count_sub_pcpu(mp, ref, 1);
482 		vfs_op_thread_exit(mp);
483 		return;
484 	}
485 
486 	MNT_ILOCK(mp);
487 	MNT_REL(mp);
488 	MNT_IUNLOCK(mp);
489 }
490 
491 /*
492  * Allocate and initialize the mount point struct.
493  */
494 struct mount *
495 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
496     struct ucred *cred)
497 {
498 	struct mount *mp;
499 
500 	mp = uma_zalloc(mount_zone, M_WAITOK);
501 	bzero(&mp->mnt_startzero,
502 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
503 	TAILQ_INIT(&mp->mnt_nvnodelist);
504 	mp->mnt_nvnodelistsize = 0;
505 	TAILQ_INIT(&mp->mnt_activevnodelist);
506 	mp->mnt_activevnodelistsize = 0;
507 	TAILQ_INIT(&mp->mnt_tmpfreevnodelist);
508 	mp->mnt_tmpfreevnodelistsize = 0;
509 	if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
510 	    mp->mnt_writeopcount != 0)
511 		panic("%s: non-zero counters on new mp %p\n", __func__, mp);
512 	if (mp->mnt_vfs_ops != 1)
513 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
514 		    mp->mnt_vfs_ops);
515 	(void) vfs_busy(mp, MBF_NOWAIT);
516 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
517 	mp->mnt_op = vfsp->vfc_vfsops;
518 	mp->mnt_vfc = vfsp;
519 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
520 	mp->mnt_gen++;
521 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
522 	mp->mnt_vnodecovered = vp;
523 	mp->mnt_cred = crdup(cred);
524 	mp->mnt_stat.f_owner = cred->cr_uid;
525 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
526 	mp->mnt_iosize_max = DFLTPHYS;
527 #ifdef MAC
528 	mac_mount_init(mp);
529 	mac_mount_create(cred, mp);
530 #endif
531 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
532 	TAILQ_INIT(&mp->mnt_uppers);
533 	return (mp);
534 }
535 
536 /*
537  * Destroy the mount struct previously allocated by vfs_mount_alloc().
538  */
539 void
540 vfs_mount_destroy(struct mount *mp)
541 {
542 
543 	if (mp->mnt_vfs_ops == 0)
544 		panic("%s: entered with zero vfs_ops\n", __func__);
545 
546 	vfs_assert_mount_counters(mp);
547 
548 	MNT_ILOCK(mp);
549 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
550 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
551 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
552 		wakeup(mp);
553 	}
554 	while (mp->mnt_ref)
555 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
556 	KASSERT(mp->mnt_ref == 0,
557 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
558 	    __FILE__, __LINE__));
559 	if (mp->mnt_writeopcount != 0)
560 		panic("vfs_mount_destroy: nonzero writeopcount");
561 	if (mp->mnt_secondary_writes != 0)
562 		panic("vfs_mount_destroy: nonzero secondary_writes");
563 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
564 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
565 		struct vnode *vp;
566 
567 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
568 			vn_printf(vp, "dangling vnode ");
569 		panic("unmount: dangling vnode");
570 	}
571 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
572 	if (mp->mnt_nvnodelistsize != 0)
573 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
574 	if (mp->mnt_activevnodelistsize != 0)
575 		panic("vfs_mount_destroy: nonzero activevnodelistsize");
576 	if (mp->mnt_tmpfreevnodelistsize != 0)
577 		panic("vfs_mount_destroy: nonzero tmpfreevnodelistsize");
578 	if (mp->mnt_lockref != 0)
579 		panic("vfs_mount_destroy: nonzero lock refcount");
580 	MNT_IUNLOCK(mp);
581 
582 	if (mp->mnt_vfs_ops != 1)
583 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
584 		    mp->mnt_vfs_ops);
585 
586 	if (mp->mnt_rootvnode != NULL)
587 		panic("%s: mount point still has a root vnode %p\n", __func__,
588 		    mp->mnt_rootvnode);
589 
590 	if (mp->mnt_vnodecovered != NULL)
591 		vrele(mp->mnt_vnodecovered);
592 #ifdef MAC
593 	mac_mount_destroy(mp);
594 #endif
595 	if (mp->mnt_opt != NULL)
596 		vfs_freeopts(mp->mnt_opt);
597 	crfree(mp->mnt_cred);
598 	uma_zfree(mount_zone, mp);
599 }
600 
601 static bool
602 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
603 {
604 	/* This is an upgrade of an exisiting mount. */
605 	if ((fsflags & MNT_UPDATE) != 0)
606 		return (false);
607 	/* This is already an R/O mount. */
608 	if ((fsflags & MNT_RDONLY) != 0)
609 		return (false);
610 
611 	switch (error) {
612 	case ENODEV:	/* generic, geom, ... */
613 	case EACCES:	/* cam/scsi, ... */
614 	case EROFS:	/* md, mmcsd, ... */
615 		/*
616 		 * These errors can be returned by the storage layer to signal
617 		 * that the media is read-only.  No harm in the R/O mount
618 		 * attempt if the error was returned for some other reason.
619 		 */
620 		return (true);
621 	default:
622 		return (false);
623 	}
624 }
625 
626 int
627 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
628 {
629 	struct vfsoptlist *optlist;
630 	struct vfsopt *opt, *tmp_opt;
631 	char *fstype, *fspath, *errmsg;
632 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
633 	bool autoro;
634 
635 	errmsg = fspath = NULL;
636 	errmsg_len = fspathlen = 0;
637 	errmsg_pos = -1;
638 	autoro = default_autoro;
639 
640 	error = vfs_buildopts(fsoptions, &optlist);
641 	if (error)
642 		return (error);
643 
644 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
645 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
646 
647 	/*
648 	 * We need these two options before the others,
649 	 * and they are mandatory for any filesystem.
650 	 * Ensure they are NUL terminated as well.
651 	 */
652 	fstypelen = 0;
653 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
654 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
655 		error = EINVAL;
656 		if (errmsg != NULL)
657 			strncpy(errmsg, "Invalid fstype", errmsg_len);
658 		goto bail;
659 	}
660 	fspathlen = 0;
661 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
662 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
663 		error = EINVAL;
664 		if (errmsg != NULL)
665 			strncpy(errmsg, "Invalid fspath", errmsg_len);
666 		goto bail;
667 	}
668 
669 	/*
670 	 * We need to see if we have the "update" option
671 	 * before we call vfs_domount(), since vfs_domount() has special
672 	 * logic based on MNT_UPDATE.  This is very important
673 	 * when we want to update the root filesystem.
674 	 */
675 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
676 		int do_freeopt = 0;
677 
678 		if (strcmp(opt->name, "update") == 0) {
679 			fsflags |= MNT_UPDATE;
680 			do_freeopt = 1;
681 		}
682 		else if (strcmp(opt->name, "async") == 0)
683 			fsflags |= MNT_ASYNC;
684 		else if (strcmp(opt->name, "force") == 0) {
685 			fsflags |= MNT_FORCE;
686 			do_freeopt = 1;
687 		}
688 		else if (strcmp(opt->name, "reload") == 0) {
689 			fsflags |= MNT_RELOAD;
690 			do_freeopt = 1;
691 		}
692 		else if (strcmp(opt->name, "multilabel") == 0)
693 			fsflags |= MNT_MULTILABEL;
694 		else if (strcmp(opt->name, "noasync") == 0)
695 			fsflags &= ~MNT_ASYNC;
696 		else if (strcmp(opt->name, "noatime") == 0)
697 			fsflags |= MNT_NOATIME;
698 		else if (strcmp(opt->name, "atime") == 0) {
699 			free(opt->name, M_MOUNT);
700 			opt->name = strdup("nonoatime", M_MOUNT);
701 		}
702 		else if (strcmp(opt->name, "noclusterr") == 0)
703 			fsflags |= MNT_NOCLUSTERR;
704 		else if (strcmp(opt->name, "clusterr") == 0) {
705 			free(opt->name, M_MOUNT);
706 			opt->name = strdup("nonoclusterr", M_MOUNT);
707 		}
708 		else if (strcmp(opt->name, "noclusterw") == 0)
709 			fsflags |= MNT_NOCLUSTERW;
710 		else if (strcmp(opt->name, "clusterw") == 0) {
711 			free(opt->name, M_MOUNT);
712 			opt->name = strdup("nonoclusterw", M_MOUNT);
713 		}
714 		else if (strcmp(opt->name, "noexec") == 0)
715 			fsflags |= MNT_NOEXEC;
716 		else if (strcmp(opt->name, "exec") == 0) {
717 			free(opt->name, M_MOUNT);
718 			opt->name = strdup("nonoexec", M_MOUNT);
719 		}
720 		else if (strcmp(opt->name, "nosuid") == 0)
721 			fsflags |= MNT_NOSUID;
722 		else if (strcmp(opt->name, "suid") == 0) {
723 			free(opt->name, M_MOUNT);
724 			opt->name = strdup("nonosuid", M_MOUNT);
725 		}
726 		else if (strcmp(opt->name, "nosymfollow") == 0)
727 			fsflags |= MNT_NOSYMFOLLOW;
728 		else if (strcmp(opt->name, "symfollow") == 0) {
729 			free(opt->name, M_MOUNT);
730 			opt->name = strdup("nonosymfollow", M_MOUNT);
731 		}
732 		else if (strcmp(opt->name, "noro") == 0) {
733 			fsflags &= ~MNT_RDONLY;
734 			autoro = false;
735 		}
736 		else if (strcmp(opt->name, "rw") == 0) {
737 			fsflags &= ~MNT_RDONLY;
738 			autoro = false;
739 		}
740 		else if (strcmp(opt->name, "ro") == 0) {
741 			fsflags |= MNT_RDONLY;
742 			autoro = false;
743 		}
744 		else if (strcmp(opt->name, "rdonly") == 0) {
745 			free(opt->name, M_MOUNT);
746 			opt->name = strdup("ro", M_MOUNT);
747 			fsflags |= MNT_RDONLY;
748 			autoro = false;
749 		}
750 		else if (strcmp(opt->name, "autoro") == 0) {
751 			do_freeopt = 1;
752 			autoro = true;
753 		}
754 		else if (strcmp(opt->name, "suiddir") == 0)
755 			fsflags |= MNT_SUIDDIR;
756 		else if (strcmp(opt->name, "sync") == 0)
757 			fsflags |= MNT_SYNCHRONOUS;
758 		else if (strcmp(opt->name, "union") == 0)
759 			fsflags |= MNT_UNION;
760 		else if (strcmp(opt->name, "automounted") == 0) {
761 			fsflags |= MNT_AUTOMOUNTED;
762 			do_freeopt = 1;
763 		} else if (strcmp(opt->name, "nocover") == 0) {
764 			fsflags |= MNT_NOCOVER;
765 			do_freeopt = 1;
766 		} else if (strcmp(opt->name, "cover") == 0) {
767 			fsflags &= ~MNT_NOCOVER;
768 			do_freeopt = 1;
769 		} else if (strcmp(opt->name, "emptydir") == 0) {
770 			fsflags |= MNT_EMPTYDIR;
771 			do_freeopt = 1;
772 		} else if (strcmp(opt->name, "noemptydir") == 0) {
773 			fsflags &= ~MNT_EMPTYDIR;
774 			do_freeopt = 1;
775 		}
776 		if (do_freeopt)
777 			vfs_freeopt(optlist, opt);
778 	}
779 
780 	/*
781 	 * Be ultra-paranoid about making sure the type and fspath
782 	 * variables will fit in our mp buffers, including the
783 	 * terminating NUL.
784 	 */
785 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
786 		error = ENAMETOOLONG;
787 		goto bail;
788 	}
789 
790 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
791 
792 	/*
793 	 * See if we can mount in the read-only mode if the error code suggests
794 	 * that it could be possible and the mount options allow for that.
795 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
796 	 * overridden by "autoro".
797 	 */
798 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
799 		printf("%s: R/W mount failed, possibly R/O media,"
800 		    " trying R/O mount\n", __func__);
801 		fsflags |= MNT_RDONLY;
802 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
803 	}
804 bail:
805 	/* copyout the errmsg */
806 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
807 	    && errmsg_len > 0 && errmsg != NULL) {
808 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
809 			bcopy(errmsg,
810 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
811 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
812 		} else {
813 			copyout(errmsg,
814 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
815 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
816 		}
817 	}
818 
819 	if (optlist != NULL)
820 		vfs_freeopts(optlist);
821 	return (error);
822 }
823 
824 /*
825  * Old mount API.
826  */
827 #ifndef _SYS_SYSPROTO_H_
828 struct mount_args {
829 	char	*type;
830 	char	*path;
831 	int	flags;
832 	caddr_t	data;
833 };
834 #endif
835 /* ARGSUSED */
836 int
837 sys_mount(struct thread *td, struct mount_args *uap)
838 {
839 	char *fstype;
840 	struct vfsconf *vfsp = NULL;
841 	struct mntarg *ma = NULL;
842 	uint64_t flags;
843 	int error;
844 
845 	/*
846 	 * Mount flags are now 64-bits. On 32-bit architectures only
847 	 * 32-bits are passed in, but from here on everything handles
848 	 * 64-bit flags correctly.
849 	 */
850 	flags = uap->flags;
851 
852 	AUDIT_ARG_FFLAGS(flags);
853 
854 	/*
855 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
856 	 * userspace to set this flag, but we must filter it out if we want
857 	 * MNT_UPDATE on the root file system to work.
858 	 * MNT_ROOTFS should only be set by the kernel when mounting its
859 	 * root file system.
860 	 */
861 	flags &= ~MNT_ROOTFS;
862 
863 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
864 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
865 	if (error) {
866 		free(fstype, M_TEMP);
867 		return (error);
868 	}
869 
870 	AUDIT_ARG_TEXT(fstype);
871 	vfsp = vfs_byname_kld(fstype, td, &error);
872 	free(fstype, M_TEMP);
873 	if (vfsp == NULL)
874 		return (ENOENT);
875 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
876 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
877 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
878 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
879 		return (EOPNOTSUPP);
880 
881 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
882 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
883 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
884 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
885 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
886 
887 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
888 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
889 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
890 }
891 
892 /*
893  * vfs_domount_first(): first file system mount (not update)
894  */
895 static int
896 vfs_domount_first(
897 	struct thread *td,		/* Calling thread. */
898 	struct vfsconf *vfsp,		/* File system type. */
899 	char *fspath,			/* Mount path. */
900 	struct vnode *vp,		/* Vnode to be covered. */
901 	uint64_t fsflags,		/* Flags common to all filesystems. */
902 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
903 	)
904 {
905 	struct vattr va;
906 	struct mount *mp;
907 	struct vnode *newdp;
908 	int error, error1;
909 
910 	ASSERT_VOP_ELOCKED(vp, __func__);
911 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
912 
913 	if ((fsflags & MNT_EMPTYDIR) != 0) {
914 		error = vfs_emptydir(vp);
915 		if (error != 0) {
916 			vput(vp);
917 			return (error);
918 		}
919 	}
920 
921 	/*
922 	 * If the jail of the calling thread lacks permission for this type of
923 	 * file system, deny immediately.
924 	 */
925 	if (jailed(td->td_ucred) && !prison_allow(td->td_ucred,
926 	    vfsp->vfc_prison_flag)) {
927 		vput(vp);
928 		return (EPERM);
929 	}
930 
931 	/*
932 	 * If the user is not root, ensure that they own the directory
933 	 * onto which we are attempting to mount.
934 	 */
935 	error = VOP_GETATTR(vp, &va, td->td_ucred);
936 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
937 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
938 	if (error == 0)
939 		error = vinvalbuf(vp, V_SAVE, 0, 0);
940 	if (error == 0 && vp->v_type != VDIR)
941 		error = ENOTDIR;
942 	if (error == 0) {
943 		VI_LOCK(vp);
944 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
945 			vp->v_iflag |= VI_MOUNT;
946 		else
947 			error = EBUSY;
948 		VI_UNLOCK(vp);
949 	}
950 	if (error != 0) {
951 		vput(vp);
952 		return (error);
953 	}
954 	VOP_UNLOCK(vp, 0);
955 
956 	/* Allocate and initialize the filesystem. */
957 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
958 	/* XXXMAC: pass to vfs_mount_alloc? */
959 	mp->mnt_optnew = *optlist;
960 	/* Set the mount level flags. */
961 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
962 
963 	/*
964 	 * Mount the filesystem.
965 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
966 	 * get.  No freeing of cn_pnbuf.
967 	 */
968 	error1 = 0;
969 	if ((error = VFS_MOUNT(mp)) != 0 ||
970 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
971 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
972 		if (error1 != 0) {
973 			error = error1;
974 			if ((error1 = VFS_UNMOUNT(mp, 0)) != 0)
975 				printf("VFS_UNMOUNT returned %d\n", error1);
976 		}
977 		vfs_unbusy(mp);
978 		mp->mnt_vnodecovered = NULL;
979 		vfs_mount_destroy(mp);
980 		VI_LOCK(vp);
981 		vp->v_iflag &= ~VI_MOUNT;
982 		VI_UNLOCK(vp);
983 		vrele(vp);
984 		return (error);
985 	}
986 	VOP_UNLOCK(newdp, 0);
987 
988 	if (mp->mnt_opt != NULL)
989 		vfs_freeopts(mp->mnt_opt);
990 	mp->mnt_opt = mp->mnt_optnew;
991 	*optlist = NULL;
992 
993 	/*
994 	 * Prevent external consumers of mount options from reading mnt_optnew.
995 	 */
996 	mp->mnt_optnew = NULL;
997 
998 	MNT_ILOCK(mp);
999 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1000 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1001 		mp->mnt_kern_flag |= MNTK_ASYNC;
1002 	else
1003 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1004 	MNT_IUNLOCK(mp);
1005 
1006 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1007 	cache_purge(vp);
1008 	VI_LOCK(vp);
1009 	vp->v_iflag &= ~VI_MOUNT;
1010 	VI_UNLOCK(vp);
1011 	vp->v_mountedhere = mp;
1012 	/* Place the new filesystem at the end of the mount list. */
1013 	mtx_lock(&mountlist_mtx);
1014 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1015 	mtx_unlock(&mountlist_mtx);
1016 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1017 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1018 	VOP_UNLOCK(vp, 0);
1019 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1020 	VOP_UNLOCK(newdp, 0);
1021 	mountcheckdirs(vp, newdp);
1022 	vrele(newdp);
1023 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1024 		vfs_allocate_syncvnode(mp);
1025 	vfs_op_exit(mp);
1026 	vfs_unbusy(mp);
1027 	return (0);
1028 }
1029 
1030 /*
1031  * vfs_domount_update(): update of mounted file system
1032  */
1033 static int
1034 vfs_domount_update(
1035 	struct thread *td,		/* Calling thread. */
1036 	struct vnode *vp,		/* Mount point vnode. */
1037 	uint64_t fsflags,		/* Flags common to all filesystems. */
1038 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1039 	)
1040 {
1041 	struct export_args export;
1042 	struct vnode *rootvp;
1043 	void *bufp;
1044 	struct mount *mp;
1045 	int error, export_error, len;
1046 	uint64_t flag;
1047 
1048 	ASSERT_VOP_ELOCKED(vp, __func__);
1049 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1050 	mp = vp->v_mount;
1051 
1052 	if ((vp->v_vflag & VV_ROOT) == 0) {
1053 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1054 		    == 0)
1055 			error = EXDEV;
1056 		else
1057 			error = EINVAL;
1058 		vput(vp);
1059 		return (error);
1060 	}
1061 
1062 	/*
1063 	 * We only allow the filesystem to be reloaded if it
1064 	 * is currently mounted read-only.
1065 	 */
1066 	flag = mp->mnt_flag;
1067 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1068 		vput(vp);
1069 		return (EOPNOTSUPP);	/* Needs translation */
1070 	}
1071 	/*
1072 	 * Only privileged root, or (if MNT_USER is set) the user that
1073 	 * did the original mount is permitted to update it.
1074 	 */
1075 	error = vfs_suser(mp, td);
1076 	if (error != 0) {
1077 		vput(vp);
1078 		return (error);
1079 	}
1080 	if (vfs_busy(mp, MBF_NOWAIT)) {
1081 		vput(vp);
1082 		return (EBUSY);
1083 	}
1084 	VI_LOCK(vp);
1085 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1086 		VI_UNLOCK(vp);
1087 		vfs_unbusy(mp);
1088 		vput(vp);
1089 		return (EBUSY);
1090 	}
1091 	vp->v_iflag |= VI_MOUNT;
1092 	VI_UNLOCK(vp);
1093 	VOP_UNLOCK(vp, 0);
1094 
1095 	vfs_op_enter(mp);
1096 
1097 	MNT_ILOCK(mp);
1098 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1099 		MNT_IUNLOCK(mp);
1100 		error = EBUSY;
1101 		goto end;
1102 	}
1103 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1104 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1105 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1106 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1107 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1108 	rootvp = vfs_cache_root_clear(mp);
1109 	MNT_IUNLOCK(mp);
1110 	if (rootvp != NULL)
1111 		vrele(rootvp);
1112 	mp->mnt_optnew = *optlist;
1113 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1114 
1115 	/*
1116 	 * Mount the filesystem.
1117 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1118 	 * get.  No freeing of cn_pnbuf.
1119 	 */
1120 	error = VFS_MOUNT(mp);
1121 
1122 	export_error = 0;
1123 	/* Process the export option. */
1124 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1125 	    &len) == 0) {
1126 		/* Assume that there is only 1 ABI for each length. */
1127 		switch (len) {
1128 		case (sizeof(struct oexport_args)):
1129 			bzero(&export, sizeof(export));
1130 			/* FALLTHROUGH */
1131 		case (sizeof(export)):
1132 			bcopy(bufp, &export, len);
1133 			export_error = vfs_export(mp, &export);
1134 			break;
1135 		default:
1136 			export_error = EINVAL;
1137 			break;
1138 		}
1139 	}
1140 
1141 	MNT_ILOCK(mp);
1142 	if (error == 0) {
1143 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1144 		    MNT_SNAPSHOT);
1145 	} else {
1146 		/*
1147 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1148 		 * because it is not part of MNT_UPDATEMASK, but it could have
1149 		 * changed in the meantime if quotactl(2) was called.
1150 		 * All in all we want current value of MNT_QUOTA, not the old
1151 		 * one.
1152 		 */
1153 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1154 	}
1155 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1156 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1157 		mp->mnt_kern_flag |= MNTK_ASYNC;
1158 	else
1159 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1160 	MNT_IUNLOCK(mp);
1161 
1162 	if (error != 0)
1163 		goto end;
1164 
1165 	if (mp->mnt_opt != NULL)
1166 		vfs_freeopts(mp->mnt_opt);
1167 	mp->mnt_opt = mp->mnt_optnew;
1168 	*optlist = NULL;
1169 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1170 	/*
1171 	 * Prevent external consumers of mount options from reading
1172 	 * mnt_optnew.
1173 	 */
1174 	mp->mnt_optnew = NULL;
1175 
1176 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1177 		vfs_allocate_syncvnode(mp);
1178 	else
1179 		vfs_deallocate_syncvnode(mp);
1180 end:
1181 	vfs_op_exit(mp);
1182 	vfs_unbusy(mp);
1183 	VI_LOCK(vp);
1184 	vp->v_iflag &= ~VI_MOUNT;
1185 	VI_UNLOCK(vp);
1186 	vrele(vp);
1187 	return (error != 0 ? error : export_error);
1188 }
1189 
1190 /*
1191  * vfs_domount(): actually attempt a filesystem mount.
1192  */
1193 static int
1194 vfs_domount(
1195 	struct thread *td,		/* Calling thread. */
1196 	const char *fstype,		/* Filesystem type. */
1197 	char *fspath,			/* Mount path. */
1198 	uint64_t fsflags,		/* Flags common to all filesystems. */
1199 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1200 	)
1201 {
1202 	struct vfsconf *vfsp;
1203 	struct nameidata nd;
1204 	struct vnode *vp;
1205 	char *pathbuf;
1206 	int error;
1207 
1208 	/*
1209 	 * Be ultra-paranoid about making sure the type and fspath
1210 	 * variables will fit in our mp buffers, including the
1211 	 * terminating NUL.
1212 	 */
1213 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1214 		return (ENAMETOOLONG);
1215 
1216 	if (jailed(td->td_ucred) || usermount == 0) {
1217 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1218 			return (error);
1219 	}
1220 
1221 	/*
1222 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1223 	 */
1224 	if (fsflags & MNT_EXPORTED) {
1225 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1226 		if (error)
1227 			return (error);
1228 	}
1229 	if (fsflags & MNT_SUIDDIR) {
1230 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1231 		if (error)
1232 			return (error);
1233 	}
1234 	/*
1235 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1236 	 */
1237 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1238 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1239 			fsflags |= MNT_NOSUID | MNT_USER;
1240 	}
1241 
1242 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1243 	vfsp = NULL;
1244 	if ((fsflags & MNT_UPDATE) == 0) {
1245 		/* Don't try to load KLDs if we're mounting the root. */
1246 		if (fsflags & MNT_ROOTFS)
1247 			vfsp = vfs_byname(fstype);
1248 		else
1249 			vfsp = vfs_byname_kld(fstype, td, &error);
1250 		if (vfsp == NULL)
1251 			return (ENODEV);
1252 	}
1253 
1254 	/*
1255 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1256 	 */
1257 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1258 	    UIO_SYSSPACE, fspath, td);
1259 	error = namei(&nd);
1260 	if (error != 0)
1261 		return (error);
1262 	NDFREE(&nd, NDF_ONLY_PNBUF);
1263 	vp = nd.ni_vp;
1264 	if ((fsflags & MNT_UPDATE) == 0) {
1265 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1266 		    (fsflags & MNT_NOCOVER) != 0) {
1267 			vput(vp);
1268 			return (EBUSY);
1269 		}
1270 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1271 		strcpy(pathbuf, fspath);
1272 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1273 		/* debug.disablefullpath == 1 results in ENODEV */
1274 		if (error == 0 || error == ENODEV) {
1275 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1276 			    fsflags, optlist);
1277 		}
1278 		free(pathbuf, M_TEMP);
1279 	} else
1280 		error = vfs_domount_update(td, vp, fsflags, optlist);
1281 
1282 	return (error);
1283 }
1284 
1285 /*
1286  * Unmount a filesystem.
1287  *
1288  * Note: unmount takes a path to the vnode mounted on as argument, not
1289  * special file (as before).
1290  */
1291 #ifndef _SYS_SYSPROTO_H_
1292 struct unmount_args {
1293 	char	*path;
1294 	int	flags;
1295 };
1296 #endif
1297 /* ARGSUSED */
1298 int
1299 sys_unmount(struct thread *td, struct unmount_args *uap)
1300 {
1301 	struct nameidata nd;
1302 	struct mount *mp;
1303 	char *pathbuf;
1304 	int error, id0, id1;
1305 
1306 	AUDIT_ARG_VALUE(uap->flags);
1307 	if (jailed(td->td_ucred) || usermount == 0) {
1308 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1309 		if (error)
1310 			return (error);
1311 	}
1312 
1313 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1314 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1315 	if (error) {
1316 		free(pathbuf, M_TEMP);
1317 		return (error);
1318 	}
1319 	if (uap->flags & MNT_BYFSID) {
1320 		AUDIT_ARG_TEXT(pathbuf);
1321 		/* Decode the filesystem ID. */
1322 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1323 			free(pathbuf, M_TEMP);
1324 			return (EINVAL);
1325 		}
1326 
1327 		mtx_lock(&mountlist_mtx);
1328 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1329 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1330 			    mp->mnt_stat.f_fsid.val[1] == id1) {
1331 				vfs_ref(mp);
1332 				break;
1333 			}
1334 		}
1335 		mtx_unlock(&mountlist_mtx);
1336 	} else {
1337 		/*
1338 		 * Try to find global path for path argument.
1339 		 */
1340 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1341 		    UIO_SYSSPACE, pathbuf, td);
1342 		if (namei(&nd) == 0) {
1343 			NDFREE(&nd, NDF_ONLY_PNBUF);
1344 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1345 			    MNAMELEN);
1346 			if (error == 0 || error == ENODEV)
1347 				vput(nd.ni_vp);
1348 		}
1349 		mtx_lock(&mountlist_mtx);
1350 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1351 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1352 				vfs_ref(mp);
1353 				break;
1354 			}
1355 		}
1356 		mtx_unlock(&mountlist_mtx);
1357 	}
1358 	free(pathbuf, M_TEMP);
1359 	if (mp == NULL) {
1360 		/*
1361 		 * Previously we returned ENOENT for a nonexistent path and
1362 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1363 		 * now, so in the !MNT_BYFSID case return the more likely
1364 		 * EINVAL for compatibility.
1365 		 */
1366 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1367 	}
1368 
1369 	/*
1370 	 * Don't allow unmounting the root filesystem.
1371 	 */
1372 	if (mp->mnt_flag & MNT_ROOTFS) {
1373 		vfs_rel(mp);
1374 		return (EINVAL);
1375 	}
1376 	error = dounmount(mp, uap->flags, td);
1377 	return (error);
1378 }
1379 
1380 /*
1381  * Return error if any of the vnodes, ignoring the root vnode
1382  * and the syncer vnode, have non-zero usecount.
1383  *
1384  * This function is purely advisory - it can return false positives
1385  * and negatives.
1386  */
1387 static int
1388 vfs_check_usecounts(struct mount *mp)
1389 {
1390 	struct vnode *vp, *mvp;
1391 
1392 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1393 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1394 		    vp->v_usecount != 0) {
1395 			VI_UNLOCK(vp);
1396 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1397 			return (EBUSY);
1398 		}
1399 		VI_UNLOCK(vp);
1400 	}
1401 
1402 	return (0);
1403 }
1404 
1405 static void
1406 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1407 {
1408 
1409 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1410 	mp->mnt_kern_flag &= ~mntkflags;
1411 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1412 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1413 		wakeup(mp);
1414 	}
1415 	vfs_op_exit_locked(mp);
1416 	MNT_IUNLOCK(mp);
1417 	if (coveredvp != NULL) {
1418 		VOP_UNLOCK(coveredvp, 0);
1419 		vdrop(coveredvp);
1420 	}
1421 	vn_finished_write(mp);
1422 }
1423 
1424 /*
1425  * There are various reference counters associated with the mount point.
1426  * Normally it is permitted to modify them without taking the mnt ilock,
1427  * but this behavior can be temporarily disabled if stable value is needed
1428  * or callers are expected to block (e.g. to not allow new users during
1429  * forced unmount).
1430  */
1431 void
1432 vfs_op_enter(struct mount *mp)
1433 {
1434 	int cpu;
1435 
1436 	MNT_ILOCK(mp);
1437 	mp->mnt_vfs_ops++;
1438 	if (mp->mnt_vfs_ops > 1) {
1439 		MNT_IUNLOCK(mp);
1440 		return;
1441 	}
1442 	/*
1443 	 * Paired with a fence in vfs_op_thread_enter(). See the comment
1444 	 * above it for details.
1445 	 */
1446 	atomic_thread_fence_seq_cst();
1447 	vfs_op_barrier_wait(mp);
1448 	/*
1449 	 * Paired with a fence in vfs_op_thread_exit().
1450 	 */
1451 	atomic_thread_fence_acq();
1452 	CPU_FOREACH(cpu) {
1453 		mp->mnt_ref +=
1454 		    zpcpu_replace_cpu(mp->mnt_ref_pcpu, 0, cpu);
1455 		mp->mnt_lockref +=
1456 		    zpcpu_replace_cpu(mp->mnt_lockref_pcpu, 0, cpu);
1457 		mp->mnt_writeopcount +=
1458 		    zpcpu_replace_cpu(mp->mnt_writeopcount_pcpu, 0, cpu);
1459 	}
1460 	MNT_IUNLOCK(mp);
1461 	vfs_assert_mount_counters(mp);
1462 }
1463 
1464 void
1465 vfs_op_exit_locked(struct mount *mp)
1466 {
1467 
1468 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1469 
1470 	if (mp->mnt_vfs_ops <= 0)
1471 		panic("%s: invalid vfs_ops count %d for mp %p\n",
1472 		    __func__, mp->mnt_vfs_ops, mp);
1473 	mp->mnt_vfs_ops--;
1474 }
1475 
1476 void
1477 vfs_op_exit(struct mount *mp)
1478 {
1479 
1480 	MNT_ILOCK(mp);
1481 	vfs_op_exit_locked(mp);
1482 	MNT_IUNLOCK(mp);
1483 }
1484 
1485 /*
1486  * It is assumed the caller already posted at least an acquire barrier.
1487  */
1488 void
1489 vfs_op_barrier_wait(struct mount *mp)
1490 {
1491 	int *in_op;
1492 	int cpu;
1493 
1494 	CPU_FOREACH(cpu) {
1495 		in_op = zpcpu_get_cpu(mp->mnt_thread_in_ops_pcpu, cpu);
1496 		while (atomic_load_int(in_op))
1497 			cpu_spinwait();
1498 	}
1499 }
1500 
1501 #ifdef DIAGNOSTIC
1502 void
1503 vfs_assert_mount_counters(struct mount *mp)
1504 {
1505 	int cpu;
1506 
1507 	if (mp->mnt_vfs_ops == 0)
1508 		return;
1509 
1510 	CPU_FOREACH(cpu) {
1511 		if (*(int *)zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu) != 0 ||
1512 		    *(int *)zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu) != 0 ||
1513 		    *(int *)zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu) != 0)
1514 			vfs_dump_mount_counters(mp);
1515 	}
1516 }
1517 
1518 void
1519 vfs_dump_mount_counters(struct mount *mp)
1520 {
1521 	int cpu, *count;
1522 	int ref, lockref, writeopcount;
1523 
1524 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1525 
1526 	printf("        ref : ");
1527 	ref = mp->mnt_ref;
1528 	CPU_FOREACH(cpu) {
1529 		count = zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu);
1530 		printf("%d ", *count);
1531 		ref += *count;
1532 	}
1533 	printf("\n");
1534 	printf("    lockref : ");
1535 	lockref = mp->mnt_lockref;
1536 	CPU_FOREACH(cpu) {
1537 		count = zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu);
1538 		printf("%d ", *count);
1539 		lockref += *count;
1540 	}
1541 	printf("\n");
1542 	printf("writeopcount: ");
1543 	writeopcount = mp->mnt_writeopcount;
1544 	CPU_FOREACH(cpu) {
1545 		count = zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu);
1546 		printf("%d ", *count);
1547 		writeopcount += *count;
1548 	}
1549 	printf("\n");
1550 
1551 	printf("counter       struct total\n");
1552 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1553 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1554 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1555 
1556 	panic("invalid counts on struct mount");
1557 }
1558 #endif
1559 
1560 int
1561 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1562 {
1563 	int *base, *pcpu;
1564 	int cpu, sum;
1565 
1566 	switch (which) {
1567 	case MNT_COUNT_REF:
1568 		base = &mp->mnt_ref;
1569 		pcpu = mp->mnt_ref_pcpu;
1570 		break;
1571 	case MNT_COUNT_LOCKREF:
1572 		base = &mp->mnt_lockref;
1573 		pcpu = mp->mnt_lockref_pcpu;
1574 		break;
1575 	case MNT_COUNT_WRITEOPCOUNT:
1576 		base = &mp->mnt_writeopcount;
1577 		pcpu = mp->mnt_writeopcount_pcpu;
1578 		break;
1579 	}
1580 
1581 	sum = *base;
1582 	CPU_FOREACH(cpu) {
1583 		sum += *(int *)zpcpu_get_cpu(pcpu, cpu);
1584 	}
1585 	return (sum);
1586 }
1587 
1588 /*
1589  * Do the actual filesystem unmount.
1590  */
1591 int
1592 dounmount(struct mount *mp, int flags, struct thread *td)
1593 {
1594 	struct vnode *coveredvp, *rootvp;
1595 	int error;
1596 	uint64_t async_flag;
1597 	int mnt_gen_r;
1598 
1599 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1600 		mnt_gen_r = mp->mnt_gen;
1601 		VI_LOCK(coveredvp);
1602 		vholdl(coveredvp);
1603 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1604 		/*
1605 		 * Check for mp being unmounted while waiting for the
1606 		 * covered vnode lock.
1607 		 */
1608 		if (coveredvp->v_mountedhere != mp ||
1609 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1610 			VOP_UNLOCK(coveredvp, 0);
1611 			vdrop(coveredvp);
1612 			vfs_rel(mp);
1613 			return (EBUSY);
1614 		}
1615 	}
1616 
1617 	/*
1618 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1619 	 * original mount is permitted to unmount this filesystem.
1620 	 */
1621 	error = vfs_suser(mp, td);
1622 	if (error != 0) {
1623 		if (coveredvp != NULL) {
1624 			VOP_UNLOCK(coveredvp, 0);
1625 			vdrop(coveredvp);
1626 		}
1627 		vfs_rel(mp);
1628 		return (error);
1629 	}
1630 
1631 	vfs_op_enter(mp);
1632 
1633 	vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1634 	MNT_ILOCK(mp);
1635 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1636 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
1637 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
1638 		dounmount_cleanup(mp, coveredvp, 0);
1639 		return (EBUSY);
1640 	}
1641 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1642 	rootvp = vfs_cache_root_clear(mp);
1643 	if (flags & MNT_NONBUSY) {
1644 		MNT_IUNLOCK(mp);
1645 		error = vfs_check_usecounts(mp);
1646 		MNT_ILOCK(mp);
1647 		if (error != 0) {
1648 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1649 			if (rootvp != NULL)
1650 				vrele(rootvp);
1651 			return (error);
1652 		}
1653 	}
1654 	/* Allow filesystems to detect that a forced unmount is in progress. */
1655 	if (flags & MNT_FORCE) {
1656 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1657 		MNT_IUNLOCK(mp);
1658 		/*
1659 		 * Must be done after setting MNTK_UNMOUNTF and before
1660 		 * waiting for mnt_lockref to become 0.
1661 		 */
1662 		VFS_PURGE(mp);
1663 		MNT_ILOCK(mp);
1664 	}
1665 	error = 0;
1666 	if (mp->mnt_lockref) {
1667 		mp->mnt_kern_flag |= MNTK_DRAINING;
1668 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1669 		    "mount drain", 0);
1670 	}
1671 	MNT_IUNLOCK(mp);
1672 	KASSERT(mp->mnt_lockref == 0,
1673 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1674 	    __func__, __FILE__, __LINE__));
1675 	KASSERT(error == 0,
1676 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1677 	    __func__, __FILE__, __LINE__));
1678 
1679 	if (rootvp != NULL)
1680 		vrele(rootvp);
1681 
1682 	if (mp->mnt_flag & MNT_EXPUBLIC)
1683 		vfs_setpublicfs(NULL, NULL, NULL);
1684 
1685 	/*
1686 	 * From now, we can claim that the use reference on the
1687 	 * coveredvp is ours, and the ref can be released only by
1688 	 * successfull unmount by us, or left for later unmount
1689 	 * attempt.  The previously acquired hold reference is no
1690 	 * longer needed to protect the vnode from reuse.
1691 	 */
1692 	if (coveredvp != NULL)
1693 		vdrop(coveredvp);
1694 
1695 	vfs_msync(mp, MNT_WAIT);
1696 	MNT_ILOCK(mp);
1697 	async_flag = mp->mnt_flag & MNT_ASYNC;
1698 	mp->mnt_flag &= ~MNT_ASYNC;
1699 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1700 	MNT_IUNLOCK(mp);
1701 	cache_purgevfs(mp, false); /* remove cache entries for this file sys */
1702 	vfs_deallocate_syncvnode(mp);
1703 	if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 ||
1704 	    (error = VFS_SYNC(mp, MNT_WAIT)) == 0)
1705 		error = VFS_UNMOUNT(mp, flags);
1706 	vn_finished_write(mp);
1707 	/*
1708 	 * If we failed to flush the dirty blocks for this mount point,
1709 	 * undo all the cdir/rdir and rootvnode changes we made above.
1710 	 * Unless we failed to do so because the device is reporting that
1711 	 * it doesn't exist anymore.
1712 	 */
1713 	if (error && error != ENXIO) {
1714 		MNT_ILOCK(mp);
1715 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1716 			MNT_IUNLOCK(mp);
1717 			vfs_allocate_syncvnode(mp);
1718 			MNT_ILOCK(mp);
1719 		}
1720 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1721 		mp->mnt_flag |= async_flag;
1722 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1723 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1724 			mp->mnt_kern_flag |= MNTK_ASYNC;
1725 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1726 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1727 			wakeup(mp);
1728 		}
1729 		vfs_op_exit_locked(mp);
1730 		MNT_IUNLOCK(mp);
1731 		if (coveredvp)
1732 			VOP_UNLOCK(coveredvp, 0);
1733 		return (error);
1734 	}
1735 	mtx_lock(&mountlist_mtx);
1736 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1737 	mtx_unlock(&mountlist_mtx);
1738 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1739 	if (coveredvp != NULL) {
1740 		coveredvp->v_mountedhere = NULL;
1741 		VOP_UNLOCK(coveredvp, 0);
1742 	}
1743 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1744 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
1745 		vrele(rootvnode);
1746 		rootvnode = NULL;
1747 	}
1748 	if (mp == rootdevmp)
1749 		rootdevmp = NULL;
1750 	vfs_mount_destroy(mp);
1751 	return (0);
1752 }
1753 
1754 /*
1755  * Report errors during filesystem mounting.
1756  */
1757 void
1758 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1759 {
1760 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1761 	va_list ap;
1762 	int error, len;
1763 	char *errmsg;
1764 
1765 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1766 	if (error || errmsg == NULL || len <= 0)
1767 		return;
1768 
1769 	va_start(ap, fmt);
1770 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1771 	va_end(ap);
1772 }
1773 
1774 void
1775 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1776 {
1777 	va_list ap;
1778 	int error, len;
1779 	char *errmsg;
1780 
1781 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1782 	if (error || errmsg == NULL || len <= 0)
1783 		return;
1784 
1785 	va_start(ap, fmt);
1786 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1787 	va_end(ap);
1788 }
1789 
1790 /*
1791  * ---------------------------------------------------------------------
1792  * Functions for querying mount options/arguments from filesystems.
1793  */
1794 
1795 /*
1796  * Check that no unknown options are given
1797  */
1798 int
1799 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1800 {
1801 	struct vfsopt *opt;
1802 	char errmsg[255];
1803 	const char **t, *p, *q;
1804 	int ret = 0;
1805 
1806 	TAILQ_FOREACH(opt, opts, link) {
1807 		p = opt->name;
1808 		q = NULL;
1809 		if (p[0] == 'n' && p[1] == 'o')
1810 			q = p + 2;
1811 		for(t = global_opts; *t != NULL; t++) {
1812 			if (strcmp(*t, p) == 0)
1813 				break;
1814 			if (q != NULL) {
1815 				if (strcmp(*t, q) == 0)
1816 					break;
1817 			}
1818 		}
1819 		if (*t != NULL)
1820 			continue;
1821 		for(t = legal; *t != NULL; t++) {
1822 			if (strcmp(*t, p) == 0)
1823 				break;
1824 			if (q != NULL) {
1825 				if (strcmp(*t, q) == 0)
1826 					break;
1827 			}
1828 		}
1829 		if (*t != NULL)
1830 			continue;
1831 		snprintf(errmsg, sizeof(errmsg),
1832 		    "mount option <%s> is unknown", p);
1833 		ret = EINVAL;
1834 	}
1835 	if (ret != 0) {
1836 		TAILQ_FOREACH(opt, opts, link) {
1837 			if (strcmp(opt->name, "errmsg") == 0) {
1838 				strncpy((char *)opt->value, errmsg, opt->len);
1839 				break;
1840 			}
1841 		}
1842 		if (opt == NULL)
1843 			printf("%s\n", errmsg);
1844 	}
1845 	return (ret);
1846 }
1847 
1848 /*
1849  * Get a mount option by its name.
1850  *
1851  * Return 0 if the option was found, ENOENT otherwise.
1852  * If len is non-NULL it will be filled with the length
1853  * of the option. If buf is non-NULL, it will be filled
1854  * with the address of the option.
1855  */
1856 int
1857 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
1858 {
1859 	struct vfsopt *opt;
1860 
1861 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1862 
1863 	TAILQ_FOREACH(opt, opts, link) {
1864 		if (strcmp(name, opt->name) == 0) {
1865 			opt->seen = 1;
1866 			if (len != NULL)
1867 				*len = opt->len;
1868 			if (buf != NULL)
1869 				*buf = opt->value;
1870 			return (0);
1871 		}
1872 	}
1873 	return (ENOENT);
1874 }
1875 
1876 int
1877 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1878 {
1879 	struct vfsopt *opt;
1880 
1881 	if (opts == NULL)
1882 		return (-1);
1883 
1884 	TAILQ_FOREACH(opt, opts, link) {
1885 		if (strcmp(name, opt->name) == 0) {
1886 			opt->seen = 1;
1887 			return (opt->pos);
1888 		}
1889 	}
1890 	return (-1);
1891 }
1892 
1893 int
1894 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1895 {
1896 	char *opt_value, *vtp;
1897 	quad_t iv;
1898 	int error, opt_len;
1899 
1900 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1901 	if (error != 0)
1902 		return (error);
1903 	if (opt_len == 0 || opt_value == NULL)
1904 		return (EINVAL);
1905 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1906 		return (EINVAL);
1907 	iv = strtoq(opt_value, &vtp, 0);
1908 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1909 		return (EINVAL);
1910 	if (iv < 0)
1911 		return (EINVAL);
1912 	switch (vtp[0]) {
1913 	case 't': case 'T':
1914 		iv *= 1024;
1915 		/* FALLTHROUGH */
1916 	case 'g': case 'G':
1917 		iv *= 1024;
1918 		/* FALLTHROUGH */
1919 	case 'm': case 'M':
1920 		iv *= 1024;
1921 		/* FALLTHROUGH */
1922 	case 'k': case 'K':
1923 		iv *= 1024;
1924 	case '\0':
1925 		break;
1926 	default:
1927 		return (EINVAL);
1928 	}
1929 	*value = iv;
1930 
1931 	return (0);
1932 }
1933 
1934 char *
1935 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1936 {
1937 	struct vfsopt *opt;
1938 
1939 	*error = 0;
1940 	TAILQ_FOREACH(opt, opts, link) {
1941 		if (strcmp(name, opt->name) != 0)
1942 			continue;
1943 		opt->seen = 1;
1944 		if (opt->len == 0 ||
1945 		    ((char *)opt->value)[opt->len - 1] != '\0') {
1946 			*error = EINVAL;
1947 			return (NULL);
1948 		}
1949 		return (opt->value);
1950 	}
1951 	*error = ENOENT;
1952 	return (NULL);
1953 }
1954 
1955 int
1956 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1957 	uint64_t val)
1958 {
1959 	struct vfsopt *opt;
1960 
1961 	TAILQ_FOREACH(opt, opts, link) {
1962 		if (strcmp(name, opt->name) == 0) {
1963 			opt->seen = 1;
1964 			if (w != NULL)
1965 				*w |= val;
1966 			return (1);
1967 		}
1968 	}
1969 	if (w != NULL)
1970 		*w &= ~val;
1971 	return (0);
1972 }
1973 
1974 int
1975 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1976 {
1977 	va_list ap;
1978 	struct vfsopt *opt;
1979 	int ret;
1980 
1981 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1982 
1983 	TAILQ_FOREACH(opt, opts, link) {
1984 		if (strcmp(name, opt->name) != 0)
1985 			continue;
1986 		opt->seen = 1;
1987 		if (opt->len == 0 || opt->value == NULL)
1988 			return (0);
1989 		if (((char *)opt->value)[opt->len - 1] != '\0')
1990 			return (0);
1991 		va_start(ap, fmt);
1992 		ret = vsscanf(opt->value, fmt, ap);
1993 		va_end(ap);
1994 		return (ret);
1995 	}
1996 	return (0);
1997 }
1998 
1999 int
2000 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2001 {
2002 	struct vfsopt *opt;
2003 
2004 	TAILQ_FOREACH(opt, opts, link) {
2005 		if (strcmp(name, opt->name) != 0)
2006 			continue;
2007 		opt->seen = 1;
2008 		if (opt->value == NULL)
2009 			opt->len = len;
2010 		else {
2011 			if (opt->len != len)
2012 				return (EINVAL);
2013 			bcopy(value, opt->value, len);
2014 		}
2015 		return (0);
2016 	}
2017 	return (ENOENT);
2018 }
2019 
2020 int
2021 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2022 {
2023 	struct vfsopt *opt;
2024 
2025 	TAILQ_FOREACH(opt, opts, link) {
2026 		if (strcmp(name, opt->name) != 0)
2027 			continue;
2028 		opt->seen = 1;
2029 		if (opt->value == NULL)
2030 			opt->len = len;
2031 		else {
2032 			if (opt->len < len)
2033 				return (EINVAL);
2034 			opt->len = len;
2035 			bcopy(value, opt->value, len);
2036 		}
2037 		return (0);
2038 	}
2039 	return (ENOENT);
2040 }
2041 
2042 int
2043 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2044 {
2045 	struct vfsopt *opt;
2046 
2047 	TAILQ_FOREACH(opt, opts, link) {
2048 		if (strcmp(name, opt->name) != 0)
2049 			continue;
2050 		opt->seen = 1;
2051 		if (opt->value == NULL)
2052 			opt->len = strlen(value) + 1;
2053 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2054 			return (EINVAL);
2055 		return (0);
2056 	}
2057 	return (ENOENT);
2058 }
2059 
2060 /*
2061  * Find and copy a mount option.
2062  *
2063  * The size of the buffer has to be specified
2064  * in len, if it is not the same length as the
2065  * mount option, EINVAL is returned.
2066  * Returns ENOENT if the option is not found.
2067  */
2068 int
2069 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2070 {
2071 	struct vfsopt *opt;
2072 
2073 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2074 
2075 	TAILQ_FOREACH(opt, opts, link) {
2076 		if (strcmp(name, opt->name) == 0) {
2077 			opt->seen = 1;
2078 			if (len != opt->len)
2079 				return (EINVAL);
2080 			bcopy(opt->value, dest, opt->len);
2081 			return (0);
2082 		}
2083 	}
2084 	return (ENOENT);
2085 }
2086 
2087 int
2088 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2089 {
2090 
2091 	/*
2092 	 * Filesystems only fill in part of the structure for updates, we
2093 	 * have to read the entirety first to get all content.
2094 	 */
2095 	memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2096 
2097 	/*
2098 	 * Set these in case the underlying filesystem fails to do so.
2099 	 */
2100 	sbp->f_version = STATFS_VERSION;
2101 	sbp->f_namemax = NAME_MAX;
2102 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2103 
2104 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2105 }
2106 
2107 void
2108 vfs_mountedfrom(struct mount *mp, const char *from)
2109 {
2110 
2111 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2112 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2113 	    sizeof mp->mnt_stat.f_mntfromname);
2114 }
2115 
2116 /*
2117  * ---------------------------------------------------------------------
2118  * This is the api for building mount args and mounting filesystems from
2119  * inside the kernel.
2120  *
2121  * The API works by accumulation of individual args.  First error is
2122  * latched.
2123  *
2124  * XXX: should be documented in new manpage kernel_mount(9)
2125  */
2126 
2127 /* A memory allocation which must be freed when we are done */
2128 struct mntaarg {
2129 	SLIST_ENTRY(mntaarg)	next;
2130 };
2131 
2132 /* The header for the mount arguments */
2133 struct mntarg {
2134 	struct iovec *v;
2135 	int len;
2136 	int error;
2137 	SLIST_HEAD(, mntaarg)	list;
2138 };
2139 
2140 /*
2141  * Add a boolean argument.
2142  *
2143  * flag is the boolean value.
2144  * name must start with "no".
2145  */
2146 struct mntarg *
2147 mount_argb(struct mntarg *ma, int flag, const char *name)
2148 {
2149 
2150 	KASSERT(name[0] == 'n' && name[1] == 'o',
2151 	    ("mount_argb(...,%s): name must start with 'no'", name));
2152 
2153 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2154 }
2155 
2156 /*
2157  * Add an argument printf style
2158  */
2159 struct mntarg *
2160 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2161 {
2162 	va_list ap;
2163 	struct mntaarg *maa;
2164 	struct sbuf *sb;
2165 	int len;
2166 
2167 	if (ma == NULL) {
2168 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2169 		SLIST_INIT(&ma->list);
2170 	}
2171 	if (ma->error)
2172 		return (ma);
2173 
2174 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2175 	    M_MOUNT, M_WAITOK);
2176 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2177 	ma->v[ma->len].iov_len = strlen(name) + 1;
2178 	ma->len++;
2179 
2180 	sb = sbuf_new_auto();
2181 	va_start(ap, fmt);
2182 	sbuf_vprintf(sb, fmt, ap);
2183 	va_end(ap);
2184 	sbuf_finish(sb);
2185 	len = sbuf_len(sb) + 1;
2186 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2187 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2188 	bcopy(sbuf_data(sb), maa + 1, len);
2189 	sbuf_delete(sb);
2190 
2191 	ma->v[ma->len].iov_base = maa + 1;
2192 	ma->v[ma->len].iov_len = len;
2193 	ma->len++;
2194 
2195 	return (ma);
2196 }
2197 
2198 /*
2199  * Add an argument which is a userland string.
2200  */
2201 struct mntarg *
2202 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2203 {
2204 	struct mntaarg *maa;
2205 	char *tbuf;
2206 
2207 	if (val == NULL)
2208 		return (ma);
2209 	if (ma == NULL) {
2210 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2211 		SLIST_INIT(&ma->list);
2212 	}
2213 	if (ma->error)
2214 		return (ma);
2215 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2216 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2217 	tbuf = (void *)(maa + 1);
2218 	ma->error = copyinstr(val, tbuf, len, NULL);
2219 	return (mount_arg(ma, name, tbuf, -1));
2220 }
2221 
2222 /*
2223  * Plain argument.
2224  *
2225  * If length is -1, treat value as a C string.
2226  */
2227 struct mntarg *
2228 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2229 {
2230 
2231 	if (ma == NULL) {
2232 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2233 		SLIST_INIT(&ma->list);
2234 	}
2235 	if (ma->error)
2236 		return (ma);
2237 
2238 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2239 	    M_MOUNT, M_WAITOK);
2240 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2241 	ma->v[ma->len].iov_len = strlen(name) + 1;
2242 	ma->len++;
2243 
2244 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2245 	if (len < 0)
2246 		ma->v[ma->len].iov_len = strlen(val) + 1;
2247 	else
2248 		ma->v[ma->len].iov_len = len;
2249 	ma->len++;
2250 	return (ma);
2251 }
2252 
2253 /*
2254  * Free a mntarg structure
2255  */
2256 static void
2257 free_mntarg(struct mntarg *ma)
2258 {
2259 	struct mntaarg *maa;
2260 
2261 	while (!SLIST_EMPTY(&ma->list)) {
2262 		maa = SLIST_FIRST(&ma->list);
2263 		SLIST_REMOVE_HEAD(&ma->list, next);
2264 		free(maa, M_MOUNT);
2265 	}
2266 	free(ma->v, M_MOUNT);
2267 	free(ma, M_MOUNT);
2268 }
2269 
2270 /*
2271  * Mount a filesystem
2272  */
2273 int
2274 kernel_mount(struct mntarg *ma, uint64_t flags)
2275 {
2276 	struct uio auio;
2277 	int error;
2278 
2279 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2280 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2281 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2282 
2283 	auio.uio_iov = ma->v;
2284 	auio.uio_iovcnt = ma->len;
2285 	auio.uio_segflg = UIO_SYSSPACE;
2286 
2287 	error = ma->error;
2288 	if (!error)
2289 		error = vfs_donmount(curthread, flags, &auio);
2290 	free_mntarg(ma);
2291 	return (error);
2292 }
2293 
2294 /*
2295  * A printflike function to mount a filesystem.
2296  */
2297 int
2298 kernel_vmount(int flags, ...)
2299 {
2300 	struct mntarg *ma = NULL;
2301 	va_list ap;
2302 	const char *cp;
2303 	const void *vp;
2304 	int error;
2305 
2306 	va_start(ap, flags);
2307 	for (;;) {
2308 		cp = va_arg(ap, const char *);
2309 		if (cp == NULL)
2310 			break;
2311 		vp = va_arg(ap, const void *);
2312 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2313 	}
2314 	va_end(ap);
2315 
2316 	error = kernel_mount(ma, flags);
2317 	return (error);
2318 }
2319 
2320 void
2321 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
2322 {
2323 
2324 	bcopy(oexp, exp, sizeof(*oexp));
2325 	exp->ex_numsecflavors = 0;
2326 }
2327