xref: /freebsd/sys/kern/vfs_mount.c (revision d01f7b29a9cf3a4f651b8d66967b447cd3856685)
1 /*-
2  * Copyright (c) 1999-2004 Poul-Henning Kamp
3  * Copyright (c) 1999 Michael Smith
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 AUTHOR 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 AUTHOR 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 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/conf.h>
42 #include <sys/fcntl.h>
43 #include <sys/jail.h>
44 #include <sys/kernel.h>
45 #include <sys/libkern.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/filedesc.h>
53 #include <sys/reboot.h>
54 #include <sys/sbuf.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysproto.h>
57 #include <sys/sx.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/systm.h>
61 #include <sys/vnode.h>
62 #include <vm/uma.h>
63 
64 #include <geom/geom.h>
65 
66 #include <machine/stdarg.h>
67 
68 #include <security/audit/audit.h>
69 #include <security/mac/mac_framework.h>
70 
71 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
72 
73 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
74 		    uint64_t fsflags, struct vfsoptlist **optlist);
75 static void	free_mntarg(struct mntarg *ma);
76 
77 static int	usermount = 0;
78 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
79     "Unprivileged users may mount and unmount file systems");
80 
81 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
82 static uma_zone_t mount_zone;
83 
84 /* List of mounted filesystems. */
85 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
86 
87 /* For any iteration/modification of mountlist */
88 struct mtx mountlist_mtx;
89 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
90 
91 /*
92  * Global opts, taken by all filesystems
93  */
94 static const char *global_opts[] = {
95 	"errmsg",
96 	"fstype",
97 	"fspath",
98 	"ro",
99 	"rw",
100 	"nosuid",
101 	"noexec",
102 	NULL
103 };
104 
105 static int
106 mount_init(void *mem, int size, int flags)
107 {
108 	struct mount *mp;
109 
110 	mp = (struct mount *)mem;
111 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
112 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
113 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
114 	return (0);
115 }
116 
117 static void
118 mount_fini(void *mem, int size)
119 {
120 	struct mount *mp;
121 
122 	mp = (struct mount *)mem;
123 	lockdestroy(&mp->mnt_explock);
124 	mtx_destroy(&mp->mnt_listmtx);
125 	mtx_destroy(&mp->mnt_mtx);
126 }
127 
128 static void
129 vfs_mount_init(void *dummy __unused)
130 {
131 
132 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
133 	    NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
134 }
135 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
136 
137 /*
138  * ---------------------------------------------------------------------
139  * Functions for building and sanitizing the mount options
140  */
141 
142 /* Remove one mount option. */
143 static void
144 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
145 {
146 
147 	TAILQ_REMOVE(opts, opt, link);
148 	free(opt->name, M_MOUNT);
149 	if (opt->value != NULL)
150 		free(opt->value, M_MOUNT);
151 	free(opt, M_MOUNT);
152 }
153 
154 /* Release all resources related to the mount options. */
155 void
156 vfs_freeopts(struct vfsoptlist *opts)
157 {
158 	struct vfsopt *opt;
159 
160 	while (!TAILQ_EMPTY(opts)) {
161 		opt = TAILQ_FIRST(opts);
162 		vfs_freeopt(opts, opt);
163 	}
164 	free(opts, M_MOUNT);
165 }
166 
167 void
168 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
169 {
170 	struct vfsopt *opt, *temp;
171 
172 	if (opts == NULL)
173 		return;
174 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
175 		if (strcmp(opt->name, name) == 0)
176 			vfs_freeopt(opts, opt);
177 	}
178 }
179 
180 static int
181 vfs_isopt_ro(const char *opt)
182 {
183 
184 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
185 	    strcmp(opt, "norw") == 0)
186 		return (1);
187 	return (0);
188 }
189 
190 static int
191 vfs_isopt_rw(const char *opt)
192 {
193 
194 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
195 		return (1);
196 	return (0);
197 }
198 
199 /*
200  * Check if options are equal (with or without the "no" prefix).
201  */
202 static int
203 vfs_equalopts(const char *opt1, const char *opt2)
204 {
205 	char *p;
206 
207 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
208 	if (strcmp(opt1, opt2) == 0)
209 		return (1);
210 	/* "noopt" vs. "opt" */
211 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
212 		return (1);
213 	/* "opt" vs. "noopt" */
214 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
215 		return (1);
216 	while ((p = strchr(opt1, '.')) != NULL &&
217 	    !strncmp(opt1, opt2, ++p - opt1)) {
218 		opt2 += p - opt1;
219 		opt1 = p;
220 		/* "foo.noopt" vs. "foo.opt" */
221 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
222 			return (1);
223 		/* "foo.opt" vs. "foo.noopt" */
224 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
225 			return (1);
226 	}
227 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
228 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
229 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
230 		return (1);
231 	return (0);
232 }
233 
234 /*
235  * If a mount option is specified several times,
236  * (with or without the "no" prefix) only keep
237  * the last occurrence of it.
238  */
239 static void
240 vfs_sanitizeopts(struct vfsoptlist *opts)
241 {
242 	struct vfsopt *opt, *opt2, *tmp;
243 
244 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
245 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
246 		while (opt2 != NULL) {
247 			if (vfs_equalopts(opt->name, opt2->name)) {
248 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
249 				vfs_freeopt(opts, opt2);
250 				opt2 = tmp;
251 			} else {
252 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
253 			}
254 		}
255 	}
256 }
257 
258 /*
259  * Build a linked list of mount options from a struct uio.
260  */
261 int
262 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
263 {
264 	struct vfsoptlist *opts;
265 	struct vfsopt *opt;
266 	size_t memused, namelen, optlen;
267 	unsigned int i, iovcnt;
268 	int error;
269 
270 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
271 	TAILQ_INIT(opts);
272 	memused = 0;
273 	iovcnt = auio->uio_iovcnt;
274 	for (i = 0; i < iovcnt; i += 2) {
275 		namelen = auio->uio_iov[i].iov_len;
276 		optlen = auio->uio_iov[i + 1].iov_len;
277 		memused += sizeof(struct vfsopt) + optlen + namelen;
278 		/*
279 		 * Avoid consuming too much memory, and attempts to overflow
280 		 * memused.
281 		 */
282 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
283 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
284 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
285 			error = EINVAL;
286 			goto bad;
287 		}
288 
289 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
290 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
291 		opt->value = NULL;
292 		opt->len = 0;
293 		opt->pos = i / 2;
294 		opt->seen = 0;
295 
296 		/*
297 		 * Do this early, so jumps to "bad" will free the current
298 		 * option.
299 		 */
300 		TAILQ_INSERT_TAIL(opts, opt, link);
301 
302 		if (auio->uio_segflg == UIO_SYSSPACE) {
303 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
304 		} else {
305 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
306 			    namelen);
307 			if (error)
308 				goto bad;
309 		}
310 		/* Ensure names are null-terminated strings. */
311 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
312 			error = EINVAL;
313 			goto bad;
314 		}
315 		if (optlen != 0) {
316 			opt->len = optlen;
317 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
318 			if (auio->uio_segflg == UIO_SYSSPACE) {
319 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
320 				    optlen);
321 			} else {
322 				error = copyin(auio->uio_iov[i + 1].iov_base,
323 				    opt->value, optlen);
324 				if (error)
325 					goto bad;
326 			}
327 		}
328 	}
329 	vfs_sanitizeopts(opts);
330 	*options = opts;
331 	return (0);
332 bad:
333 	vfs_freeopts(opts);
334 	return (error);
335 }
336 
337 /*
338  * Merge the old mount options with the new ones passed
339  * in the MNT_UPDATE case.
340  *
341  * XXX: This function will keep a "nofoo" option in the new
342  * options.  E.g, if the option's canonical name is "foo",
343  * "nofoo" ends up in the mount point's active options.
344  */
345 static void
346 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
347 {
348 	struct vfsopt *opt, *new;
349 
350 	TAILQ_FOREACH(opt, oldopts, link) {
351 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
352 		new->name = strdup(opt->name, M_MOUNT);
353 		if (opt->len != 0) {
354 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
355 			bcopy(opt->value, new->value, opt->len);
356 		} else
357 			new->value = NULL;
358 		new->len = opt->len;
359 		new->seen = opt->seen;
360 		TAILQ_INSERT_HEAD(toopts, new, link);
361 	}
362 	vfs_sanitizeopts(toopts);
363 }
364 
365 /*
366  * Mount a filesystem.
367  */
368 int
369 sys_nmount(td, uap)
370 	struct thread *td;
371 	struct nmount_args /* {
372 		struct iovec *iovp;
373 		unsigned int iovcnt;
374 		int flags;
375 	} */ *uap;
376 {
377 	struct uio *auio;
378 	int error;
379 	u_int iovcnt;
380 	uint64_t flags;
381 
382 	/*
383 	 * Mount flags are now 64-bits. On 32-bit archtectures only
384 	 * 32-bits are passed in, but from here on everything handles
385 	 * 64-bit flags correctly.
386 	 */
387 	flags = uap->flags;
388 
389 	AUDIT_ARG_FFLAGS(flags);
390 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
391 	    uap->iovp, uap->iovcnt, flags);
392 
393 	/*
394 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
395 	 * userspace to set this flag, but we must filter it out if we want
396 	 * MNT_UPDATE on the root file system to work.
397 	 * MNT_ROOTFS should only be set by the kernel when mounting its
398 	 * root file system.
399 	 */
400 	flags &= ~MNT_ROOTFS;
401 
402 	iovcnt = uap->iovcnt;
403 	/*
404 	 * Check that we have an even number of iovec's
405 	 * and that we have at least two options.
406 	 */
407 	if ((iovcnt & 1) || (iovcnt < 4)) {
408 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
409 		    uap->iovcnt);
410 		return (EINVAL);
411 	}
412 
413 	error = copyinuio(uap->iovp, iovcnt, &auio);
414 	if (error) {
415 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
416 		    __func__, error);
417 		return (error);
418 	}
419 	error = vfs_donmount(td, flags, auio);
420 
421 	free(auio, M_IOV);
422 	return (error);
423 }
424 
425 /*
426  * ---------------------------------------------------------------------
427  * Various utility functions
428  */
429 
430 void
431 vfs_ref(struct mount *mp)
432 {
433 
434 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
435 	MNT_ILOCK(mp);
436 	MNT_REF(mp);
437 	MNT_IUNLOCK(mp);
438 }
439 
440 void
441 vfs_rel(struct mount *mp)
442 {
443 
444 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
445 	MNT_ILOCK(mp);
446 	MNT_REL(mp);
447 	MNT_IUNLOCK(mp);
448 }
449 
450 /*
451  * Allocate and initialize the mount point struct.
452  */
453 struct mount *
454 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
455     struct ucred *cred)
456 {
457 	struct mount *mp;
458 
459 	mp = uma_zalloc(mount_zone, M_WAITOK);
460 	bzero(&mp->mnt_startzero,
461 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
462 	TAILQ_INIT(&mp->mnt_nvnodelist);
463 	mp->mnt_nvnodelistsize = 0;
464 	TAILQ_INIT(&mp->mnt_activevnodelist);
465 	mp->mnt_activevnodelistsize = 0;
466 	TAILQ_INIT(&mp->mnt_tmpfreevnodelist);
467 	mp->mnt_tmpfreevnodelistsize = 0;
468 	mp->mnt_ref = 0;
469 	(void) vfs_busy(mp, MBF_NOWAIT);
470 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
471 	mp->mnt_op = vfsp->vfc_vfsops;
472 	mp->mnt_vfc = vfsp;
473 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
474 	mp->mnt_gen++;
475 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
476 	mp->mnt_vnodecovered = vp;
477 	mp->mnt_cred = crdup(cred);
478 	mp->mnt_stat.f_owner = cred->cr_uid;
479 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
480 	mp->mnt_iosize_max = DFLTPHYS;
481 #ifdef MAC
482 	mac_mount_init(mp);
483 	mac_mount_create(cred, mp);
484 #endif
485 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
486 	TAILQ_INIT(&mp->mnt_uppers);
487 	return (mp);
488 }
489 
490 /*
491  * Destroy the mount struct previously allocated by vfs_mount_alloc().
492  */
493 void
494 vfs_mount_destroy(struct mount *mp)
495 {
496 
497 	MNT_ILOCK(mp);
498 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
499 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
500 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
501 		wakeup(mp);
502 	}
503 	while (mp->mnt_ref)
504 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
505 	KASSERT(mp->mnt_ref == 0,
506 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
507 	    __FILE__, __LINE__));
508 	if (mp->mnt_writeopcount != 0)
509 		panic("vfs_mount_destroy: nonzero writeopcount");
510 	if (mp->mnt_secondary_writes != 0)
511 		panic("vfs_mount_destroy: nonzero secondary_writes");
512 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
513 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
514 		struct vnode *vp;
515 
516 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
517 			vn_printf(vp, "dangling vnode ");
518 		panic("unmount: dangling vnode");
519 	}
520 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
521 	if (mp->mnt_nvnodelistsize != 0)
522 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
523 	if (mp->mnt_activevnodelistsize != 0)
524 		panic("vfs_mount_destroy: nonzero activevnodelistsize");
525 	if (mp->mnt_tmpfreevnodelistsize != 0)
526 		panic("vfs_mount_destroy: nonzero tmpfreevnodelistsize");
527 	if (mp->mnt_lockref != 0)
528 		panic("vfs_mount_destroy: nonzero lock refcount");
529 	MNT_IUNLOCK(mp);
530 #ifdef MAC
531 	mac_mount_destroy(mp);
532 #endif
533 	if (mp->mnt_opt != NULL)
534 		vfs_freeopts(mp->mnt_opt);
535 	crfree(mp->mnt_cred);
536 	uma_zfree(mount_zone, mp);
537 }
538 
539 int
540 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
541 {
542 	struct vfsoptlist *optlist;
543 	struct vfsopt *opt, *tmp_opt;
544 	char *fstype, *fspath, *errmsg;
545 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
546 
547 	errmsg = fspath = NULL;
548 	errmsg_len = fspathlen = 0;
549 	errmsg_pos = -1;
550 
551 	error = vfs_buildopts(fsoptions, &optlist);
552 	if (error)
553 		return (error);
554 
555 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
556 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
557 
558 	/*
559 	 * We need these two options before the others,
560 	 * and they are mandatory for any filesystem.
561 	 * Ensure they are NUL terminated as well.
562 	 */
563 	fstypelen = 0;
564 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
565 	if (error || fstype[fstypelen - 1] != '\0') {
566 		error = EINVAL;
567 		if (errmsg != NULL)
568 			strncpy(errmsg, "Invalid fstype", errmsg_len);
569 		goto bail;
570 	}
571 	fspathlen = 0;
572 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
573 	if (error || fspath[fspathlen - 1] != '\0') {
574 		error = EINVAL;
575 		if (errmsg != NULL)
576 			strncpy(errmsg, "Invalid fspath", errmsg_len);
577 		goto bail;
578 	}
579 
580 	/*
581 	 * We need to see if we have the "update" option
582 	 * before we call vfs_domount(), since vfs_domount() has special
583 	 * logic based on MNT_UPDATE.  This is very important
584 	 * when we want to update the root filesystem.
585 	 */
586 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
587 		if (strcmp(opt->name, "update") == 0) {
588 			fsflags |= MNT_UPDATE;
589 			vfs_freeopt(optlist, opt);
590 		}
591 		else if (strcmp(opt->name, "async") == 0)
592 			fsflags |= MNT_ASYNC;
593 		else if (strcmp(opt->name, "force") == 0) {
594 			fsflags |= MNT_FORCE;
595 			vfs_freeopt(optlist, opt);
596 		}
597 		else if (strcmp(opt->name, "reload") == 0) {
598 			fsflags |= MNT_RELOAD;
599 			vfs_freeopt(optlist, opt);
600 		}
601 		else if (strcmp(opt->name, "multilabel") == 0)
602 			fsflags |= MNT_MULTILABEL;
603 		else if (strcmp(opt->name, "noasync") == 0)
604 			fsflags &= ~MNT_ASYNC;
605 		else if (strcmp(opt->name, "noatime") == 0)
606 			fsflags |= MNT_NOATIME;
607 		else if (strcmp(opt->name, "atime") == 0) {
608 			free(opt->name, M_MOUNT);
609 			opt->name = strdup("nonoatime", M_MOUNT);
610 		}
611 		else if (strcmp(opt->name, "noclusterr") == 0)
612 			fsflags |= MNT_NOCLUSTERR;
613 		else if (strcmp(opt->name, "clusterr") == 0) {
614 			free(opt->name, M_MOUNT);
615 			opt->name = strdup("nonoclusterr", M_MOUNT);
616 		}
617 		else if (strcmp(opt->name, "noclusterw") == 0)
618 			fsflags |= MNT_NOCLUSTERW;
619 		else if (strcmp(opt->name, "clusterw") == 0) {
620 			free(opt->name, M_MOUNT);
621 			opt->name = strdup("nonoclusterw", M_MOUNT);
622 		}
623 		else if (strcmp(opt->name, "noexec") == 0)
624 			fsflags |= MNT_NOEXEC;
625 		else if (strcmp(opt->name, "exec") == 0) {
626 			free(opt->name, M_MOUNT);
627 			opt->name = strdup("nonoexec", M_MOUNT);
628 		}
629 		else if (strcmp(opt->name, "nosuid") == 0)
630 			fsflags |= MNT_NOSUID;
631 		else if (strcmp(opt->name, "suid") == 0) {
632 			free(opt->name, M_MOUNT);
633 			opt->name = strdup("nonosuid", M_MOUNT);
634 		}
635 		else if (strcmp(opt->name, "nosymfollow") == 0)
636 			fsflags |= MNT_NOSYMFOLLOW;
637 		else if (strcmp(opt->name, "symfollow") == 0) {
638 			free(opt->name, M_MOUNT);
639 			opt->name = strdup("nonosymfollow", M_MOUNT);
640 		}
641 		else if (strcmp(opt->name, "noro") == 0)
642 			fsflags &= ~MNT_RDONLY;
643 		else if (strcmp(opt->name, "rw") == 0)
644 			fsflags &= ~MNT_RDONLY;
645 		else if (strcmp(opt->name, "ro") == 0)
646 			fsflags |= MNT_RDONLY;
647 		else if (strcmp(opt->name, "rdonly") == 0) {
648 			free(opt->name, M_MOUNT);
649 			opt->name = strdup("ro", M_MOUNT);
650 			fsflags |= MNT_RDONLY;
651 		}
652 		else if (strcmp(opt->name, "suiddir") == 0)
653 			fsflags |= MNT_SUIDDIR;
654 		else if (strcmp(opt->name, "sync") == 0)
655 			fsflags |= MNT_SYNCHRONOUS;
656 		else if (strcmp(opt->name, "union") == 0)
657 			fsflags |= MNT_UNION;
658 		else if (strcmp(opt->name, "automounted") == 0) {
659 			fsflags |= MNT_AUTOMOUNTED;
660 			vfs_freeopt(optlist, opt);
661 		}
662 	}
663 
664 	/*
665 	 * Be ultra-paranoid about making sure the type and fspath
666 	 * variables will fit in our mp buffers, including the
667 	 * terminating NUL.
668 	 */
669 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
670 		error = ENAMETOOLONG;
671 		goto bail;
672 	}
673 
674 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
675 bail:
676 	/* copyout the errmsg */
677 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
678 	    && errmsg_len > 0 && errmsg != NULL) {
679 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
680 			bcopy(errmsg,
681 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
682 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
683 		} else {
684 			copyout(errmsg,
685 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
686 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
687 		}
688 	}
689 
690 	if (optlist != NULL)
691 		vfs_freeopts(optlist);
692 	return (error);
693 }
694 
695 /*
696  * Old mount API.
697  */
698 #ifndef _SYS_SYSPROTO_H_
699 struct mount_args {
700 	char	*type;
701 	char	*path;
702 	int	flags;
703 	caddr_t	data;
704 };
705 #endif
706 /* ARGSUSED */
707 int
708 sys_mount(td, uap)
709 	struct thread *td;
710 	struct mount_args /* {
711 		char *type;
712 		char *path;
713 		int flags;
714 		caddr_t data;
715 	} */ *uap;
716 {
717 	char *fstype;
718 	struct vfsconf *vfsp = NULL;
719 	struct mntarg *ma = NULL;
720 	uint64_t flags;
721 	int error;
722 
723 	/*
724 	 * Mount flags are now 64-bits. On 32-bit architectures only
725 	 * 32-bits are passed in, but from here on everything handles
726 	 * 64-bit flags correctly.
727 	 */
728 	flags = uap->flags;
729 
730 	AUDIT_ARG_FFLAGS(flags);
731 
732 	/*
733 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
734 	 * userspace to set this flag, but we must filter it out if we want
735 	 * MNT_UPDATE on the root file system to work.
736 	 * MNT_ROOTFS should only be set by the kernel when mounting its
737 	 * root file system.
738 	 */
739 	flags &= ~MNT_ROOTFS;
740 
741 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
742 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
743 	if (error) {
744 		free(fstype, M_TEMP);
745 		return (error);
746 	}
747 
748 	AUDIT_ARG_TEXT(fstype);
749 	vfsp = vfs_byname_kld(fstype, td, &error);
750 	free(fstype, M_TEMP);
751 	if (vfsp == NULL)
752 		return (ENOENT);
753 	if (vfsp->vfc_vfsops->vfs_cmount == NULL)
754 		return (EOPNOTSUPP);
755 
756 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
757 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
758 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
759 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
760 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
761 
762 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags);
763 	return (error);
764 }
765 
766 /*
767  * vfs_domount_first(): first file system mount (not update)
768  */
769 static int
770 vfs_domount_first(
771 	struct thread *td,		/* Calling thread. */
772 	struct vfsconf *vfsp,		/* File system type. */
773 	char *fspath,			/* Mount path. */
774 	struct vnode *vp,		/* Vnode to be covered. */
775 	uint64_t fsflags,		/* Flags common to all filesystems. */
776 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
777 	)
778 {
779 	struct vattr va;
780 	struct mount *mp;
781 	struct vnode *newdp;
782 	int error;
783 
784 	ASSERT_VOP_ELOCKED(vp, __func__);
785 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
786 
787 	/*
788 	 * If the user is not root, ensure that they own the directory
789 	 * onto which we are attempting to mount.
790 	 */
791 	error = VOP_GETATTR(vp, &va, td->td_ucred);
792 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
793 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0);
794 	if (error == 0)
795 		error = vinvalbuf(vp, V_SAVE, 0, 0);
796 	if (error == 0 && vp->v_type != VDIR)
797 		error = ENOTDIR;
798 	if (error == 0) {
799 		VI_LOCK(vp);
800 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
801 			vp->v_iflag |= VI_MOUNT;
802 		else
803 			error = EBUSY;
804 		VI_UNLOCK(vp);
805 	}
806 	if (error != 0) {
807 		vput(vp);
808 		return (error);
809 	}
810 	VOP_UNLOCK(vp, 0);
811 
812 	/* Allocate and initialize the filesystem. */
813 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
814 	/* XXXMAC: pass to vfs_mount_alloc? */
815 	mp->mnt_optnew = *optlist;
816 	/* Set the mount level flags. */
817 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
818 
819 	/*
820 	 * Mount the filesystem.
821 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
822 	 * get.  No freeing of cn_pnbuf.
823 	 */
824 	error = VFS_MOUNT(mp);
825 	if (error != 0) {
826 		vfs_unbusy(mp);
827 		vfs_mount_destroy(mp);
828 		VI_LOCK(vp);
829 		vp->v_iflag &= ~VI_MOUNT;
830 		VI_UNLOCK(vp);
831 		vrele(vp);
832 		return (error);
833 	}
834 
835 	if (mp->mnt_opt != NULL)
836 		vfs_freeopts(mp->mnt_opt);
837 	mp->mnt_opt = mp->mnt_optnew;
838 	*optlist = NULL;
839 	(void)VFS_STATFS(mp, &mp->mnt_stat);
840 
841 	/*
842 	 * Prevent external consumers of mount options from reading mnt_optnew.
843 	 */
844 	mp->mnt_optnew = NULL;
845 
846 	MNT_ILOCK(mp);
847 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
848 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
849 		mp->mnt_kern_flag |= MNTK_ASYNC;
850 	else
851 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
852 	MNT_IUNLOCK(mp);
853 
854 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
855 	cache_purge(vp);
856 	VI_LOCK(vp);
857 	vp->v_iflag &= ~VI_MOUNT;
858 	VI_UNLOCK(vp);
859 	vp->v_mountedhere = mp;
860 	/* Place the new filesystem at the end of the mount list. */
861 	mtx_lock(&mountlist_mtx);
862 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
863 	mtx_unlock(&mountlist_mtx);
864 	vfs_event_signal(NULL, VQ_MOUNT, 0);
865 	if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp))
866 		panic("mount: lost mount");
867 	VOP_UNLOCK(vp, 0);
868 	EVENTHANDLER_INVOKE(vfs_mounted, mp, newdp, td);
869 	VOP_UNLOCK(newdp, 0);
870 	mountcheckdirs(vp, newdp);
871 	vrele(newdp);
872 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
873 		vfs_allocate_syncvnode(mp);
874 	vfs_unbusy(mp);
875 	return (0);
876 }
877 
878 /*
879  * vfs_domount_update(): update of mounted file system
880  */
881 static int
882 vfs_domount_update(
883 	struct thread *td,		/* Calling thread. */
884 	struct vnode *vp,		/* Mount point vnode. */
885 	uint64_t fsflags,		/* Flags common to all filesystems. */
886 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
887 	)
888 {
889 	struct export_args export;
890 	void *bufp;
891 	struct mount *mp;
892 	int error, export_error, len;
893 	uint64_t flag;
894 
895 	ASSERT_VOP_ELOCKED(vp, __func__);
896 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
897 	mp = vp->v_mount;
898 
899 	if ((vp->v_vflag & VV_ROOT) == 0) {
900 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
901 		    == 0)
902 			error = EXDEV;
903 		else
904 			error = EINVAL;
905 		vput(vp);
906 		return (error);
907 	}
908 
909 	/*
910 	 * We only allow the filesystem to be reloaded if it
911 	 * is currently mounted read-only.
912 	 */
913 	flag = mp->mnt_flag;
914 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
915 		vput(vp);
916 		return (EOPNOTSUPP);	/* Needs translation */
917 	}
918 	/*
919 	 * Only privileged root, or (if MNT_USER is set) the user that
920 	 * did the original mount is permitted to update it.
921 	 */
922 	error = vfs_suser(mp, td);
923 	if (error != 0) {
924 		vput(vp);
925 		return (error);
926 	}
927 	if (vfs_busy(mp, MBF_NOWAIT)) {
928 		vput(vp);
929 		return (EBUSY);
930 	}
931 	VI_LOCK(vp);
932 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
933 		VI_UNLOCK(vp);
934 		vfs_unbusy(mp);
935 		vput(vp);
936 		return (EBUSY);
937 	}
938 	vp->v_iflag |= VI_MOUNT;
939 	VI_UNLOCK(vp);
940 	VOP_UNLOCK(vp, 0);
941 
942 	MNT_ILOCK(mp);
943 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
944 		MNT_IUNLOCK(mp);
945 		error = EBUSY;
946 		goto end;
947 	}
948 	mp->mnt_flag &= ~MNT_UPDATEMASK;
949 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
950 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
951 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
952 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
953 	MNT_IUNLOCK(mp);
954 	mp->mnt_optnew = *optlist;
955 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
956 
957 	/*
958 	 * Mount the filesystem.
959 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
960 	 * get.  No freeing of cn_pnbuf.
961 	 */
962 	error = VFS_MOUNT(mp);
963 
964 	export_error = 0;
965 	/* Process the export option. */
966 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
967 	    &len) == 0) {
968 		/* Assume that there is only 1 ABI for each length. */
969 		switch (len) {
970 		case (sizeof(struct oexport_args)):
971 			bzero(&export, sizeof(export));
972 			/* FALLTHROUGH */
973 		case (sizeof(export)):
974 			bcopy(bufp, &export, len);
975 			export_error = vfs_export(mp, &export);
976 			break;
977 		default:
978 			export_error = EINVAL;
979 			break;
980 		}
981 	}
982 
983 	MNT_ILOCK(mp);
984 	if (error == 0) {
985 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
986 		    MNT_SNAPSHOT);
987 	} else {
988 		/*
989 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
990 		 * because it is not part of MNT_UPDATEMASK, but it could have
991 		 * changed in the meantime if quotactl(2) was called.
992 		 * All in all we want current value of MNT_QUOTA, not the old
993 		 * one.
994 		 */
995 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
996 	}
997 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
998 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
999 		mp->mnt_kern_flag |= MNTK_ASYNC;
1000 	else
1001 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1002 	MNT_IUNLOCK(mp);
1003 
1004 	if (error != 0)
1005 		goto end;
1006 
1007 	if (mp->mnt_opt != NULL)
1008 		vfs_freeopts(mp->mnt_opt);
1009 	mp->mnt_opt = mp->mnt_optnew;
1010 	*optlist = NULL;
1011 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1012 	/*
1013 	 * Prevent external consumers of mount options from reading
1014 	 * mnt_optnew.
1015 	 */
1016 	mp->mnt_optnew = NULL;
1017 
1018 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1019 		vfs_allocate_syncvnode(mp);
1020 	else
1021 		vfs_deallocate_syncvnode(mp);
1022 end:
1023 	vfs_unbusy(mp);
1024 	VI_LOCK(vp);
1025 	vp->v_iflag &= ~VI_MOUNT;
1026 	VI_UNLOCK(vp);
1027 	vrele(vp);
1028 	return (error != 0 ? error : export_error);
1029 }
1030 
1031 /*
1032  * vfs_domount(): actually attempt a filesystem mount.
1033  */
1034 static int
1035 vfs_domount(
1036 	struct thread *td,		/* Calling thread. */
1037 	const char *fstype,		/* Filesystem type. */
1038 	char *fspath,			/* Mount path. */
1039 	uint64_t fsflags,		/* Flags common to all filesystems. */
1040 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1041 	)
1042 {
1043 	struct vfsconf *vfsp;
1044 	struct nameidata nd;
1045 	struct vnode *vp;
1046 	char *pathbuf;
1047 	int error;
1048 
1049 	/*
1050 	 * Be ultra-paranoid about making sure the type and fspath
1051 	 * variables will fit in our mp buffers, including the
1052 	 * terminating NUL.
1053 	 */
1054 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1055 		return (ENAMETOOLONG);
1056 
1057 	if (jailed(td->td_ucred) || usermount == 0) {
1058 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1059 			return (error);
1060 	}
1061 
1062 	/*
1063 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1064 	 */
1065 	if (fsflags & MNT_EXPORTED) {
1066 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1067 		if (error)
1068 			return (error);
1069 	}
1070 	if (fsflags & MNT_SUIDDIR) {
1071 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1072 		if (error)
1073 			return (error);
1074 	}
1075 	/*
1076 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1077 	 */
1078 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1079 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1080 			fsflags |= MNT_NOSUID | MNT_USER;
1081 	}
1082 
1083 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1084 	vfsp = NULL;
1085 	if ((fsflags & MNT_UPDATE) == 0) {
1086 		/* Don't try to load KLDs if we're mounting the root. */
1087 		if (fsflags & MNT_ROOTFS)
1088 			vfsp = vfs_byname(fstype);
1089 		else
1090 			vfsp = vfs_byname_kld(fstype, td, &error);
1091 		if (vfsp == NULL)
1092 			return (ENODEV);
1093 		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
1094 			return (EPERM);
1095 	}
1096 
1097 	/*
1098 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1099 	 */
1100 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1101 	    UIO_SYSSPACE, fspath, td);
1102 	error = namei(&nd);
1103 	if (error != 0)
1104 		return (error);
1105 	NDFREE(&nd, NDF_ONLY_PNBUF);
1106 	vp = nd.ni_vp;
1107 	if ((fsflags & MNT_UPDATE) == 0) {
1108 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1109 		strcpy(pathbuf, fspath);
1110 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1111 		/* debug.disablefullpath == 1 results in ENODEV */
1112 		if (error == 0 || error == ENODEV) {
1113 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1114 			    fsflags, optlist);
1115 		}
1116 		free(pathbuf, M_TEMP);
1117 	} else
1118 		error = vfs_domount_update(td, vp, fsflags, optlist);
1119 
1120 	return (error);
1121 }
1122 
1123 /*
1124  * Unmount a filesystem.
1125  *
1126  * Note: unmount takes a path to the vnode mounted on as argument, not
1127  * special file (as before).
1128  */
1129 #ifndef _SYS_SYSPROTO_H_
1130 struct unmount_args {
1131 	char	*path;
1132 	int	flags;
1133 };
1134 #endif
1135 /* ARGSUSED */
1136 int
1137 sys_unmount(struct thread *td, struct unmount_args *uap)
1138 {
1139 	struct nameidata nd;
1140 	struct mount *mp;
1141 	char *pathbuf;
1142 	int error, id0, id1;
1143 
1144 	AUDIT_ARG_VALUE(uap->flags);
1145 	if (jailed(td->td_ucred) || usermount == 0) {
1146 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1147 		if (error)
1148 			return (error);
1149 	}
1150 
1151 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1152 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1153 	if (error) {
1154 		free(pathbuf, M_TEMP);
1155 		return (error);
1156 	}
1157 	if (uap->flags & MNT_BYFSID) {
1158 		AUDIT_ARG_TEXT(pathbuf);
1159 		/* Decode the filesystem ID. */
1160 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1161 			free(pathbuf, M_TEMP);
1162 			return (EINVAL);
1163 		}
1164 
1165 		mtx_lock(&mountlist_mtx);
1166 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1167 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1168 			    mp->mnt_stat.f_fsid.val[1] == id1) {
1169 				vfs_ref(mp);
1170 				break;
1171 			}
1172 		}
1173 		mtx_unlock(&mountlist_mtx);
1174 	} else {
1175 		/*
1176 		 * Try to find global path for path argument.
1177 		 */
1178 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1179 		    UIO_SYSSPACE, pathbuf, td);
1180 		if (namei(&nd) == 0) {
1181 			NDFREE(&nd, NDF_ONLY_PNBUF);
1182 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1183 			    MNAMELEN);
1184 			if (error == 0 || error == ENODEV)
1185 				vput(nd.ni_vp);
1186 		}
1187 		mtx_lock(&mountlist_mtx);
1188 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1189 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1190 				vfs_ref(mp);
1191 				break;
1192 			}
1193 		}
1194 		mtx_unlock(&mountlist_mtx);
1195 	}
1196 	free(pathbuf, M_TEMP);
1197 	if (mp == NULL) {
1198 		/*
1199 		 * Previously we returned ENOENT for a nonexistent path and
1200 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1201 		 * now, so in the !MNT_BYFSID case return the more likely
1202 		 * EINVAL for compatibility.
1203 		 */
1204 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1205 	}
1206 
1207 	/*
1208 	 * Don't allow unmounting the root filesystem.
1209 	 */
1210 	if (mp->mnt_flag & MNT_ROOTFS) {
1211 		vfs_rel(mp);
1212 		return (EINVAL);
1213 	}
1214 	error = dounmount(mp, uap->flags, td);
1215 	return (error);
1216 }
1217 
1218 /*
1219  * Return error if any of the vnodes, ignoring the root vnode
1220  * and the syncer vnode, have non-zero usecount.
1221  *
1222  * This function is purely advisory - it can return false positives
1223  * and negatives.
1224  */
1225 static int
1226 vfs_check_usecounts(struct mount *mp)
1227 {
1228 	struct vnode *vp, *mvp;
1229 
1230 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1231 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1232 		    vp->v_usecount != 0) {
1233 			VI_UNLOCK(vp);
1234 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1235 			return (EBUSY);
1236 		}
1237 		VI_UNLOCK(vp);
1238 	}
1239 
1240 	return (0);
1241 }
1242 
1243 static void
1244 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1245 {
1246 
1247 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1248 	mp->mnt_kern_flag &= ~mntkflags;
1249 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1250 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1251 		wakeup(mp);
1252 	}
1253 	MNT_IUNLOCK(mp);
1254 	if (coveredvp != NULL) {
1255 		VOP_UNLOCK(coveredvp, 0);
1256 		vdrop(coveredvp);
1257 	}
1258 	vn_finished_write(mp);
1259 }
1260 
1261 /*
1262  * Do the actual filesystem unmount.
1263  */
1264 int
1265 dounmount(struct mount *mp, int flags, struct thread *td)
1266 {
1267 	struct vnode *coveredvp, *fsrootvp;
1268 	int error;
1269 	uint64_t async_flag;
1270 	int mnt_gen_r;
1271 
1272 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1273 		mnt_gen_r = mp->mnt_gen;
1274 		VI_LOCK(coveredvp);
1275 		vholdl(coveredvp);
1276 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1277 		/*
1278 		 * Check for mp being unmounted while waiting for the
1279 		 * covered vnode lock.
1280 		 */
1281 		if (coveredvp->v_mountedhere != mp ||
1282 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1283 			VOP_UNLOCK(coveredvp, 0);
1284 			vdrop(coveredvp);
1285 			vfs_rel(mp);
1286 			return (EBUSY);
1287 		}
1288 	}
1289 
1290 	/*
1291 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1292 	 * original mount is permitted to unmount this filesystem.
1293 	 */
1294 	error = vfs_suser(mp, td);
1295 	if (error != 0) {
1296 		if (coveredvp != NULL) {
1297 			VOP_UNLOCK(coveredvp, 0);
1298 			vdrop(coveredvp);
1299 		}
1300 		vfs_rel(mp);
1301 		return (error);
1302 	}
1303 
1304 	vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1305 	MNT_ILOCK(mp);
1306 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1307 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
1308 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
1309 		dounmount_cleanup(mp, coveredvp, 0);
1310 		return (EBUSY);
1311 	}
1312 	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1313 	if (flags & MNT_NONBUSY) {
1314 		MNT_IUNLOCK(mp);
1315 		error = vfs_check_usecounts(mp);
1316 		MNT_ILOCK(mp);
1317 		if (error != 0) {
1318 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT |
1319 			    MNTK_NOINSMNTQ);
1320 			return (error);
1321 		}
1322 	}
1323 	/* Allow filesystems to detect that a forced unmount is in progress. */
1324 	if (flags & MNT_FORCE) {
1325 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1326 		MNT_IUNLOCK(mp);
1327 		/*
1328 		 * Must be done after setting MNTK_UNMOUNTF and before
1329 		 * waiting for mnt_lockref to become 0.
1330 		 */
1331 		VFS_PURGE(mp);
1332 		MNT_ILOCK(mp);
1333 	}
1334 	error = 0;
1335 	if (mp->mnt_lockref) {
1336 		mp->mnt_kern_flag |= MNTK_DRAINING;
1337 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1338 		    "mount drain", 0);
1339 	}
1340 	MNT_IUNLOCK(mp);
1341 	KASSERT(mp->mnt_lockref == 0,
1342 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1343 	    __func__, __FILE__, __LINE__));
1344 	KASSERT(error == 0,
1345 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1346 	    __func__, __FILE__, __LINE__));
1347 
1348 	if (mp->mnt_flag & MNT_EXPUBLIC)
1349 		vfs_setpublicfs(NULL, NULL, NULL);
1350 
1351 	/*
1352 	 * From now, we can claim that the use reference on the
1353 	 * coveredvp is ours, and the ref can be released only by
1354 	 * successfull unmount by us, or left for later unmount
1355 	 * attempt.  The previously acquired hold reference is no
1356 	 * longer needed to protect the vnode from reuse.
1357 	 */
1358 	if (coveredvp != NULL)
1359 		vdrop(coveredvp);
1360 
1361 	vfs_msync(mp, MNT_WAIT);
1362 	MNT_ILOCK(mp);
1363 	async_flag = mp->mnt_flag & MNT_ASYNC;
1364 	mp->mnt_flag &= ~MNT_ASYNC;
1365 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1366 	MNT_IUNLOCK(mp);
1367 	cache_purgevfs(mp, false); /* remove cache entries for this file sys */
1368 	vfs_deallocate_syncvnode(mp);
1369 	/*
1370 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1371 	 * vnode to the covered vnode.  For non-forced unmounts we want
1372 	 * such references to cause an EBUSY error.
1373 	 */
1374 	if ((flags & MNT_FORCE) &&
1375 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1376 		if (mp->mnt_vnodecovered != NULL &&
1377 		    (mp->mnt_flag & MNT_IGNORE) == 0)
1378 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1379 		if (fsrootvp == rootvnode) {
1380 			vrele(rootvnode);
1381 			rootvnode = NULL;
1382 		}
1383 		vput(fsrootvp);
1384 	}
1385 	if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 ||
1386 	    (error = VFS_SYNC(mp, MNT_WAIT)) == 0)
1387 		error = VFS_UNMOUNT(mp, flags);
1388 	vn_finished_write(mp);
1389 	/*
1390 	 * If we failed to flush the dirty blocks for this mount point,
1391 	 * undo all the cdir/rdir and rootvnode changes we made above.
1392 	 * Unless we failed to do so because the device is reporting that
1393 	 * it doesn't exist anymore.
1394 	 */
1395 	if (error && error != ENXIO) {
1396 		if ((flags & MNT_FORCE) &&
1397 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1398 			if (mp->mnt_vnodecovered != NULL &&
1399 			    (mp->mnt_flag & MNT_IGNORE) == 0)
1400 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1401 			if (rootvnode == NULL) {
1402 				rootvnode = fsrootvp;
1403 				vref(rootvnode);
1404 			}
1405 			vput(fsrootvp);
1406 		}
1407 		MNT_ILOCK(mp);
1408 		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1409 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1410 			MNT_IUNLOCK(mp);
1411 			vfs_allocate_syncvnode(mp);
1412 			MNT_ILOCK(mp);
1413 		}
1414 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1415 		mp->mnt_flag |= async_flag;
1416 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1417 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1418 			mp->mnt_kern_flag |= MNTK_ASYNC;
1419 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1420 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1421 			wakeup(mp);
1422 		}
1423 		MNT_IUNLOCK(mp);
1424 		if (coveredvp)
1425 			VOP_UNLOCK(coveredvp, 0);
1426 		return (error);
1427 	}
1428 	mtx_lock(&mountlist_mtx);
1429 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1430 	mtx_unlock(&mountlist_mtx);
1431 	EVENTHANDLER_INVOKE(vfs_unmounted, mp, td);
1432 	if (coveredvp != NULL) {
1433 		coveredvp->v_mountedhere = NULL;
1434 		vput(coveredvp);
1435 	}
1436 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1437 	if (mp == rootdevmp)
1438 		rootdevmp = NULL;
1439 	vfs_mount_destroy(mp);
1440 	return (0);
1441 }
1442 
1443 /*
1444  * Report errors during filesystem mounting.
1445  */
1446 void
1447 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1448 {
1449 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1450 	va_list ap;
1451 	int error, len;
1452 	char *errmsg;
1453 
1454 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1455 	if (error || errmsg == NULL || len <= 0)
1456 		return;
1457 
1458 	va_start(ap, fmt);
1459 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1460 	va_end(ap);
1461 }
1462 
1463 void
1464 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1465 {
1466 	va_list ap;
1467 	int error, len;
1468 	char *errmsg;
1469 
1470 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1471 	if (error || errmsg == NULL || len <= 0)
1472 		return;
1473 
1474 	va_start(ap, fmt);
1475 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1476 	va_end(ap);
1477 }
1478 
1479 /*
1480  * ---------------------------------------------------------------------
1481  * Functions for querying mount options/arguments from filesystems.
1482  */
1483 
1484 /*
1485  * Check that no unknown options are given
1486  */
1487 int
1488 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1489 {
1490 	struct vfsopt *opt;
1491 	char errmsg[255];
1492 	const char **t, *p, *q;
1493 	int ret = 0;
1494 
1495 	TAILQ_FOREACH(opt, opts, link) {
1496 		p = opt->name;
1497 		q = NULL;
1498 		if (p[0] == 'n' && p[1] == 'o')
1499 			q = p + 2;
1500 		for(t = global_opts; *t != NULL; t++) {
1501 			if (strcmp(*t, p) == 0)
1502 				break;
1503 			if (q != NULL) {
1504 				if (strcmp(*t, q) == 0)
1505 					break;
1506 			}
1507 		}
1508 		if (*t != NULL)
1509 			continue;
1510 		for(t = legal; *t != NULL; t++) {
1511 			if (strcmp(*t, p) == 0)
1512 				break;
1513 			if (q != NULL) {
1514 				if (strcmp(*t, q) == 0)
1515 					break;
1516 			}
1517 		}
1518 		if (*t != NULL)
1519 			continue;
1520 		snprintf(errmsg, sizeof(errmsg),
1521 		    "mount option <%s> is unknown", p);
1522 		ret = EINVAL;
1523 	}
1524 	if (ret != 0) {
1525 		TAILQ_FOREACH(opt, opts, link) {
1526 			if (strcmp(opt->name, "errmsg") == 0) {
1527 				strncpy((char *)opt->value, errmsg, opt->len);
1528 				break;
1529 			}
1530 		}
1531 		if (opt == NULL)
1532 			printf("%s\n", errmsg);
1533 	}
1534 	return (ret);
1535 }
1536 
1537 /*
1538  * Get a mount option by its name.
1539  *
1540  * Return 0 if the option was found, ENOENT otherwise.
1541  * If len is non-NULL it will be filled with the length
1542  * of the option. If buf is non-NULL, it will be filled
1543  * with the address of the option.
1544  */
1545 int
1546 vfs_getopt(opts, name, buf, len)
1547 	struct vfsoptlist *opts;
1548 	const char *name;
1549 	void **buf;
1550 	int *len;
1551 {
1552 	struct vfsopt *opt;
1553 
1554 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1555 
1556 	TAILQ_FOREACH(opt, opts, link) {
1557 		if (strcmp(name, opt->name) == 0) {
1558 			opt->seen = 1;
1559 			if (len != NULL)
1560 				*len = opt->len;
1561 			if (buf != NULL)
1562 				*buf = opt->value;
1563 			return (0);
1564 		}
1565 	}
1566 	return (ENOENT);
1567 }
1568 
1569 int
1570 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1571 {
1572 	struct vfsopt *opt;
1573 
1574 	if (opts == NULL)
1575 		return (-1);
1576 
1577 	TAILQ_FOREACH(opt, opts, link) {
1578 		if (strcmp(name, opt->name) == 0) {
1579 			opt->seen = 1;
1580 			return (opt->pos);
1581 		}
1582 	}
1583 	return (-1);
1584 }
1585 
1586 int
1587 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1588 {
1589 	char *opt_value, *vtp;
1590 	quad_t iv;
1591 	int error, opt_len;
1592 
1593 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1594 	if (error != 0)
1595 		return (error);
1596 	if (opt_len == 0 || opt_value == NULL)
1597 		return (EINVAL);
1598 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1599 		return (EINVAL);
1600 	iv = strtoq(opt_value, &vtp, 0);
1601 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1602 		return (EINVAL);
1603 	if (iv < 0)
1604 		return (EINVAL);
1605 	switch (vtp[0]) {
1606 	case 't':
1607 	case 'T':
1608 		iv *= 1024;
1609 	case 'g':
1610 	case 'G':
1611 		iv *= 1024;
1612 	case 'm':
1613 	case 'M':
1614 		iv *= 1024;
1615 	case 'k':
1616 	case 'K':
1617 		iv *= 1024;
1618 	case '\0':
1619 		break;
1620 	default:
1621 		return (EINVAL);
1622 	}
1623 	*value = iv;
1624 
1625 	return (0);
1626 }
1627 
1628 char *
1629 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1630 {
1631 	struct vfsopt *opt;
1632 
1633 	*error = 0;
1634 	TAILQ_FOREACH(opt, opts, link) {
1635 		if (strcmp(name, opt->name) != 0)
1636 			continue;
1637 		opt->seen = 1;
1638 		if (opt->len == 0 ||
1639 		    ((char *)opt->value)[opt->len - 1] != '\0') {
1640 			*error = EINVAL;
1641 			return (NULL);
1642 		}
1643 		return (opt->value);
1644 	}
1645 	*error = ENOENT;
1646 	return (NULL);
1647 }
1648 
1649 int
1650 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1651 	uint64_t val)
1652 {
1653 	struct vfsopt *opt;
1654 
1655 	TAILQ_FOREACH(opt, opts, link) {
1656 		if (strcmp(name, opt->name) == 0) {
1657 			opt->seen = 1;
1658 			if (w != NULL)
1659 				*w |= val;
1660 			return (1);
1661 		}
1662 	}
1663 	if (w != NULL)
1664 		*w &= ~val;
1665 	return (0);
1666 }
1667 
1668 int
1669 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1670 {
1671 	va_list ap;
1672 	struct vfsopt *opt;
1673 	int ret;
1674 
1675 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1676 
1677 	TAILQ_FOREACH(opt, opts, link) {
1678 		if (strcmp(name, opt->name) != 0)
1679 			continue;
1680 		opt->seen = 1;
1681 		if (opt->len == 0 || opt->value == NULL)
1682 			return (0);
1683 		if (((char *)opt->value)[opt->len - 1] != '\0')
1684 			return (0);
1685 		va_start(ap, fmt);
1686 		ret = vsscanf(opt->value, fmt, ap);
1687 		va_end(ap);
1688 		return (ret);
1689 	}
1690 	return (0);
1691 }
1692 
1693 int
1694 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
1695 {
1696 	struct vfsopt *opt;
1697 
1698 	TAILQ_FOREACH(opt, opts, link) {
1699 		if (strcmp(name, opt->name) != 0)
1700 			continue;
1701 		opt->seen = 1;
1702 		if (opt->value == NULL)
1703 			opt->len = len;
1704 		else {
1705 			if (opt->len != len)
1706 				return (EINVAL);
1707 			bcopy(value, opt->value, len);
1708 		}
1709 		return (0);
1710 	}
1711 	return (ENOENT);
1712 }
1713 
1714 int
1715 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
1716 {
1717 	struct vfsopt *opt;
1718 
1719 	TAILQ_FOREACH(opt, opts, link) {
1720 		if (strcmp(name, opt->name) != 0)
1721 			continue;
1722 		opt->seen = 1;
1723 		if (opt->value == NULL)
1724 			opt->len = len;
1725 		else {
1726 			if (opt->len < len)
1727 				return (EINVAL);
1728 			opt->len = len;
1729 			bcopy(value, opt->value, len);
1730 		}
1731 		return (0);
1732 	}
1733 	return (ENOENT);
1734 }
1735 
1736 int
1737 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
1738 {
1739 	struct vfsopt *opt;
1740 
1741 	TAILQ_FOREACH(opt, opts, link) {
1742 		if (strcmp(name, opt->name) != 0)
1743 			continue;
1744 		opt->seen = 1;
1745 		if (opt->value == NULL)
1746 			opt->len = strlen(value) + 1;
1747 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
1748 			return (EINVAL);
1749 		return (0);
1750 	}
1751 	return (ENOENT);
1752 }
1753 
1754 /*
1755  * Find and copy a mount option.
1756  *
1757  * The size of the buffer has to be specified
1758  * in len, if it is not the same length as the
1759  * mount option, EINVAL is returned.
1760  * Returns ENOENT if the option is not found.
1761  */
1762 int
1763 vfs_copyopt(opts, name, dest, len)
1764 	struct vfsoptlist *opts;
1765 	const char *name;
1766 	void *dest;
1767 	int len;
1768 {
1769 	struct vfsopt *opt;
1770 
1771 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1772 
1773 	TAILQ_FOREACH(opt, opts, link) {
1774 		if (strcmp(name, opt->name) == 0) {
1775 			opt->seen = 1;
1776 			if (len != opt->len)
1777 				return (EINVAL);
1778 			bcopy(opt->value, dest, opt->len);
1779 			return (0);
1780 		}
1781 	}
1782 	return (ENOENT);
1783 }
1784 
1785 int
1786 __vfs_statfs(struct mount *mp, struct statfs *sbp)
1787 {
1788 	int error;
1789 
1790 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat);
1791 	if (sbp != &mp->mnt_stat)
1792 		*sbp = mp->mnt_stat;
1793 	return (error);
1794 }
1795 
1796 void
1797 vfs_mountedfrom(struct mount *mp, const char *from)
1798 {
1799 
1800 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1801 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1802 	    sizeof mp->mnt_stat.f_mntfromname);
1803 }
1804 
1805 /*
1806  * ---------------------------------------------------------------------
1807  * This is the api for building mount args and mounting filesystems from
1808  * inside the kernel.
1809  *
1810  * The API works by accumulation of individual args.  First error is
1811  * latched.
1812  *
1813  * XXX: should be documented in new manpage kernel_mount(9)
1814  */
1815 
1816 /* A memory allocation which must be freed when we are done */
1817 struct mntaarg {
1818 	SLIST_ENTRY(mntaarg)	next;
1819 };
1820 
1821 /* The header for the mount arguments */
1822 struct mntarg {
1823 	struct iovec *v;
1824 	int len;
1825 	int error;
1826 	SLIST_HEAD(, mntaarg)	list;
1827 };
1828 
1829 /*
1830  * Add a boolean argument.
1831  *
1832  * flag is the boolean value.
1833  * name must start with "no".
1834  */
1835 struct mntarg *
1836 mount_argb(struct mntarg *ma, int flag, const char *name)
1837 {
1838 
1839 	KASSERT(name[0] == 'n' && name[1] == 'o',
1840 	    ("mount_argb(...,%s): name must start with 'no'", name));
1841 
1842 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1843 }
1844 
1845 /*
1846  * Add an argument printf style
1847  */
1848 struct mntarg *
1849 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1850 {
1851 	va_list ap;
1852 	struct mntaarg *maa;
1853 	struct sbuf *sb;
1854 	int len;
1855 
1856 	if (ma == NULL) {
1857 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1858 		SLIST_INIT(&ma->list);
1859 	}
1860 	if (ma->error)
1861 		return (ma);
1862 
1863 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1864 	    M_MOUNT, M_WAITOK);
1865 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1866 	ma->v[ma->len].iov_len = strlen(name) + 1;
1867 	ma->len++;
1868 
1869 	sb = sbuf_new_auto();
1870 	va_start(ap, fmt);
1871 	sbuf_vprintf(sb, fmt, ap);
1872 	va_end(ap);
1873 	sbuf_finish(sb);
1874 	len = sbuf_len(sb) + 1;
1875 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1876 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1877 	bcopy(sbuf_data(sb), maa + 1, len);
1878 	sbuf_delete(sb);
1879 
1880 	ma->v[ma->len].iov_base = maa + 1;
1881 	ma->v[ma->len].iov_len = len;
1882 	ma->len++;
1883 
1884 	return (ma);
1885 }
1886 
1887 /*
1888  * Add an argument which is a userland string.
1889  */
1890 struct mntarg *
1891 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1892 {
1893 	struct mntaarg *maa;
1894 	char *tbuf;
1895 
1896 	if (val == NULL)
1897 		return (ma);
1898 	if (ma == NULL) {
1899 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1900 		SLIST_INIT(&ma->list);
1901 	}
1902 	if (ma->error)
1903 		return (ma);
1904 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1905 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1906 	tbuf = (void *)(maa + 1);
1907 	ma->error = copyinstr(val, tbuf, len, NULL);
1908 	return (mount_arg(ma, name, tbuf, -1));
1909 }
1910 
1911 /*
1912  * Plain argument.
1913  *
1914  * If length is -1, treat value as a C string.
1915  */
1916 struct mntarg *
1917 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
1918 {
1919 
1920 	if (ma == NULL) {
1921 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1922 		SLIST_INIT(&ma->list);
1923 	}
1924 	if (ma->error)
1925 		return (ma);
1926 
1927 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1928 	    M_MOUNT, M_WAITOK);
1929 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1930 	ma->v[ma->len].iov_len = strlen(name) + 1;
1931 	ma->len++;
1932 
1933 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
1934 	if (len < 0)
1935 		ma->v[ma->len].iov_len = strlen(val) + 1;
1936 	else
1937 		ma->v[ma->len].iov_len = len;
1938 	ma->len++;
1939 	return (ma);
1940 }
1941 
1942 /*
1943  * Free a mntarg structure
1944  */
1945 static void
1946 free_mntarg(struct mntarg *ma)
1947 {
1948 	struct mntaarg *maa;
1949 
1950 	while (!SLIST_EMPTY(&ma->list)) {
1951 		maa = SLIST_FIRST(&ma->list);
1952 		SLIST_REMOVE_HEAD(&ma->list, next);
1953 		free(maa, M_MOUNT);
1954 	}
1955 	free(ma->v, M_MOUNT);
1956 	free(ma, M_MOUNT);
1957 }
1958 
1959 /*
1960  * Mount a filesystem
1961  */
1962 int
1963 kernel_mount(struct mntarg *ma, uint64_t flags)
1964 {
1965 	struct uio auio;
1966 	int error;
1967 
1968 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
1969 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
1970 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
1971 
1972 	auio.uio_iov = ma->v;
1973 	auio.uio_iovcnt = ma->len;
1974 	auio.uio_segflg = UIO_SYSSPACE;
1975 
1976 	error = ma->error;
1977 	if (!error)
1978 		error = vfs_donmount(curthread, flags, &auio);
1979 	free_mntarg(ma);
1980 	return (error);
1981 }
1982 
1983 /*
1984  * A printflike function to mount a filesystem.
1985  */
1986 int
1987 kernel_vmount(int flags, ...)
1988 {
1989 	struct mntarg *ma = NULL;
1990 	va_list ap;
1991 	const char *cp;
1992 	const void *vp;
1993 	int error;
1994 
1995 	va_start(ap, flags);
1996 	for (;;) {
1997 		cp = va_arg(ap, const char *);
1998 		if (cp == NULL)
1999 			break;
2000 		vp = va_arg(ap, const void *);
2001 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2002 	}
2003 	va_end(ap);
2004 
2005 	error = kernel_mount(ma, flags);
2006 	return (error);
2007 }
2008 
2009 void
2010 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
2011 {
2012 
2013 	bcopy(oexp, exp, sizeof(*oexp));
2014 	exp->ex_numsecflavors = 0;
2015 }
2016