xref: /freebsd/sys/kern/vfs_mount.c (revision 273c26a3c3bea87a241d6879abd4f991db180bf0)
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_lockref != 0)
526 		panic("vfs_mount_destroy: nonzero lock refcount");
527 	MNT_IUNLOCK(mp);
528 #ifdef MAC
529 	mac_mount_destroy(mp);
530 #endif
531 	if (mp->mnt_opt != NULL)
532 		vfs_freeopts(mp->mnt_opt);
533 	crfree(mp->mnt_cred);
534 	uma_zfree(mount_zone, mp);
535 }
536 
537 int
538 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
539 {
540 	struct vfsoptlist *optlist;
541 	struct vfsopt *opt, *tmp_opt;
542 	char *fstype, *fspath, *errmsg;
543 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
544 
545 	errmsg = fspath = NULL;
546 	errmsg_len = fspathlen = 0;
547 	errmsg_pos = -1;
548 
549 	error = vfs_buildopts(fsoptions, &optlist);
550 	if (error)
551 		return (error);
552 
553 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
554 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
555 
556 	/*
557 	 * We need these two options before the others,
558 	 * and they are mandatory for any filesystem.
559 	 * Ensure they are NUL terminated as well.
560 	 */
561 	fstypelen = 0;
562 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
563 	if (error || fstype[fstypelen - 1] != '\0') {
564 		error = EINVAL;
565 		if (errmsg != NULL)
566 			strncpy(errmsg, "Invalid fstype", errmsg_len);
567 		goto bail;
568 	}
569 	fspathlen = 0;
570 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
571 	if (error || fspath[fspathlen - 1] != '\0') {
572 		error = EINVAL;
573 		if (errmsg != NULL)
574 			strncpy(errmsg, "Invalid fspath", errmsg_len);
575 		goto bail;
576 	}
577 
578 	/*
579 	 * We need to see if we have the "update" option
580 	 * before we call vfs_domount(), since vfs_domount() has special
581 	 * logic based on MNT_UPDATE.  This is very important
582 	 * when we want to update the root filesystem.
583 	 */
584 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
585 		if (strcmp(opt->name, "update") == 0) {
586 			fsflags |= MNT_UPDATE;
587 			vfs_freeopt(optlist, opt);
588 		}
589 		else if (strcmp(opt->name, "async") == 0)
590 			fsflags |= MNT_ASYNC;
591 		else if (strcmp(opt->name, "force") == 0) {
592 			fsflags |= MNT_FORCE;
593 			vfs_freeopt(optlist, opt);
594 		}
595 		else if (strcmp(opt->name, "reload") == 0) {
596 			fsflags |= MNT_RELOAD;
597 			vfs_freeopt(optlist, opt);
598 		}
599 		else if (strcmp(opt->name, "multilabel") == 0)
600 			fsflags |= MNT_MULTILABEL;
601 		else if (strcmp(opt->name, "noasync") == 0)
602 			fsflags &= ~MNT_ASYNC;
603 		else if (strcmp(opt->name, "noatime") == 0)
604 			fsflags |= MNT_NOATIME;
605 		else if (strcmp(opt->name, "atime") == 0) {
606 			free(opt->name, M_MOUNT);
607 			opt->name = strdup("nonoatime", M_MOUNT);
608 		}
609 		else if (strcmp(opt->name, "noclusterr") == 0)
610 			fsflags |= MNT_NOCLUSTERR;
611 		else if (strcmp(opt->name, "clusterr") == 0) {
612 			free(opt->name, M_MOUNT);
613 			opt->name = strdup("nonoclusterr", M_MOUNT);
614 		}
615 		else if (strcmp(opt->name, "noclusterw") == 0)
616 			fsflags |= MNT_NOCLUSTERW;
617 		else if (strcmp(opt->name, "clusterw") == 0) {
618 			free(opt->name, M_MOUNT);
619 			opt->name = strdup("nonoclusterw", M_MOUNT);
620 		}
621 		else if (strcmp(opt->name, "noexec") == 0)
622 			fsflags |= MNT_NOEXEC;
623 		else if (strcmp(opt->name, "exec") == 0) {
624 			free(opt->name, M_MOUNT);
625 			opt->name = strdup("nonoexec", M_MOUNT);
626 		}
627 		else if (strcmp(opt->name, "nosuid") == 0)
628 			fsflags |= MNT_NOSUID;
629 		else if (strcmp(opt->name, "suid") == 0) {
630 			free(opt->name, M_MOUNT);
631 			opt->name = strdup("nonosuid", M_MOUNT);
632 		}
633 		else if (strcmp(opt->name, "nosymfollow") == 0)
634 			fsflags |= MNT_NOSYMFOLLOW;
635 		else if (strcmp(opt->name, "symfollow") == 0) {
636 			free(opt->name, M_MOUNT);
637 			opt->name = strdup("nonosymfollow", M_MOUNT);
638 		}
639 		else if (strcmp(opt->name, "noro") == 0)
640 			fsflags &= ~MNT_RDONLY;
641 		else if (strcmp(opt->name, "rw") == 0)
642 			fsflags &= ~MNT_RDONLY;
643 		else if (strcmp(opt->name, "ro") == 0)
644 			fsflags |= MNT_RDONLY;
645 		else if (strcmp(opt->name, "rdonly") == 0) {
646 			free(opt->name, M_MOUNT);
647 			opt->name = strdup("ro", M_MOUNT);
648 			fsflags |= MNT_RDONLY;
649 		}
650 		else if (strcmp(opt->name, "suiddir") == 0)
651 			fsflags |= MNT_SUIDDIR;
652 		else if (strcmp(opt->name, "sync") == 0)
653 			fsflags |= MNT_SYNCHRONOUS;
654 		else if (strcmp(opt->name, "union") == 0)
655 			fsflags |= MNT_UNION;
656 		else if (strcmp(opt->name, "automounted") == 0) {
657 			fsflags |= MNT_AUTOMOUNTED;
658 			vfs_freeopt(optlist, opt);
659 		}
660 	}
661 
662 	/*
663 	 * Be ultra-paranoid about making sure the type and fspath
664 	 * variables will fit in our mp buffers, including the
665 	 * terminating NUL.
666 	 */
667 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
668 		error = ENAMETOOLONG;
669 		goto bail;
670 	}
671 
672 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
673 bail:
674 	/* copyout the errmsg */
675 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
676 	    && errmsg_len > 0 && errmsg != NULL) {
677 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
678 			bcopy(errmsg,
679 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
680 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
681 		} else {
682 			copyout(errmsg,
683 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
684 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
685 		}
686 	}
687 
688 	if (optlist != NULL)
689 		vfs_freeopts(optlist);
690 	return (error);
691 }
692 
693 /*
694  * Old mount API.
695  */
696 #ifndef _SYS_SYSPROTO_H_
697 struct mount_args {
698 	char	*type;
699 	char	*path;
700 	int	flags;
701 	caddr_t	data;
702 };
703 #endif
704 /* ARGSUSED */
705 int
706 sys_mount(td, uap)
707 	struct thread *td;
708 	struct mount_args /* {
709 		char *type;
710 		char *path;
711 		int flags;
712 		caddr_t data;
713 	} */ *uap;
714 {
715 	char *fstype;
716 	struct vfsconf *vfsp = NULL;
717 	struct mntarg *ma = NULL;
718 	uint64_t flags;
719 	int error;
720 
721 	/*
722 	 * Mount flags are now 64-bits. On 32-bit architectures only
723 	 * 32-bits are passed in, but from here on everything handles
724 	 * 64-bit flags correctly.
725 	 */
726 	flags = uap->flags;
727 
728 	AUDIT_ARG_FFLAGS(flags);
729 
730 	/*
731 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
732 	 * userspace to set this flag, but we must filter it out if we want
733 	 * MNT_UPDATE on the root file system to work.
734 	 * MNT_ROOTFS should only be set by the kernel when mounting its
735 	 * root file system.
736 	 */
737 	flags &= ~MNT_ROOTFS;
738 
739 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
740 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
741 	if (error) {
742 		free(fstype, M_TEMP);
743 		return (error);
744 	}
745 
746 	AUDIT_ARG_TEXT(fstype);
747 	vfsp = vfs_byname_kld(fstype, td, &error);
748 	free(fstype, M_TEMP);
749 	if (vfsp == NULL)
750 		return (ENOENT);
751 	if (vfsp->vfc_vfsops->vfs_cmount == NULL)
752 		return (EOPNOTSUPP);
753 
754 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
755 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
756 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
757 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
758 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
759 
760 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags);
761 	return (error);
762 }
763 
764 /*
765  * vfs_domount_first(): first file system mount (not update)
766  */
767 static int
768 vfs_domount_first(
769 	struct thread *td,		/* Calling thread. */
770 	struct vfsconf *vfsp,		/* File system type. */
771 	char *fspath,			/* Mount path. */
772 	struct vnode *vp,		/* Vnode to be covered. */
773 	uint64_t fsflags,		/* Flags common to all filesystems. */
774 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
775 	)
776 {
777 	struct vattr va;
778 	struct mount *mp;
779 	struct vnode *newdp;
780 	int error;
781 
782 	ASSERT_VOP_ELOCKED(vp, __func__);
783 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
784 
785 	/*
786 	 * If the user is not root, ensure that they own the directory
787 	 * onto which we are attempting to mount.
788 	 */
789 	error = VOP_GETATTR(vp, &va, td->td_ucred);
790 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
791 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0);
792 	if (error == 0)
793 		error = vinvalbuf(vp, V_SAVE, 0, 0);
794 	if (error == 0 && vp->v_type != VDIR)
795 		error = ENOTDIR;
796 	if (error == 0) {
797 		VI_LOCK(vp);
798 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
799 			vp->v_iflag |= VI_MOUNT;
800 		else
801 			error = EBUSY;
802 		VI_UNLOCK(vp);
803 	}
804 	if (error != 0) {
805 		vput(vp);
806 		return (error);
807 	}
808 	VOP_UNLOCK(vp, 0);
809 
810 	/* Allocate and initialize the filesystem. */
811 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
812 	/* XXXMAC: pass to vfs_mount_alloc? */
813 	mp->mnt_optnew = *optlist;
814 	/* Set the mount level flags. */
815 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
816 
817 	/*
818 	 * Mount the filesystem.
819 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
820 	 * get.  No freeing of cn_pnbuf.
821 	 */
822 	error = VFS_MOUNT(mp);
823 	if (error != 0) {
824 		vfs_unbusy(mp);
825 		vfs_mount_destroy(mp);
826 		VI_LOCK(vp);
827 		vp->v_iflag &= ~VI_MOUNT;
828 		VI_UNLOCK(vp);
829 		vrele(vp);
830 		return (error);
831 	}
832 
833 	if (mp->mnt_opt != NULL)
834 		vfs_freeopts(mp->mnt_opt);
835 	mp->mnt_opt = mp->mnt_optnew;
836 	*optlist = NULL;
837 	(void)VFS_STATFS(mp, &mp->mnt_stat);
838 
839 	/*
840 	 * Prevent external consumers of mount options from reading mnt_optnew.
841 	 */
842 	mp->mnt_optnew = NULL;
843 
844 	MNT_ILOCK(mp);
845 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
846 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
847 		mp->mnt_kern_flag |= MNTK_ASYNC;
848 	else
849 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
850 	MNT_IUNLOCK(mp);
851 
852 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
853 	cache_purge(vp);
854 	VI_LOCK(vp);
855 	vp->v_iflag &= ~VI_MOUNT;
856 	VI_UNLOCK(vp);
857 	vp->v_mountedhere = mp;
858 	/* Place the new filesystem at the end of the mount list. */
859 	mtx_lock(&mountlist_mtx);
860 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
861 	mtx_unlock(&mountlist_mtx);
862 	vfs_event_signal(NULL, VQ_MOUNT, 0);
863 	if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp))
864 		panic("mount: lost mount");
865 	VOP_UNLOCK(vp, 0);
866 	EVENTHANDLER_INVOKE(vfs_mounted, mp, newdp, td);
867 	VOP_UNLOCK(newdp, 0);
868 	mountcheckdirs(vp, newdp);
869 	vrele(newdp);
870 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
871 		vfs_allocate_syncvnode(mp);
872 	vfs_unbusy(mp);
873 	return (0);
874 }
875 
876 /*
877  * vfs_domount_update(): update of mounted file system
878  */
879 static int
880 vfs_domount_update(
881 	struct thread *td,		/* Calling thread. */
882 	struct vnode *vp,		/* Mount point vnode. */
883 	uint64_t fsflags,		/* Flags common to all filesystems. */
884 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
885 	)
886 {
887 	struct export_args export;
888 	void *bufp;
889 	struct mount *mp;
890 	int error, export_error, len;
891 	uint64_t flag;
892 
893 	ASSERT_VOP_ELOCKED(vp, __func__);
894 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
895 	mp = vp->v_mount;
896 
897 	if ((vp->v_vflag & VV_ROOT) == 0) {
898 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
899 		    == 0)
900 			error = EXDEV;
901 		else
902 			error = EINVAL;
903 		vput(vp);
904 		return (error);
905 	}
906 
907 	/*
908 	 * We only allow the filesystem to be reloaded if it
909 	 * is currently mounted read-only.
910 	 */
911 	flag = mp->mnt_flag;
912 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
913 		vput(vp);
914 		return (EOPNOTSUPP);	/* Needs translation */
915 	}
916 	/*
917 	 * Only privileged root, or (if MNT_USER is set) the user that
918 	 * did the original mount is permitted to update it.
919 	 */
920 	error = vfs_suser(mp, td);
921 	if (error != 0) {
922 		vput(vp);
923 		return (error);
924 	}
925 	if (vfs_busy(mp, MBF_NOWAIT)) {
926 		vput(vp);
927 		return (EBUSY);
928 	}
929 	VI_LOCK(vp);
930 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
931 		VI_UNLOCK(vp);
932 		vfs_unbusy(mp);
933 		vput(vp);
934 		return (EBUSY);
935 	}
936 	vp->v_iflag |= VI_MOUNT;
937 	VI_UNLOCK(vp);
938 	VOP_UNLOCK(vp, 0);
939 
940 	MNT_ILOCK(mp);
941 	mp->mnt_flag &= ~MNT_UPDATEMASK;
942 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
943 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
944 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
945 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
946 	MNT_IUNLOCK(mp);
947 	mp->mnt_optnew = *optlist;
948 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
949 
950 	/*
951 	 * Mount the filesystem.
952 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
953 	 * get.  No freeing of cn_pnbuf.
954 	 */
955 	error = VFS_MOUNT(mp);
956 
957 	export_error = 0;
958 	/* Process the export option. */
959 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
960 	    &len) == 0) {
961 		/* Assume that there is only 1 ABI for each length. */
962 		switch (len) {
963 		case (sizeof(struct oexport_args)):
964 			bzero(&export, sizeof(export));
965 			/* FALLTHROUGH */
966 		case (sizeof(export)):
967 			bcopy(bufp, &export, len);
968 			export_error = vfs_export(mp, &export);
969 			break;
970 		default:
971 			export_error = EINVAL;
972 			break;
973 		}
974 	}
975 
976 	MNT_ILOCK(mp);
977 	if (error == 0) {
978 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
979 		    MNT_SNAPSHOT);
980 	} else {
981 		/*
982 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
983 		 * because it is not part of MNT_UPDATEMASK, but it could have
984 		 * changed in the meantime if quotactl(2) was called.
985 		 * All in all we want current value of MNT_QUOTA, not the old
986 		 * one.
987 		 */
988 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
989 	}
990 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
991 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
992 		mp->mnt_kern_flag |= MNTK_ASYNC;
993 	else
994 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
995 	MNT_IUNLOCK(mp);
996 
997 	if (error != 0)
998 		goto end;
999 
1000 	if (mp->mnt_opt != NULL)
1001 		vfs_freeopts(mp->mnt_opt);
1002 	mp->mnt_opt = mp->mnt_optnew;
1003 	*optlist = NULL;
1004 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1005 	/*
1006 	 * Prevent external consumers of mount options from reading
1007 	 * mnt_optnew.
1008 	 */
1009 	mp->mnt_optnew = NULL;
1010 
1011 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1012 		vfs_allocate_syncvnode(mp);
1013 	else
1014 		vfs_deallocate_syncvnode(mp);
1015 end:
1016 	vfs_unbusy(mp);
1017 	VI_LOCK(vp);
1018 	vp->v_iflag &= ~VI_MOUNT;
1019 	VI_UNLOCK(vp);
1020 	vrele(vp);
1021 	return (error != 0 ? error : export_error);
1022 }
1023 
1024 /*
1025  * vfs_domount(): actually attempt a filesystem mount.
1026  */
1027 static int
1028 vfs_domount(
1029 	struct thread *td,		/* Calling thread. */
1030 	const char *fstype,		/* Filesystem type. */
1031 	char *fspath,			/* Mount path. */
1032 	uint64_t fsflags,		/* Flags common to all filesystems. */
1033 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1034 	)
1035 {
1036 	struct vfsconf *vfsp;
1037 	struct nameidata nd;
1038 	struct vnode *vp;
1039 	char *pathbuf;
1040 	int error;
1041 
1042 	/*
1043 	 * Be ultra-paranoid about making sure the type and fspath
1044 	 * variables will fit in our mp buffers, including the
1045 	 * terminating NUL.
1046 	 */
1047 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1048 		return (ENAMETOOLONG);
1049 
1050 	if (jailed(td->td_ucred) || usermount == 0) {
1051 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1052 			return (error);
1053 	}
1054 
1055 	/*
1056 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1057 	 */
1058 	if (fsflags & MNT_EXPORTED) {
1059 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1060 		if (error)
1061 			return (error);
1062 	}
1063 	if (fsflags & MNT_SUIDDIR) {
1064 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1065 		if (error)
1066 			return (error);
1067 	}
1068 	/*
1069 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1070 	 */
1071 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1072 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1073 			fsflags |= MNT_NOSUID | MNT_USER;
1074 	}
1075 
1076 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1077 	vfsp = NULL;
1078 	if ((fsflags & MNT_UPDATE) == 0) {
1079 		/* Don't try to load KLDs if we're mounting the root. */
1080 		if (fsflags & MNT_ROOTFS)
1081 			vfsp = vfs_byname(fstype);
1082 		else
1083 			vfsp = vfs_byname_kld(fstype, td, &error);
1084 		if (vfsp == NULL)
1085 			return (ENODEV);
1086 		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
1087 			return (EPERM);
1088 	}
1089 
1090 	/*
1091 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1092 	 */
1093 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1094 	    UIO_SYSSPACE, fspath, td);
1095 	error = namei(&nd);
1096 	if (error != 0)
1097 		return (error);
1098 	NDFREE(&nd, NDF_ONLY_PNBUF);
1099 	vp = nd.ni_vp;
1100 	if ((fsflags & MNT_UPDATE) == 0) {
1101 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1102 		strcpy(pathbuf, fspath);
1103 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1104 		/* debug.disablefullpath == 1 results in ENODEV */
1105 		if (error == 0 || error == ENODEV) {
1106 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1107 			    fsflags, optlist);
1108 		}
1109 		free(pathbuf, M_TEMP);
1110 	} else
1111 		error = vfs_domount_update(td, vp, fsflags, optlist);
1112 
1113 	return (error);
1114 }
1115 
1116 /*
1117  * Unmount a filesystem.
1118  *
1119  * Note: unmount takes a path to the vnode mounted on as argument, not
1120  * special file (as before).
1121  */
1122 #ifndef _SYS_SYSPROTO_H_
1123 struct unmount_args {
1124 	char	*path;
1125 	int	flags;
1126 };
1127 #endif
1128 /* ARGSUSED */
1129 int
1130 sys_unmount(struct thread *td, struct unmount_args *uap)
1131 {
1132 	struct nameidata nd;
1133 	struct mount *mp;
1134 	char *pathbuf;
1135 	int error, id0, id1;
1136 
1137 	AUDIT_ARG_VALUE(uap->flags);
1138 	if (jailed(td->td_ucred) || usermount == 0) {
1139 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1140 		if (error)
1141 			return (error);
1142 	}
1143 
1144 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1145 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1146 	if (error) {
1147 		free(pathbuf, M_TEMP);
1148 		return (error);
1149 	}
1150 	if (uap->flags & MNT_BYFSID) {
1151 		AUDIT_ARG_TEXT(pathbuf);
1152 		/* Decode the filesystem ID. */
1153 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1154 			free(pathbuf, M_TEMP);
1155 			return (EINVAL);
1156 		}
1157 
1158 		mtx_lock(&mountlist_mtx);
1159 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1160 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1161 			    mp->mnt_stat.f_fsid.val[1] == id1) {
1162 				vfs_ref(mp);
1163 				break;
1164 			}
1165 		}
1166 		mtx_unlock(&mountlist_mtx);
1167 	} else {
1168 		/*
1169 		 * Try to find global path for path argument.
1170 		 */
1171 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1172 		    UIO_SYSSPACE, pathbuf, td);
1173 		if (namei(&nd) == 0) {
1174 			NDFREE(&nd, NDF_ONLY_PNBUF);
1175 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1176 			    MNAMELEN);
1177 			if (error == 0 || error == ENODEV)
1178 				vput(nd.ni_vp);
1179 		}
1180 		mtx_lock(&mountlist_mtx);
1181 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1182 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1183 				vfs_ref(mp);
1184 				break;
1185 			}
1186 		}
1187 		mtx_unlock(&mountlist_mtx);
1188 	}
1189 	free(pathbuf, M_TEMP);
1190 	if (mp == NULL) {
1191 		/*
1192 		 * Previously we returned ENOENT for a nonexistent path and
1193 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1194 		 * now, so in the !MNT_BYFSID case return the more likely
1195 		 * EINVAL for compatibility.
1196 		 */
1197 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1198 	}
1199 
1200 	/*
1201 	 * Don't allow unmounting the root filesystem.
1202 	 */
1203 	if (mp->mnt_flag & MNT_ROOTFS) {
1204 		vfs_rel(mp);
1205 		return (EINVAL);
1206 	}
1207 	error = dounmount(mp, uap->flags, td);
1208 	return (error);
1209 }
1210 
1211 /*
1212  * Return error if any of the vnodes, ignoring the root vnode
1213  * and the syncer vnode, have non-zero usecount.
1214  *
1215  * This function is purely advisory - it can return false positives
1216  * and negatives.
1217  */
1218 static int
1219 vfs_check_usecounts(struct mount *mp)
1220 {
1221 	struct vnode *vp, *mvp;
1222 
1223 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1224 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1225 		    vp->v_usecount != 0) {
1226 			VI_UNLOCK(vp);
1227 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1228 			return (EBUSY);
1229 		}
1230 		VI_UNLOCK(vp);
1231 	}
1232 
1233 	return (0);
1234 }
1235 
1236 /*
1237  * Do the actual filesystem unmount.
1238  */
1239 int
1240 dounmount(struct mount *mp, int flags, struct thread *td)
1241 {
1242 	struct vnode *coveredvp, *fsrootvp;
1243 	int error;
1244 	uint64_t async_flag;
1245 	int mnt_gen_r;
1246 
1247 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1248 		mnt_gen_r = mp->mnt_gen;
1249 		VI_LOCK(coveredvp);
1250 		vholdl(coveredvp);
1251 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1252 		/*
1253 		 * Check for mp being unmounted while waiting for the
1254 		 * covered vnode lock.
1255 		 */
1256 		if (coveredvp->v_mountedhere != mp ||
1257 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1258 			VOP_UNLOCK(coveredvp, 0);
1259 			vdrop(coveredvp);
1260 			vfs_rel(mp);
1261 			return (EBUSY);
1262 		}
1263 	}
1264 
1265 	/*
1266 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1267 	 * original mount is permitted to unmount this filesystem.
1268 	 */
1269 	error = vfs_suser(mp, td);
1270 	if (error != 0) {
1271 		if (coveredvp != NULL) {
1272 			VOP_UNLOCK(coveredvp, 0);
1273 			vdrop(coveredvp);
1274 		}
1275 		vfs_rel(mp);
1276 		return (error);
1277 	}
1278 
1279 	vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1280 	MNT_ILOCK(mp);
1281 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1282 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
1283 		MNT_IUNLOCK(mp);
1284 		if (coveredvp != NULL) {
1285 			VOP_UNLOCK(coveredvp, 0);
1286 			vdrop(coveredvp);
1287 		}
1288 		vn_finished_write(mp);
1289 		return (EBUSY);
1290 	}
1291 	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1292 	if (flags & MNT_NONBUSY) {
1293 		MNT_IUNLOCK(mp);
1294 		error = vfs_check_usecounts(mp);
1295 		MNT_ILOCK(mp);
1296 		if (error != 0) {
1297 			mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ);
1298 			if (mp->mnt_kern_flag & MNTK_MWAIT) {
1299 				mp->mnt_kern_flag &= ~MNTK_MWAIT;
1300 				wakeup(mp);
1301 			}
1302 			MNT_IUNLOCK(mp);
1303 			if (coveredvp != NULL) {
1304 				VOP_UNLOCK(coveredvp, 0);
1305 				vdrop(coveredvp);
1306 			}
1307 			vn_finished_write(mp);
1308 			return (error);
1309 		}
1310 	}
1311 	/* Allow filesystems to detect that a forced unmount is in progress. */
1312 	if (flags & MNT_FORCE) {
1313 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1314 		MNT_IUNLOCK(mp);
1315 		/*
1316 		 * Must be done after setting MNTK_UNMOUNTF and before
1317 		 * waiting for mnt_lockref to become 0.
1318 		 */
1319 		VFS_PURGE(mp);
1320 		MNT_ILOCK(mp);
1321 	}
1322 	error = 0;
1323 	if (mp->mnt_lockref) {
1324 		mp->mnt_kern_flag |= MNTK_DRAINING;
1325 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1326 		    "mount drain", 0);
1327 	}
1328 	MNT_IUNLOCK(mp);
1329 	KASSERT(mp->mnt_lockref == 0,
1330 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1331 	    __func__, __FILE__, __LINE__));
1332 	KASSERT(error == 0,
1333 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1334 	    __func__, __FILE__, __LINE__));
1335 
1336 	if (mp->mnt_flag & MNT_EXPUBLIC)
1337 		vfs_setpublicfs(NULL, NULL, NULL);
1338 
1339 	/*
1340 	 * From now, we can claim that the use reference on the
1341 	 * coveredvp is ours, and the ref can be released only by
1342 	 * successfull unmount by us, or left for later unmount
1343 	 * attempt.  The previously acquired hold reference is no
1344 	 * longer needed to protect the vnode from reuse.
1345 	 */
1346 	if (coveredvp != NULL)
1347 		vdrop(coveredvp);
1348 
1349 	vfs_msync(mp, MNT_WAIT);
1350 	MNT_ILOCK(mp);
1351 	async_flag = mp->mnt_flag & MNT_ASYNC;
1352 	mp->mnt_flag &= ~MNT_ASYNC;
1353 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1354 	MNT_IUNLOCK(mp);
1355 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1356 	vfs_deallocate_syncvnode(mp);
1357 	/*
1358 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1359 	 * vnode to the covered vnode.  For non-forced unmounts we want
1360 	 * such references to cause an EBUSY error.
1361 	 */
1362 	if ((flags & MNT_FORCE) &&
1363 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1364 		if (mp->mnt_vnodecovered != NULL &&
1365 		    (mp->mnt_flag & MNT_IGNORE) == 0)
1366 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1367 		if (fsrootvp == rootvnode) {
1368 			vrele(rootvnode);
1369 			rootvnode = NULL;
1370 		}
1371 		vput(fsrootvp);
1372 	}
1373 	if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 ||
1374 	    (error = VFS_SYNC(mp, MNT_WAIT)) == 0)
1375 		error = VFS_UNMOUNT(mp, flags);
1376 	vn_finished_write(mp);
1377 	/*
1378 	 * If we failed to flush the dirty blocks for this mount point,
1379 	 * undo all the cdir/rdir and rootvnode changes we made above.
1380 	 * Unless we failed to do so because the device is reporting that
1381 	 * it doesn't exist anymore.
1382 	 */
1383 	if (error && error != ENXIO) {
1384 		if ((flags & MNT_FORCE) &&
1385 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1386 			if (mp->mnt_vnodecovered != NULL &&
1387 			    (mp->mnt_flag & MNT_IGNORE) == 0)
1388 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1389 			if (rootvnode == NULL) {
1390 				rootvnode = fsrootvp;
1391 				vref(rootvnode);
1392 			}
1393 			vput(fsrootvp);
1394 		}
1395 		MNT_ILOCK(mp);
1396 		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1397 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1398 			MNT_IUNLOCK(mp);
1399 			vfs_allocate_syncvnode(mp);
1400 			MNT_ILOCK(mp);
1401 		}
1402 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1403 		mp->mnt_flag |= async_flag;
1404 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1405 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1406 			mp->mnt_kern_flag |= MNTK_ASYNC;
1407 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1408 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1409 			wakeup(mp);
1410 		}
1411 		MNT_IUNLOCK(mp);
1412 		if (coveredvp)
1413 			VOP_UNLOCK(coveredvp, 0);
1414 		return (error);
1415 	}
1416 	mtx_lock(&mountlist_mtx);
1417 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1418 	mtx_unlock(&mountlist_mtx);
1419 	EVENTHANDLER_INVOKE(vfs_unmounted, mp, td);
1420 	if (coveredvp != NULL) {
1421 		coveredvp->v_mountedhere = NULL;
1422 		vput(coveredvp);
1423 	}
1424 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1425 	if (mp == rootdevmp)
1426 		rootdevmp = NULL;
1427 	vfs_mount_destroy(mp);
1428 	return (0);
1429 }
1430 
1431 /*
1432  * Report errors during filesystem mounting.
1433  */
1434 void
1435 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1436 {
1437 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1438 	va_list ap;
1439 	int error, len;
1440 	char *errmsg;
1441 
1442 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1443 	if (error || errmsg == NULL || len <= 0)
1444 		return;
1445 
1446 	va_start(ap, fmt);
1447 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1448 	va_end(ap);
1449 }
1450 
1451 void
1452 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1453 {
1454 	va_list ap;
1455 	int error, len;
1456 	char *errmsg;
1457 
1458 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1459 	if (error || errmsg == NULL || len <= 0)
1460 		return;
1461 
1462 	va_start(ap, fmt);
1463 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1464 	va_end(ap);
1465 }
1466 
1467 /*
1468  * ---------------------------------------------------------------------
1469  * Functions for querying mount options/arguments from filesystems.
1470  */
1471 
1472 /*
1473  * Check that no unknown options are given
1474  */
1475 int
1476 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1477 {
1478 	struct vfsopt *opt;
1479 	char errmsg[255];
1480 	const char **t, *p, *q;
1481 	int ret = 0;
1482 
1483 	TAILQ_FOREACH(opt, opts, link) {
1484 		p = opt->name;
1485 		q = NULL;
1486 		if (p[0] == 'n' && p[1] == 'o')
1487 			q = p + 2;
1488 		for(t = global_opts; *t != NULL; t++) {
1489 			if (strcmp(*t, p) == 0)
1490 				break;
1491 			if (q != NULL) {
1492 				if (strcmp(*t, q) == 0)
1493 					break;
1494 			}
1495 		}
1496 		if (*t != NULL)
1497 			continue;
1498 		for(t = legal; *t != NULL; t++) {
1499 			if (strcmp(*t, p) == 0)
1500 				break;
1501 			if (q != NULL) {
1502 				if (strcmp(*t, q) == 0)
1503 					break;
1504 			}
1505 		}
1506 		if (*t != NULL)
1507 			continue;
1508 		snprintf(errmsg, sizeof(errmsg),
1509 		    "mount option <%s> is unknown", p);
1510 		ret = EINVAL;
1511 	}
1512 	if (ret != 0) {
1513 		TAILQ_FOREACH(opt, opts, link) {
1514 			if (strcmp(opt->name, "errmsg") == 0) {
1515 				strncpy((char *)opt->value, errmsg, opt->len);
1516 				break;
1517 			}
1518 		}
1519 		if (opt == NULL)
1520 			printf("%s\n", errmsg);
1521 	}
1522 	return (ret);
1523 }
1524 
1525 /*
1526  * Get a mount option by its name.
1527  *
1528  * Return 0 if the option was found, ENOENT otherwise.
1529  * If len is non-NULL it will be filled with the length
1530  * of the option. If buf is non-NULL, it will be filled
1531  * with the address of the option.
1532  */
1533 int
1534 vfs_getopt(opts, name, buf, len)
1535 	struct vfsoptlist *opts;
1536 	const char *name;
1537 	void **buf;
1538 	int *len;
1539 {
1540 	struct vfsopt *opt;
1541 
1542 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1543 
1544 	TAILQ_FOREACH(opt, opts, link) {
1545 		if (strcmp(name, opt->name) == 0) {
1546 			opt->seen = 1;
1547 			if (len != NULL)
1548 				*len = opt->len;
1549 			if (buf != NULL)
1550 				*buf = opt->value;
1551 			return (0);
1552 		}
1553 	}
1554 	return (ENOENT);
1555 }
1556 
1557 int
1558 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1559 {
1560 	struct vfsopt *opt;
1561 
1562 	if (opts == NULL)
1563 		return (-1);
1564 
1565 	TAILQ_FOREACH(opt, opts, link) {
1566 		if (strcmp(name, opt->name) == 0) {
1567 			opt->seen = 1;
1568 			return (opt->pos);
1569 		}
1570 	}
1571 	return (-1);
1572 }
1573 
1574 int
1575 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1576 {
1577 	char *opt_value, *vtp;
1578 	quad_t iv;
1579 	int error, opt_len;
1580 
1581 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1582 	if (error != 0)
1583 		return (error);
1584 	if (opt_len == 0 || opt_value == NULL)
1585 		return (EINVAL);
1586 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1587 		return (EINVAL);
1588 	iv = strtoq(opt_value, &vtp, 0);
1589 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1590 		return (EINVAL);
1591 	if (iv < 0)
1592 		return (EINVAL);
1593 	switch (vtp[0]) {
1594 	case 't':
1595 	case 'T':
1596 		iv *= 1024;
1597 	case 'g':
1598 	case 'G':
1599 		iv *= 1024;
1600 	case 'm':
1601 	case 'M':
1602 		iv *= 1024;
1603 	case 'k':
1604 	case 'K':
1605 		iv *= 1024;
1606 	case '\0':
1607 		break;
1608 	default:
1609 		return (EINVAL);
1610 	}
1611 	*value = iv;
1612 
1613 	return (0);
1614 }
1615 
1616 char *
1617 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1618 {
1619 	struct vfsopt *opt;
1620 
1621 	*error = 0;
1622 	TAILQ_FOREACH(opt, opts, link) {
1623 		if (strcmp(name, opt->name) != 0)
1624 			continue;
1625 		opt->seen = 1;
1626 		if (opt->len == 0 ||
1627 		    ((char *)opt->value)[opt->len - 1] != '\0') {
1628 			*error = EINVAL;
1629 			return (NULL);
1630 		}
1631 		return (opt->value);
1632 	}
1633 	*error = ENOENT;
1634 	return (NULL);
1635 }
1636 
1637 int
1638 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1639 	uint64_t val)
1640 {
1641 	struct vfsopt *opt;
1642 
1643 	TAILQ_FOREACH(opt, opts, link) {
1644 		if (strcmp(name, opt->name) == 0) {
1645 			opt->seen = 1;
1646 			if (w != NULL)
1647 				*w |= val;
1648 			return (1);
1649 		}
1650 	}
1651 	if (w != NULL)
1652 		*w &= ~val;
1653 	return (0);
1654 }
1655 
1656 int
1657 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1658 {
1659 	va_list ap;
1660 	struct vfsopt *opt;
1661 	int ret;
1662 
1663 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1664 
1665 	TAILQ_FOREACH(opt, opts, link) {
1666 		if (strcmp(name, opt->name) != 0)
1667 			continue;
1668 		opt->seen = 1;
1669 		if (opt->len == 0 || opt->value == NULL)
1670 			return (0);
1671 		if (((char *)opt->value)[opt->len - 1] != '\0')
1672 			return (0);
1673 		va_start(ap, fmt);
1674 		ret = vsscanf(opt->value, fmt, ap);
1675 		va_end(ap);
1676 		return (ret);
1677 	}
1678 	return (0);
1679 }
1680 
1681 int
1682 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
1683 {
1684 	struct vfsopt *opt;
1685 
1686 	TAILQ_FOREACH(opt, opts, link) {
1687 		if (strcmp(name, opt->name) != 0)
1688 			continue;
1689 		opt->seen = 1;
1690 		if (opt->value == NULL)
1691 			opt->len = len;
1692 		else {
1693 			if (opt->len != len)
1694 				return (EINVAL);
1695 			bcopy(value, opt->value, len);
1696 		}
1697 		return (0);
1698 	}
1699 	return (ENOENT);
1700 }
1701 
1702 int
1703 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
1704 {
1705 	struct vfsopt *opt;
1706 
1707 	TAILQ_FOREACH(opt, opts, link) {
1708 		if (strcmp(name, opt->name) != 0)
1709 			continue;
1710 		opt->seen = 1;
1711 		if (opt->value == NULL)
1712 			opt->len = len;
1713 		else {
1714 			if (opt->len < len)
1715 				return (EINVAL);
1716 			opt->len = len;
1717 			bcopy(value, opt->value, len);
1718 		}
1719 		return (0);
1720 	}
1721 	return (ENOENT);
1722 }
1723 
1724 int
1725 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
1726 {
1727 	struct vfsopt *opt;
1728 
1729 	TAILQ_FOREACH(opt, opts, link) {
1730 		if (strcmp(name, opt->name) != 0)
1731 			continue;
1732 		opt->seen = 1;
1733 		if (opt->value == NULL)
1734 			opt->len = strlen(value) + 1;
1735 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
1736 			return (EINVAL);
1737 		return (0);
1738 	}
1739 	return (ENOENT);
1740 }
1741 
1742 /*
1743  * Find and copy a mount option.
1744  *
1745  * The size of the buffer has to be specified
1746  * in len, if it is not the same length as the
1747  * mount option, EINVAL is returned.
1748  * Returns ENOENT if the option is not found.
1749  */
1750 int
1751 vfs_copyopt(opts, name, dest, len)
1752 	struct vfsoptlist *opts;
1753 	const char *name;
1754 	void *dest;
1755 	int len;
1756 {
1757 	struct vfsopt *opt;
1758 
1759 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1760 
1761 	TAILQ_FOREACH(opt, opts, link) {
1762 		if (strcmp(name, opt->name) == 0) {
1763 			opt->seen = 1;
1764 			if (len != opt->len)
1765 				return (EINVAL);
1766 			bcopy(opt->value, dest, opt->len);
1767 			return (0);
1768 		}
1769 	}
1770 	return (ENOENT);
1771 }
1772 
1773 int
1774 __vfs_statfs(struct mount *mp, struct statfs *sbp)
1775 {
1776 	int error;
1777 
1778 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat);
1779 	if (sbp != &mp->mnt_stat)
1780 		*sbp = mp->mnt_stat;
1781 	return (error);
1782 }
1783 
1784 void
1785 vfs_mountedfrom(struct mount *mp, const char *from)
1786 {
1787 
1788 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1789 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1790 	    sizeof mp->mnt_stat.f_mntfromname);
1791 }
1792 
1793 /*
1794  * ---------------------------------------------------------------------
1795  * This is the api for building mount args and mounting filesystems from
1796  * inside the kernel.
1797  *
1798  * The API works by accumulation of individual args.  First error is
1799  * latched.
1800  *
1801  * XXX: should be documented in new manpage kernel_mount(9)
1802  */
1803 
1804 /* A memory allocation which must be freed when we are done */
1805 struct mntaarg {
1806 	SLIST_ENTRY(mntaarg)	next;
1807 };
1808 
1809 /* The header for the mount arguments */
1810 struct mntarg {
1811 	struct iovec *v;
1812 	int len;
1813 	int error;
1814 	SLIST_HEAD(, mntaarg)	list;
1815 };
1816 
1817 /*
1818  * Add a boolean argument.
1819  *
1820  * flag is the boolean value.
1821  * name must start with "no".
1822  */
1823 struct mntarg *
1824 mount_argb(struct mntarg *ma, int flag, const char *name)
1825 {
1826 
1827 	KASSERT(name[0] == 'n' && name[1] == 'o',
1828 	    ("mount_argb(...,%s): name must start with 'no'", name));
1829 
1830 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1831 }
1832 
1833 /*
1834  * Add an argument printf style
1835  */
1836 struct mntarg *
1837 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1838 {
1839 	va_list ap;
1840 	struct mntaarg *maa;
1841 	struct sbuf *sb;
1842 	int len;
1843 
1844 	if (ma == NULL) {
1845 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1846 		SLIST_INIT(&ma->list);
1847 	}
1848 	if (ma->error)
1849 		return (ma);
1850 
1851 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1852 	    M_MOUNT, M_WAITOK);
1853 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1854 	ma->v[ma->len].iov_len = strlen(name) + 1;
1855 	ma->len++;
1856 
1857 	sb = sbuf_new_auto();
1858 	va_start(ap, fmt);
1859 	sbuf_vprintf(sb, fmt, ap);
1860 	va_end(ap);
1861 	sbuf_finish(sb);
1862 	len = sbuf_len(sb) + 1;
1863 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1864 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1865 	bcopy(sbuf_data(sb), maa + 1, len);
1866 	sbuf_delete(sb);
1867 
1868 	ma->v[ma->len].iov_base = maa + 1;
1869 	ma->v[ma->len].iov_len = len;
1870 	ma->len++;
1871 
1872 	return (ma);
1873 }
1874 
1875 /*
1876  * Add an argument which is a userland string.
1877  */
1878 struct mntarg *
1879 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1880 {
1881 	struct mntaarg *maa;
1882 	char *tbuf;
1883 
1884 	if (val == NULL)
1885 		return (ma);
1886 	if (ma == NULL) {
1887 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1888 		SLIST_INIT(&ma->list);
1889 	}
1890 	if (ma->error)
1891 		return (ma);
1892 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1893 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1894 	tbuf = (void *)(maa + 1);
1895 	ma->error = copyinstr(val, tbuf, len, NULL);
1896 	return (mount_arg(ma, name, tbuf, -1));
1897 }
1898 
1899 /*
1900  * Plain argument.
1901  *
1902  * If length is -1, treat value as a C string.
1903  */
1904 struct mntarg *
1905 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
1906 {
1907 
1908 	if (ma == NULL) {
1909 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1910 		SLIST_INIT(&ma->list);
1911 	}
1912 	if (ma->error)
1913 		return (ma);
1914 
1915 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1916 	    M_MOUNT, M_WAITOK);
1917 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1918 	ma->v[ma->len].iov_len = strlen(name) + 1;
1919 	ma->len++;
1920 
1921 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
1922 	if (len < 0)
1923 		ma->v[ma->len].iov_len = strlen(val) + 1;
1924 	else
1925 		ma->v[ma->len].iov_len = len;
1926 	ma->len++;
1927 	return (ma);
1928 }
1929 
1930 /*
1931  * Free a mntarg structure
1932  */
1933 static void
1934 free_mntarg(struct mntarg *ma)
1935 {
1936 	struct mntaarg *maa;
1937 
1938 	while (!SLIST_EMPTY(&ma->list)) {
1939 		maa = SLIST_FIRST(&ma->list);
1940 		SLIST_REMOVE_HEAD(&ma->list, next);
1941 		free(maa, M_MOUNT);
1942 	}
1943 	free(ma->v, M_MOUNT);
1944 	free(ma, M_MOUNT);
1945 }
1946 
1947 /*
1948  * Mount a filesystem
1949  */
1950 int
1951 kernel_mount(struct mntarg *ma, uint64_t flags)
1952 {
1953 	struct uio auio;
1954 	int error;
1955 
1956 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
1957 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
1958 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
1959 
1960 	auio.uio_iov = ma->v;
1961 	auio.uio_iovcnt = ma->len;
1962 	auio.uio_segflg = UIO_SYSSPACE;
1963 
1964 	error = ma->error;
1965 	if (!error)
1966 		error = vfs_donmount(curthread, flags, &auio);
1967 	free_mntarg(ma);
1968 	return (error);
1969 }
1970 
1971 /*
1972  * A printflike function to mount a filesystem.
1973  */
1974 int
1975 kernel_vmount(int flags, ...)
1976 {
1977 	struct mntarg *ma = NULL;
1978 	va_list ap;
1979 	const char *cp;
1980 	const void *vp;
1981 	int error;
1982 
1983 	va_start(ap, flags);
1984 	for (;;) {
1985 		cp = va_arg(ap, const char *);
1986 		if (cp == NULL)
1987 			break;
1988 		vp = va_arg(ap, const void *);
1989 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
1990 	}
1991 	va_end(ap);
1992 
1993 	error = kernel_mount(ma, flags);
1994 	return (error);
1995 }
1996 
1997 void
1998 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
1999 {
2000 
2001 	bcopy(oexp, exp, sizeof(*oexp));
2002 	exp->ex_numsecflavors = 0;
2003 }
2004