xref: /freebsd/sys/kern/vfs_mount.c (revision 6be3386466ab79a84b48429ae66244f21526d3df)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1999-2004 Poul-Henning Kamp
5  * Copyright (c) 1999 Michael Smith
6  * Copyright (c) 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/conf.h>
44 #include <sys/smp.h>
45 #include <sys/devctl.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
48 #include <sys/jail.h>
49 #include <sys/kernel.h>
50 #include <sys/ktr.h>
51 #include <sys/libkern.h>
52 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/filedesc.h>
59 #include <sys/reboot.h>
60 #include <sys/sbuf.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysproto.h>
63 #include <sys/sx.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/systm.h>
67 #include <sys/vnode.h>
68 #include <vm/uma.h>
69 
70 #include <geom/geom.h>
71 
72 #include <machine/stdarg.h>
73 
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
76 
77 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
78 
79 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
80 		    uint64_t fsflags, struct vfsoptlist **optlist);
81 static void	free_mntarg(struct mntarg *ma);
82 
83 static int	usermount = 0;
84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
85     "Unprivileged users may mount and unmount file systems");
86 
87 static bool	default_autoro = false;
88 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
89     "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
90 
91 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
92 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
93 static uma_zone_t mount_zone;
94 
95 /* List of mounted filesystems. */
96 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
97 
98 /* For any iteration/modification of mountlist */
99 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
100 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
101 
102 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
103 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
104 
105 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
106 
107 /*
108  * Global opts, taken by all filesystems
109  */
110 static const char *global_opts[] = {
111 	"errmsg",
112 	"fstype",
113 	"fspath",
114 	"ro",
115 	"rw",
116 	"nosuid",
117 	"noexec",
118 	NULL
119 };
120 
121 static int
122 mount_init(void *mem, int size, int flags)
123 {
124 	struct mount *mp;
125 
126 	mp = (struct mount *)mem;
127 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
128 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
129 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
130 	mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
131 	mp->mnt_ref = 0;
132 	mp->mnt_vfs_ops = 1;
133 	mp->mnt_rootvnode = NULL;
134 	return (0);
135 }
136 
137 static void
138 mount_fini(void *mem, int size)
139 {
140 	struct mount *mp;
141 
142 	mp = (struct mount *)mem;
143 	uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
144 	lockdestroy(&mp->mnt_explock);
145 	mtx_destroy(&mp->mnt_listmtx);
146 	mtx_destroy(&mp->mnt_mtx);
147 }
148 
149 static void
150 vfs_mount_init(void *dummy __unused)
151 {
152 
153 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
154 	    NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
155 }
156 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
157 
158 /*
159  * ---------------------------------------------------------------------
160  * Functions for building and sanitizing the mount options
161  */
162 
163 /* Remove one mount option. */
164 static void
165 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
166 {
167 
168 	TAILQ_REMOVE(opts, opt, link);
169 	free(opt->name, M_MOUNT);
170 	if (opt->value != NULL)
171 		free(opt->value, M_MOUNT);
172 	free(opt, M_MOUNT);
173 }
174 
175 /* Release all resources related to the mount options. */
176 void
177 vfs_freeopts(struct vfsoptlist *opts)
178 {
179 	struct vfsopt *opt;
180 
181 	while (!TAILQ_EMPTY(opts)) {
182 		opt = TAILQ_FIRST(opts);
183 		vfs_freeopt(opts, opt);
184 	}
185 	free(opts, M_MOUNT);
186 }
187 
188 void
189 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
190 {
191 	struct vfsopt *opt, *temp;
192 
193 	if (opts == NULL)
194 		return;
195 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
196 		if (strcmp(opt->name, name) == 0)
197 			vfs_freeopt(opts, opt);
198 	}
199 }
200 
201 static int
202 vfs_isopt_ro(const char *opt)
203 {
204 
205 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
206 	    strcmp(opt, "norw") == 0)
207 		return (1);
208 	return (0);
209 }
210 
211 static int
212 vfs_isopt_rw(const char *opt)
213 {
214 
215 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
216 		return (1);
217 	return (0);
218 }
219 
220 /*
221  * Check if options are equal (with or without the "no" prefix).
222  */
223 static int
224 vfs_equalopts(const char *opt1, const char *opt2)
225 {
226 	char *p;
227 
228 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
229 	if (strcmp(opt1, opt2) == 0)
230 		return (1);
231 	/* "noopt" vs. "opt" */
232 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
233 		return (1);
234 	/* "opt" vs. "noopt" */
235 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
236 		return (1);
237 	while ((p = strchr(opt1, '.')) != NULL &&
238 	    !strncmp(opt1, opt2, ++p - opt1)) {
239 		opt2 += p - opt1;
240 		opt1 = p;
241 		/* "foo.noopt" vs. "foo.opt" */
242 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
243 			return (1);
244 		/* "foo.opt" vs. "foo.noopt" */
245 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
246 			return (1);
247 	}
248 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
249 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
250 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
251 		return (1);
252 	return (0);
253 }
254 
255 /*
256  * If a mount option is specified several times,
257  * (with or without the "no" prefix) only keep
258  * the last occurrence of it.
259  */
260 static void
261 vfs_sanitizeopts(struct vfsoptlist *opts)
262 {
263 	struct vfsopt *opt, *opt2, *tmp;
264 
265 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
266 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
267 		while (opt2 != NULL) {
268 			if (vfs_equalopts(opt->name, opt2->name)) {
269 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
270 				vfs_freeopt(opts, opt2);
271 				opt2 = tmp;
272 			} else {
273 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
274 			}
275 		}
276 	}
277 }
278 
279 /*
280  * Build a linked list of mount options from a struct uio.
281  */
282 int
283 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
284 {
285 	struct vfsoptlist *opts;
286 	struct vfsopt *opt;
287 	size_t memused, namelen, optlen;
288 	unsigned int i, iovcnt;
289 	int error;
290 
291 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
292 	TAILQ_INIT(opts);
293 	memused = 0;
294 	iovcnt = auio->uio_iovcnt;
295 	for (i = 0; i < iovcnt; i += 2) {
296 		namelen = auio->uio_iov[i].iov_len;
297 		optlen = auio->uio_iov[i + 1].iov_len;
298 		memused += sizeof(struct vfsopt) + optlen + namelen;
299 		/*
300 		 * Avoid consuming too much memory, and attempts to overflow
301 		 * memused.
302 		 */
303 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
304 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
305 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
306 			error = EINVAL;
307 			goto bad;
308 		}
309 
310 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
311 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
312 		opt->value = NULL;
313 		opt->len = 0;
314 		opt->pos = i / 2;
315 		opt->seen = 0;
316 
317 		/*
318 		 * Do this early, so jumps to "bad" will free the current
319 		 * option.
320 		 */
321 		TAILQ_INSERT_TAIL(opts, opt, link);
322 
323 		if (auio->uio_segflg == UIO_SYSSPACE) {
324 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
325 		} else {
326 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
327 			    namelen);
328 			if (error)
329 				goto bad;
330 		}
331 		/* Ensure names are null-terminated strings. */
332 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
333 			error = EINVAL;
334 			goto bad;
335 		}
336 		if (optlen != 0) {
337 			opt->len = optlen;
338 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
339 			if (auio->uio_segflg == UIO_SYSSPACE) {
340 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
341 				    optlen);
342 			} else {
343 				error = copyin(auio->uio_iov[i + 1].iov_base,
344 				    opt->value, optlen);
345 				if (error)
346 					goto bad;
347 			}
348 		}
349 	}
350 	vfs_sanitizeopts(opts);
351 	*options = opts;
352 	return (0);
353 bad:
354 	vfs_freeopts(opts);
355 	return (error);
356 }
357 
358 /*
359  * Merge the old mount options with the new ones passed
360  * in the MNT_UPDATE case.
361  *
362  * XXX: This function will keep a "nofoo" option in the new
363  * options.  E.g, if the option's canonical name is "foo",
364  * "nofoo" ends up in the mount point's active options.
365  */
366 static void
367 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
368 {
369 	struct vfsopt *opt, *new;
370 
371 	TAILQ_FOREACH(opt, oldopts, link) {
372 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
373 		new->name = strdup(opt->name, M_MOUNT);
374 		if (opt->len != 0) {
375 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
376 			bcopy(opt->value, new->value, opt->len);
377 		} else
378 			new->value = NULL;
379 		new->len = opt->len;
380 		new->seen = opt->seen;
381 		TAILQ_INSERT_HEAD(toopts, new, link);
382 	}
383 	vfs_sanitizeopts(toopts);
384 }
385 
386 /*
387  * Mount a filesystem.
388  */
389 #ifndef _SYS_SYSPROTO_H_
390 struct nmount_args {
391 	struct iovec *iovp;
392 	unsigned int iovcnt;
393 	int flags;
394 };
395 #endif
396 int
397 sys_nmount(struct thread *td, struct nmount_args *uap)
398 {
399 	struct uio *auio;
400 	int error;
401 	u_int iovcnt;
402 	uint64_t flags;
403 
404 	/*
405 	 * Mount flags are now 64-bits. On 32-bit archtectures only
406 	 * 32-bits are passed in, but from here on everything handles
407 	 * 64-bit flags correctly.
408 	 */
409 	flags = uap->flags;
410 
411 	AUDIT_ARG_FFLAGS(flags);
412 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
413 	    uap->iovp, uap->iovcnt, flags);
414 
415 	/*
416 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
417 	 * userspace to set this flag, but we must filter it out if we want
418 	 * MNT_UPDATE on the root file system to work.
419 	 * MNT_ROOTFS should only be set by the kernel when mounting its
420 	 * root file system.
421 	 */
422 	flags &= ~MNT_ROOTFS;
423 
424 	iovcnt = uap->iovcnt;
425 	/*
426 	 * Check that we have an even number of iovec's
427 	 * and that we have at least two options.
428 	 */
429 	if ((iovcnt & 1) || (iovcnt < 4)) {
430 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
431 		    uap->iovcnt);
432 		return (EINVAL);
433 	}
434 
435 	error = copyinuio(uap->iovp, iovcnt, &auio);
436 	if (error) {
437 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
438 		    __func__, error);
439 		return (error);
440 	}
441 	error = vfs_donmount(td, flags, auio);
442 
443 	free(auio, M_IOV);
444 	return (error);
445 }
446 
447 /*
448  * ---------------------------------------------------------------------
449  * Various utility functions
450  */
451 
452 /*
453  * Get a reference on a mount point from a vnode.
454  *
455  * The vnode is allowed to be passed unlocked and race against dooming. Note in
456  * such case there are no guarantees the referenced mount point will still be
457  * associated with it after the function returns.
458  */
459 struct mount *
460 vfs_ref_from_vp(struct vnode *vp)
461 {
462 	struct mount *mp;
463 	struct mount_pcpu *mpcpu;
464 
465 	mp = atomic_load_ptr(&vp->v_mount);
466 	if (__predict_false(mp == NULL)) {
467 		return (mp);
468 	}
469 	if (vfs_op_thread_enter(mp, mpcpu)) {
470 		if (__predict_true(mp == vp->v_mount)) {
471 			vfs_mp_count_add_pcpu(mpcpu, ref, 1);
472 			vfs_op_thread_exit(mp, mpcpu);
473 		} else {
474 			vfs_op_thread_exit(mp, mpcpu);
475 			mp = NULL;
476 		}
477 	} else {
478 		MNT_ILOCK(mp);
479 		if (mp == vp->v_mount) {
480 			MNT_REF(mp);
481 			MNT_IUNLOCK(mp);
482 		} else {
483 			MNT_IUNLOCK(mp);
484 			mp = NULL;
485 		}
486 	}
487 	return (mp);
488 }
489 
490 void
491 vfs_ref(struct mount *mp)
492 {
493 	struct mount_pcpu *mpcpu;
494 
495 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
496 	if (vfs_op_thread_enter(mp, mpcpu)) {
497 		vfs_mp_count_add_pcpu(mpcpu, ref, 1);
498 		vfs_op_thread_exit(mp, mpcpu);
499 		return;
500 	}
501 
502 	MNT_ILOCK(mp);
503 	MNT_REF(mp);
504 	MNT_IUNLOCK(mp);
505 }
506 
507 void
508 vfs_rel(struct mount *mp)
509 {
510 	struct mount_pcpu *mpcpu;
511 
512 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
513 	if (vfs_op_thread_enter(mp, mpcpu)) {
514 		vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
515 		vfs_op_thread_exit(mp, mpcpu);
516 		return;
517 	}
518 
519 	MNT_ILOCK(mp);
520 	MNT_REL(mp);
521 	MNT_IUNLOCK(mp);
522 }
523 
524 /*
525  * Allocate and initialize the mount point struct.
526  */
527 struct mount *
528 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
529     struct ucred *cred)
530 {
531 	struct mount *mp;
532 
533 	mp = uma_zalloc(mount_zone, M_WAITOK);
534 	bzero(&mp->mnt_startzero,
535 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
536 	mp->mnt_kern_flag = 0;
537 	mp->mnt_flag = 0;
538 	mp->mnt_rootvnode = NULL;
539 	mp->mnt_vnodecovered = NULL;
540 	mp->mnt_op = NULL;
541 	mp->mnt_vfc = NULL;
542 	TAILQ_INIT(&mp->mnt_nvnodelist);
543 	mp->mnt_nvnodelistsize = 0;
544 	TAILQ_INIT(&mp->mnt_lazyvnodelist);
545 	mp->mnt_lazyvnodelistsize = 0;
546 	if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
547 	    mp->mnt_writeopcount != 0)
548 		panic("%s: non-zero counters on new mp %p\n", __func__, mp);
549 	if (mp->mnt_vfs_ops != 1)
550 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
551 		    mp->mnt_vfs_ops);
552 	(void) vfs_busy(mp, MBF_NOWAIT);
553 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
554 	mp->mnt_op = vfsp->vfc_vfsops;
555 	mp->mnt_vfc = vfsp;
556 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
557 	mp->mnt_gen++;
558 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
559 	mp->mnt_vnodecovered = vp;
560 	mp->mnt_cred = crdup(cred);
561 	mp->mnt_stat.f_owner = cred->cr_uid;
562 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
563 	mp->mnt_iosize_max = DFLTPHYS;
564 #ifdef MAC
565 	mac_mount_init(mp);
566 	mac_mount_create(cred, mp);
567 #endif
568 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
569 	TAILQ_INIT(&mp->mnt_uppers);
570 	return (mp);
571 }
572 
573 /*
574  * Destroy the mount struct previously allocated by vfs_mount_alloc().
575  */
576 void
577 vfs_mount_destroy(struct mount *mp)
578 {
579 
580 	if (mp->mnt_vfs_ops == 0)
581 		panic("%s: entered with zero vfs_ops\n", __func__);
582 
583 	vfs_assert_mount_counters(mp);
584 
585 	MNT_ILOCK(mp);
586 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
587 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
588 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
589 		wakeup(mp);
590 	}
591 	while (mp->mnt_ref)
592 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
593 	KASSERT(mp->mnt_ref == 0,
594 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
595 	    __FILE__, __LINE__));
596 	if (mp->mnt_writeopcount != 0)
597 		panic("vfs_mount_destroy: nonzero writeopcount");
598 	if (mp->mnt_secondary_writes != 0)
599 		panic("vfs_mount_destroy: nonzero secondary_writes");
600 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
601 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
602 		struct vnode *vp;
603 
604 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
605 			vn_printf(vp, "dangling vnode ");
606 		panic("unmount: dangling vnode");
607 	}
608 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
609 	if (mp->mnt_nvnodelistsize != 0)
610 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
611 	if (mp->mnt_lazyvnodelistsize != 0)
612 		panic("vfs_mount_destroy: nonzero lazyvnodelistsize");
613 	if (mp->mnt_lockref != 0)
614 		panic("vfs_mount_destroy: nonzero lock refcount");
615 	MNT_IUNLOCK(mp);
616 
617 	if (mp->mnt_vfs_ops != 1)
618 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
619 		    mp->mnt_vfs_ops);
620 
621 	if (mp->mnt_rootvnode != NULL)
622 		panic("%s: mount point still has a root vnode %p\n", __func__,
623 		    mp->mnt_rootvnode);
624 
625 	if (mp->mnt_vnodecovered != NULL)
626 		vrele(mp->mnt_vnodecovered);
627 #ifdef MAC
628 	mac_mount_destroy(mp);
629 #endif
630 	if (mp->mnt_opt != NULL)
631 		vfs_freeopts(mp->mnt_opt);
632 	crfree(mp->mnt_cred);
633 	uma_zfree(mount_zone, mp);
634 }
635 
636 static bool
637 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
638 {
639 	/* This is an upgrade of an exisiting mount. */
640 	if ((fsflags & MNT_UPDATE) != 0)
641 		return (false);
642 	/* This is already an R/O mount. */
643 	if ((fsflags & MNT_RDONLY) != 0)
644 		return (false);
645 
646 	switch (error) {
647 	case ENODEV:	/* generic, geom, ... */
648 	case EACCES:	/* cam/scsi, ... */
649 	case EROFS:	/* md, mmcsd, ... */
650 		/*
651 		 * These errors can be returned by the storage layer to signal
652 		 * that the media is read-only.  No harm in the R/O mount
653 		 * attempt if the error was returned for some other reason.
654 		 */
655 		return (true);
656 	default:
657 		return (false);
658 	}
659 }
660 
661 int
662 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
663 {
664 	struct vfsoptlist *optlist;
665 	struct vfsopt *opt, *tmp_opt;
666 	char *fstype, *fspath, *errmsg;
667 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
668 	bool autoro;
669 
670 	errmsg = fspath = NULL;
671 	errmsg_len = fspathlen = 0;
672 	errmsg_pos = -1;
673 	autoro = default_autoro;
674 
675 	error = vfs_buildopts(fsoptions, &optlist);
676 	if (error)
677 		return (error);
678 
679 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
680 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
681 
682 	/*
683 	 * We need these two options before the others,
684 	 * and they are mandatory for any filesystem.
685 	 * Ensure they are NUL terminated as well.
686 	 */
687 	fstypelen = 0;
688 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
689 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
690 		error = EINVAL;
691 		if (errmsg != NULL)
692 			strncpy(errmsg, "Invalid fstype", errmsg_len);
693 		goto bail;
694 	}
695 	fspathlen = 0;
696 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
697 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
698 		error = EINVAL;
699 		if (errmsg != NULL)
700 			strncpy(errmsg, "Invalid fspath", errmsg_len);
701 		goto bail;
702 	}
703 
704 	/*
705 	 * We need to see if we have the "update" option
706 	 * before we call vfs_domount(), since vfs_domount() has special
707 	 * logic based on MNT_UPDATE.  This is very important
708 	 * when we want to update the root filesystem.
709 	 */
710 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
711 		int do_freeopt = 0;
712 
713 		if (strcmp(opt->name, "update") == 0) {
714 			fsflags |= MNT_UPDATE;
715 			do_freeopt = 1;
716 		}
717 		else if (strcmp(opt->name, "async") == 0)
718 			fsflags |= MNT_ASYNC;
719 		else if (strcmp(opt->name, "force") == 0) {
720 			fsflags |= MNT_FORCE;
721 			do_freeopt = 1;
722 		}
723 		else if (strcmp(opt->name, "reload") == 0) {
724 			fsflags |= MNT_RELOAD;
725 			do_freeopt = 1;
726 		}
727 		else if (strcmp(opt->name, "multilabel") == 0)
728 			fsflags |= MNT_MULTILABEL;
729 		else if (strcmp(opt->name, "noasync") == 0)
730 			fsflags &= ~MNT_ASYNC;
731 		else if (strcmp(opt->name, "noatime") == 0)
732 			fsflags |= MNT_NOATIME;
733 		else if (strcmp(opt->name, "atime") == 0) {
734 			free(opt->name, M_MOUNT);
735 			opt->name = strdup("nonoatime", M_MOUNT);
736 		}
737 		else if (strcmp(opt->name, "noclusterr") == 0)
738 			fsflags |= MNT_NOCLUSTERR;
739 		else if (strcmp(opt->name, "clusterr") == 0) {
740 			free(opt->name, M_MOUNT);
741 			opt->name = strdup("nonoclusterr", M_MOUNT);
742 		}
743 		else if (strcmp(opt->name, "noclusterw") == 0)
744 			fsflags |= MNT_NOCLUSTERW;
745 		else if (strcmp(opt->name, "clusterw") == 0) {
746 			free(opt->name, M_MOUNT);
747 			opt->name = strdup("nonoclusterw", M_MOUNT);
748 		}
749 		else if (strcmp(opt->name, "noexec") == 0)
750 			fsflags |= MNT_NOEXEC;
751 		else if (strcmp(opt->name, "exec") == 0) {
752 			free(opt->name, M_MOUNT);
753 			opt->name = strdup("nonoexec", M_MOUNT);
754 		}
755 		else if (strcmp(opt->name, "nosuid") == 0)
756 			fsflags |= MNT_NOSUID;
757 		else if (strcmp(opt->name, "suid") == 0) {
758 			free(opt->name, M_MOUNT);
759 			opt->name = strdup("nonosuid", M_MOUNT);
760 		}
761 		else if (strcmp(opt->name, "nosymfollow") == 0)
762 			fsflags |= MNT_NOSYMFOLLOW;
763 		else if (strcmp(opt->name, "symfollow") == 0) {
764 			free(opt->name, M_MOUNT);
765 			opt->name = strdup("nonosymfollow", M_MOUNT);
766 		}
767 		else if (strcmp(opt->name, "noro") == 0) {
768 			fsflags &= ~MNT_RDONLY;
769 			autoro = false;
770 		}
771 		else if (strcmp(opt->name, "rw") == 0) {
772 			fsflags &= ~MNT_RDONLY;
773 			autoro = false;
774 		}
775 		else if (strcmp(opt->name, "ro") == 0) {
776 			fsflags |= MNT_RDONLY;
777 			autoro = false;
778 		}
779 		else if (strcmp(opt->name, "rdonly") == 0) {
780 			free(opt->name, M_MOUNT);
781 			opt->name = strdup("ro", M_MOUNT);
782 			fsflags |= MNT_RDONLY;
783 			autoro = false;
784 		}
785 		else if (strcmp(opt->name, "autoro") == 0) {
786 			do_freeopt = 1;
787 			autoro = true;
788 		}
789 		else if (strcmp(opt->name, "suiddir") == 0)
790 			fsflags |= MNT_SUIDDIR;
791 		else if (strcmp(opt->name, "sync") == 0)
792 			fsflags |= MNT_SYNCHRONOUS;
793 		else if (strcmp(opt->name, "union") == 0)
794 			fsflags |= MNT_UNION;
795 		else if (strcmp(opt->name, "automounted") == 0) {
796 			fsflags |= MNT_AUTOMOUNTED;
797 			do_freeopt = 1;
798 		} else if (strcmp(opt->name, "nocover") == 0) {
799 			fsflags |= MNT_NOCOVER;
800 			do_freeopt = 1;
801 		} else if (strcmp(opt->name, "cover") == 0) {
802 			fsflags &= ~MNT_NOCOVER;
803 			do_freeopt = 1;
804 		} else if (strcmp(opt->name, "emptydir") == 0) {
805 			fsflags |= MNT_EMPTYDIR;
806 			do_freeopt = 1;
807 		} else if (strcmp(opt->name, "noemptydir") == 0) {
808 			fsflags &= ~MNT_EMPTYDIR;
809 			do_freeopt = 1;
810 		}
811 		if (do_freeopt)
812 			vfs_freeopt(optlist, opt);
813 	}
814 
815 	/*
816 	 * Be ultra-paranoid about making sure the type and fspath
817 	 * variables will fit in our mp buffers, including the
818 	 * terminating NUL.
819 	 */
820 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
821 		error = ENAMETOOLONG;
822 		goto bail;
823 	}
824 
825 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
826 
827 	/*
828 	 * See if we can mount in the read-only mode if the error code suggests
829 	 * that it could be possible and the mount options allow for that.
830 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
831 	 * overridden by "autoro".
832 	 */
833 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
834 		printf("%s: R/W mount failed, possibly R/O media,"
835 		    " trying R/O mount\n", __func__);
836 		fsflags |= MNT_RDONLY;
837 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
838 	}
839 bail:
840 	/* copyout the errmsg */
841 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
842 	    && errmsg_len > 0 && errmsg != NULL) {
843 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
844 			bcopy(errmsg,
845 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
846 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
847 		} else {
848 			copyout(errmsg,
849 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
850 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
851 		}
852 	}
853 
854 	if (optlist != NULL)
855 		vfs_freeopts(optlist);
856 	return (error);
857 }
858 
859 /*
860  * Old mount API.
861  */
862 #ifndef _SYS_SYSPROTO_H_
863 struct mount_args {
864 	char	*type;
865 	char	*path;
866 	int	flags;
867 	caddr_t	data;
868 };
869 #endif
870 /* ARGSUSED */
871 int
872 sys_mount(struct thread *td, struct mount_args *uap)
873 {
874 	char *fstype;
875 	struct vfsconf *vfsp = NULL;
876 	struct mntarg *ma = NULL;
877 	uint64_t flags;
878 	int error;
879 
880 	/*
881 	 * Mount flags are now 64-bits. On 32-bit architectures only
882 	 * 32-bits are passed in, but from here on everything handles
883 	 * 64-bit flags correctly.
884 	 */
885 	flags = uap->flags;
886 
887 	AUDIT_ARG_FFLAGS(flags);
888 
889 	/*
890 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
891 	 * userspace to set this flag, but we must filter it out if we want
892 	 * MNT_UPDATE on the root file system to work.
893 	 * MNT_ROOTFS should only be set by the kernel when mounting its
894 	 * root file system.
895 	 */
896 	flags &= ~MNT_ROOTFS;
897 
898 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
899 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
900 	if (error) {
901 		free(fstype, M_TEMP);
902 		return (error);
903 	}
904 
905 	AUDIT_ARG_TEXT(fstype);
906 	vfsp = vfs_byname_kld(fstype, td, &error);
907 	free(fstype, M_TEMP);
908 	if (vfsp == NULL)
909 		return (ENOENT);
910 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
911 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
912 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
913 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
914 		return (EOPNOTSUPP);
915 
916 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
917 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
918 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
919 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
920 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
921 
922 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
923 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
924 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
925 }
926 
927 /*
928  * vfs_domount_first(): first file system mount (not update)
929  */
930 static int
931 vfs_domount_first(
932 	struct thread *td,		/* Calling thread. */
933 	struct vfsconf *vfsp,		/* File system type. */
934 	char *fspath,			/* Mount path. */
935 	struct vnode *vp,		/* Vnode to be covered. */
936 	uint64_t fsflags,		/* Flags common to all filesystems. */
937 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
938 	)
939 {
940 	struct vattr va;
941 	struct mount *mp;
942 	struct vnode *newdp, *rootvp;
943 	int error, error1;
944 	bool unmounted;
945 
946 	ASSERT_VOP_ELOCKED(vp, __func__);
947 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
948 
949 	if ((fsflags & MNT_EMPTYDIR) != 0) {
950 		error = vfs_emptydir(vp);
951 		if (error != 0) {
952 			vput(vp);
953 			return (error);
954 		}
955 	}
956 
957 	/*
958 	 * If the jail of the calling thread lacks permission for this type of
959 	 * file system, deny immediately.
960 	 */
961 	if (jailed(td->td_ucred) && !prison_allow(td->td_ucred,
962 	    vfsp->vfc_prison_flag)) {
963 		vput(vp);
964 		return (EPERM);
965 	}
966 
967 	/*
968 	 * If the user is not root, ensure that they own the directory
969 	 * onto which we are attempting to mount.
970 	 */
971 	error = VOP_GETATTR(vp, &va, td->td_ucred);
972 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
973 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
974 	if (error == 0)
975 		error = vinvalbuf(vp, V_SAVE, 0, 0);
976 	if (error == 0 && vp->v_type != VDIR)
977 		error = ENOTDIR;
978 	if (error == 0) {
979 		VI_LOCK(vp);
980 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
981 			vp->v_iflag |= VI_MOUNT;
982 		else
983 			error = EBUSY;
984 		VI_UNLOCK(vp);
985 	}
986 	if (error != 0) {
987 		vput(vp);
988 		return (error);
989 	}
990 	vn_seqc_write_begin(vp);
991 	VOP_UNLOCK(vp);
992 
993 	/* Allocate and initialize the filesystem. */
994 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
995 	/* XXXMAC: pass to vfs_mount_alloc? */
996 	mp->mnt_optnew = *optlist;
997 	/* Set the mount level flags. */
998 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
999 
1000 	/*
1001 	 * Mount the filesystem.
1002 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1003 	 * get.  No freeing of cn_pnbuf.
1004 	 */
1005 	error1 = 0;
1006 	unmounted = true;
1007 	if ((error = VFS_MOUNT(mp)) != 0 ||
1008 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1009 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1010 		rootvp = NULL;
1011 		if (error1 != 0) {
1012 			MPASS(error == 0);
1013 			rootvp = vfs_cache_root_clear(mp);
1014 			if (rootvp != NULL) {
1015 				vhold(rootvp);
1016 				vrele(rootvp);
1017 			}
1018 			(void)vn_start_write(NULL, &mp, V_WAIT);
1019 			MNT_ILOCK(mp);
1020 			mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1021 			MNT_IUNLOCK(mp);
1022 			VFS_PURGE(mp);
1023 			error = VFS_UNMOUNT(mp, 0);
1024 			vn_finished_write(mp);
1025 			if (error != 0) {
1026 				printf(
1027 		    "failed post-mount (%d): rollback unmount returned %d\n",
1028 				    error1, error);
1029 				unmounted = false;
1030 			}
1031 			error = error1;
1032 		}
1033 		vfs_unbusy(mp);
1034 		mp->mnt_vnodecovered = NULL;
1035 		if (unmounted) {
1036 			/* XXXKIB wait for mnt_lockref drain? */
1037 			vfs_mount_destroy(mp);
1038 		}
1039 		VI_LOCK(vp);
1040 		vp->v_iflag &= ~VI_MOUNT;
1041 		VI_UNLOCK(vp);
1042 		if (rootvp != NULL) {
1043 			vn_seqc_write_end(rootvp);
1044 			vdrop(rootvp);
1045 		}
1046 		vn_seqc_write_end(vp);
1047 		vrele(vp);
1048 		return (error);
1049 	}
1050 	vn_seqc_write_begin(newdp);
1051 	VOP_UNLOCK(newdp);
1052 
1053 	if (mp->mnt_opt != NULL)
1054 		vfs_freeopts(mp->mnt_opt);
1055 	mp->mnt_opt = mp->mnt_optnew;
1056 	*optlist = NULL;
1057 
1058 	/*
1059 	 * Prevent external consumers of mount options from reading mnt_optnew.
1060 	 */
1061 	mp->mnt_optnew = NULL;
1062 
1063 	MNT_ILOCK(mp);
1064 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1065 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1066 		mp->mnt_kern_flag |= MNTK_ASYNC;
1067 	else
1068 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1069 	MNT_IUNLOCK(mp);
1070 
1071 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1072 	cache_purge(vp);
1073 	VI_LOCK(vp);
1074 	vp->v_iflag &= ~VI_MOUNT;
1075 	vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1076 	vp->v_mountedhere = mp;
1077 	VI_UNLOCK(vp);
1078 	/* Place the new filesystem at the end of the mount list. */
1079 	mtx_lock(&mountlist_mtx);
1080 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1081 	mtx_unlock(&mountlist_mtx);
1082 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1083 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1084 	VOP_UNLOCK(vp);
1085 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1086 	VOP_UNLOCK(newdp);
1087 	mount_devctl_event("MOUNT", mp, false);
1088 	mountcheckdirs(vp, newdp);
1089 	vn_seqc_write_end(vp);
1090 	vn_seqc_write_end(newdp);
1091 	vrele(newdp);
1092 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1093 		vfs_allocate_syncvnode(mp);
1094 	vfs_op_exit(mp);
1095 	vfs_unbusy(mp);
1096 	return (0);
1097 }
1098 
1099 /*
1100  * vfs_domount_update(): update of mounted file system
1101  */
1102 static int
1103 vfs_domount_update(
1104 	struct thread *td,		/* Calling thread. */
1105 	struct vnode *vp,		/* Mount point vnode. */
1106 	uint64_t fsflags,		/* Flags common to all filesystems. */
1107 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1108 	)
1109 {
1110 	struct export_args export;
1111 	struct o2export_args o2export;
1112 	struct vnode *rootvp;
1113 	void *bufp;
1114 	struct mount *mp;
1115 	int error, export_error, i, len;
1116 	uint64_t flag;
1117 	gid_t *grps;
1118 
1119 	ASSERT_VOP_ELOCKED(vp, __func__);
1120 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1121 	mp = vp->v_mount;
1122 
1123 	if ((vp->v_vflag & VV_ROOT) == 0) {
1124 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1125 		    == 0)
1126 			error = EXDEV;
1127 		else
1128 			error = EINVAL;
1129 		vput(vp);
1130 		return (error);
1131 	}
1132 
1133 	/*
1134 	 * We only allow the filesystem to be reloaded if it
1135 	 * is currently mounted read-only.
1136 	 */
1137 	flag = mp->mnt_flag;
1138 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1139 		vput(vp);
1140 		return (EOPNOTSUPP);	/* Needs translation */
1141 	}
1142 	/*
1143 	 * Only privileged root, or (if MNT_USER is set) the user that
1144 	 * did the original mount is permitted to update it.
1145 	 */
1146 	error = vfs_suser(mp, td);
1147 	if (error != 0) {
1148 		vput(vp);
1149 		return (error);
1150 	}
1151 	if (vfs_busy(mp, MBF_NOWAIT)) {
1152 		vput(vp);
1153 		return (EBUSY);
1154 	}
1155 	VI_LOCK(vp);
1156 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1157 		VI_UNLOCK(vp);
1158 		vfs_unbusy(mp);
1159 		vput(vp);
1160 		return (EBUSY);
1161 	}
1162 	vp->v_iflag |= VI_MOUNT;
1163 	VI_UNLOCK(vp);
1164 	VOP_UNLOCK(vp);
1165 
1166 	vfs_op_enter(mp);
1167 	vn_seqc_write_begin(vp);
1168 
1169 	rootvp = NULL;
1170 	MNT_ILOCK(mp);
1171 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1172 		MNT_IUNLOCK(mp);
1173 		error = EBUSY;
1174 		goto end;
1175 	}
1176 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1177 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1178 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1179 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1180 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1181 	rootvp = vfs_cache_root_clear(mp);
1182 	MNT_IUNLOCK(mp);
1183 	mp->mnt_optnew = *optlist;
1184 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1185 
1186 	/*
1187 	 * Mount the filesystem.
1188 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1189 	 * get.  No freeing of cn_pnbuf.
1190 	 */
1191 	error = VFS_MOUNT(mp);
1192 
1193 	export_error = 0;
1194 	/* Process the export option. */
1195 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1196 	    &len) == 0) {
1197 		/* Assume that there is only 1 ABI for each length. */
1198 		switch (len) {
1199 		case (sizeof(struct oexport_args)):
1200 			bzero(&o2export, sizeof(o2export));
1201 			/* FALLTHROUGH */
1202 		case (sizeof(o2export)):
1203 			bcopy(bufp, &o2export, len);
1204 			export.ex_flags = (uint64_t)o2export.ex_flags;
1205 			export.ex_root = o2export.ex_root;
1206 			export.ex_uid = o2export.ex_anon.cr_uid;
1207 			export.ex_groups = NULL;
1208 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1209 			if (export.ex_ngroups > 0) {
1210 				if (export.ex_ngroups <= XU_NGROUPS) {
1211 					export.ex_groups = malloc(
1212 					    export.ex_ngroups * sizeof(gid_t),
1213 					    M_TEMP, M_WAITOK);
1214 					for (i = 0; i < export.ex_ngroups; i++)
1215 						export.ex_groups[i] =
1216 						  o2export.ex_anon.cr_groups[i];
1217 				} else
1218 					export_error = EINVAL;
1219 			} else if (export.ex_ngroups < 0)
1220 				export_error = EINVAL;
1221 			export.ex_addr = o2export.ex_addr;
1222 			export.ex_addrlen = o2export.ex_addrlen;
1223 			export.ex_mask = o2export.ex_mask;
1224 			export.ex_masklen = o2export.ex_masklen;
1225 			export.ex_indexfile = o2export.ex_indexfile;
1226 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1227 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1228 				for (i = 0; i < export.ex_numsecflavors; i++)
1229 					export.ex_secflavors[i] =
1230 					    o2export.ex_secflavors[i];
1231 			} else
1232 				export_error = EINVAL;
1233 			if (export_error == 0)
1234 				export_error = vfs_export(mp, &export);
1235 			free(export.ex_groups, M_TEMP);
1236 			break;
1237 		case (sizeof(export)):
1238 			bcopy(bufp, &export, len);
1239 			grps = NULL;
1240 			if (export.ex_ngroups > 0) {
1241 				if (export.ex_ngroups <= NGROUPS_MAX) {
1242 					grps = malloc(export.ex_ngroups *
1243 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1244 					export_error = copyin(export.ex_groups,
1245 					    grps, export.ex_ngroups *
1246 					    sizeof(gid_t));
1247 					if (export_error == 0)
1248 						export.ex_groups = grps;
1249 				} else
1250 					export_error = EINVAL;
1251 			} else if (export.ex_ngroups == 0)
1252 				export.ex_groups = NULL;
1253 			else
1254 				export_error = EINVAL;
1255 			if (export_error == 0)
1256 				export_error = vfs_export(mp, &export);
1257 			free(grps, M_TEMP);
1258 			break;
1259 		default:
1260 			export_error = EINVAL;
1261 			break;
1262 		}
1263 	}
1264 
1265 	MNT_ILOCK(mp);
1266 	if (error == 0) {
1267 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1268 		    MNT_SNAPSHOT);
1269 	} else {
1270 		/*
1271 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1272 		 * because it is not part of MNT_UPDATEMASK, but it could have
1273 		 * changed in the meantime if quotactl(2) was called.
1274 		 * All in all we want current value of MNT_QUOTA, not the old
1275 		 * one.
1276 		 */
1277 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1278 	}
1279 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1280 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1281 		mp->mnt_kern_flag |= MNTK_ASYNC;
1282 	else
1283 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1284 	MNT_IUNLOCK(mp);
1285 
1286 	if (error != 0)
1287 		goto end;
1288 
1289 	mount_devctl_event("REMOUNT", mp, true);
1290 	if (mp->mnt_opt != NULL)
1291 		vfs_freeopts(mp->mnt_opt);
1292 	mp->mnt_opt = mp->mnt_optnew;
1293 	*optlist = NULL;
1294 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1295 	/*
1296 	 * Prevent external consumers of mount options from reading
1297 	 * mnt_optnew.
1298 	 */
1299 	mp->mnt_optnew = NULL;
1300 
1301 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1302 		vfs_allocate_syncvnode(mp);
1303 	else
1304 		vfs_deallocate_syncvnode(mp);
1305 end:
1306 	vfs_op_exit(mp);
1307 	if (rootvp != NULL) {
1308 		vn_seqc_write_end(rootvp);
1309 		vrele(rootvp);
1310 	}
1311 	vn_seqc_write_end(vp);
1312 	vfs_unbusy(mp);
1313 	VI_LOCK(vp);
1314 	vp->v_iflag &= ~VI_MOUNT;
1315 	VI_UNLOCK(vp);
1316 	vrele(vp);
1317 	return (error != 0 ? error : export_error);
1318 }
1319 
1320 /*
1321  * vfs_domount(): actually attempt a filesystem mount.
1322  */
1323 static int
1324 vfs_domount(
1325 	struct thread *td,		/* Calling thread. */
1326 	const char *fstype,		/* Filesystem type. */
1327 	char *fspath,			/* Mount path. */
1328 	uint64_t fsflags,		/* Flags common to all filesystems. */
1329 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1330 	)
1331 {
1332 	struct vfsconf *vfsp;
1333 	struct nameidata nd;
1334 	struct vnode *vp;
1335 	char *pathbuf;
1336 	int error;
1337 
1338 	/*
1339 	 * Be ultra-paranoid about making sure the type and fspath
1340 	 * variables will fit in our mp buffers, including the
1341 	 * terminating NUL.
1342 	 */
1343 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1344 		return (ENAMETOOLONG);
1345 
1346 	if (jailed(td->td_ucred) || usermount == 0) {
1347 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1348 			return (error);
1349 	}
1350 
1351 	/*
1352 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1353 	 */
1354 	if (fsflags & MNT_EXPORTED) {
1355 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1356 		if (error)
1357 			return (error);
1358 	}
1359 	if (fsflags & MNT_SUIDDIR) {
1360 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1361 		if (error)
1362 			return (error);
1363 	}
1364 	/*
1365 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1366 	 */
1367 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1368 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1369 			fsflags |= MNT_NOSUID | MNT_USER;
1370 	}
1371 
1372 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1373 	vfsp = NULL;
1374 	if ((fsflags & MNT_UPDATE) == 0) {
1375 		/* Don't try to load KLDs if we're mounting the root. */
1376 		if (fsflags & MNT_ROOTFS)
1377 			vfsp = vfs_byname(fstype);
1378 		else
1379 			vfsp = vfs_byname_kld(fstype, td, &error);
1380 		if (vfsp == NULL)
1381 			return (ENODEV);
1382 	}
1383 
1384 	/*
1385 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1386 	 */
1387 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1388 	    UIO_SYSSPACE, fspath, td);
1389 	error = namei(&nd);
1390 	if (error != 0)
1391 		return (error);
1392 	NDFREE(&nd, NDF_ONLY_PNBUF);
1393 	vp = nd.ni_vp;
1394 	if ((fsflags & MNT_UPDATE) == 0) {
1395 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1396 		    (fsflags & MNT_NOCOVER) != 0) {
1397 			vput(vp);
1398 			return (EBUSY);
1399 		}
1400 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1401 		strcpy(pathbuf, fspath);
1402 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1403 		if (error == 0) {
1404 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1405 			    fsflags, optlist);
1406 		}
1407 		free(pathbuf, M_TEMP);
1408 	} else
1409 		error = vfs_domount_update(td, vp, fsflags, optlist);
1410 
1411 	return (error);
1412 }
1413 
1414 /*
1415  * Unmount a filesystem.
1416  *
1417  * Note: unmount takes a path to the vnode mounted on as argument, not
1418  * special file (as before).
1419  */
1420 #ifndef _SYS_SYSPROTO_H_
1421 struct unmount_args {
1422 	char	*path;
1423 	int	flags;
1424 };
1425 #endif
1426 /* ARGSUSED */
1427 int
1428 sys_unmount(struct thread *td, struct unmount_args *uap)
1429 {
1430 
1431 	return (kern_unmount(td, uap->path, uap->flags));
1432 }
1433 
1434 int
1435 kern_unmount(struct thread *td, const char *path, int flags)
1436 {
1437 	struct nameidata nd;
1438 	struct mount *mp;
1439 	char *pathbuf;
1440 	int error, id0, id1;
1441 
1442 	AUDIT_ARG_VALUE(flags);
1443 	if (jailed(td->td_ucred) || usermount == 0) {
1444 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1445 		if (error)
1446 			return (error);
1447 	}
1448 
1449 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1450 	error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1451 	if (error) {
1452 		free(pathbuf, M_TEMP);
1453 		return (error);
1454 	}
1455 	if (flags & MNT_BYFSID) {
1456 		AUDIT_ARG_TEXT(pathbuf);
1457 		/* Decode the filesystem ID. */
1458 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1459 			free(pathbuf, M_TEMP);
1460 			return (EINVAL);
1461 		}
1462 
1463 		mtx_lock(&mountlist_mtx);
1464 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1465 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1466 			    mp->mnt_stat.f_fsid.val[1] == id1) {
1467 				vfs_ref(mp);
1468 				break;
1469 			}
1470 		}
1471 		mtx_unlock(&mountlist_mtx);
1472 	} else {
1473 		/*
1474 		 * Try to find global path for path argument.
1475 		 */
1476 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1477 		    UIO_SYSSPACE, pathbuf, td);
1478 		if (namei(&nd) == 0) {
1479 			NDFREE(&nd, NDF_ONLY_PNBUF);
1480 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1481 			    MNAMELEN);
1482 			if (error == 0)
1483 				vput(nd.ni_vp);
1484 		}
1485 		mtx_lock(&mountlist_mtx);
1486 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1487 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1488 				vfs_ref(mp);
1489 				break;
1490 			}
1491 		}
1492 		mtx_unlock(&mountlist_mtx);
1493 	}
1494 	free(pathbuf, M_TEMP);
1495 	if (mp == NULL) {
1496 		/*
1497 		 * Previously we returned ENOENT for a nonexistent path and
1498 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1499 		 * now, so in the !MNT_BYFSID case return the more likely
1500 		 * EINVAL for compatibility.
1501 		 */
1502 		return ((flags & MNT_BYFSID) ? ENOENT : EINVAL);
1503 	}
1504 
1505 	/*
1506 	 * Don't allow unmounting the root filesystem.
1507 	 */
1508 	if (mp->mnt_flag & MNT_ROOTFS) {
1509 		vfs_rel(mp);
1510 		return (EINVAL);
1511 	}
1512 	error = dounmount(mp, flags, td);
1513 	return (error);
1514 }
1515 
1516 /*
1517  * Return error if any of the vnodes, ignoring the root vnode
1518  * and the syncer vnode, have non-zero usecount.
1519  *
1520  * This function is purely advisory - it can return false positives
1521  * and negatives.
1522  */
1523 static int
1524 vfs_check_usecounts(struct mount *mp)
1525 {
1526 	struct vnode *vp, *mvp;
1527 
1528 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1529 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1530 		    vp->v_usecount != 0) {
1531 			VI_UNLOCK(vp);
1532 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1533 			return (EBUSY);
1534 		}
1535 		VI_UNLOCK(vp);
1536 	}
1537 
1538 	return (0);
1539 }
1540 
1541 static void
1542 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1543 {
1544 
1545 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1546 	mp->mnt_kern_flag &= ~mntkflags;
1547 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1548 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1549 		wakeup(mp);
1550 	}
1551 	vfs_op_exit_locked(mp);
1552 	MNT_IUNLOCK(mp);
1553 	if (coveredvp != NULL) {
1554 		VOP_UNLOCK(coveredvp);
1555 		vdrop(coveredvp);
1556 	}
1557 	vn_finished_write(mp);
1558 }
1559 
1560 /*
1561  * There are various reference counters associated with the mount point.
1562  * Normally it is permitted to modify them without taking the mnt ilock,
1563  * but this behavior can be temporarily disabled if stable value is needed
1564  * or callers are expected to block (e.g. to not allow new users during
1565  * forced unmount).
1566  */
1567 void
1568 vfs_op_enter(struct mount *mp)
1569 {
1570 	struct mount_pcpu *mpcpu;
1571 	int cpu;
1572 
1573 	MNT_ILOCK(mp);
1574 	mp->mnt_vfs_ops++;
1575 	if (mp->mnt_vfs_ops > 1) {
1576 		MNT_IUNLOCK(mp);
1577 		return;
1578 	}
1579 	vfs_op_barrier_wait(mp);
1580 	CPU_FOREACH(cpu) {
1581 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1582 
1583 		mp->mnt_ref += mpcpu->mntp_ref;
1584 		mpcpu->mntp_ref = 0;
1585 
1586 		mp->mnt_lockref += mpcpu->mntp_lockref;
1587 		mpcpu->mntp_lockref = 0;
1588 
1589 		mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1590 		mpcpu->mntp_writeopcount = 0;
1591 	}
1592 	if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0)
1593 		panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n",
1594 		    __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount);
1595 	MNT_IUNLOCK(mp);
1596 	vfs_assert_mount_counters(mp);
1597 }
1598 
1599 void
1600 vfs_op_exit_locked(struct mount *mp)
1601 {
1602 
1603 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1604 
1605 	if (mp->mnt_vfs_ops <= 0)
1606 		panic("%s: invalid vfs_ops count %d for mp %p\n",
1607 		    __func__, mp->mnt_vfs_ops, mp);
1608 	mp->mnt_vfs_ops--;
1609 }
1610 
1611 void
1612 vfs_op_exit(struct mount *mp)
1613 {
1614 
1615 	MNT_ILOCK(mp);
1616 	vfs_op_exit_locked(mp);
1617 	MNT_IUNLOCK(mp);
1618 }
1619 
1620 struct vfs_op_barrier_ipi {
1621 	struct mount *mp;
1622 	struct smp_rendezvous_cpus_retry_arg srcra;
1623 };
1624 
1625 static void
1626 vfs_op_action_func(void *arg)
1627 {
1628 	struct vfs_op_barrier_ipi *vfsopipi;
1629 	struct mount *mp;
1630 
1631 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1632 	mp = vfsopipi->mp;
1633 
1634 	if (!vfs_op_thread_entered(mp))
1635 		smp_rendezvous_cpus_done(arg);
1636 }
1637 
1638 static void
1639 vfs_op_wait_func(void *arg, int cpu)
1640 {
1641 	struct vfs_op_barrier_ipi *vfsopipi;
1642 	struct mount *mp;
1643 	struct mount_pcpu *mpcpu;
1644 
1645 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1646 	mp = vfsopipi->mp;
1647 
1648 	mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1649 	while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1650 		cpu_spinwait();
1651 }
1652 
1653 void
1654 vfs_op_barrier_wait(struct mount *mp)
1655 {
1656 	struct vfs_op_barrier_ipi vfsopipi;
1657 
1658 	vfsopipi.mp = mp;
1659 
1660 	smp_rendezvous_cpus_retry(all_cpus,
1661 	    smp_no_rendezvous_barrier,
1662 	    vfs_op_action_func,
1663 	    smp_no_rendezvous_barrier,
1664 	    vfs_op_wait_func,
1665 	    &vfsopipi.srcra);
1666 }
1667 
1668 #ifdef DIAGNOSTIC
1669 void
1670 vfs_assert_mount_counters(struct mount *mp)
1671 {
1672 	struct mount_pcpu *mpcpu;
1673 	int cpu;
1674 
1675 	if (mp->mnt_vfs_ops == 0)
1676 		return;
1677 
1678 	CPU_FOREACH(cpu) {
1679 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1680 		if (mpcpu->mntp_ref != 0 ||
1681 		    mpcpu->mntp_lockref != 0 ||
1682 		    mpcpu->mntp_writeopcount != 0)
1683 			vfs_dump_mount_counters(mp);
1684 	}
1685 }
1686 
1687 void
1688 vfs_dump_mount_counters(struct mount *mp)
1689 {
1690 	struct mount_pcpu *mpcpu;
1691 	int ref, lockref, writeopcount;
1692 	int cpu;
1693 
1694 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1695 
1696 	printf("        ref : ");
1697 	ref = mp->mnt_ref;
1698 	CPU_FOREACH(cpu) {
1699 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1700 		printf("%d ", mpcpu->mntp_ref);
1701 		ref += mpcpu->mntp_ref;
1702 	}
1703 	printf("\n");
1704 	printf("    lockref : ");
1705 	lockref = mp->mnt_lockref;
1706 	CPU_FOREACH(cpu) {
1707 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1708 		printf("%d ", mpcpu->mntp_lockref);
1709 		lockref += mpcpu->mntp_lockref;
1710 	}
1711 	printf("\n");
1712 	printf("writeopcount: ");
1713 	writeopcount = mp->mnt_writeopcount;
1714 	CPU_FOREACH(cpu) {
1715 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1716 		printf("%d ", mpcpu->mntp_writeopcount);
1717 		writeopcount += mpcpu->mntp_writeopcount;
1718 	}
1719 	printf("\n");
1720 
1721 	printf("counter       struct total\n");
1722 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1723 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1724 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1725 
1726 	panic("invalid counts on struct mount");
1727 }
1728 #endif
1729 
1730 int
1731 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1732 {
1733 	struct mount_pcpu *mpcpu;
1734 	int cpu, sum;
1735 
1736 	switch (which) {
1737 	case MNT_COUNT_REF:
1738 		sum = mp->mnt_ref;
1739 		break;
1740 	case MNT_COUNT_LOCKREF:
1741 		sum = mp->mnt_lockref;
1742 		break;
1743 	case MNT_COUNT_WRITEOPCOUNT:
1744 		sum = mp->mnt_writeopcount;
1745 		break;
1746 	}
1747 
1748 	CPU_FOREACH(cpu) {
1749 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1750 		switch (which) {
1751 		case MNT_COUNT_REF:
1752 			sum += mpcpu->mntp_ref;
1753 			break;
1754 		case MNT_COUNT_LOCKREF:
1755 			sum += mpcpu->mntp_lockref;
1756 			break;
1757 		case MNT_COUNT_WRITEOPCOUNT:
1758 			sum += mpcpu->mntp_writeopcount;
1759 			break;
1760 		}
1761 	}
1762 	return (sum);
1763 }
1764 
1765 /*
1766  * Do the actual filesystem unmount.
1767  */
1768 int
1769 dounmount(struct mount *mp, int flags, struct thread *td)
1770 {
1771 	struct vnode *coveredvp, *rootvp;
1772 	int error;
1773 	uint64_t async_flag;
1774 	int mnt_gen_r;
1775 
1776 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1777 		mnt_gen_r = mp->mnt_gen;
1778 		VI_LOCK(coveredvp);
1779 		vholdl(coveredvp);
1780 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1781 		/*
1782 		 * Check for mp being unmounted while waiting for the
1783 		 * covered vnode lock.
1784 		 */
1785 		if (coveredvp->v_mountedhere != mp ||
1786 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1787 			VOP_UNLOCK(coveredvp);
1788 			vdrop(coveredvp);
1789 			vfs_rel(mp);
1790 			return (EBUSY);
1791 		}
1792 	}
1793 
1794 	/*
1795 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1796 	 * original mount is permitted to unmount this filesystem.
1797 	 */
1798 	error = vfs_suser(mp, td);
1799 	if (error != 0) {
1800 		if (coveredvp != NULL) {
1801 			VOP_UNLOCK(coveredvp);
1802 			vdrop(coveredvp);
1803 		}
1804 		vfs_rel(mp);
1805 		return (error);
1806 	}
1807 
1808 	vfs_op_enter(mp);
1809 
1810 	vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1811 	MNT_ILOCK(mp);
1812 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1813 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
1814 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
1815 		dounmount_cleanup(mp, coveredvp, 0);
1816 		return (EBUSY);
1817 	}
1818 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1819 	rootvp = vfs_cache_root_clear(mp);
1820 	if (coveredvp != NULL)
1821 		vn_seqc_write_begin(coveredvp);
1822 	if (flags & MNT_NONBUSY) {
1823 		MNT_IUNLOCK(mp);
1824 		error = vfs_check_usecounts(mp);
1825 		MNT_ILOCK(mp);
1826 		if (error != 0) {
1827 			vn_seqc_write_end(coveredvp);
1828 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1829 			if (rootvp != NULL) {
1830 				vn_seqc_write_end(rootvp);
1831 				vrele(rootvp);
1832 			}
1833 			return (error);
1834 		}
1835 	}
1836 	/* Allow filesystems to detect that a forced unmount is in progress. */
1837 	if (flags & MNT_FORCE) {
1838 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1839 		MNT_IUNLOCK(mp);
1840 		/*
1841 		 * Must be done after setting MNTK_UNMOUNTF and before
1842 		 * waiting for mnt_lockref to become 0.
1843 		 */
1844 		VFS_PURGE(mp);
1845 		MNT_ILOCK(mp);
1846 	}
1847 	error = 0;
1848 	if (mp->mnt_lockref) {
1849 		mp->mnt_kern_flag |= MNTK_DRAINING;
1850 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1851 		    "mount drain", 0);
1852 	}
1853 	MNT_IUNLOCK(mp);
1854 	KASSERT(mp->mnt_lockref == 0,
1855 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1856 	    __func__, __FILE__, __LINE__));
1857 	KASSERT(error == 0,
1858 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1859 	    __func__, __FILE__, __LINE__));
1860 
1861 	/*
1862 	 * We want to keep the vnode around so that we can vn_seqc_write_end
1863 	 * after we are done with unmount. Downgrade our reference to a mere
1864 	 * hold count so that we don't interefere with anything.
1865 	 */
1866 	if (rootvp != NULL) {
1867 		vhold(rootvp);
1868 		vrele(rootvp);
1869 	}
1870 
1871 	if (mp->mnt_flag & MNT_EXPUBLIC)
1872 		vfs_setpublicfs(NULL, NULL, NULL);
1873 
1874 	vfs_periodic(mp, MNT_WAIT);
1875 	MNT_ILOCK(mp);
1876 	async_flag = mp->mnt_flag & MNT_ASYNC;
1877 	mp->mnt_flag &= ~MNT_ASYNC;
1878 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1879 	MNT_IUNLOCK(mp);
1880 	vfs_deallocate_syncvnode(mp);
1881 	error = VFS_UNMOUNT(mp, flags);
1882 	vn_finished_write(mp);
1883 	/*
1884 	 * If we failed to flush the dirty blocks for this mount point,
1885 	 * undo all the cdir/rdir and rootvnode changes we made above.
1886 	 * Unless we failed to do so because the device is reporting that
1887 	 * it doesn't exist anymore.
1888 	 */
1889 	if (error && error != ENXIO) {
1890 		MNT_ILOCK(mp);
1891 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1892 			MNT_IUNLOCK(mp);
1893 			vfs_allocate_syncvnode(mp);
1894 			MNT_ILOCK(mp);
1895 		}
1896 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1897 		mp->mnt_flag |= async_flag;
1898 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1899 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1900 			mp->mnt_kern_flag |= MNTK_ASYNC;
1901 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1902 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1903 			wakeup(mp);
1904 		}
1905 		vfs_op_exit_locked(mp);
1906 		MNT_IUNLOCK(mp);
1907 		if (coveredvp) {
1908 			vn_seqc_write_end(coveredvp);
1909 			VOP_UNLOCK(coveredvp);
1910 			vdrop(coveredvp);
1911 		}
1912 		if (rootvp != NULL) {
1913 			vn_seqc_write_end(rootvp);
1914 			vdrop(rootvp);
1915 		}
1916 		return (error);
1917 	}
1918 	mtx_lock(&mountlist_mtx);
1919 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1920 	mtx_unlock(&mountlist_mtx);
1921 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1922 	if (coveredvp != NULL) {
1923 		VI_LOCK(coveredvp);
1924 		vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
1925 		coveredvp->v_mountedhere = NULL;
1926 		vn_seqc_write_end_locked(coveredvp);
1927 		VI_UNLOCK(coveredvp);
1928 		VOP_UNLOCK(coveredvp);
1929 		vdrop(coveredvp);
1930 	}
1931 	mount_devctl_event("UNMOUNT", mp, false);
1932 	if (rootvp != NULL) {
1933 		vn_seqc_write_end(rootvp);
1934 		vdrop(rootvp);
1935 	}
1936 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1937 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
1938 		vrele(rootvnode);
1939 		rootvnode = NULL;
1940 	}
1941 	if (mp == rootdevmp)
1942 		rootdevmp = NULL;
1943 	vfs_mount_destroy(mp);
1944 	return (0);
1945 }
1946 
1947 /*
1948  * Report errors during filesystem mounting.
1949  */
1950 void
1951 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1952 {
1953 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1954 	va_list ap;
1955 	int error, len;
1956 	char *errmsg;
1957 
1958 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1959 	if (error || errmsg == NULL || len <= 0)
1960 		return;
1961 
1962 	va_start(ap, fmt);
1963 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1964 	va_end(ap);
1965 }
1966 
1967 void
1968 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1969 {
1970 	va_list ap;
1971 	int error, len;
1972 	char *errmsg;
1973 
1974 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1975 	if (error || errmsg == NULL || len <= 0)
1976 		return;
1977 
1978 	va_start(ap, fmt);
1979 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1980 	va_end(ap);
1981 }
1982 
1983 /*
1984  * ---------------------------------------------------------------------
1985  * Functions for querying mount options/arguments from filesystems.
1986  */
1987 
1988 /*
1989  * Check that no unknown options are given
1990  */
1991 int
1992 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1993 {
1994 	struct vfsopt *opt;
1995 	char errmsg[255];
1996 	const char **t, *p, *q;
1997 	int ret = 0;
1998 
1999 	TAILQ_FOREACH(opt, opts, link) {
2000 		p = opt->name;
2001 		q = NULL;
2002 		if (p[0] == 'n' && p[1] == 'o')
2003 			q = p + 2;
2004 		for(t = global_opts; *t != NULL; t++) {
2005 			if (strcmp(*t, p) == 0)
2006 				break;
2007 			if (q != NULL) {
2008 				if (strcmp(*t, q) == 0)
2009 					break;
2010 			}
2011 		}
2012 		if (*t != NULL)
2013 			continue;
2014 		for(t = legal; *t != NULL; t++) {
2015 			if (strcmp(*t, p) == 0)
2016 				break;
2017 			if (q != NULL) {
2018 				if (strcmp(*t, q) == 0)
2019 					break;
2020 			}
2021 		}
2022 		if (*t != NULL)
2023 			continue;
2024 		snprintf(errmsg, sizeof(errmsg),
2025 		    "mount option <%s> is unknown", p);
2026 		ret = EINVAL;
2027 	}
2028 	if (ret != 0) {
2029 		TAILQ_FOREACH(opt, opts, link) {
2030 			if (strcmp(opt->name, "errmsg") == 0) {
2031 				strncpy((char *)opt->value, errmsg, opt->len);
2032 				break;
2033 			}
2034 		}
2035 		if (opt == NULL)
2036 			printf("%s\n", errmsg);
2037 	}
2038 	return (ret);
2039 }
2040 
2041 /*
2042  * Get a mount option by its name.
2043  *
2044  * Return 0 if the option was found, ENOENT otherwise.
2045  * If len is non-NULL it will be filled with the length
2046  * of the option. If buf is non-NULL, it will be filled
2047  * with the address of the option.
2048  */
2049 int
2050 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2051 {
2052 	struct vfsopt *opt;
2053 
2054 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2055 
2056 	TAILQ_FOREACH(opt, opts, link) {
2057 		if (strcmp(name, opt->name) == 0) {
2058 			opt->seen = 1;
2059 			if (len != NULL)
2060 				*len = opt->len;
2061 			if (buf != NULL)
2062 				*buf = opt->value;
2063 			return (0);
2064 		}
2065 	}
2066 	return (ENOENT);
2067 }
2068 
2069 int
2070 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2071 {
2072 	struct vfsopt *opt;
2073 
2074 	if (opts == NULL)
2075 		return (-1);
2076 
2077 	TAILQ_FOREACH(opt, opts, link) {
2078 		if (strcmp(name, opt->name) == 0) {
2079 			opt->seen = 1;
2080 			return (opt->pos);
2081 		}
2082 	}
2083 	return (-1);
2084 }
2085 
2086 int
2087 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2088 {
2089 	char *opt_value, *vtp;
2090 	quad_t iv;
2091 	int error, opt_len;
2092 
2093 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2094 	if (error != 0)
2095 		return (error);
2096 	if (opt_len == 0 || opt_value == NULL)
2097 		return (EINVAL);
2098 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2099 		return (EINVAL);
2100 	iv = strtoq(opt_value, &vtp, 0);
2101 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2102 		return (EINVAL);
2103 	if (iv < 0)
2104 		return (EINVAL);
2105 	switch (vtp[0]) {
2106 	case 't': case 'T':
2107 		iv *= 1024;
2108 		/* FALLTHROUGH */
2109 	case 'g': case 'G':
2110 		iv *= 1024;
2111 		/* FALLTHROUGH */
2112 	case 'm': case 'M':
2113 		iv *= 1024;
2114 		/* FALLTHROUGH */
2115 	case 'k': case 'K':
2116 		iv *= 1024;
2117 	case '\0':
2118 		break;
2119 	default:
2120 		return (EINVAL);
2121 	}
2122 	*value = iv;
2123 
2124 	return (0);
2125 }
2126 
2127 char *
2128 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2129 {
2130 	struct vfsopt *opt;
2131 
2132 	*error = 0;
2133 	TAILQ_FOREACH(opt, opts, link) {
2134 		if (strcmp(name, opt->name) != 0)
2135 			continue;
2136 		opt->seen = 1;
2137 		if (opt->len == 0 ||
2138 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2139 			*error = EINVAL;
2140 			return (NULL);
2141 		}
2142 		return (opt->value);
2143 	}
2144 	*error = ENOENT;
2145 	return (NULL);
2146 }
2147 
2148 int
2149 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2150 	uint64_t val)
2151 {
2152 	struct vfsopt *opt;
2153 
2154 	TAILQ_FOREACH(opt, opts, link) {
2155 		if (strcmp(name, opt->name) == 0) {
2156 			opt->seen = 1;
2157 			if (w != NULL)
2158 				*w |= val;
2159 			return (1);
2160 		}
2161 	}
2162 	if (w != NULL)
2163 		*w &= ~val;
2164 	return (0);
2165 }
2166 
2167 int
2168 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2169 {
2170 	va_list ap;
2171 	struct vfsopt *opt;
2172 	int ret;
2173 
2174 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2175 
2176 	TAILQ_FOREACH(opt, opts, link) {
2177 		if (strcmp(name, opt->name) != 0)
2178 			continue;
2179 		opt->seen = 1;
2180 		if (opt->len == 0 || opt->value == NULL)
2181 			return (0);
2182 		if (((char *)opt->value)[opt->len - 1] != '\0')
2183 			return (0);
2184 		va_start(ap, fmt);
2185 		ret = vsscanf(opt->value, fmt, ap);
2186 		va_end(ap);
2187 		return (ret);
2188 	}
2189 	return (0);
2190 }
2191 
2192 int
2193 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2194 {
2195 	struct vfsopt *opt;
2196 
2197 	TAILQ_FOREACH(opt, opts, link) {
2198 		if (strcmp(name, opt->name) != 0)
2199 			continue;
2200 		opt->seen = 1;
2201 		if (opt->value == NULL)
2202 			opt->len = len;
2203 		else {
2204 			if (opt->len != len)
2205 				return (EINVAL);
2206 			bcopy(value, opt->value, len);
2207 		}
2208 		return (0);
2209 	}
2210 	return (ENOENT);
2211 }
2212 
2213 int
2214 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2215 {
2216 	struct vfsopt *opt;
2217 
2218 	TAILQ_FOREACH(opt, opts, link) {
2219 		if (strcmp(name, opt->name) != 0)
2220 			continue;
2221 		opt->seen = 1;
2222 		if (opt->value == NULL)
2223 			opt->len = len;
2224 		else {
2225 			if (opt->len < len)
2226 				return (EINVAL);
2227 			opt->len = len;
2228 			bcopy(value, opt->value, len);
2229 		}
2230 		return (0);
2231 	}
2232 	return (ENOENT);
2233 }
2234 
2235 int
2236 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2237 {
2238 	struct vfsopt *opt;
2239 
2240 	TAILQ_FOREACH(opt, opts, link) {
2241 		if (strcmp(name, opt->name) != 0)
2242 			continue;
2243 		opt->seen = 1;
2244 		if (opt->value == NULL)
2245 			opt->len = strlen(value) + 1;
2246 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2247 			return (EINVAL);
2248 		return (0);
2249 	}
2250 	return (ENOENT);
2251 }
2252 
2253 /*
2254  * Find and copy a mount option.
2255  *
2256  * The size of the buffer has to be specified
2257  * in len, if it is not the same length as the
2258  * mount option, EINVAL is returned.
2259  * Returns ENOENT if the option is not found.
2260  */
2261 int
2262 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2263 {
2264 	struct vfsopt *opt;
2265 
2266 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2267 
2268 	TAILQ_FOREACH(opt, opts, link) {
2269 		if (strcmp(name, opt->name) == 0) {
2270 			opt->seen = 1;
2271 			if (len != opt->len)
2272 				return (EINVAL);
2273 			bcopy(opt->value, dest, opt->len);
2274 			return (0);
2275 		}
2276 	}
2277 	return (ENOENT);
2278 }
2279 
2280 int
2281 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2282 {
2283 
2284 	/*
2285 	 * Filesystems only fill in part of the structure for updates, we
2286 	 * have to read the entirety first to get all content.
2287 	 */
2288 	if (sbp != &mp->mnt_stat)
2289 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2290 
2291 	/*
2292 	 * Set these in case the underlying filesystem fails to do so.
2293 	 */
2294 	sbp->f_version = STATFS_VERSION;
2295 	sbp->f_namemax = NAME_MAX;
2296 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2297 
2298 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2299 }
2300 
2301 void
2302 vfs_mountedfrom(struct mount *mp, const char *from)
2303 {
2304 
2305 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2306 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2307 	    sizeof mp->mnt_stat.f_mntfromname);
2308 }
2309 
2310 /*
2311  * ---------------------------------------------------------------------
2312  * This is the api for building mount args and mounting filesystems from
2313  * inside the kernel.
2314  *
2315  * The API works by accumulation of individual args.  First error is
2316  * latched.
2317  *
2318  * XXX: should be documented in new manpage kernel_mount(9)
2319  */
2320 
2321 /* A memory allocation which must be freed when we are done */
2322 struct mntaarg {
2323 	SLIST_ENTRY(mntaarg)	next;
2324 };
2325 
2326 /* The header for the mount arguments */
2327 struct mntarg {
2328 	struct iovec *v;
2329 	int len;
2330 	int error;
2331 	SLIST_HEAD(, mntaarg)	list;
2332 };
2333 
2334 /*
2335  * Add a boolean argument.
2336  *
2337  * flag is the boolean value.
2338  * name must start with "no".
2339  */
2340 struct mntarg *
2341 mount_argb(struct mntarg *ma, int flag, const char *name)
2342 {
2343 
2344 	KASSERT(name[0] == 'n' && name[1] == 'o',
2345 	    ("mount_argb(...,%s): name must start with 'no'", name));
2346 
2347 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2348 }
2349 
2350 /*
2351  * Add an argument printf style
2352  */
2353 struct mntarg *
2354 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2355 {
2356 	va_list ap;
2357 	struct mntaarg *maa;
2358 	struct sbuf *sb;
2359 	int len;
2360 
2361 	if (ma == NULL) {
2362 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2363 		SLIST_INIT(&ma->list);
2364 	}
2365 	if (ma->error)
2366 		return (ma);
2367 
2368 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2369 	    M_MOUNT, M_WAITOK);
2370 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2371 	ma->v[ma->len].iov_len = strlen(name) + 1;
2372 	ma->len++;
2373 
2374 	sb = sbuf_new_auto();
2375 	va_start(ap, fmt);
2376 	sbuf_vprintf(sb, fmt, ap);
2377 	va_end(ap);
2378 	sbuf_finish(sb);
2379 	len = sbuf_len(sb) + 1;
2380 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2381 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2382 	bcopy(sbuf_data(sb), maa + 1, len);
2383 	sbuf_delete(sb);
2384 
2385 	ma->v[ma->len].iov_base = maa + 1;
2386 	ma->v[ma->len].iov_len = len;
2387 	ma->len++;
2388 
2389 	return (ma);
2390 }
2391 
2392 /*
2393  * Add an argument which is a userland string.
2394  */
2395 struct mntarg *
2396 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2397 {
2398 	struct mntaarg *maa;
2399 	char *tbuf;
2400 
2401 	if (val == NULL)
2402 		return (ma);
2403 	if (ma == NULL) {
2404 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2405 		SLIST_INIT(&ma->list);
2406 	}
2407 	if (ma->error)
2408 		return (ma);
2409 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2410 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2411 	tbuf = (void *)(maa + 1);
2412 	ma->error = copyinstr(val, tbuf, len, NULL);
2413 	return (mount_arg(ma, name, tbuf, -1));
2414 }
2415 
2416 /*
2417  * Plain argument.
2418  *
2419  * If length is -1, treat value as a C string.
2420  */
2421 struct mntarg *
2422 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2423 {
2424 
2425 	if (ma == NULL) {
2426 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2427 		SLIST_INIT(&ma->list);
2428 	}
2429 	if (ma->error)
2430 		return (ma);
2431 
2432 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2433 	    M_MOUNT, M_WAITOK);
2434 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2435 	ma->v[ma->len].iov_len = strlen(name) + 1;
2436 	ma->len++;
2437 
2438 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2439 	if (len < 0)
2440 		ma->v[ma->len].iov_len = strlen(val) + 1;
2441 	else
2442 		ma->v[ma->len].iov_len = len;
2443 	ma->len++;
2444 	return (ma);
2445 }
2446 
2447 /*
2448  * Free a mntarg structure
2449  */
2450 static void
2451 free_mntarg(struct mntarg *ma)
2452 {
2453 	struct mntaarg *maa;
2454 
2455 	while (!SLIST_EMPTY(&ma->list)) {
2456 		maa = SLIST_FIRST(&ma->list);
2457 		SLIST_REMOVE_HEAD(&ma->list, next);
2458 		free(maa, M_MOUNT);
2459 	}
2460 	free(ma->v, M_MOUNT);
2461 	free(ma, M_MOUNT);
2462 }
2463 
2464 /*
2465  * Mount a filesystem
2466  */
2467 int
2468 kernel_mount(struct mntarg *ma, uint64_t flags)
2469 {
2470 	struct uio auio;
2471 	int error;
2472 
2473 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2474 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2475 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2476 
2477 	auio.uio_iov = ma->v;
2478 	auio.uio_iovcnt = ma->len;
2479 	auio.uio_segflg = UIO_SYSSPACE;
2480 
2481 	error = ma->error;
2482 	if (!error)
2483 		error = vfs_donmount(curthread, flags, &auio);
2484 	free_mntarg(ma);
2485 	return (error);
2486 }
2487 
2488 /*
2489  * A printflike function to mount a filesystem.
2490  */
2491 int
2492 kernel_vmount(int flags, ...)
2493 {
2494 	struct mntarg *ma = NULL;
2495 	va_list ap;
2496 	const char *cp;
2497 	const void *vp;
2498 	int error;
2499 
2500 	va_start(ap, flags);
2501 	for (;;) {
2502 		cp = va_arg(ap, const char *);
2503 		if (cp == NULL)
2504 			break;
2505 		vp = va_arg(ap, const void *);
2506 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2507 	}
2508 	va_end(ap);
2509 
2510 	error = kernel_mount(ma, flags);
2511 	return (error);
2512 }
2513 
2514 /* Map from mount options to printable formats. */
2515 static struct mntoptnames optnames[] = {
2516 	MNTOPT_NAMES
2517 };
2518 
2519 static void
2520 mount_devctl_event_mntopt(struct sbuf *sb, const char *what, struct vfsoptlist *opts)
2521 {
2522 	struct vfsopt *opt;
2523 
2524 	if (opts == NULL || TAILQ_EMPTY(opts))
2525 		return;
2526 	sbuf_printf(sb, " %s=\"", what);
2527 	TAILQ_FOREACH(opt, opts, link) {
2528 		if (opt->name[0] == '\0' || (opt->len > 0 && *(char *)opt->value == '\0'))
2529 			continue;
2530 		devctl_safe_quote_sb(sb, opt->name);
2531 		if (opt->len > 0) {
2532 			sbuf_putc(sb, '=');
2533 			devctl_safe_quote_sb(sb, opt->value);
2534 		}
2535 		sbuf_putc(sb, ';');
2536 	}
2537 	sbuf_putc(sb, '"');
2538 }
2539 
2540 #define DEVCTL_LEN 1024
2541 static void
2542 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2543 {
2544 	const uint8_t *cp;
2545 	struct mntoptnames *fp;
2546 	struct sbuf sb;
2547 	struct statfs *sfp = &mp->mnt_stat;
2548 	char *buf;
2549 
2550 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2551 	if (buf == NULL)
2552 		return;
2553 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2554 	sbuf_cpy(&sb, "mount-point=\"");
2555 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2556 	sbuf_cat(&sb, "\" mount-dev=\"");
2557 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2558 	sbuf_cat(&sb, "\" mount-type=\"");
2559 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2560 	sbuf_cat(&sb, "\" fsid=0x");
2561 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2562 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2563 		sbuf_printf(&sb, "%02x", cp[i]);
2564 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2565 	for (fp = optnames; fp->o_opt != 0; fp++) {
2566 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2567 			sbuf_cat(&sb, fp->o_name);
2568 			sbuf_putc(&sb, ';');
2569 		}
2570 	}
2571 	sbuf_putc(&sb, '"');
2572 	mount_devctl_event_mntopt(&sb, "opt", mp->mnt_opt);
2573 	if (donew)
2574 		mount_devctl_event_mntopt(&sb, "optnew", mp->mnt_optnew);
2575 	sbuf_finish(&sb);
2576 
2577 	if (sbuf_error(&sb) == 0)
2578 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2579 	sbuf_delete(&sb);
2580 	free(buf, M_MOUNT);
2581 }
2582 
2583 /*
2584  * Suspend write operations on all local writeable filesystems.  Does
2585  * full sync of them in the process.
2586  *
2587  * Iterate over the mount points in reverse order, suspending most
2588  * recently mounted filesystems first.  It handles a case where a
2589  * filesystem mounted from a md(4) vnode-backed device should be
2590  * suspended before the filesystem that owns the vnode.
2591  */
2592 void
2593 suspend_all_fs(void)
2594 {
2595 	struct mount *mp;
2596 	int error;
2597 
2598 	mtx_lock(&mountlist_mtx);
2599 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2600 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2601 		if (error != 0)
2602 			continue;
2603 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2604 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2605 			mtx_lock(&mountlist_mtx);
2606 			vfs_unbusy(mp);
2607 			continue;
2608 		}
2609 		error = vfs_write_suspend(mp, 0);
2610 		if (error == 0) {
2611 			MNT_ILOCK(mp);
2612 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2613 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2614 			MNT_IUNLOCK(mp);
2615 			mtx_lock(&mountlist_mtx);
2616 		} else {
2617 			printf("suspend of %s failed, error %d\n",
2618 			    mp->mnt_stat.f_mntonname, error);
2619 			mtx_lock(&mountlist_mtx);
2620 			vfs_unbusy(mp);
2621 		}
2622 	}
2623 	mtx_unlock(&mountlist_mtx);
2624 }
2625 
2626 void
2627 resume_all_fs(void)
2628 {
2629 	struct mount *mp;
2630 
2631 	mtx_lock(&mountlist_mtx);
2632 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2633 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2634 			continue;
2635 		mtx_unlock(&mountlist_mtx);
2636 		MNT_ILOCK(mp);
2637 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2638 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2639 		MNT_IUNLOCK(mp);
2640 		vfs_write_resume(mp, 0);
2641 		mtx_lock(&mountlist_mtx);
2642 		vfs_unbusy(mp);
2643 	}
2644 	mtx_unlock(&mountlist_mtx);
2645 }
2646