xref: /freebsd/sys/kern/vfs_mount.c (revision ae41709ab46305df80f7f35bb478a3c8ebf22ebb)
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_thread_in_ops_pcpu = uma_zalloc_pcpu(pcpu_zone_4,
131 	    M_WAITOK | M_ZERO);
132 	mp->mnt_ref_pcpu = uma_zalloc_pcpu(pcpu_zone_4,
133 	    M_WAITOK | M_ZERO);
134 	mp->mnt_lockref_pcpu = uma_zalloc_pcpu(pcpu_zone_4,
135 	    M_WAITOK | M_ZERO);
136 	mp->mnt_writeopcount_pcpu = uma_zalloc_pcpu(pcpu_zone_4,
137 	    M_WAITOK | M_ZERO);
138 	mp->mnt_ref = 0;
139 	mp->mnt_vfs_ops = 1;
140 	mp->mnt_rootvnode = NULL;
141 	return (0);
142 }
143 
144 static void
145 mount_fini(void *mem, int size)
146 {
147 	struct mount *mp;
148 
149 	mp = (struct mount *)mem;
150 	uma_zfree_pcpu(pcpu_zone_4, mp->mnt_writeopcount_pcpu);
151 	uma_zfree_pcpu(pcpu_zone_4, mp->mnt_lockref_pcpu);
152 	uma_zfree_pcpu(pcpu_zone_4, mp->mnt_ref_pcpu);
153 	uma_zfree_pcpu(pcpu_zone_4, mp->mnt_thread_in_ops_pcpu);
154 	lockdestroy(&mp->mnt_explock);
155 	mtx_destroy(&mp->mnt_listmtx);
156 	mtx_destroy(&mp->mnt_mtx);
157 }
158 
159 static void
160 vfs_mount_init(void *dummy __unused)
161 {
162 
163 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
164 	    NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
165 }
166 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
167 
168 /*
169  * ---------------------------------------------------------------------
170  * Functions for building and sanitizing the mount options
171  */
172 
173 /* Remove one mount option. */
174 static void
175 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
176 {
177 
178 	TAILQ_REMOVE(opts, opt, link);
179 	free(opt->name, M_MOUNT);
180 	if (opt->value != NULL)
181 		free(opt->value, M_MOUNT);
182 	free(opt, M_MOUNT);
183 }
184 
185 /* Release all resources related to the mount options. */
186 void
187 vfs_freeopts(struct vfsoptlist *opts)
188 {
189 	struct vfsopt *opt;
190 
191 	while (!TAILQ_EMPTY(opts)) {
192 		opt = TAILQ_FIRST(opts);
193 		vfs_freeopt(opts, opt);
194 	}
195 	free(opts, M_MOUNT);
196 }
197 
198 void
199 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
200 {
201 	struct vfsopt *opt, *temp;
202 
203 	if (opts == NULL)
204 		return;
205 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
206 		if (strcmp(opt->name, name) == 0)
207 			vfs_freeopt(opts, opt);
208 	}
209 }
210 
211 static int
212 vfs_isopt_ro(const char *opt)
213 {
214 
215 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
216 	    strcmp(opt, "norw") == 0)
217 		return (1);
218 	return (0);
219 }
220 
221 static int
222 vfs_isopt_rw(const char *opt)
223 {
224 
225 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
226 		return (1);
227 	return (0);
228 }
229 
230 /*
231  * Check if options are equal (with or without the "no" prefix).
232  */
233 static int
234 vfs_equalopts(const char *opt1, const char *opt2)
235 {
236 	char *p;
237 
238 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
239 	if (strcmp(opt1, opt2) == 0)
240 		return (1);
241 	/* "noopt" vs. "opt" */
242 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
243 		return (1);
244 	/* "opt" vs. "noopt" */
245 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
246 		return (1);
247 	while ((p = strchr(opt1, '.')) != NULL &&
248 	    !strncmp(opt1, opt2, ++p - opt1)) {
249 		opt2 += p - opt1;
250 		opt1 = p;
251 		/* "foo.noopt" vs. "foo.opt" */
252 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
253 			return (1);
254 		/* "foo.opt" vs. "foo.noopt" */
255 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
256 			return (1);
257 	}
258 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
259 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
260 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
261 		return (1);
262 	return (0);
263 }
264 
265 /*
266  * If a mount option is specified several times,
267  * (with or without the "no" prefix) only keep
268  * the last occurrence of it.
269  */
270 static void
271 vfs_sanitizeopts(struct vfsoptlist *opts)
272 {
273 	struct vfsopt *opt, *opt2, *tmp;
274 
275 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
276 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
277 		while (opt2 != NULL) {
278 			if (vfs_equalopts(opt->name, opt2->name)) {
279 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
280 				vfs_freeopt(opts, opt2);
281 				opt2 = tmp;
282 			} else {
283 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
284 			}
285 		}
286 	}
287 }
288 
289 /*
290  * Build a linked list of mount options from a struct uio.
291  */
292 int
293 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
294 {
295 	struct vfsoptlist *opts;
296 	struct vfsopt *opt;
297 	size_t memused, namelen, optlen;
298 	unsigned int i, iovcnt;
299 	int error;
300 
301 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
302 	TAILQ_INIT(opts);
303 	memused = 0;
304 	iovcnt = auio->uio_iovcnt;
305 	for (i = 0; i < iovcnt; i += 2) {
306 		namelen = auio->uio_iov[i].iov_len;
307 		optlen = auio->uio_iov[i + 1].iov_len;
308 		memused += sizeof(struct vfsopt) + optlen + namelen;
309 		/*
310 		 * Avoid consuming too much memory, and attempts to overflow
311 		 * memused.
312 		 */
313 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
314 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
315 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
316 			error = EINVAL;
317 			goto bad;
318 		}
319 
320 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
321 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
322 		opt->value = NULL;
323 		opt->len = 0;
324 		opt->pos = i / 2;
325 		opt->seen = 0;
326 
327 		/*
328 		 * Do this early, so jumps to "bad" will free the current
329 		 * option.
330 		 */
331 		TAILQ_INSERT_TAIL(opts, opt, link);
332 
333 		if (auio->uio_segflg == UIO_SYSSPACE) {
334 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
335 		} else {
336 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
337 			    namelen);
338 			if (error)
339 				goto bad;
340 		}
341 		/* Ensure names are null-terminated strings. */
342 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
343 			error = EINVAL;
344 			goto bad;
345 		}
346 		if (optlen != 0) {
347 			opt->len = optlen;
348 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
349 			if (auio->uio_segflg == UIO_SYSSPACE) {
350 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
351 				    optlen);
352 			} else {
353 				error = copyin(auio->uio_iov[i + 1].iov_base,
354 				    opt->value, optlen);
355 				if (error)
356 					goto bad;
357 			}
358 		}
359 	}
360 	vfs_sanitizeopts(opts);
361 	*options = opts;
362 	return (0);
363 bad:
364 	vfs_freeopts(opts);
365 	return (error);
366 }
367 
368 /*
369  * Merge the old mount options with the new ones passed
370  * in the MNT_UPDATE case.
371  *
372  * XXX: This function will keep a "nofoo" option in the new
373  * options.  E.g, if the option's canonical name is "foo",
374  * "nofoo" ends up in the mount point's active options.
375  */
376 static void
377 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
378 {
379 	struct vfsopt *opt, *new;
380 
381 	TAILQ_FOREACH(opt, oldopts, link) {
382 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
383 		new->name = strdup(opt->name, M_MOUNT);
384 		if (opt->len != 0) {
385 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
386 			bcopy(opt->value, new->value, opt->len);
387 		} else
388 			new->value = NULL;
389 		new->len = opt->len;
390 		new->seen = opt->seen;
391 		TAILQ_INSERT_HEAD(toopts, new, link);
392 	}
393 	vfs_sanitizeopts(toopts);
394 }
395 
396 /*
397  * Mount a filesystem.
398  */
399 #ifndef _SYS_SYSPROTO_H_
400 struct nmount_args {
401 	struct iovec *iovp;
402 	unsigned int iovcnt;
403 	int flags;
404 };
405 #endif
406 int
407 sys_nmount(struct thread *td, struct nmount_args *uap)
408 {
409 	struct uio *auio;
410 	int error;
411 	u_int iovcnt;
412 	uint64_t flags;
413 
414 	/*
415 	 * Mount flags are now 64-bits. On 32-bit archtectures only
416 	 * 32-bits are passed in, but from here on everything handles
417 	 * 64-bit flags correctly.
418 	 */
419 	flags = uap->flags;
420 
421 	AUDIT_ARG_FFLAGS(flags);
422 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
423 	    uap->iovp, uap->iovcnt, flags);
424 
425 	/*
426 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
427 	 * userspace to set this flag, but we must filter it out if we want
428 	 * MNT_UPDATE on the root file system to work.
429 	 * MNT_ROOTFS should only be set by the kernel when mounting its
430 	 * root file system.
431 	 */
432 	flags &= ~MNT_ROOTFS;
433 
434 	iovcnt = uap->iovcnt;
435 	/*
436 	 * Check that we have an even number of iovec's
437 	 * and that we have at least two options.
438 	 */
439 	if ((iovcnt & 1) || (iovcnt < 4)) {
440 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
441 		    uap->iovcnt);
442 		return (EINVAL);
443 	}
444 
445 	error = copyinuio(uap->iovp, iovcnt, &auio);
446 	if (error) {
447 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
448 		    __func__, error);
449 		return (error);
450 	}
451 	error = vfs_donmount(td, flags, auio);
452 
453 	free(auio, M_IOV);
454 	return (error);
455 }
456 
457 /*
458  * ---------------------------------------------------------------------
459  * Various utility functions
460  */
461 
462 void
463 vfs_ref(struct mount *mp)
464 {
465 
466 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
467 	if (vfs_op_thread_enter(mp)) {
468 		vfs_mp_count_add_pcpu(mp, ref, 1);
469 		vfs_op_thread_exit(mp);
470 		return;
471 	}
472 
473 	MNT_ILOCK(mp);
474 	MNT_REF(mp);
475 	MNT_IUNLOCK(mp);
476 }
477 
478 void
479 vfs_rel(struct mount *mp)
480 {
481 
482 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
483 	if (vfs_op_thread_enter(mp)) {
484 		vfs_mp_count_sub_pcpu(mp, ref, 1);
485 		vfs_op_thread_exit(mp);
486 		return;
487 	}
488 
489 	MNT_ILOCK(mp);
490 	MNT_REL(mp);
491 	MNT_IUNLOCK(mp);
492 }
493 
494 /*
495  * Allocate and initialize the mount point struct.
496  */
497 struct mount *
498 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
499     struct ucred *cred)
500 {
501 	struct mount *mp;
502 
503 	mp = uma_zalloc(mount_zone, M_WAITOK);
504 	bzero(&mp->mnt_startzero,
505 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
506 	TAILQ_INIT(&mp->mnt_nvnodelist);
507 	mp->mnt_nvnodelistsize = 0;
508 	TAILQ_INIT(&mp->mnt_lazyvnodelist);
509 	mp->mnt_lazyvnodelistsize = 0;
510 	if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
511 	    mp->mnt_writeopcount != 0)
512 		panic("%s: non-zero counters on new mp %p\n", __func__, mp);
513 	if (mp->mnt_vfs_ops != 1)
514 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
515 		    mp->mnt_vfs_ops);
516 	(void) vfs_busy(mp, MBF_NOWAIT);
517 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
518 	mp->mnt_op = vfsp->vfc_vfsops;
519 	mp->mnt_vfc = vfsp;
520 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
521 	mp->mnt_gen++;
522 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
523 	mp->mnt_vnodecovered = vp;
524 	mp->mnt_cred = crdup(cred);
525 	mp->mnt_stat.f_owner = cred->cr_uid;
526 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
527 	mp->mnt_iosize_max = DFLTPHYS;
528 #ifdef MAC
529 	mac_mount_init(mp);
530 	mac_mount_create(cred, mp);
531 #endif
532 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
533 	TAILQ_INIT(&mp->mnt_uppers);
534 	return (mp);
535 }
536 
537 /*
538  * Destroy the mount struct previously allocated by vfs_mount_alloc().
539  */
540 void
541 vfs_mount_destroy(struct mount *mp)
542 {
543 
544 	if (mp->mnt_vfs_ops == 0)
545 		panic("%s: entered with zero vfs_ops\n", __func__);
546 
547 	vfs_assert_mount_counters(mp);
548 
549 	MNT_ILOCK(mp);
550 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
551 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
552 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
553 		wakeup(mp);
554 	}
555 	while (mp->mnt_ref)
556 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
557 	KASSERT(mp->mnt_ref == 0,
558 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
559 	    __FILE__, __LINE__));
560 	if (mp->mnt_writeopcount != 0)
561 		panic("vfs_mount_destroy: nonzero writeopcount");
562 	if (mp->mnt_secondary_writes != 0)
563 		panic("vfs_mount_destroy: nonzero secondary_writes");
564 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
565 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
566 		struct vnode *vp;
567 
568 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
569 			vn_printf(vp, "dangling vnode ");
570 		panic("unmount: dangling vnode");
571 	}
572 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
573 	if (mp->mnt_nvnodelistsize != 0)
574 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
575 	if (mp->mnt_lazyvnodelistsize != 0)
576 		panic("vfs_mount_destroy: nonzero lazyvnodelistsize");
577 	if (mp->mnt_lockref != 0)
578 		panic("vfs_mount_destroy: nonzero lock refcount");
579 	MNT_IUNLOCK(mp);
580 
581 	if (mp->mnt_vfs_ops != 1)
582 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
583 		    mp->mnt_vfs_ops);
584 
585 	if (mp->mnt_rootvnode != NULL)
586 		panic("%s: mount point still has a root vnode %p\n", __func__,
587 		    mp->mnt_rootvnode);
588 
589 	if (mp->mnt_vnodecovered != NULL)
590 		vrele(mp->mnt_vnodecovered);
591 #ifdef MAC
592 	mac_mount_destroy(mp);
593 #endif
594 	if (mp->mnt_opt != NULL)
595 		vfs_freeopts(mp->mnt_opt);
596 	crfree(mp->mnt_cred);
597 	uma_zfree(mount_zone, mp);
598 }
599 
600 static bool
601 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
602 {
603 	/* This is an upgrade of an exisiting mount. */
604 	if ((fsflags & MNT_UPDATE) != 0)
605 		return (false);
606 	/* This is already an R/O mount. */
607 	if ((fsflags & MNT_RDONLY) != 0)
608 		return (false);
609 
610 	switch (error) {
611 	case ENODEV:	/* generic, geom, ... */
612 	case EACCES:	/* cam/scsi, ... */
613 	case EROFS:	/* md, mmcsd, ... */
614 		/*
615 		 * These errors can be returned by the storage layer to signal
616 		 * that the media is read-only.  No harm in the R/O mount
617 		 * attempt if the error was returned for some other reason.
618 		 */
619 		return (true);
620 	default:
621 		return (false);
622 	}
623 }
624 
625 int
626 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
627 {
628 	struct vfsoptlist *optlist;
629 	struct vfsopt *opt, *tmp_opt;
630 	char *fstype, *fspath, *errmsg;
631 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
632 	bool autoro;
633 
634 	errmsg = fspath = NULL;
635 	errmsg_len = fspathlen = 0;
636 	errmsg_pos = -1;
637 	autoro = default_autoro;
638 
639 	error = vfs_buildopts(fsoptions, &optlist);
640 	if (error)
641 		return (error);
642 
643 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
644 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
645 
646 	/*
647 	 * We need these two options before the others,
648 	 * and they are mandatory for any filesystem.
649 	 * Ensure they are NUL terminated as well.
650 	 */
651 	fstypelen = 0;
652 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
653 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
654 		error = EINVAL;
655 		if (errmsg != NULL)
656 			strncpy(errmsg, "Invalid fstype", errmsg_len);
657 		goto bail;
658 	}
659 	fspathlen = 0;
660 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
661 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
662 		error = EINVAL;
663 		if (errmsg != NULL)
664 			strncpy(errmsg, "Invalid fspath", errmsg_len);
665 		goto bail;
666 	}
667 
668 	/*
669 	 * We need to see if we have the "update" option
670 	 * before we call vfs_domount(), since vfs_domount() has special
671 	 * logic based on MNT_UPDATE.  This is very important
672 	 * when we want to update the root filesystem.
673 	 */
674 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
675 		int do_freeopt = 0;
676 
677 		if (strcmp(opt->name, "update") == 0) {
678 			fsflags |= MNT_UPDATE;
679 			do_freeopt = 1;
680 		}
681 		else if (strcmp(opt->name, "async") == 0)
682 			fsflags |= MNT_ASYNC;
683 		else if (strcmp(opt->name, "force") == 0) {
684 			fsflags |= MNT_FORCE;
685 			do_freeopt = 1;
686 		}
687 		else if (strcmp(opt->name, "reload") == 0) {
688 			fsflags |= MNT_RELOAD;
689 			do_freeopt = 1;
690 		}
691 		else if (strcmp(opt->name, "multilabel") == 0)
692 			fsflags |= MNT_MULTILABEL;
693 		else if (strcmp(opt->name, "noasync") == 0)
694 			fsflags &= ~MNT_ASYNC;
695 		else if (strcmp(opt->name, "noatime") == 0)
696 			fsflags |= MNT_NOATIME;
697 		else if (strcmp(opt->name, "atime") == 0) {
698 			free(opt->name, M_MOUNT);
699 			opt->name = strdup("nonoatime", M_MOUNT);
700 		}
701 		else if (strcmp(opt->name, "noclusterr") == 0)
702 			fsflags |= MNT_NOCLUSTERR;
703 		else if (strcmp(opt->name, "clusterr") == 0) {
704 			free(opt->name, M_MOUNT);
705 			opt->name = strdup("nonoclusterr", M_MOUNT);
706 		}
707 		else if (strcmp(opt->name, "noclusterw") == 0)
708 			fsflags |= MNT_NOCLUSTERW;
709 		else if (strcmp(opt->name, "clusterw") == 0) {
710 			free(opt->name, M_MOUNT);
711 			opt->name = strdup("nonoclusterw", M_MOUNT);
712 		}
713 		else if (strcmp(opt->name, "noexec") == 0)
714 			fsflags |= MNT_NOEXEC;
715 		else if (strcmp(opt->name, "exec") == 0) {
716 			free(opt->name, M_MOUNT);
717 			opt->name = strdup("nonoexec", M_MOUNT);
718 		}
719 		else if (strcmp(opt->name, "nosuid") == 0)
720 			fsflags |= MNT_NOSUID;
721 		else if (strcmp(opt->name, "suid") == 0) {
722 			free(opt->name, M_MOUNT);
723 			opt->name = strdup("nonosuid", M_MOUNT);
724 		}
725 		else if (strcmp(opt->name, "nosymfollow") == 0)
726 			fsflags |= MNT_NOSYMFOLLOW;
727 		else if (strcmp(opt->name, "symfollow") == 0) {
728 			free(opt->name, M_MOUNT);
729 			opt->name = strdup("nonosymfollow", M_MOUNT);
730 		}
731 		else if (strcmp(opt->name, "noro") == 0) {
732 			fsflags &= ~MNT_RDONLY;
733 			autoro = false;
734 		}
735 		else if (strcmp(opt->name, "rw") == 0) {
736 			fsflags &= ~MNT_RDONLY;
737 			autoro = false;
738 		}
739 		else if (strcmp(opt->name, "ro") == 0) {
740 			fsflags |= MNT_RDONLY;
741 			autoro = false;
742 		}
743 		else if (strcmp(opt->name, "rdonly") == 0) {
744 			free(opt->name, M_MOUNT);
745 			opt->name = strdup("ro", M_MOUNT);
746 			fsflags |= MNT_RDONLY;
747 			autoro = false;
748 		}
749 		else if (strcmp(opt->name, "autoro") == 0) {
750 			do_freeopt = 1;
751 			autoro = true;
752 		}
753 		else if (strcmp(opt->name, "suiddir") == 0)
754 			fsflags |= MNT_SUIDDIR;
755 		else if (strcmp(opt->name, "sync") == 0)
756 			fsflags |= MNT_SYNCHRONOUS;
757 		else if (strcmp(opt->name, "union") == 0)
758 			fsflags |= MNT_UNION;
759 		else if (strcmp(opt->name, "automounted") == 0) {
760 			fsflags |= MNT_AUTOMOUNTED;
761 			do_freeopt = 1;
762 		} else if (strcmp(opt->name, "nocover") == 0) {
763 			fsflags |= MNT_NOCOVER;
764 			do_freeopt = 1;
765 		} else if (strcmp(opt->name, "cover") == 0) {
766 			fsflags &= ~MNT_NOCOVER;
767 			do_freeopt = 1;
768 		} else if (strcmp(opt->name, "emptydir") == 0) {
769 			fsflags |= MNT_EMPTYDIR;
770 			do_freeopt = 1;
771 		} else if (strcmp(opt->name, "noemptydir") == 0) {
772 			fsflags &= ~MNT_EMPTYDIR;
773 			do_freeopt = 1;
774 		}
775 		if (do_freeopt)
776 			vfs_freeopt(optlist, opt);
777 	}
778 
779 	/*
780 	 * Be ultra-paranoid about making sure the type and fspath
781 	 * variables will fit in our mp buffers, including the
782 	 * terminating NUL.
783 	 */
784 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
785 		error = ENAMETOOLONG;
786 		goto bail;
787 	}
788 
789 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
790 
791 	/*
792 	 * See if we can mount in the read-only mode if the error code suggests
793 	 * that it could be possible and the mount options allow for that.
794 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
795 	 * overridden by "autoro".
796 	 */
797 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
798 		printf("%s: R/W mount failed, possibly R/O media,"
799 		    " trying R/O mount\n", __func__);
800 		fsflags |= MNT_RDONLY;
801 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
802 	}
803 bail:
804 	/* copyout the errmsg */
805 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
806 	    && errmsg_len > 0 && errmsg != NULL) {
807 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
808 			bcopy(errmsg,
809 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
810 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
811 		} else {
812 			copyout(errmsg,
813 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
814 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
815 		}
816 	}
817 
818 	if (optlist != NULL)
819 		vfs_freeopts(optlist);
820 	return (error);
821 }
822 
823 /*
824  * Old mount API.
825  */
826 #ifndef _SYS_SYSPROTO_H_
827 struct mount_args {
828 	char	*type;
829 	char	*path;
830 	int	flags;
831 	caddr_t	data;
832 };
833 #endif
834 /* ARGSUSED */
835 int
836 sys_mount(struct thread *td, struct mount_args *uap)
837 {
838 	char *fstype;
839 	struct vfsconf *vfsp = NULL;
840 	struct mntarg *ma = NULL;
841 	uint64_t flags;
842 	int error;
843 
844 	/*
845 	 * Mount flags are now 64-bits. On 32-bit architectures only
846 	 * 32-bits are passed in, but from here on everything handles
847 	 * 64-bit flags correctly.
848 	 */
849 	flags = uap->flags;
850 
851 	AUDIT_ARG_FFLAGS(flags);
852 
853 	/*
854 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
855 	 * userspace to set this flag, but we must filter it out if we want
856 	 * MNT_UPDATE on the root file system to work.
857 	 * MNT_ROOTFS should only be set by the kernel when mounting its
858 	 * root file system.
859 	 */
860 	flags &= ~MNT_ROOTFS;
861 
862 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
863 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
864 	if (error) {
865 		free(fstype, M_TEMP);
866 		return (error);
867 	}
868 
869 	AUDIT_ARG_TEXT(fstype);
870 	vfsp = vfs_byname_kld(fstype, td, &error);
871 	free(fstype, M_TEMP);
872 	if (vfsp == NULL)
873 		return (ENOENT);
874 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
875 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
876 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
877 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
878 		return (EOPNOTSUPP);
879 
880 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
881 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
882 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
883 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
884 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
885 
886 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
887 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
888 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
889 }
890 
891 /*
892  * vfs_domount_first(): first file system mount (not update)
893  */
894 static int
895 vfs_domount_first(
896 	struct thread *td,		/* Calling thread. */
897 	struct vfsconf *vfsp,		/* File system type. */
898 	char *fspath,			/* Mount path. */
899 	struct vnode *vp,		/* Vnode to be covered. */
900 	uint64_t fsflags,		/* Flags common to all filesystems. */
901 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
902 	)
903 {
904 	struct vattr va;
905 	struct mount *mp;
906 	struct vnode *newdp, *rootvp;
907 	int error, error1;
908 
909 	ASSERT_VOP_ELOCKED(vp, __func__);
910 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
911 
912 	if ((fsflags & MNT_EMPTYDIR) != 0) {
913 		error = vfs_emptydir(vp);
914 		if (error != 0) {
915 			vput(vp);
916 			return (error);
917 		}
918 	}
919 
920 	/*
921 	 * If the jail of the calling thread lacks permission for this type of
922 	 * file system, deny immediately.
923 	 */
924 	if (jailed(td->td_ucred) && !prison_allow(td->td_ucred,
925 	    vfsp->vfc_prison_flag)) {
926 		vput(vp);
927 		return (EPERM);
928 	}
929 
930 	/*
931 	 * If the user is not root, ensure that they own the directory
932 	 * onto which we are attempting to mount.
933 	 */
934 	error = VOP_GETATTR(vp, &va, td->td_ucred);
935 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
936 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
937 	if (error == 0)
938 		error = vinvalbuf(vp, V_SAVE, 0, 0);
939 	if (error == 0 && vp->v_type != VDIR)
940 		error = ENOTDIR;
941 	if (error == 0) {
942 		VI_LOCK(vp);
943 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
944 			vp->v_iflag |= VI_MOUNT;
945 		else
946 			error = EBUSY;
947 		VI_UNLOCK(vp);
948 	}
949 	if (error != 0) {
950 		vput(vp);
951 		return (error);
952 	}
953 	vn_seqc_write_begin(vp);
954 	VOP_UNLOCK(vp);
955 
956 	/* Allocate and initialize the filesystem. */
957 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
958 	/* XXXMAC: pass to vfs_mount_alloc? */
959 	mp->mnt_optnew = *optlist;
960 	/* Set the mount level flags. */
961 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
962 
963 	/*
964 	 * Mount the filesystem.
965 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
966 	 * get.  No freeing of cn_pnbuf.
967 	 */
968 	error1 = 0;
969 	if ((error = VFS_MOUNT(mp)) != 0 ||
970 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
971 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
972 		rootvp = NULL;
973 		if (error1 != 0) {
974 			error = error1;
975 			rootvp = vfs_cache_root_clear(mp);
976 			if (rootvp != NULL) {
977 				vhold(rootvp);
978 				vrele(rootvp);
979 			}
980 			if ((error1 = VFS_UNMOUNT(mp, 0)) != 0)
981 				printf("VFS_UNMOUNT returned %d\n", error1);
982 		}
983 		vfs_unbusy(mp);
984 		mp->mnt_vnodecovered = NULL;
985 		vfs_mount_destroy(mp);
986 		VI_LOCK(vp);
987 		vp->v_iflag &= ~VI_MOUNT;
988 		VI_UNLOCK(vp);
989 		if (rootvp != NULL) {
990 			vn_seqc_write_end(rootvp);
991 			vdrop(rootvp);
992 		}
993 		vn_seqc_write_end(vp);
994 		vrele(vp);
995 		return (error);
996 	}
997 	vn_seqc_write_begin(newdp);
998 	VOP_UNLOCK(newdp);
999 
1000 	if (mp->mnt_opt != NULL)
1001 		vfs_freeopts(mp->mnt_opt);
1002 	mp->mnt_opt = mp->mnt_optnew;
1003 	*optlist = NULL;
1004 
1005 	/*
1006 	 * Prevent external consumers of mount options from reading mnt_optnew.
1007 	 */
1008 	mp->mnt_optnew = NULL;
1009 
1010 	MNT_ILOCK(mp);
1011 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1012 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1013 		mp->mnt_kern_flag |= MNTK_ASYNC;
1014 	else
1015 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1016 	MNT_IUNLOCK(mp);
1017 
1018 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1019 	cache_purge(vp);
1020 	VI_LOCK(vp);
1021 	vp->v_iflag &= ~VI_MOUNT;
1022 	VI_UNLOCK(vp);
1023 	vp->v_mountedhere = mp;
1024 	/* Place the new filesystem at the end of the mount list. */
1025 	mtx_lock(&mountlist_mtx);
1026 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1027 	mtx_unlock(&mountlist_mtx);
1028 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1029 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1030 	VOP_UNLOCK(vp);
1031 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1032 	VOP_UNLOCK(newdp);
1033 	mount_devctl_event("MOUNT", mp, false);
1034 	mountcheckdirs(vp, newdp);
1035 	vn_seqc_write_end(vp);
1036 	vn_seqc_write_end(newdp);
1037 	vrele(newdp);
1038 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1039 		vfs_allocate_syncvnode(mp);
1040 	vfs_op_exit(mp);
1041 	vfs_unbusy(mp);
1042 	return (0);
1043 }
1044 
1045 /*
1046  * vfs_domount_update(): update of mounted file system
1047  */
1048 static int
1049 vfs_domount_update(
1050 	struct thread *td,		/* Calling thread. */
1051 	struct vnode *vp,		/* Mount point vnode. */
1052 	uint64_t fsflags,		/* Flags common to all filesystems. */
1053 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1054 	)
1055 {
1056 	struct export_args export;
1057 	struct o2export_args o2export;
1058 	struct vnode *rootvp;
1059 	void *bufp;
1060 	struct mount *mp;
1061 	int error, export_error, i, len;
1062 	uint64_t flag;
1063 	gid_t *grps;
1064 
1065 	ASSERT_VOP_ELOCKED(vp, __func__);
1066 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1067 	mp = vp->v_mount;
1068 
1069 	if ((vp->v_vflag & VV_ROOT) == 0) {
1070 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1071 		    == 0)
1072 			error = EXDEV;
1073 		else
1074 			error = EINVAL;
1075 		vput(vp);
1076 		return (error);
1077 	}
1078 
1079 	/*
1080 	 * We only allow the filesystem to be reloaded if it
1081 	 * is currently mounted read-only.
1082 	 */
1083 	flag = mp->mnt_flag;
1084 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1085 		vput(vp);
1086 		return (EOPNOTSUPP);	/* Needs translation */
1087 	}
1088 	/*
1089 	 * Only privileged root, or (if MNT_USER is set) the user that
1090 	 * did the original mount is permitted to update it.
1091 	 */
1092 	error = vfs_suser(mp, td);
1093 	if (error != 0) {
1094 		vput(vp);
1095 		return (error);
1096 	}
1097 	if (vfs_busy(mp, MBF_NOWAIT)) {
1098 		vput(vp);
1099 		return (EBUSY);
1100 	}
1101 	VI_LOCK(vp);
1102 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1103 		VI_UNLOCK(vp);
1104 		vfs_unbusy(mp);
1105 		vput(vp);
1106 		return (EBUSY);
1107 	}
1108 	vp->v_iflag |= VI_MOUNT;
1109 	VI_UNLOCK(vp);
1110 	VOP_UNLOCK(vp);
1111 
1112 	vfs_op_enter(mp);
1113 	vn_seqc_write_begin(vp);
1114 
1115 	rootvp = NULL;
1116 	MNT_ILOCK(mp);
1117 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1118 		MNT_IUNLOCK(mp);
1119 		error = EBUSY;
1120 		goto end;
1121 	}
1122 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1123 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1124 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1125 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1126 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1127 	rootvp = vfs_cache_root_clear(mp);
1128 	MNT_IUNLOCK(mp);
1129 	mp->mnt_optnew = *optlist;
1130 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1131 
1132 	/*
1133 	 * Mount the filesystem.
1134 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1135 	 * get.  No freeing of cn_pnbuf.
1136 	 */
1137 	error = VFS_MOUNT(mp);
1138 
1139 	export_error = 0;
1140 	/* Process the export option. */
1141 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1142 	    &len) == 0) {
1143 		/* Assume that there is only 1 ABI for each length. */
1144 		switch (len) {
1145 		case (sizeof(struct oexport_args)):
1146 			bzero(&o2export, sizeof(o2export));
1147 			/* FALLTHROUGH */
1148 		case (sizeof(o2export)):
1149 			bcopy(bufp, &o2export, len);
1150 			export.ex_flags = (uint64_t)o2export.ex_flags;
1151 			export.ex_root = o2export.ex_root;
1152 			export.ex_uid = o2export.ex_anon.cr_uid;
1153 			export.ex_groups = NULL;
1154 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1155 			if (export.ex_ngroups > 0) {
1156 				if (export.ex_ngroups <= XU_NGROUPS) {
1157 					export.ex_groups = malloc(
1158 					    export.ex_ngroups * sizeof(gid_t),
1159 					    M_TEMP, M_WAITOK);
1160 					for (i = 0; i < export.ex_ngroups; i++)
1161 						export.ex_groups[i] =
1162 						  o2export.ex_anon.cr_groups[i];
1163 				} else
1164 					export_error = EINVAL;
1165 			} else if (export.ex_ngroups < 0)
1166 				export_error = EINVAL;
1167 			export.ex_addr = o2export.ex_addr;
1168 			export.ex_addrlen = o2export.ex_addrlen;
1169 			export.ex_mask = o2export.ex_mask;
1170 			export.ex_masklen = o2export.ex_masklen;
1171 			export.ex_indexfile = o2export.ex_indexfile;
1172 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1173 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1174 				for (i = 0; i < export.ex_numsecflavors; i++)
1175 					export.ex_secflavors[i] =
1176 					    o2export.ex_secflavors[i];
1177 			} else
1178 				export_error = EINVAL;
1179 			if (export_error == 0)
1180 				export_error = vfs_export(mp, &export);
1181 			free(export.ex_groups, M_TEMP);
1182 			break;
1183 		case (sizeof(export)):
1184 			bcopy(bufp, &export, len);
1185 			grps = NULL;
1186 			if (export.ex_ngroups > 0) {
1187 				if (export.ex_ngroups <= NGROUPS_MAX) {
1188 					grps = malloc(export.ex_ngroups *
1189 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1190 					export_error = copyin(export.ex_groups,
1191 					    grps, export.ex_ngroups *
1192 					    sizeof(gid_t));
1193 					if (export_error == 0)
1194 						export.ex_groups = grps;
1195 				} else
1196 					export_error = EINVAL;
1197 			} else if (export.ex_ngroups == 0)
1198 				export.ex_groups = NULL;
1199 			else
1200 				export_error = EINVAL;
1201 			if (export_error == 0)
1202 				export_error = vfs_export(mp, &export);
1203 			free(grps, M_TEMP);
1204 			break;
1205 		default:
1206 			export_error = EINVAL;
1207 			break;
1208 		}
1209 	}
1210 
1211 	MNT_ILOCK(mp);
1212 	if (error == 0) {
1213 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1214 		    MNT_SNAPSHOT);
1215 	} else {
1216 		/*
1217 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1218 		 * because it is not part of MNT_UPDATEMASK, but it could have
1219 		 * changed in the meantime if quotactl(2) was called.
1220 		 * All in all we want current value of MNT_QUOTA, not the old
1221 		 * one.
1222 		 */
1223 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1224 	}
1225 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1226 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1227 		mp->mnt_kern_flag |= MNTK_ASYNC;
1228 	else
1229 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1230 	MNT_IUNLOCK(mp);
1231 
1232 	if (error != 0)
1233 		goto end;
1234 
1235 	mount_devctl_event("REMOUNT", mp, true);
1236 	if (mp->mnt_opt != NULL)
1237 		vfs_freeopts(mp->mnt_opt);
1238 	mp->mnt_opt = mp->mnt_optnew;
1239 	*optlist = NULL;
1240 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1241 	/*
1242 	 * Prevent external consumers of mount options from reading
1243 	 * mnt_optnew.
1244 	 */
1245 	mp->mnt_optnew = NULL;
1246 
1247 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1248 		vfs_allocate_syncvnode(mp);
1249 	else
1250 		vfs_deallocate_syncvnode(mp);
1251 end:
1252 	vfs_op_exit(mp);
1253 	if (rootvp != NULL) {
1254 		vn_seqc_write_end(rootvp);
1255 		vrele(rootvp);
1256 	}
1257 	vn_seqc_write_end(vp);
1258 	vfs_unbusy(mp);
1259 	VI_LOCK(vp);
1260 	vp->v_iflag &= ~VI_MOUNT;
1261 	VI_UNLOCK(vp);
1262 	vrele(vp);
1263 	return (error != 0 ? error : export_error);
1264 }
1265 
1266 /*
1267  * vfs_domount(): actually attempt a filesystem mount.
1268  */
1269 static int
1270 vfs_domount(
1271 	struct thread *td,		/* Calling thread. */
1272 	const char *fstype,		/* Filesystem type. */
1273 	char *fspath,			/* Mount path. */
1274 	uint64_t fsflags,		/* Flags common to all filesystems. */
1275 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1276 	)
1277 {
1278 	struct vfsconf *vfsp;
1279 	struct nameidata nd;
1280 	struct vnode *vp;
1281 	char *pathbuf;
1282 	int error;
1283 
1284 	/*
1285 	 * Be ultra-paranoid about making sure the type and fspath
1286 	 * variables will fit in our mp buffers, including the
1287 	 * terminating NUL.
1288 	 */
1289 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1290 		return (ENAMETOOLONG);
1291 
1292 	if (jailed(td->td_ucred) || usermount == 0) {
1293 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1294 			return (error);
1295 	}
1296 
1297 	/*
1298 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1299 	 */
1300 	if (fsflags & MNT_EXPORTED) {
1301 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1302 		if (error)
1303 			return (error);
1304 	}
1305 	if (fsflags & MNT_SUIDDIR) {
1306 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1307 		if (error)
1308 			return (error);
1309 	}
1310 	/*
1311 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1312 	 */
1313 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1314 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1315 			fsflags |= MNT_NOSUID | MNT_USER;
1316 	}
1317 
1318 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1319 	vfsp = NULL;
1320 	if ((fsflags & MNT_UPDATE) == 0) {
1321 		/* Don't try to load KLDs if we're mounting the root. */
1322 		if (fsflags & MNT_ROOTFS)
1323 			vfsp = vfs_byname(fstype);
1324 		else
1325 			vfsp = vfs_byname_kld(fstype, td, &error);
1326 		if (vfsp == NULL)
1327 			return (ENODEV);
1328 	}
1329 
1330 	/*
1331 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1332 	 */
1333 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1334 	    UIO_SYSSPACE, fspath, td);
1335 	error = namei(&nd);
1336 	if (error != 0)
1337 		return (error);
1338 	NDFREE(&nd, NDF_ONLY_PNBUF);
1339 	vp = nd.ni_vp;
1340 	if ((fsflags & MNT_UPDATE) == 0) {
1341 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1342 		    (fsflags & MNT_NOCOVER) != 0) {
1343 			vput(vp);
1344 			return (EBUSY);
1345 		}
1346 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1347 		strcpy(pathbuf, fspath);
1348 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1349 		if (error == 0) {
1350 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1351 			    fsflags, optlist);
1352 		}
1353 		free(pathbuf, M_TEMP);
1354 	} else
1355 		error = vfs_domount_update(td, vp, fsflags, optlist);
1356 
1357 	return (error);
1358 }
1359 
1360 /*
1361  * Unmount a filesystem.
1362  *
1363  * Note: unmount takes a path to the vnode mounted on as argument, not
1364  * special file (as before).
1365  */
1366 #ifndef _SYS_SYSPROTO_H_
1367 struct unmount_args {
1368 	char	*path;
1369 	int	flags;
1370 };
1371 #endif
1372 /* ARGSUSED */
1373 int
1374 sys_unmount(struct thread *td, struct unmount_args *uap)
1375 {
1376 
1377 	return (kern_unmount(td, uap->path, uap->flags));
1378 }
1379 
1380 int
1381 kern_unmount(struct thread *td, const char *path, int flags)
1382 {
1383 	struct nameidata nd;
1384 	struct mount *mp;
1385 	char *pathbuf;
1386 	int error, id0, id1;
1387 
1388 	AUDIT_ARG_VALUE(flags);
1389 	if (jailed(td->td_ucred) || usermount == 0) {
1390 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1391 		if (error)
1392 			return (error);
1393 	}
1394 
1395 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1396 	error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1397 	if (error) {
1398 		free(pathbuf, M_TEMP);
1399 		return (error);
1400 	}
1401 	if (flags & MNT_BYFSID) {
1402 		AUDIT_ARG_TEXT(pathbuf);
1403 		/* Decode the filesystem ID. */
1404 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1405 			free(pathbuf, M_TEMP);
1406 			return (EINVAL);
1407 		}
1408 
1409 		mtx_lock(&mountlist_mtx);
1410 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1411 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1412 			    mp->mnt_stat.f_fsid.val[1] == id1) {
1413 				vfs_ref(mp);
1414 				break;
1415 			}
1416 		}
1417 		mtx_unlock(&mountlist_mtx);
1418 	} else {
1419 		/*
1420 		 * Try to find global path for path argument.
1421 		 */
1422 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1423 		    UIO_SYSSPACE, pathbuf, td);
1424 		if (namei(&nd) == 0) {
1425 			NDFREE(&nd, NDF_ONLY_PNBUF);
1426 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1427 			    MNAMELEN);
1428 			if (error == 0)
1429 				vput(nd.ni_vp);
1430 		}
1431 		mtx_lock(&mountlist_mtx);
1432 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1433 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1434 				vfs_ref(mp);
1435 				break;
1436 			}
1437 		}
1438 		mtx_unlock(&mountlist_mtx);
1439 	}
1440 	free(pathbuf, M_TEMP);
1441 	if (mp == NULL) {
1442 		/*
1443 		 * Previously we returned ENOENT for a nonexistent path and
1444 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1445 		 * now, so in the !MNT_BYFSID case return the more likely
1446 		 * EINVAL for compatibility.
1447 		 */
1448 		return ((flags & MNT_BYFSID) ? ENOENT : EINVAL);
1449 	}
1450 
1451 	/*
1452 	 * Don't allow unmounting the root filesystem.
1453 	 */
1454 	if (mp->mnt_flag & MNT_ROOTFS) {
1455 		vfs_rel(mp);
1456 		return (EINVAL);
1457 	}
1458 	error = dounmount(mp, flags, td);
1459 	return (error);
1460 }
1461 
1462 /*
1463  * Return error if any of the vnodes, ignoring the root vnode
1464  * and the syncer vnode, have non-zero usecount.
1465  *
1466  * This function is purely advisory - it can return false positives
1467  * and negatives.
1468  */
1469 static int
1470 vfs_check_usecounts(struct mount *mp)
1471 {
1472 	struct vnode *vp, *mvp;
1473 
1474 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1475 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1476 		    vp->v_usecount != 0) {
1477 			VI_UNLOCK(vp);
1478 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1479 			return (EBUSY);
1480 		}
1481 		VI_UNLOCK(vp);
1482 	}
1483 
1484 	return (0);
1485 }
1486 
1487 static void
1488 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1489 {
1490 
1491 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1492 	mp->mnt_kern_flag &= ~mntkflags;
1493 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1494 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1495 		wakeup(mp);
1496 	}
1497 	vfs_op_exit_locked(mp);
1498 	MNT_IUNLOCK(mp);
1499 	if (coveredvp != NULL) {
1500 		VOP_UNLOCK(coveredvp);
1501 		vdrop(coveredvp);
1502 	}
1503 	vn_finished_write(mp);
1504 }
1505 
1506 /*
1507  * There are various reference counters associated with the mount point.
1508  * Normally it is permitted to modify them without taking the mnt ilock,
1509  * but this behavior can be temporarily disabled if stable value is needed
1510  * or callers are expected to block (e.g. to not allow new users during
1511  * forced unmount).
1512  */
1513 void
1514 vfs_op_enter(struct mount *mp)
1515 {
1516 	int cpu;
1517 
1518 	MNT_ILOCK(mp);
1519 	mp->mnt_vfs_ops++;
1520 	if (mp->mnt_vfs_ops > 1) {
1521 		MNT_IUNLOCK(mp);
1522 		return;
1523 	}
1524 	vfs_op_barrier_wait(mp);
1525 	CPU_FOREACH(cpu) {
1526 		mp->mnt_ref +=
1527 		    zpcpu_replace_cpu(mp->mnt_ref_pcpu, 0, cpu);
1528 		mp->mnt_lockref +=
1529 		    zpcpu_replace_cpu(mp->mnt_lockref_pcpu, 0, cpu);
1530 		mp->mnt_writeopcount +=
1531 		    zpcpu_replace_cpu(mp->mnt_writeopcount_pcpu, 0, cpu);
1532 	}
1533 	if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0)
1534 		panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n",
1535 		    __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount);
1536 	MNT_IUNLOCK(mp);
1537 	vfs_assert_mount_counters(mp);
1538 }
1539 
1540 void
1541 vfs_op_exit_locked(struct mount *mp)
1542 {
1543 
1544 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1545 
1546 	if (mp->mnt_vfs_ops <= 0)
1547 		panic("%s: invalid vfs_ops count %d for mp %p\n",
1548 		    __func__, mp->mnt_vfs_ops, mp);
1549 	mp->mnt_vfs_ops--;
1550 }
1551 
1552 void
1553 vfs_op_exit(struct mount *mp)
1554 {
1555 
1556 	MNT_ILOCK(mp);
1557 	vfs_op_exit_locked(mp);
1558 	MNT_IUNLOCK(mp);
1559 }
1560 
1561 struct vfs_op_barrier_ipi {
1562 	struct mount *mp;
1563 	struct smp_rendezvous_cpus_retry_arg srcra;
1564 };
1565 
1566 static void
1567 vfs_op_action_func(void *arg)
1568 {
1569 	struct vfs_op_barrier_ipi *vfsopipi;
1570 	struct mount *mp;
1571 
1572 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1573 	mp = vfsopipi->mp;
1574 
1575 	if (!vfs_op_thread_entered(mp))
1576 		smp_rendezvous_cpus_done(arg);
1577 }
1578 
1579 static void
1580 vfs_op_wait_func(void *arg, int cpu)
1581 {
1582 	struct vfs_op_barrier_ipi *vfsopipi;
1583 	struct mount *mp;
1584 	int *in_op;
1585 
1586 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1587 	mp = vfsopipi->mp;
1588 
1589 	in_op = zpcpu_get_cpu(mp->mnt_thread_in_ops_pcpu, cpu);
1590 	while (atomic_load_int(in_op))
1591 		cpu_spinwait();
1592 }
1593 
1594 void
1595 vfs_op_barrier_wait(struct mount *mp)
1596 {
1597 	struct vfs_op_barrier_ipi vfsopipi;
1598 
1599 	vfsopipi.mp = mp;
1600 
1601 	smp_rendezvous_cpus_retry(all_cpus,
1602 	    smp_no_rendezvous_barrier,
1603 	    vfs_op_action_func,
1604 	    smp_no_rendezvous_barrier,
1605 	    vfs_op_wait_func,
1606 	    &vfsopipi.srcra);
1607 }
1608 
1609 #ifdef DIAGNOSTIC
1610 void
1611 vfs_assert_mount_counters(struct mount *mp)
1612 {
1613 	int cpu;
1614 
1615 	if (mp->mnt_vfs_ops == 0)
1616 		return;
1617 
1618 	CPU_FOREACH(cpu) {
1619 		if (*zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu) != 0 ||
1620 		    *zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu) != 0 ||
1621 		    *zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu) != 0)
1622 			vfs_dump_mount_counters(mp);
1623 	}
1624 }
1625 
1626 void
1627 vfs_dump_mount_counters(struct mount *mp)
1628 {
1629 	int cpu, *count;
1630 	int ref, lockref, writeopcount;
1631 
1632 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1633 
1634 	printf("        ref : ");
1635 	ref = mp->mnt_ref;
1636 	CPU_FOREACH(cpu) {
1637 		count = zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu);
1638 		printf("%d ", *count);
1639 		ref += *count;
1640 	}
1641 	printf("\n");
1642 	printf("    lockref : ");
1643 	lockref = mp->mnt_lockref;
1644 	CPU_FOREACH(cpu) {
1645 		count = zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu);
1646 		printf("%d ", *count);
1647 		lockref += *count;
1648 	}
1649 	printf("\n");
1650 	printf("writeopcount: ");
1651 	writeopcount = mp->mnt_writeopcount;
1652 	CPU_FOREACH(cpu) {
1653 		count = zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu);
1654 		printf("%d ", *count);
1655 		writeopcount += *count;
1656 	}
1657 	printf("\n");
1658 
1659 	printf("counter       struct total\n");
1660 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1661 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1662 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1663 
1664 	panic("invalid counts on struct mount");
1665 }
1666 #endif
1667 
1668 int
1669 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1670 {
1671 	int *base, *pcpu;
1672 	int cpu, sum;
1673 
1674 	switch (which) {
1675 	case MNT_COUNT_REF:
1676 		base = &mp->mnt_ref;
1677 		pcpu = mp->mnt_ref_pcpu;
1678 		break;
1679 	case MNT_COUNT_LOCKREF:
1680 		base = &mp->mnt_lockref;
1681 		pcpu = mp->mnt_lockref_pcpu;
1682 		break;
1683 	case MNT_COUNT_WRITEOPCOUNT:
1684 		base = &mp->mnt_writeopcount;
1685 		pcpu = mp->mnt_writeopcount_pcpu;
1686 		break;
1687 	}
1688 
1689 	sum = *base;
1690 	CPU_FOREACH(cpu) {
1691 		sum += *zpcpu_get_cpu(pcpu, cpu);
1692 	}
1693 	return (sum);
1694 }
1695 
1696 /*
1697  * Do the actual filesystem unmount.
1698  */
1699 int
1700 dounmount(struct mount *mp, int flags, struct thread *td)
1701 {
1702 	struct vnode *coveredvp, *rootvp;
1703 	int error;
1704 	uint64_t async_flag;
1705 	int mnt_gen_r;
1706 
1707 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1708 		mnt_gen_r = mp->mnt_gen;
1709 		VI_LOCK(coveredvp);
1710 		vholdl(coveredvp);
1711 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1712 		/*
1713 		 * Check for mp being unmounted while waiting for the
1714 		 * covered vnode lock.
1715 		 */
1716 		if (coveredvp->v_mountedhere != mp ||
1717 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1718 			VOP_UNLOCK(coveredvp);
1719 			vdrop(coveredvp);
1720 			vfs_rel(mp);
1721 			return (EBUSY);
1722 		}
1723 	}
1724 
1725 	/*
1726 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1727 	 * original mount is permitted to unmount this filesystem.
1728 	 */
1729 	error = vfs_suser(mp, td);
1730 	if (error != 0) {
1731 		if (coveredvp != NULL) {
1732 			VOP_UNLOCK(coveredvp);
1733 			vdrop(coveredvp);
1734 		}
1735 		vfs_rel(mp);
1736 		return (error);
1737 	}
1738 
1739 	vfs_op_enter(mp);
1740 
1741 	vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1742 	MNT_ILOCK(mp);
1743 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1744 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
1745 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
1746 		dounmount_cleanup(mp, coveredvp, 0);
1747 		return (EBUSY);
1748 	}
1749 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1750 	rootvp = vfs_cache_root_clear(mp);
1751 	if (coveredvp != NULL)
1752 		vn_seqc_write_begin(coveredvp);
1753 	if (flags & MNT_NONBUSY) {
1754 		MNT_IUNLOCK(mp);
1755 		error = vfs_check_usecounts(mp);
1756 		MNT_ILOCK(mp);
1757 		if (error != 0) {
1758 			vn_seqc_write_end(coveredvp);
1759 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1760 			if (rootvp != NULL) {
1761 				vn_seqc_write_end(rootvp);
1762 				vrele(rootvp);
1763 			}
1764 			return (error);
1765 		}
1766 	}
1767 	/* Allow filesystems to detect that a forced unmount is in progress. */
1768 	if (flags & MNT_FORCE) {
1769 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1770 		MNT_IUNLOCK(mp);
1771 		/*
1772 		 * Must be done after setting MNTK_UNMOUNTF and before
1773 		 * waiting for mnt_lockref to become 0.
1774 		 */
1775 		VFS_PURGE(mp);
1776 		MNT_ILOCK(mp);
1777 	}
1778 	error = 0;
1779 	if (mp->mnt_lockref) {
1780 		mp->mnt_kern_flag |= MNTK_DRAINING;
1781 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1782 		    "mount drain", 0);
1783 	}
1784 	MNT_IUNLOCK(mp);
1785 	KASSERT(mp->mnt_lockref == 0,
1786 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1787 	    __func__, __FILE__, __LINE__));
1788 	KASSERT(error == 0,
1789 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1790 	    __func__, __FILE__, __LINE__));
1791 
1792 	/*
1793 	 * We want to keep the vnode around so that we can vn_seqc_write_end
1794 	 * after we are done with unmount. Downgrade our reference to a mere
1795 	 * hold count so that we don't interefere with anything.
1796 	 */
1797 	if (rootvp != NULL) {
1798 		vhold(rootvp);
1799 		vrele(rootvp);
1800 	}
1801 
1802 	if (mp->mnt_flag & MNT_EXPUBLIC)
1803 		vfs_setpublicfs(NULL, NULL, NULL);
1804 
1805 	vfs_periodic(mp, MNT_WAIT);
1806 	MNT_ILOCK(mp);
1807 	async_flag = mp->mnt_flag & MNT_ASYNC;
1808 	mp->mnt_flag &= ~MNT_ASYNC;
1809 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1810 	MNT_IUNLOCK(mp);
1811 	vfs_deallocate_syncvnode(mp);
1812 	error = VFS_UNMOUNT(mp, flags);
1813 	vn_finished_write(mp);
1814 	/*
1815 	 * If we failed to flush the dirty blocks for this mount point,
1816 	 * undo all the cdir/rdir and rootvnode changes we made above.
1817 	 * Unless we failed to do so because the device is reporting that
1818 	 * it doesn't exist anymore.
1819 	 */
1820 	if (error && error != ENXIO) {
1821 		MNT_ILOCK(mp);
1822 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1823 			MNT_IUNLOCK(mp);
1824 			vfs_allocate_syncvnode(mp);
1825 			MNT_ILOCK(mp);
1826 		}
1827 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1828 		mp->mnt_flag |= async_flag;
1829 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1830 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1831 			mp->mnt_kern_flag |= MNTK_ASYNC;
1832 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1833 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1834 			wakeup(mp);
1835 		}
1836 		vfs_op_exit_locked(mp);
1837 		MNT_IUNLOCK(mp);
1838 		if (coveredvp) {
1839 			vn_seqc_write_end(coveredvp);
1840 			VOP_UNLOCK(coveredvp);
1841 			vdrop(coveredvp);
1842 		}
1843 		if (rootvp != NULL) {
1844 			vn_seqc_write_end(rootvp);
1845 			vdrop(rootvp);
1846 		}
1847 		return (error);
1848 	}
1849 	mtx_lock(&mountlist_mtx);
1850 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1851 	mtx_unlock(&mountlist_mtx);
1852 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1853 	if (coveredvp != NULL) {
1854 		coveredvp->v_mountedhere = NULL;
1855 		vn_seqc_write_end(coveredvp);
1856 		VOP_UNLOCK(coveredvp);
1857 		vdrop(coveredvp);
1858 	}
1859 	mount_devctl_event("UNMOUNT", mp, false);
1860 	if (rootvp != NULL) {
1861 		vn_seqc_write_end(rootvp);
1862 		vdrop(rootvp);
1863 	}
1864 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1865 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
1866 		vrele(rootvnode);
1867 		rootvnode = NULL;
1868 	}
1869 	if (mp == rootdevmp)
1870 		rootdevmp = NULL;
1871 	vfs_mount_destroy(mp);
1872 	return (0);
1873 }
1874 
1875 /*
1876  * Report errors during filesystem mounting.
1877  */
1878 void
1879 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1880 {
1881 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1882 	va_list ap;
1883 	int error, len;
1884 	char *errmsg;
1885 
1886 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1887 	if (error || errmsg == NULL || len <= 0)
1888 		return;
1889 
1890 	va_start(ap, fmt);
1891 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1892 	va_end(ap);
1893 }
1894 
1895 void
1896 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1897 {
1898 	va_list ap;
1899 	int error, len;
1900 	char *errmsg;
1901 
1902 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1903 	if (error || errmsg == NULL || len <= 0)
1904 		return;
1905 
1906 	va_start(ap, fmt);
1907 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1908 	va_end(ap);
1909 }
1910 
1911 /*
1912  * ---------------------------------------------------------------------
1913  * Functions for querying mount options/arguments from filesystems.
1914  */
1915 
1916 /*
1917  * Check that no unknown options are given
1918  */
1919 int
1920 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1921 {
1922 	struct vfsopt *opt;
1923 	char errmsg[255];
1924 	const char **t, *p, *q;
1925 	int ret = 0;
1926 
1927 	TAILQ_FOREACH(opt, opts, link) {
1928 		p = opt->name;
1929 		q = NULL;
1930 		if (p[0] == 'n' && p[1] == 'o')
1931 			q = p + 2;
1932 		for(t = global_opts; *t != NULL; t++) {
1933 			if (strcmp(*t, p) == 0)
1934 				break;
1935 			if (q != NULL) {
1936 				if (strcmp(*t, q) == 0)
1937 					break;
1938 			}
1939 		}
1940 		if (*t != NULL)
1941 			continue;
1942 		for(t = legal; *t != NULL; t++) {
1943 			if (strcmp(*t, p) == 0)
1944 				break;
1945 			if (q != NULL) {
1946 				if (strcmp(*t, q) == 0)
1947 					break;
1948 			}
1949 		}
1950 		if (*t != NULL)
1951 			continue;
1952 		snprintf(errmsg, sizeof(errmsg),
1953 		    "mount option <%s> is unknown", p);
1954 		ret = EINVAL;
1955 	}
1956 	if (ret != 0) {
1957 		TAILQ_FOREACH(opt, opts, link) {
1958 			if (strcmp(opt->name, "errmsg") == 0) {
1959 				strncpy((char *)opt->value, errmsg, opt->len);
1960 				break;
1961 			}
1962 		}
1963 		if (opt == NULL)
1964 			printf("%s\n", errmsg);
1965 	}
1966 	return (ret);
1967 }
1968 
1969 /*
1970  * Get a mount option by its name.
1971  *
1972  * Return 0 if the option was found, ENOENT otherwise.
1973  * If len is non-NULL it will be filled with the length
1974  * of the option. If buf is non-NULL, it will be filled
1975  * with the address of the option.
1976  */
1977 int
1978 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
1979 {
1980 	struct vfsopt *opt;
1981 
1982 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1983 
1984 	TAILQ_FOREACH(opt, opts, link) {
1985 		if (strcmp(name, opt->name) == 0) {
1986 			opt->seen = 1;
1987 			if (len != NULL)
1988 				*len = opt->len;
1989 			if (buf != NULL)
1990 				*buf = opt->value;
1991 			return (0);
1992 		}
1993 	}
1994 	return (ENOENT);
1995 }
1996 
1997 int
1998 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1999 {
2000 	struct vfsopt *opt;
2001 
2002 	if (opts == NULL)
2003 		return (-1);
2004 
2005 	TAILQ_FOREACH(opt, opts, link) {
2006 		if (strcmp(name, opt->name) == 0) {
2007 			opt->seen = 1;
2008 			return (opt->pos);
2009 		}
2010 	}
2011 	return (-1);
2012 }
2013 
2014 int
2015 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2016 {
2017 	char *opt_value, *vtp;
2018 	quad_t iv;
2019 	int error, opt_len;
2020 
2021 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2022 	if (error != 0)
2023 		return (error);
2024 	if (opt_len == 0 || opt_value == NULL)
2025 		return (EINVAL);
2026 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2027 		return (EINVAL);
2028 	iv = strtoq(opt_value, &vtp, 0);
2029 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2030 		return (EINVAL);
2031 	if (iv < 0)
2032 		return (EINVAL);
2033 	switch (vtp[0]) {
2034 	case 't': case 'T':
2035 		iv *= 1024;
2036 		/* FALLTHROUGH */
2037 	case 'g': case 'G':
2038 		iv *= 1024;
2039 		/* FALLTHROUGH */
2040 	case 'm': case 'M':
2041 		iv *= 1024;
2042 		/* FALLTHROUGH */
2043 	case 'k': case 'K':
2044 		iv *= 1024;
2045 	case '\0':
2046 		break;
2047 	default:
2048 		return (EINVAL);
2049 	}
2050 	*value = iv;
2051 
2052 	return (0);
2053 }
2054 
2055 char *
2056 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2057 {
2058 	struct vfsopt *opt;
2059 
2060 	*error = 0;
2061 	TAILQ_FOREACH(opt, opts, link) {
2062 		if (strcmp(name, opt->name) != 0)
2063 			continue;
2064 		opt->seen = 1;
2065 		if (opt->len == 0 ||
2066 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2067 			*error = EINVAL;
2068 			return (NULL);
2069 		}
2070 		return (opt->value);
2071 	}
2072 	*error = ENOENT;
2073 	return (NULL);
2074 }
2075 
2076 int
2077 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2078 	uint64_t val)
2079 {
2080 	struct vfsopt *opt;
2081 
2082 	TAILQ_FOREACH(opt, opts, link) {
2083 		if (strcmp(name, opt->name) == 0) {
2084 			opt->seen = 1;
2085 			if (w != NULL)
2086 				*w |= val;
2087 			return (1);
2088 		}
2089 	}
2090 	if (w != NULL)
2091 		*w &= ~val;
2092 	return (0);
2093 }
2094 
2095 int
2096 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2097 {
2098 	va_list ap;
2099 	struct vfsopt *opt;
2100 	int ret;
2101 
2102 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2103 
2104 	TAILQ_FOREACH(opt, opts, link) {
2105 		if (strcmp(name, opt->name) != 0)
2106 			continue;
2107 		opt->seen = 1;
2108 		if (opt->len == 0 || opt->value == NULL)
2109 			return (0);
2110 		if (((char *)opt->value)[opt->len - 1] != '\0')
2111 			return (0);
2112 		va_start(ap, fmt);
2113 		ret = vsscanf(opt->value, fmt, ap);
2114 		va_end(ap);
2115 		return (ret);
2116 	}
2117 	return (0);
2118 }
2119 
2120 int
2121 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2122 {
2123 	struct vfsopt *opt;
2124 
2125 	TAILQ_FOREACH(opt, opts, link) {
2126 		if (strcmp(name, opt->name) != 0)
2127 			continue;
2128 		opt->seen = 1;
2129 		if (opt->value == NULL)
2130 			opt->len = len;
2131 		else {
2132 			if (opt->len != len)
2133 				return (EINVAL);
2134 			bcopy(value, opt->value, len);
2135 		}
2136 		return (0);
2137 	}
2138 	return (ENOENT);
2139 }
2140 
2141 int
2142 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2143 {
2144 	struct vfsopt *opt;
2145 
2146 	TAILQ_FOREACH(opt, opts, link) {
2147 		if (strcmp(name, opt->name) != 0)
2148 			continue;
2149 		opt->seen = 1;
2150 		if (opt->value == NULL)
2151 			opt->len = len;
2152 		else {
2153 			if (opt->len < len)
2154 				return (EINVAL);
2155 			opt->len = len;
2156 			bcopy(value, opt->value, len);
2157 		}
2158 		return (0);
2159 	}
2160 	return (ENOENT);
2161 }
2162 
2163 int
2164 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2165 {
2166 	struct vfsopt *opt;
2167 
2168 	TAILQ_FOREACH(opt, opts, link) {
2169 		if (strcmp(name, opt->name) != 0)
2170 			continue;
2171 		opt->seen = 1;
2172 		if (opt->value == NULL)
2173 			opt->len = strlen(value) + 1;
2174 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2175 			return (EINVAL);
2176 		return (0);
2177 	}
2178 	return (ENOENT);
2179 }
2180 
2181 /*
2182  * Find and copy a mount option.
2183  *
2184  * The size of the buffer has to be specified
2185  * in len, if it is not the same length as the
2186  * mount option, EINVAL is returned.
2187  * Returns ENOENT if the option is not found.
2188  */
2189 int
2190 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2191 {
2192 	struct vfsopt *opt;
2193 
2194 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2195 
2196 	TAILQ_FOREACH(opt, opts, link) {
2197 		if (strcmp(name, opt->name) == 0) {
2198 			opt->seen = 1;
2199 			if (len != opt->len)
2200 				return (EINVAL);
2201 			bcopy(opt->value, dest, opt->len);
2202 			return (0);
2203 		}
2204 	}
2205 	return (ENOENT);
2206 }
2207 
2208 int
2209 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2210 {
2211 
2212 	/*
2213 	 * Filesystems only fill in part of the structure for updates, we
2214 	 * have to read the entirety first to get all content.
2215 	 */
2216 	if (sbp != &mp->mnt_stat)
2217 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2218 
2219 	/*
2220 	 * Set these in case the underlying filesystem fails to do so.
2221 	 */
2222 	sbp->f_version = STATFS_VERSION;
2223 	sbp->f_namemax = NAME_MAX;
2224 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2225 
2226 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2227 }
2228 
2229 void
2230 vfs_mountedfrom(struct mount *mp, const char *from)
2231 {
2232 
2233 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2234 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2235 	    sizeof mp->mnt_stat.f_mntfromname);
2236 }
2237 
2238 /*
2239  * ---------------------------------------------------------------------
2240  * This is the api for building mount args and mounting filesystems from
2241  * inside the kernel.
2242  *
2243  * The API works by accumulation of individual args.  First error is
2244  * latched.
2245  *
2246  * XXX: should be documented in new manpage kernel_mount(9)
2247  */
2248 
2249 /* A memory allocation which must be freed when we are done */
2250 struct mntaarg {
2251 	SLIST_ENTRY(mntaarg)	next;
2252 };
2253 
2254 /* The header for the mount arguments */
2255 struct mntarg {
2256 	struct iovec *v;
2257 	int len;
2258 	int error;
2259 	SLIST_HEAD(, mntaarg)	list;
2260 };
2261 
2262 /*
2263  * Add a boolean argument.
2264  *
2265  * flag is the boolean value.
2266  * name must start with "no".
2267  */
2268 struct mntarg *
2269 mount_argb(struct mntarg *ma, int flag, const char *name)
2270 {
2271 
2272 	KASSERT(name[0] == 'n' && name[1] == 'o',
2273 	    ("mount_argb(...,%s): name must start with 'no'", name));
2274 
2275 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2276 }
2277 
2278 /*
2279  * Add an argument printf style
2280  */
2281 struct mntarg *
2282 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2283 {
2284 	va_list ap;
2285 	struct mntaarg *maa;
2286 	struct sbuf *sb;
2287 	int len;
2288 
2289 	if (ma == NULL) {
2290 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2291 		SLIST_INIT(&ma->list);
2292 	}
2293 	if (ma->error)
2294 		return (ma);
2295 
2296 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2297 	    M_MOUNT, M_WAITOK);
2298 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2299 	ma->v[ma->len].iov_len = strlen(name) + 1;
2300 	ma->len++;
2301 
2302 	sb = sbuf_new_auto();
2303 	va_start(ap, fmt);
2304 	sbuf_vprintf(sb, fmt, ap);
2305 	va_end(ap);
2306 	sbuf_finish(sb);
2307 	len = sbuf_len(sb) + 1;
2308 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2309 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2310 	bcopy(sbuf_data(sb), maa + 1, len);
2311 	sbuf_delete(sb);
2312 
2313 	ma->v[ma->len].iov_base = maa + 1;
2314 	ma->v[ma->len].iov_len = len;
2315 	ma->len++;
2316 
2317 	return (ma);
2318 }
2319 
2320 /*
2321  * Add an argument which is a userland string.
2322  */
2323 struct mntarg *
2324 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2325 {
2326 	struct mntaarg *maa;
2327 	char *tbuf;
2328 
2329 	if (val == NULL)
2330 		return (ma);
2331 	if (ma == NULL) {
2332 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2333 		SLIST_INIT(&ma->list);
2334 	}
2335 	if (ma->error)
2336 		return (ma);
2337 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2338 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2339 	tbuf = (void *)(maa + 1);
2340 	ma->error = copyinstr(val, tbuf, len, NULL);
2341 	return (mount_arg(ma, name, tbuf, -1));
2342 }
2343 
2344 /*
2345  * Plain argument.
2346  *
2347  * If length is -1, treat value as a C string.
2348  */
2349 struct mntarg *
2350 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2351 {
2352 
2353 	if (ma == NULL) {
2354 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2355 		SLIST_INIT(&ma->list);
2356 	}
2357 	if (ma->error)
2358 		return (ma);
2359 
2360 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2361 	    M_MOUNT, M_WAITOK);
2362 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2363 	ma->v[ma->len].iov_len = strlen(name) + 1;
2364 	ma->len++;
2365 
2366 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2367 	if (len < 0)
2368 		ma->v[ma->len].iov_len = strlen(val) + 1;
2369 	else
2370 		ma->v[ma->len].iov_len = len;
2371 	ma->len++;
2372 	return (ma);
2373 }
2374 
2375 /*
2376  * Free a mntarg structure
2377  */
2378 static void
2379 free_mntarg(struct mntarg *ma)
2380 {
2381 	struct mntaarg *maa;
2382 
2383 	while (!SLIST_EMPTY(&ma->list)) {
2384 		maa = SLIST_FIRST(&ma->list);
2385 		SLIST_REMOVE_HEAD(&ma->list, next);
2386 		free(maa, M_MOUNT);
2387 	}
2388 	free(ma->v, M_MOUNT);
2389 	free(ma, M_MOUNT);
2390 }
2391 
2392 /*
2393  * Mount a filesystem
2394  */
2395 int
2396 kernel_mount(struct mntarg *ma, uint64_t flags)
2397 {
2398 	struct uio auio;
2399 	int error;
2400 
2401 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2402 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2403 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2404 
2405 	auio.uio_iov = ma->v;
2406 	auio.uio_iovcnt = ma->len;
2407 	auio.uio_segflg = UIO_SYSSPACE;
2408 
2409 	error = ma->error;
2410 	if (!error)
2411 		error = vfs_donmount(curthread, flags, &auio);
2412 	free_mntarg(ma);
2413 	return (error);
2414 }
2415 
2416 /*
2417  * A printflike function to mount a filesystem.
2418  */
2419 int
2420 kernel_vmount(int flags, ...)
2421 {
2422 	struct mntarg *ma = NULL;
2423 	va_list ap;
2424 	const char *cp;
2425 	const void *vp;
2426 	int error;
2427 
2428 	va_start(ap, flags);
2429 	for (;;) {
2430 		cp = va_arg(ap, const char *);
2431 		if (cp == NULL)
2432 			break;
2433 		vp = va_arg(ap, const void *);
2434 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2435 	}
2436 	va_end(ap);
2437 
2438 	error = kernel_mount(ma, flags);
2439 	return (error);
2440 }
2441 
2442 /* Map from mount options to printable formats. */
2443 static struct mntoptnames optnames[] = {
2444 	MNTOPT_NAMES
2445 };
2446 
2447 static void
2448 mount_devctl_event_mntopt(struct sbuf *sb, const char *what, struct vfsoptlist *opts)
2449 {
2450 	struct vfsopt *opt;
2451 
2452 	if (opts == NULL || TAILQ_EMPTY(opts))
2453 		return;
2454 	sbuf_printf(sb, " %s=\"", what);
2455 	TAILQ_FOREACH(opt, opts, link) {
2456 		if (opt->name[0] == '\0' || (opt->len > 0 && *(char *)opt->value == '\0'))
2457 			continue;
2458 		devctl_safe_quote_sb(sb, opt->name);
2459 		if (opt->len > 0) {
2460 			sbuf_putc(sb, '=');
2461 			devctl_safe_quote_sb(sb, opt->value);
2462 		}
2463 		sbuf_putc(sb, ';');
2464 	}
2465 	sbuf_putc(sb, '"');
2466 }
2467 
2468 #define DEVCTL_LEN 1024
2469 static void
2470 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2471 {
2472 	const uint8_t *cp;
2473 	struct mntoptnames *fp;
2474 	struct sbuf sb;
2475 	struct statfs *sfp = &mp->mnt_stat;
2476 	char *buf;
2477 
2478 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2479 	if (buf == NULL)
2480 		return;
2481 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2482 	sbuf_cpy(&sb, "mount-point=\"");
2483 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2484 	sbuf_cat(&sb, "\" mount-dev=\"");
2485 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2486 	sbuf_cat(&sb, "\" mount-type=\"");
2487 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2488 	sbuf_cat(&sb, "\" fsid=0x");
2489 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2490 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2491 		sbuf_printf(&sb, "%02x", cp[i]);
2492 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2493 	for (fp = optnames; fp->o_opt != 0; fp++) {
2494 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2495 			sbuf_cat(&sb, fp->o_name);
2496 			sbuf_putc(&sb, ';');
2497 		}
2498 	}
2499 	sbuf_putc(&sb, '"');
2500 	mount_devctl_event_mntopt(&sb, "opt", mp->mnt_opt);
2501 	if (donew)
2502 		mount_devctl_event_mntopt(&sb, "optnew", mp->mnt_optnew);
2503 	sbuf_finish(&sb);
2504 
2505 	if (sbuf_error(&sb) == 0)
2506 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2507 	sbuf_delete(&sb);
2508 	free(buf, M_MOUNT);
2509 }
2510 
2511 /*
2512  * Suspend write operations on all local writeable filesystems.  Does
2513  * full sync of them in the process.
2514  *
2515  * Iterate over the mount points in reverse order, suspending most
2516  * recently mounted filesystems first.  It handles a case where a
2517  * filesystem mounted from a md(4) vnode-backed device should be
2518  * suspended before the filesystem that owns the vnode.
2519  */
2520 void
2521 suspend_all_fs(void)
2522 {
2523 	struct mount *mp;
2524 	int error;
2525 
2526 	mtx_lock(&mountlist_mtx);
2527 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2528 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2529 		if (error != 0)
2530 			continue;
2531 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2532 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2533 			mtx_lock(&mountlist_mtx);
2534 			vfs_unbusy(mp);
2535 			continue;
2536 		}
2537 		error = vfs_write_suspend(mp, 0);
2538 		if (error == 0) {
2539 			MNT_ILOCK(mp);
2540 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2541 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2542 			MNT_IUNLOCK(mp);
2543 			mtx_lock(&mountlist_mtx);
2544 		} else {
2545 			printf("suspend of %s failed, error %d\n",
2546 			    mp->mnt_stat.f_mntonname, error);
2547 			mtx_lock(&mountlist_mtx);
2548 			vfs_unbusy(mp);
2549 		}
2550 	}
2551 	mtx_unlock(&mountlist_mtx);
2552 }
2553 
2554 void
2555 resume_all_fs(void)
2556 {
2557 	struct mount *mp;
2558 
2559 	mtx_lock(&mountlist_mtx);
2560 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2561 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2562 			continue;
2563 		mtx_unlock(&mountlist_mtx);
2564 		MNT_ILOCK(mp);
2565 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2566 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2567 		MNT_IUNLOCK(mp);
2568 		vfs_write_resume(mp, 0);
2569 		mtx_lock(&mountlist_mtx);
2570 		vfs_unbusy(mp);
2571 	}
2572 	mtx_unlock(&mountlist_mtx);
2573 }
2574