xref: /freebsd/sys/kern/vfs_mount.c (revision e0c4386e7e71d93b0edc0c8fa156263fc4a8b0b6)
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/param.h>
40 #include <sys/conf.h>
41 #include <sys/smp.h>
42 #include <sys/devctl.h>
43 #include <sys/eventhandler.h>
44 #include <sys/fcntl.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/ktr.h>
48 #include <sys/libkern.h>
49 #include <sys/limits.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/filedesc.h>
57 #include <sys/reboot.h>
58 #include <sys/sbuf.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysproto.h>
61 #include <sys/sx.h>
62 #include <sys/sysctl.h>
63 #include <sys/systm.h>
64 #include <sys/taskqueue.h>
65 #include <sys/vnode.h>
66 #include <vm/uma.h>
67 
68 #include <geom/geom.h>
69 
70 #include <machine/stdarg.h>
71 
72 #include <security/audit/audit.h>
73 #include <security/mac/mac_framework.h>
74 
75 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
76 
77 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
78 		    uint64_t fsflags, bool jail_export,
79 		    struct vfsoptlist **optlist);
80 static void	free_mntarg(struct mntarg *ma);
81 
82 static int	usermount = 0;
83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
84     "Unprivileged users may mount and unmount file systems");
85 
86 static bool	default_autoro = false;
87 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
88     "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
89 
90 static bool	recursive_forced_unmount = false;
91 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW,
92     &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts"
93     " when a file system is forcibly unmounted");
94 
95 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount,
96     CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls");
97 
98 static unsigned int	deferred_unmount_retry_limit = 10;
99 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW,
100     &deferred_unmount_retry_limit, 0,
101     "Maximum number of retries for deferred unmount failure");
102 
103 static int	deferred_unmount_retry_delay_hz;
104 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW,
105     &deferred_unmount_retry_delay_hz, 0,
106     "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount");
107 
108 static int	deferred_unmount_total_retries = 0;
109 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD,
110     &deferred_unmount_total_retries, 0,
111     "Total number of retried deferred unmounts");
112 
113 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
114 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
115 static uma_zone_t mount_zone;
116 
117 /* List of mounted filesystems. */
118 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
119 
120 /* For any iteration/modification of mountlist */
121 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
122 
123 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
124 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
125 
126 static void vfs_deferred_unmount(void *arg, int pending);
127 static struct timeout_task deferred_unmount_task;
128 static struct mtx deferred_unmount_lock;
129 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount",
130     MTX_DEF);
131 static STAILQ_HEAD(, mount) deferred_unmount_list =
132     STAILQ_HEAD_INITIALIZER(deferred_unmount_list);
133 TASKQUEUE_DEFINE_THREAD(deferred_unmount);
134 
135 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
136 
137 /*
138  * Global opts, taken by all filesystems
139  */
140 static const char *global_opts[] = {
141 	"errmsg",
142 	"fstype",
143 	"fspath",
144 	"ro",
145 	"rw",
146 	"nosuid",
147 	"noexec",
148 	NULL
149 };
150 
151 static int
152 mount_init(void *mem, int size, int flags)
153 {
154 	struct mount *mp;
155 
156 	mp = (struct mount *)mem;
157 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
158 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
159 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
160 	mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
161 	mp->mnt_ref = 0;
162 	mp->mnt_vfs_ops = 1;
163 	mp->mnt_rootvnode = NULL;
164 	return (0);
165 }
166 
167 static void
168 mount_fini(void *mem, int size)
169 {
170 	struct mount *mp;
171 
172 	mp = (struct mount *)mem;
173 	uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
174 	lockdestroy(&mp->mnt_explock);
175 	mtx_destroy(&mp->mnt_listmtx);
176 	mtx_destroy(&mp->mnt_mtx);
177 }
178 
179 static void
180 vfs_mount_init(void *dummy __unused)
181 {
182 	TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task,
183 	    0, vfs_deferred_unmount, NULL);
184 	deferred_unmount_retry_delay_hz = hz;
185 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
186 	    NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
187 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
188 }
189 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
190 
191 /*
192  * ---------------------------------------------------------------------
193  * Functions for building and sanitizing the mount options
194  */
195 
196 /* Remove one mount option. */
197 static void
198 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
199 {
200 
201 	TAILQ_REMOVE(opts, opt, link);
202 	free(opt->name, M_MOUNT);
203 	if (opt->value != NULL)
204 		free(opt->value, M_MOUNT);
205 	free(opt, M_MOUNT);
206 }
207 
208 /* Release all resources related to the mount options. */
209 void
210 vfs_freeopts(struct vfsoptlist *opts)
211 {
212 	struct vfsopt *opt;
213 
214 	while (!TAILQ_EMPTY(opts)) {
215 		opt = TAILQ_FIRST(opts);
216 		vfs_freeopt(opts, opt);
217 	}
218 	free(opts, M_MOUNT);
219 }
220 
221 void
222 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
223 {
224 	struct vfsopt *opt, *temp;
225 
226 	if (opts == NULL)
227 		return;
228 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
229 		if (strcmp(opt->name, name) == 0)
230 			vfs_freeopt(opts, opt);
231 	}
232 }
233 
234 static int
235 vfs_isopt_ro(const char *opt)
236 {
237 
238 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
239 	    strcmp(opt, "norw") == 0)
240 		return (1);
241 	return (0);
242 }
243 
244 static int
245 vfs_isopt_rw(const char *opt)
246 {
247 
248 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
249 		return (1);
250 	return (0);
251 }
252 
253 /*
254  * Check if options are equal (with or without the "no" prefix).
255  */
256 static int
257 vfs_equalopts(const char *opt1, const char *opt2)
258 {
259 	char *p;
260 
261 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
262 	if (strcmp(opt1, opt2) == 0)
263 		return (1);
264 	/* "noopt" vs. "opt" */
265 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
266 		return (1);
267 	/* "opt" vs. "noopt" */
268 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
269 		return (1);
270 	while ((p = strchr(opt1, '.')) != NULL &&
271 	    !strncmp(opt1, opt2, ++p - opt1)) {
272 		opt2 += p - opt1;
273 		opt1 = p;
274 		/* "foo.noopt" vs. "foo.opt" */
275 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
276 			return (1);
277 		/* "foo.opt" vs. "foo.noopt" */
278 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
279 			return (1);
280 	}
281 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
282 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
283 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
284 		return (1);
285 	return (0);
286 }
287 
288 /*
289  * If a mount option is specified several times,
290  * (with or without the "no" prefix) only keep
291  * the last occurrence of it.
292  */
293 static void
294 vfs_sanitizeopts(struct vfsoptlist *opts)
295 {
296 	struct vfsopt *opt, *opt2, *tmp;
297 
298 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
299 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
300 		while (opt2 != NULL) {
301 			if (vfs_equalopts(opt->name, opt2->name)) {
302 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
303 				vfs_freeopt(opts, opt2);
304 				opt2 = tmp;
305 			} else {
306 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
307 			}
308 		}
309 	}
310 }
311 
312 /*
313  * Build a linked list of mount options from a struct uio.
314  */
315 int
316 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
317 {
318 	struct vfsoptlist *opts;
319 	struct vfsopt *opt;
320 	size_t memused, namelen, optlen;
321 	unsigned int i, iovcnt;
322 	int error;
323 
324 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
325 	TAILQ_INIT(opts);
326 	memused = 0;
327 	iovcnt = auio->uio_iovcnt;
328 	for (i = 0; i < iovcnt; i += 2) {
329 		namelen = auio->uio_iov[i].iov_len;
330 		optlen = auio->uio_iov[i + 1].iov_len;
331 		memused += sizeof(struct vfsopt) + optlen + namelen;
332 		/*
333 		 * Avoid consuming too much memory, and attempts to overflow
334 		 * memused.
335 		 */
336 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
337 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
338 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
339 			error = EINVAL;
340 			goto bad;
341 		}
342 
343 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
344 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
345 		opt->value = NULL;
346 		opt->len = 0;
347 		opt->pos = i / 2;
348 		opt->seen = 0;
349 
350 		/*
351 		 * Do this early, so jumps to "bad" will free the current
352 		 * option.
353 		 */
354 		TAILQ_INSERT_TAIL(opts, opt, link);
355 
356 		if (auio->uio_segflg == UIO_SYSSPACE) {
357 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
358 		} else {
359 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
360 			    namelen);
361 			if (error)
362 				goto bad;
363 		}
364 		/* Ensure names are null-terminated strings. */
365 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
366 			error = EINVAL;
367 			goto bad;
368 		}
369 		if (optlen != 0) {
370 			opt->len = optlen;
371 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
372 			if (auio->uio_segflg == UIO_SYSSPACE) {
373 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
374 				    optlen);
375 			} else {
376 				error = copyin(auio->uio_iov[i + 1].iov_base,
377 				    opt->value, optlen);
378 				if (error)
379 					goto bad;
380 			}
381 		}
382 	}
383 	vfs_sanitizeopts(opts);
384 	*options = opts;
385 	return (0);
386 bad:
387 	vfs_freeopts(opts);
388 	return (error);
389 }
390 
391 /*
392  * Merge the old mount options with the new ones passed
393  * in the MNT_UPDATE case.
394  *
395  * XXX: This function will keep a "nofoo" option in the new
396  * options.  E.g, if the option's canonical name is "foo",
397  * "nofoo" ends up in the mount point's active options.
398  */
399 static void
400 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
401 {
402 	struct vfsopt *opt, *new;
403 
404 	TAILQ_FOREACH(opt, oldopts, link) {
405 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
406 		new->name = strdup(opt->name, M_MOUNT);
407 		if (opt->len != 0) {
408 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
409 			bcopy(opt->value, new->value, opt->len);
410 		} else
411 			new->value = NULL;
412 		new->len = opt->len;
413 		new->seen = opt->seen;
414 		TAILQ_INSERT_HEAD(toopts, new, link);
415 	}
416 	vfs_sanitizeopts(toopts);
417 }
418 
419 /*
420  * Mount a filesystem.
421  */
422 #ifndef _SYS_SYSPROTO_H_
423 struct nmount_args {
424 	struct iovec *iovp;
425 	unsigned int iovcnt;
426 	int flags;
427 };
428 #endif
429 int
430 sys_nmount(struct thread *td, struct nmount_args *uap)
431 {
432 	struct uio *auio;
433 	int error;
434 	u_int iovcnt;
435 	uint64_t flags;
436 
437 	/*
438 	 * Mount flags are now 64-bits. On 32-bit archtectures only
439 	 * 32-bits are passed in, but from here on everything handles
440 	 * 64-bit flags correctly.
441 	 */
442 	flags = uap->flags;
443 
444 	AUDIT_ARG_FFLAGS(flags);
445 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
446 	    uap->iovp, uap->iovcnt, flags);
447 
448 	/*
449 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
450 	 * userspace to set this flag, but we must filter it out if we want
451 	 * MNT_UPDATE on the root file system to work.
452 	 * MNT_ROOTFS should only be set by the kernel when mounting its
453 	 * root file system.
454 	 */
455 	flags &= ~MNT_ROOTFS;
456 
457 	iovcnt = uap->iovcnt;
458 	/*
459 	 * Check that we have an even number of iovec's
460 	 * and that we have at least two options.
461 	 */
462 	if ((iovcnt & 1) || (iovcnt < 4)) {
463 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
464 		    uap->iovcnt);
465 		return (EINVAL);
466 	}
467 
468 	error = copyinuio(uap->iovp, iovcnt, &auio);
469 	if (error) {
470 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
471 		    __func__, error);
472 		return (error);
473 	}
474 	error = vfs_donmount(td, flags, auio);
475 
476 	free(auio, M_IOV);
477 	return (error);
478 }
479 
480 /*
481  * ---------------------------------------------------------------------
482  * Various utility functions
483  */
484 
485 /*
486  * Get a reference on a mount point from a vnode.
487  *
488  * The vnode is allowed to be passed unlocked and race against dooming. Note in
489  * such case there are no guarantees the referenced mount point will still be
490  * associated with it after the function returns.
491  */
492 struct mount *
493 vfs_ref_from_vp(struct vnode *vp)
494 {
495 	struct mount *mp;
496 	struct mount_pcpu *mpcpu;
497 
498 	mp = atomic_load_ptr(&vp->v_mount);
499 	if (__predict_false(mp == NULL)) {
500 		return (mp);
501 	}
502 	if (vfs_op_thread_enter(mp, mpcpu)) {
503 		if (__predict_true(mp == vp->v_mount)) {
504 			vfs_mp_count_add_pcpu(mpcpu, ref, 1);
505 			vfs_op_thread_exit(mp, mpcpu);
506 		} else {
507 			vfs_op_thread_exit(mp, mpcpu);
508 			mp = NULL;
509 		}
510 	} else {
511 		MNT_ILOCK(mp);
512 		if (mp == vp->v_mount) {
513 			MNT_REF(mp);
514 			MNT_IUNLOCK(mp);
515 		} else {
516 			MNT_IUNLOCK(mp);
517 			mp = NULL;
518 		}
519 	}
520 	return (mp);
521 }
522 
523 void
524 vfs_ref(struct mount *mp)
525 {
526 	struct mount_pcpu *mpcpu;
527 
528 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
529 	if (vfs_op_thread_enter(mp, mpcpu)) {
530 		vfs_mp_count_add_pcpu(mpcpu, ref, 1);
531 		vfs_op_thread_exit(mp, mpcpu);
532 		return;
533 	}
534 
535 	MNT_ILOCK(mp);
536 	MNT_REF(mp);
537 	MNT_IUNLOCK(mp);
538 }
539 
540 /*
541  * Register ump as an upper mount of the mount associated with
542  * vnode vp.  This registration will be tracked through
543  * mount_upper_node upper, which should be allocated by the
544  * caller and stored in per-mount data associated with mp.
545  *
546  * If successful, this function will return the mount associated
547  * with vp, and will ensure that it cannot be unmounted until
548  * ump has been unregistered as one of its upper mounts.
549  *
550  * Upon failure this function will return NULL.
551  */
552 struct mount *
553 vfs_register_upper_from_vp(struct vnode *vp, struct mount *ump,
554     struct mount_upper_node *upper)
555 {
556 	struct mount *mp;
557 
558 	mp = atomic_load_ptr(&vp->v_mount);
559 	if (mp == NULL)
560 		return (NULL);
561 	MNT_ILOCK(mp);
562 	if (mp != vp->v_mount ||
563 	    ((mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_RECURSE)) != 0)) {
564 		MNT_IUNLOCK(mp);
565 		return (NULL);
566 	}
567 	KASSERT(ump != mp, ("upper and lower mounts are identical"));
568 	upper->mp = ump;
569 	MNT_REF(mp);
570 	TAILQ_INSERT_TAIL(&mp->mnt_uppers, upper, mnt_upper_link);
571 	MNT_IUNLOCK(mp);
572 	return (mp);
573 }
574 
575 /*
576  * Register upper mount ump to receive vnode unlink/reclaim
577  * notifications from lower mount mp. This registration will
578  * be tracked through mount_upper_node upper, which should be
579  * allocated by the caller and stored in per-mount data
580  * associated with mp.
581  *
582  * ump must already be registered as an upper mount of mp
583  * through a call to vfs_register_upper_from_vp().
584  */
585 void
586 vfs_register_for_notification(struct mount *mp, struct mount *ump,
587     struct mount_upper_node *upper)
588 {
589 	upper->mp = ump;
590 	MNT_ILOCK(mp);
591 	TAILQ_INSERT_TAIL(&mp->mnt_notify, upper, mnt_upper_link);
592 	MNT_IUNLOCK(mp);
593 }
594 
595 static void
596 vfs_drain_upper_locked(struct mount *mp)
597 {
598 	mtx_assert(MNT_MTX(mp), MA_OWNED);
599 	while (mp->mnt_upper_pending != 0) {
600 		mp->mnt_kern_flag |= MNTK_UPPER_WAITER;
601 		msleep(&mp->mnt_uppers, MNT_MTX(mp), 0, "mntupw", 0);
602 	}
603 }
604 
605 /*
606  * Undo a previous call to vfs_register_for_notification().
607  * The mount represented by upper must be currently registered
608  * as an upper mount for mp.
609  */
610 void
611 vfs_unregister_for_notification(struct mount *mp,
612     struct mount_upper_node *upper)
613 {
614 	MNT_ILOCK(mp);
615 	vfs_drain_upper_locked(mp);
616 	TAILQ_REMOVE(&mp->mnt_notify, upper, mnt_upper_link);
617 	MNT_IUNLOCK(mp);
618 }
619 
620 /*
621  * Undo a previous call to vfs_register_upper_from_vp().
622  * This must be done before mp can be unmounted.
623  */
624 void
625 vfs_unregister_upper(struct mount *mp, struct mount_upper_node *upper)
626 {
627 	MNT_ILOCK(mp);
628 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
629 	    ("registered upper with pending unmount"));
630 	vfs_drain_upper_locked(mp);
631 	TAILQ_REMOVE(&mp->mnt_uppers, upper, mnt_upper_link);
632 	if ((mp->mnt_kern_flag & MNTK_TASKQUEUE_WAITER) != 0 &&
633 	    TAILQ_EMPTY(&mp->mnt_uppers)) {
634 		mp->mnt_kern_flag &= ~MNTK_TASKQUEUE_WAITER;
635 		wakeup(&mp->mnt_taskqueue_link);
636 	}
637 	MNT_REL(mp);
638 	MNT_IUNLOCK(mp);
639 }
640 
641 void
642 vfs_rel(struct mount *mp)
643 {
644 	struct mount_pcpu *mpcpu;
645 
646 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
647 	if (vfs_op_thread_enter(mp, mpcpu)) {
648 		vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
649 		vfs_op_thread_exit(mp, mpcpu);
650 		return;
651 	}
652 
653 	MNT_ILOCK(mp);
654 	MNT_REL(mp);
655 	MNT_IUNLOCK(mp);
656 }
657 
658 /*
659  * Allocate and initialize the mount point struct.
660  */
661 struct mount *
662 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
663     struct ucred *cred)
664 {
665 	struct mount *mp;
666 
667 	mp = uma_zalloc(mount_zone, M_WAITOK);
668 	bzero(&mp->mnt_startzero,
669 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
670 	mp->mnt_kern_flag = 0;
671 	mp->mnt_flag = 0;
672 	mp->mnt_rootvnode = NULL;
673 	mp->mnt_vnodecovered = NULL;
674 	mp->mnt_op = NULL;
675 	mp->mnt_vfc = NULL;
676 	TAILQ_INIT(&mp->mnt_nvnodelist);
677 	mp->mnt_nvnodelistsize = 0;
678 	TAILQ_INIT(&mp->mnt_lazyvnodelist);
679 	mp->mnt_lazyvnodelistsize = 0;
680 	MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 &&
681 	    mp->mnt_writeopcount == 0, mp);
682 	MPASSERT(mp->mnt_vfs_ops == 1, mp,
683 	    ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
684 	(void) vfs_busy(mp, MBF_NOWAIT);
685 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
686 	mp->mnt_op = vfsp->vfc_vfsops;
687 	mp->mnt_vfc = vfsp;
688 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
689 	mp->mnt_gen++;
690 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
691 	mp->mnt_vnodecovered = vp;
692 	mp->mnt_cred = crdup(cred);
693 	mp->mnt_stat.f_owner = cred->cr_uid;
694 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
695 	mp->mnt_iosize_max = DFLTPHYS;
696 #ifdef MAC
697 	mac_mount_init(mp);
698 	mac_mount_create(cred, mp);
699 #endif
700 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
701 	mp->mnt_upper_pending = 0;
702 	TAILQ_INIT(&mp->mnt_uppers);
703 	TAILQ_INIT(&mp->mnt_notify);
704 	mp->mnt_taskqueue_flags = 0;
705 	mp->mnt_unmount_retries = 0;
706 	return (mp);
707 }
708 
709 /*
710  * Destroy the mount struct previously allocated by vfs_mount_alloc().
711  */
712 void
713 vfs_mount_destroy(struct mount *mp)
714 {
715 
716 	MPPASS(mp->mnt_vfs_ops != 0, mp);
717 
718 	vfs_assert_mount_counters(mp);
719 
720 	MNT_ILOCK(mp);
721 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
722 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
723 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
724 		wakeup(mp);
725 	}
726 	while (mp->mnt_ref)
727 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
728 	KASSERT(mp->mnt_ref == 0,
729 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
730 	    __FILE__, __LINE__));
731 	MPPASS(mp->mnt_writeopcount == 0, mp);
732 	MPPASS(mp->mnt_secondary_writes == 0, mp);
733 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
734 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
735 		struct vnode *vp;
736 
737 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
738 			vn_printf(vp, "dangling vnode ");
739 		panic("unmount: dangling vnode");
740 	}
741 	KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending"));
742 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
743 	KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify"));
744 	MPPASS(mp->mnt_nvnodelistsize == 0, mp);
745 	MPPASS(mp->mnt_lazyvnodelistsize == 0, mp);
746 	MPPASS(mp->mnt_lockref == 0, mp);
747 	MNT_IUNLOCK(mp);
748 
749 	MPASSERT(mp->mnt_vfs_ops == 1, mp,
750 	    ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
751 
752 	MPASSERT(mp->mnt_rootvnode == NULL, mp,
753 	    ("mount point still has a root vnode %p", mp->mnt_rootvnode));
754 
755 	if (mp->mnt_vnodecovered != NULL)
756 		vrele(mp->mnt_vnodecovered);
757 #ifdef MAC
758 	mac_mount_destroy(mp);
759 #endif
760 	if (mp->mnt_opt != NULL)
761 		vfs_freeopts(mp->mnt_opt);
762 	if (mp->mnt_exjail != NULL) {
763 		atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt,
764 		    1);
765 		crfree(mp->mnt_exjail);
766 	}
767 	if (mp->mnt_export != NULL) {
768 		vfs_free_addrlist(mp->mnt_export);
769 		free(mp->mnt_export, M_MOUNT);
770 	}
771 	crfree(mp->mnt_cred);
772 	uma_zfree(mount_zone, mp);
773 }
774 
775 static bool
776 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
777 {
778 	/* This is an upgrade of an exisiting mount. */
779 	if ((fsflags & MNT_UPDATE) != 0)
780 		return (false);
781 	/* This is already an R/O mount. */
782 	if ((fsflags & MNT_RDONLY) != 0)
783 		return (false);
784 
785 	switch (error) {
786 	case ENODEV:	/* generic, geom, ... */
787 	case EACCES:	/* cam/scsi, ... */
788 	case EROFS:	/* md, mmcsd, ... */
789 		/*
790 		 * These errors can be returned by the storage layer to signal
791 		 * that the media is read-only.  No harm in the R/O mount
792 		 * attempt if the error was returned for some other reason.
793 		 */
794 		return (true);
795 	default:
796 		return (false);
797 	}
798 }
799 
800 int
801 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
802 {
803 	struct vfsoptlist *optlist;
804 	struct vfsopt *opt, *tmp_opt;
805 	char *fstype, *fspath, *errmsg;
806 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
807 	bool autoro, has_nonexport, jail_export;
808 
809 	errmsg = fspath = NULL;
810 	errmsg_len = fspathlen = 0;
811 	errmsg_pos = -1;
812 	autoro = default_autoro;
813 
814 	error = vfs_buildopts(fsoptions, &optlist);
815 	if (error)
816 		return (error);
817 
818 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
819 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
820 
821 	/*
822 	 * We need these two options before the others,
823 	 * and they are mandatory for any filesystem.
824 	 * Ensure they are NUL terminated as well.
825 	 */
826 	fstypelen = 0;
827 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
828 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
829 		error = EINVAL;
830 		if (errmsg != NULL)
831 			strncpy(errmsg, "Invalid fstype", errmsg_len);
832 		goto bail;
833 	}
834 	fspathlen = 0;
835 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
836 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
837 		error = EINVAL;
838 		if (errmsg != NULL)
839 			strncpy(errmsg, "Invalid fspath", errmsg_len);
840 		goto bail;
841 	}
842 
843 	/*
844 	 * Check to see that "export" is only used with the "update", "fstype",
845 	 * "fspath", "from" and "errmsg" options when in a vnet jail.
846 	 * These are the ones used to set/update exports by mountd(8).
847 	 * If only the above options are set in a jail that can run mountd(8),
848 	 * then the jail_export argument of vfs_domount() will be true.
849 	 * When jail_export is true, the vfs_suser() check does not cause
850 	 * failure, but limits the update to exports only.
851 	 * This allows mountd(8) running within the vnet jail
852 	 * to export file systems visible within the jail, but
853 	 * mounted outside of the jail.
854 	 */
855 	/*
856 	 * We need to see if we have the "update" option
857 	 * before we call vfs_domount(), since vfs_domount() has special
858 	 * logic based on MNT_UPDATE.  This is very important
859 	 * when we want to update the root filesystem.
860 	 */
861 	has_nonexport = false;
862 	jail_export = false;
863 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
864 		int do_freeopt = 0;
865 
866 		if (jailed(td->td_ucred) &&
867 		    strcmp(opt->name, "export") != 0 &&
868 		    strcmp(opt->name, "update") != 0 &&
869 		    strcmp(opt->name, "fstype") != 0 &&
870 		    strcmp(opt->name, "fspath") != 0 &&
871 		    strcmp(opt->name, "from") != 0 &&
872 		    strcmp(opt->name, "errmsg") != 0)
873 			has_nonexport = true;
874 		if (strcmp(opt->name, "update") == 0) {
875 			fsflags |= MNT_UPDATE;
876 			do_freeopt = 1;
877 		}
878 		else if (strcmp(opt->name, "async") == 0)
879 			fsflags |= MNT_ASYNC;
880 		else if (strcmp(opt->name, "force") == 0) {
881 			fsflags |= MNT_FORCE;
882 			do_freeopt = 1;
883 		}
884 		else if (strcmp(opt->name, "reload") == 0) {
885 			fsflags |= MNT_RELOAD;
886 			do_freeopt = 1;
887 		}
888 		else if (strcmp(opt->name, "multilabel") == 0)
889 			fsflags |= MNT_MULTILABEL;
890 		else if (strcmp(opt->name, "noasync") == 0)
891 			fsflags &= ~MNT_ASYNC;
892 		else if (strcmp(opt->name, "noatime") == 0)
893 			fsflags |= MNT_NOATIME;
894 		else if (strcmp(opt->name, "atime") == 0) {
895 			free(opt->name, M_MOUNT);
896 			opt->name = strdup("nonoatime", M_MOUNT);
897 		}
898 		else if (strcmp(opt->name, "noclusterr") == 0)
899 			fsflags |= MNT_NOCLUSTERR;
900 		else if (strcmp(opt->name, "clusterr") == 0) {
901 			free(opt->name, M_MOUNT);
902 			opt->name = strdup("nonoclusterr", M_MOUNT);
903 		}
904 		else if (strcmp(opt->name, "noclusterw") == 0)
905 			fsflags |= MNT_NOCLUSTERW;
906 		else if (strcmp(opt->name, "clusterw") == 0) {
907 			free(opt->name, M_MOUNT);
908 			opt->name = strdup("nonoclusterw", M_MOUNT);
909 		}
910 		else if (strcmp(opt->name, "noexec") == 0)
911 			fsflags |= MNT_NOEXEC;
912 		else if (strcmp(opt->name, "exec") == 0) {
913 			free(opt->name, M_MOUNT);
914 			opt->name = strdup("nonoexec", M_MOUNT);
915 		}
916 		else if (strcmp(opt->name, "nosuid") == 0)
917 			fsflags |= MNT_NOSUID;
918 		else if (strcmp(opt->name, "suid") == 0) {
919 			free(opt->name, M_MOUNT);
920 			opt->name = strdup("nonosuid", M_MOUNT);
921 		}
922 		else if (strcmp(opt->name, "nosymfollow") == 0)
923 			fsflags |= MNT_NOSYMFOLLOW;
924 		else if (strcmp(opt->name, "symfollow") == 0) {
925 			free(opt->name, M_MOUNT);
926 			opt->name = strdup("nonosymfollow", M_MOUNT);
927 		}
928 		else if (strcmp(opt->name, "noro") == 0) {
929 			fsflags &= ~MNT_RDONLY;
930 			autoro = false;
931 		}
932 		else if (strcmp(opt->name, "rw") == 0) {
933 			fsflags &= ~MNT_RDONLY;
934 			autoro = false;
935 		}
936 		else if (strcmp(opt->name, "ro") == 0) {
937 			fsflags |= MNT_RDONLY;
938 			autoro = false;
939 		}
940 		else if (strcmp(opt->name, "rdonly") == 0) {
941 			free(opt->name, M_MOUNT);
942 			opt->name = strdup("ro", M_MOUNT);
943 			fsflags |= MNT_RDONLY;
944 			autoro = false;
945 		}
946 		else if (strcmp(opt->name, "autoro") == 0) {
947 			do_freeopt = 1;
948 			autoro = true;
949 		}
950 		else if (strcmp(opt->name, "suiddir") == 0)
951 			fsflags |= MNT_SUIDDIR;
952 		else if (strcmp(opt->name, "sync") == 0)
953 			fsflags |= MNT_SYNCHRONOUS;
954 		else if (strcmp(opt->name, "union") == 0)
955 			fsflags |= MNT_UNION;
956 		else if (strcmp(opt->name, "export") == 0) {
957 			fsflags |= MNT_EXPORTED;
958 			jail_export = true;
959 		} else if (strcmp(opt->name, "automounted") == 0) {
960 			fsflags |= MNT_AUTOMOUNTED;
961 			do_freeopt = 1;
962 		} else if (strcmp(opt->name, "nocover") == 0) {
963 			fsflags |= MNT_NOCOVER;
964 			do_freeopt = 1;
965 		} else if (strcmp(opt->name, "cover") == 0) {
966 			fsflags &= ~MNT_NOCOVER;
967 			do_freeopt = 1;
968 		} else if (strcmp(opt->name, "emptydir") == 0) {
969 			fsflags |= MNT_EMPTYDIR;
970 			do_freeopt = 1;
971 		} else if (strcmp(opt->name, "noemptydir") == 0) {
972 			fsflags &= ~MNT_EMPTYDIR;
973 			do_freeopt = 1;
974 		}
975 		if (do_freeopt)
976 			vfs_freeopt(optlist, opt);
977 	}
978 
979 	/*
980 	 * Be ultra-paranoid about making sure the type and fspath
981 	 * variables will fit in our mp buffers, including the
982 	 * terminating NUL.
983 	 */
984 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
985 		error = ENAMETOOLONG;
986 		goto bail;
987 	}
988 
989 	/*
990 	 * If has_nonexport is true or the caller is not running within a
991 	 * vnet prison that can run mountd(8), set jail_export false.
992 	 */
993 	if (has_nonexport || !jailed(td->td_ucred) ||
994 	    !prison_check_nfsd(td->td_ucred))
995 		jail_export = false;
996 
997 	error = vfs_domount(td, fstype, fspath, fsflags, jail_export, &optlist);
998 	if (error == ENODEV) {
999 		error = EINVAL;
1000 		if (errmsg != NULL)
1001 			strncpy(errmsg, "Invalid fstype", errmsg_len);
1002 		goto bail;
1003 	}
1004 
1005 	/*
1006 	 * See if we can mount in the read-only mode if the error code suggests
1007 	 * that it could be possible and the mount options allow for that.
1008 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
1009 	 * overridden by "autoro".
1010 	 */
1011 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
1012 		printf("%s: R/W mount failed, possibly R/O media,"
1013 		    " trying R/O mount\n", __func__);
1014 		fsflags |= MNT_RDONLY;
1015 		error = vfs_domount(td, fstype, fspath, fsflags, jail_export,
1016 		    &optlist);
1017 	}
1018 bail:
1019 	/* copyout the errmsg */
1020 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
1021 	    && errmsg_len > 0 && errmsg != NULL) {
1022 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
1023 			bcopy(errmsg,
1024 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1025 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1026 		} else {
1027 			(void)copyout(errmsg,
1028 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1029 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1030 		}
1031 	}
1032 
1033 	if (optlist != NULL)
1034 		vfs_freeopts(optlist);
1035 	return (error);
1036 }
1037 
1038 /*
1039  * Old mount API.
1040  */
1041 #ifndef _SYS_SYSPROTO_H_
1042 struct mount_args {
1043 	char	*type;
1044 	char	*path;
1045 	int	flags;
1046 	caddr_t	data;
1047 };
1048 #endif
1049 /* ARGSUSED */
1050 int
1051 sys_mount(struct thread *td, struct mount_args *uap)
1052 {
1053 	char *fstype;
1054 	struct vfsconf *vfsp = NULL;
1055 	struct mntarg *ma = NULL;
1056 	uint64_t flags;
1057 	int error;
1058 
1059 	/*
1060 	 * Mount flags are now 64-bits. On 32-bit architectures only
1061 	 * 32-bits are passed in, but from here on everything handles
1062 	 * 64-bit flags correctly.
1063 	 */
1064 	flags = uap->flags;
1065 
1066 	AUDIT_ARG_FFLAGS(flags);
1067 
1068 	/*
1069 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
1070 	 * userspace to set this flag, but we must filter it out if we want
1071 	 * MNT_UPDATE on the root file system to work.
1072 	 * MNT_ROOTFS should only be set by the kernel when mounting its
1073 	 * root file system.
1074 	 */
1075 	flags &= ~MNT_ROOTFS;
1076 
1077 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1078 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1079 	if (error) {
1080 		free(fstype, M_TEMP);
1081 		return (error);
1082 	}
1083 
1084 	AUDIT_ARG_TEXT(fstype);
1085 	vfsp = vfs_byname_kld(fstype, td, &error);
1086 	free(fstype, M_TEMP);
1087 	if (vfsp == NULL)
1088 		return (EINVAL);
1089 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1090 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1091 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1092 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
1093 		return (EOPNOTSUPP);
1094 
1095 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1096 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1097 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1098 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1099 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1100 
1101 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1102 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1103 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1104 }
1105 
1106 /*
1107  * vfs_domount_first(): first file system mount (not update)
1108  */
1109 static int
1110 vfs_domount_first(
1111 	struct thread *td,		/* Calling thread. */
1112 	struct vfsconf *vfsp,		/* File system type. */
1113 	char *fspath,			/* Mount path. */
1114 	struct vnode *vp,		/* Vnode to be covered. */
1115 	uint64_t fsflags,		/* Flags common to all filesystems. */
1116 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1117 	)
1118 {
1119 	struct vattr va;
1120 	struct mount *mp;
1121 	struct vnode *newdp, *rootvp;
1122 	int error, error1;
1123 	bool unmounted;
1124 
1125 	ASSERT_VOP_ELOCKED(vp, __func__);
1126 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1127 
1128 	/*
1129 	 * If the jail of the calling thread lacks permission for this type of
1130 	 * file system, or is trying to cover its own root, deny immediately.
1131 	 */
1132 	if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1133 	    vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1134 		vput(vp);
1135 		return (EPERM);
1136 	}
1137 
1138 	/*
1139 	 * If the user is not root, ensure that they own the directory
1140 	 * onto which we are attempting to mount.
1141 	 */
1142 	error = VOP_GETATTR(vp, &va, td->td_ucred);
1143 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1144 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1145 	if (error == 0)
1146 		error = vinvalbuf(vp, V_SAVE, 0, 0);
1147 	if (vfsp->vfc_flags & VFCF_FILEMOUNT) {
1148 		if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG)
1149 			error = EINVAL;
1150 		/*
1151 		 * For file mounts, ensure that there is only one hardlink to the file.
1152 		 */
1153 		if (error == 0 && vp->v_type == VREG && va.va_nlink != 1)
1154 			error = EINVAL;
1155 	} else {
1156 		if (error == 0 && vp->v_type != VDIR)
1157 			error = ENOTDIR;
1158 	}
1159 	if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
1160 		error = vn_dir_check_empty(vp);
1161 	if (error == 0) {
1162 		VI_LOCK(vp);
1163 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1164 			vp->v_iflag |= VI_MOUNT;
1165 		else
1166 			error = EBUSY;
1167 		VI_UNLOCK(vp);
1168 	}
1169 	if (error != 0) {
1170 		vput(vp);
1171 		return (error);
1172 	}
1173 	vn_seqc_write_begin(vp);
1174 	VOP_UNLOCK(vp);
1175 
1176 	/* Allocate and initialize the filesystem. */
1177 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1178 	/* XXXMAC: pass to vfs_mount_alloc? */
1179 	mp->mnt_optnew = *optlist;
1180 	/* Set the mount level flags. */
1181 	mp->mnt_flag = (fsflags &
1182 	    (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE));
1183 
1184 	/*
1185 	 * Mount the filesystem.
1186 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1187 	 * get.  No freeing of cn_pnbuf.
1188 	 */
1189 	error1 = 0;
1190 	unmounted = true;
1191 	if ((error = VFS_MOUNT(mp)) != 0 ||
1192 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1193 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1194 		rootvp = NULL;
1195 		if (error1 != 0) {
1196 			MPASS(error == 0);
1197 			rootvp = vfs_cache_root_clear(mp);
1198 			if (rootvp != NULL) {
1199 				vhold(rootvp);
1200 				vrele(rootvp);
1201 			}
1202 			(void)vn_start_write(NULL, &mp, V_WAIT);
1203 			MNT_ILOCK(mp);
1204 			mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1205 			MNT_IUNLOCK(mp);
1206 			VFS_PURGE(mp);
1207 			error = VFS_UNMOUNT(mp, 0);
1208 			vn_finished_write(mp);
1209 			if (error != 0) {
1210 				printf(
1211 		    "failed post-mount (%d): rollback unmount returned %d\n",
1212 				    error1, error);
1213 				unmounted = false;
1214 			}
1215 			error = error1;
1216 		}
1217 		vfs_unbusy(mp);
1218 		mp->mnt_vnodecovered = NULL;
1219 		if (unmounted) {
1220 			/* XXXKIB wait for mnt_lockref drain? */
1221 			vfs_mount_destroy(mp);
1222 		}
1223 		VI_LOCK(vp);
1224 		vp->v_iflag &= ~VI_MOUNT;
1225 		VI_UNLOCK(vp);
1226 		if (rootvp != NULL) {
1227 			vn_seqc_write_end(rootvp);
1228 			vdrop(rootvp);
1229 		}
1230 		vn_seqc_write_end(vp);
1231 		vrele(vp);
1232 		return (error);
1233 	}
1234 	vn_seqc_write_begin(newdp);
1235 	VOP_UNLOCK(newdp);
1236 
1237 	if (mp->mnt_opt != NULL)
1238 		vfs_freeopts(mp->mnt_opt);
1239 	mp->mnt_opt = mp->mnt_optnew;
1240 	*optlist = NULL;
1241 
1242 	/*
1243 	 * Prevent external consumers of mount options from reading mnt_optnew.
1244 	 */
1245 	mp->mnt_optnew = NULL;
1246 
1247 	MNT_ILOCK(mp);
1248 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1249 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1250 		mp->mnt_kern_flag |= MNTK_ASYNC;
1251 	else
1252 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1253 	MNT_IUNLOCK(mp);
1254 
1255 	/*
1256 	 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the
1257 	 * vp lock to satisfy vfs_lookup() requirements.
1258 	 */
1259 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1260 	VI_LOCK(vp);
1261 	vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1262 	vp->v_mountedhere = mp;
1263 	VI_UNLOCK(vp);
1264 	VOP_UNLOCK(vp);
1265 	cache_purge(vp);
1266 
1267 	/*
1268 	 * We need to lock both vnodes.
1269 	 *
1270 	 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1271 	 * from different filesystems.
1272 	 */
1273 	vn_lock_pair(vp, false, LK_EXCLUSIVE, newdp, false, LK_EXCLUSIVE);
1274 
1275 	VI_LOCK(vp);
1276 	vp->v_iflag &= ~VI_MOUNT;
1277 	VI_UNLOCK(vp);
1278 	/* Place the new filesystem at the end of the mount list. */
1279 	mtx_lock(&mountlist_mtx);
1280 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1281 	mtx_unlock(&mountlist_mtx);
1282 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1283 	VOP_UNLOCK(vp);
1284 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1285 	VOP_UNLOCK(newdp);
1286 	mount_devctl_event("MOUNT", mp, false);
1287 	mountcheckdirs(vp, newdp);
1288 	vn_seqc_write_end(vp);
1289 	vn_seqc_write_end(newdp);
1290 	vrele(newdp);
1291 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1292 		vfs_allocate_syncvnode(mp);
1293 	vfs_op_exit(mp);
1294 	vfs_unbusy(mp);
1295 	return (0);
1296 }
1297 
1298 /*
1299  * vfs_domount_update(): update of mounted file system
1300  */
1301 static int
1302 vfs_domount_update(
1303 	struct thread *td,		/* Calling thread. */
1304 	struct vnode *vp,		/* Mount point vnode. */
1305 	uint64_t fsflags,		/* Flags common to all filesystems. */
1306 	bool jail_export,		/* Got export option in vnet prison. */
1307 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1308 	)
1309 {
1310 	struct export_args export;
1311 	struct o2export_args o2export;
1312 	struct vnode *rootvp;
1313 	void *bufp;
1314 	struct mount *mp;
1315 	int error, export_error, i, len, fsid_up_len;
1316 	uint64_t flag;
1317 	gid_t *grps;
1318 	fsid_t *fsid_up;
1319 	bool vfs_suser_failed;
1320 
1321 	ASSERT_VOP_ELOCKED(vp, __func__);
1322 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1323 	mp = vp->v_mount;
1324 
1325 	if ((vp->v_vflag & VV_ROOT) == 0) {
1326 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1327 		    == 0)
1328 			error = EXDEV;
1329 		else
1330 			error = EINVAL;
1331 		vput(vp);
1332 		return (error);
1333 	}
1334 
1335 	/*
1336 	 * We only allow the filesystem to be reloaded if it
1337 	 * is currently mounted read-only.
1338 	 */
1339 	flag = mp->mnt_flag;
1340 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1341 		vput(vp);
1342 		return (EOPNOTSUPP);	/* Needs translation */
1343 	}
1344 	/*
1345 	 * Only privileged root, or (if MNT_USER is set) the user that
1346 	 * did the original mount is permitted to update it.
1347 	 */
1348 	/*
1349 	 * For the case of mountd(8) doing exports in a jail, the vfs_suser()
1350 	 * call does not cause failure.  vfs_domount() has already checked
1351 	 * that "root" is doing this and vfs_suser() will fail when
1352 	 * the file system has been mounted outside the jail.
1353 	 * jail_export set true indicates that "export" is not mixed
1354 	 * with other options that change mount behaviour.
1355 	 */
1356 	vfs_suser_failed = false;
1357 	error = vfs_suser(mp, td);
1358 	if (jail_export && error != 0) {
1359 		error = 0;
1360 		vfs_suser_failed = true;
1361 	}
1362 	if (error != 0) {
1363 		vput(vp);
1364 		return (error);
1365 	}
1366 	if (vfs_busy(mp, MBF_NOWAIT)) {
1367 		vput(vp);
1368 		return (EBUSY);
1369 	}
1370 	VI_LOCK(vp);
1371 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1372 		VI_UNLOCK(vp);
1373 		vfs_unbusy(mp);
1374 		vput(vp);
1375 		return (EBUSY);
1376 	}
1377 	vp->v_iflag |= VI_MOUNT;
1378 	VI_UNLOCK(vp);
1379 	VOP_UNLOCK(vp);
1380 
1381 	rootvp = NULL;
1382 	vfs_op_enter(mp);
1383 	vn_seqc_write_begin(vp);
1384 
1385 	if (vfs_getopt(*optlist, "fsid", (void **)&fsid_up,
1386 	    &fsid_up_len) == 0) {
1387 		if (fsid_up_len != sizeof(*fsid_up)) {
1388 			error = EINVAL;
1389 			goto end;
1390 		}
1391 		if (fsidcmp(fsid_up, &mp->mnt_stat.f_fsid) != 0) {
1392 			error = ENOENT;
1393 			goto end;
1394 		}
1395 		vfs_deleteopt(*optlist, "fsid");
1396 	}
1397 
1398 	MNT_ILOCK(mp);
1399 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1400 		MNT_IUNLOCK(mp);
1401 		error = EBUSY;
1402 		goto end;
1403 	}
1404 	if (vfs_suser_failed) {
1405 		KASSERT((fsflags & (MNT_EXPORTED | MNT_UPDATE)) ==
1406 		    (MNT_EXPORTED | MNT_UPDATE),
1407 		    ("%s: jailed export did not set expected fsflags",
1408 		     __func__));
1409 		/*
1410 		 * For this case, only MNT_UPDATE and
1411 		 * MNT_EXPORTED have been set in fsflags
1412 		 * by the options.  Only set MNT_UPDATE,
1413 		 * since that is the one that would be set
1414 		 * when set in fsflags, below.
1415 		 */
1416 		mp->mnt_flag |= MNT_UPDATE;
1417 	} else {
1418 		mp->mnt_flag &= ~MNT_UPDATEMASK;
1419 		mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1420 		    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1421 		if ((mp->mnt_flag & MNT_ASYNC) == 0)
1422 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1423 	}
1424 	rootvp = vfs_cache_root_clear(mp);
1425 	MNT_IUNLOCK(mp);
1426 	mp->mnt_optnew = *optlist;
1427 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1428 
1429 	/*
1430 	 * Mount the filesystem.
1431 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1432 	 * get.  No freeing of cn_pnbuf.
1433 	 */
1434 	/*
1435 	 * For the case of mountd(8) doing exports from within a vnet jail,
1436 	 * "from" is typically not set correctly such that VFS_MOUNT() will
1437 	 * return ENOENT. It is not obvious that VFS_MOUNT() ever needs to be
1438 	 * called when mountd is doing exports, but this check only applies to
1439 	 * the specific case where it is running inside a vnet jail, to
1440 	 * avoid any POLA violation.
1441 	 */
1442 	error = 0;
1443 	if (!jail_export)
1444 		error = VFS_MOUNT(mp);
1445 
1446 	export_error = 0;
1447 	/* Process the export option. */
1448 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1449 	    &len) == 0) {
1450 		/* Assume that there is only 1 ABI for each length. */
1451 		switch (len) {
1452 		case (sizeof(struct oexport_args)):
1453 			bzero(&o2export, sizeof(o2export));
1454 			/* FALLTHROUGH */
1455 		case (sizeof(o2export)):
1456 			bcopy(bufp, &o2export, len);
1457 			export.ex_flags = (uint64_t)o2export.ex_flags;
1458 			export.ex_root = o2export.ex_root;
1459 			export.ex_uid = o2export.ex_anon.cr_uid;
1460 			export.ex_groups = NULL;
1461 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1462 			if (export.ex_ngroups > 0) {
1463 				if (export.ex_ngroups <= XU_NGROUPS) {
1464 					export.ex_groups = malloc(
1465 					    export.ex_ngroups * sizeof(gid_t),
1466 					    M_TEMP, M_WAITOK);
1467 					for (i = 0; i < export.ex_ngroups; i++)
1468 						export.ex_groups[i] =
1469 						  o2export.ex_anon.cr_groups[i];
1470 				} else
1471 					export_error = EINVAL;
1472 			} else if (export.ex_ngroups < 0)
1473 				export_error = EINVAL;
1474 			export.ex_addr = o2export.ex_addr;
1475 			export.ex_addrlen = o2export.ex_addrlen;
1476 			export.ex_mask = o2export.ex_mask;
1477 			export.ex_masklen = o2export.ex_masklen;
1478 			export.ex_indexfile = o2export.ex_indexfile;
1479 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1480 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1481 				for (i = 0; i < export.ex_numsecflavors; i++)
1482 					export.ex_secflavors[i] =
1483 					    o2export.ex_secflavors[i];
1484 			} else
1485 				export_error = EINVAL;
1486 			if (export_error == 0)
1487 				export_error = vfs_export(mp, &export, true);
1488 			free(export.ex_groups, M_TEMP);
1489 			break;
1490 		case (sizeof(export)):
1491 			bcopy(bufp, &export, len);
1492 			grps = NULL;
1493 			if (export.ex_ngroups > 0) {
1494 				if (export.ex_ngroups <= NGROUPS_MAX) {
1495 					grps = malloc(export.ex_ngroups *
1496 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1497 					export_error = copyin(export.ex_groups,
1498 					    grps, export.ex_ngroups *
1499 					    sizeof(gid_t));
1500 					if (export_error == 0)
1501 						export.ex_groups = grps;
1502 				} else
1503 					export_error = EINVAL;
1504 			} else if (export.ex_ngroups == 0)
1505 				export.ex_groups = NULL;
1506 			else
1507 				export_error = EINVAL;
1508 			if (export_error == 0)
1509 				export_error = vfs_export(mp, &export, true);
1510 			free(grps, M_TEMP);
1511 			break;
1512 		default:
1513 			export_error = EINVAL;
1514 			break;
1515 		}
1516 	}
1517 
1518 	MNT_ILOCK(mp);
1519 	if (error == 0) {
1520 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1521 		    MNT_SNAPSHOT);
1522 	} else {
1523 		/*
1524 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1525 		 * because it is not part of MNT_UPDATEMASK, but it could have
1526 		 * changed in the meantime if quotactl(2) was called.
1527 		 * All in all we want current value of MNT_QUOTA, not the old
1528 		 * one.
1529 		 */
1530 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1531 	}
1532 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1533 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1534 		mp->mnt_kern_flag |= MNTK_ASYNC;
1535 	else
1536 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1537 	MNT_IUNLOCK(mp);
1538 
1539 	if (error != 0)
1540 		goto end;
1541 
1542 	mount_devctl_event("REMOUNT", mp, true);
1543 	if (mp->mnt_opt != NULL)
1544 		vfs_freeopts(mp->mnt_opt);
1545 	mp->mnt_opt = mp->mnt_optnew;
1546 	*optlist = NULL;
1547 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1548 	/*
1549 	 * Prevent external consumers of mount options from reading
1550 	 * mnt_optnew.
1551 	 */
1552 	mp->mnt_optnew = NULL;
1553 
1554 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1555 		vfs_allocate_syncvnode(mp);
1556 	else
1557 		vfs_deallocate_syncvnode(mp);
1558 end:
1559 	vfs_op_exit(mp);
1560 	if (rootvp != NULL) {
1561 		vn_seqc_write_end(rootvp);
1562 		vrele(rootvp);
1563 	}
1564 	vn_seqc_write_end(vp);
1565 	vfs_unbusy(mp);
1566 	VI_LOCK(vp);
1567 	vp->v_iflag &= ~VI_MOUNT;
1568 	VI_UNLOCK(vp);
1569 	vrele(vp);
1570 	return (error != 0 ? error : export_error);
1571 }
1572 
1573 /*
1574  * vfs_domount(): actually attempt a filesystem mount.
1575  */
1576 static int
1577 vfs_domount(
1578 	struct thread *td,		/* Calling thread. */
1579 	const char *fstype,		/* Filesystem type. */
1580 	char *fspath,			/* Mount path. */
1581 	uint64_t fsflags,		/* Flags common to all filesystems. */
1582 	bool jail_export,		/* Got export option in vnet prison. */
1583 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1584 	)
1585 {
1586 	struct vfsconf *vfsp;
1587 	struct nameidata nd;
1588 	struct vnode *vp;
1589 	char *pathbuf;
1590 	int error;
1591 
1592 	/*
1593 	 * Be ultra-paranoid about making sure the type and fspath
1594 	 * variables will fit in our mp buffers, including the
1595 	 * terminating NUL.
1596 	 */
1597 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1598 		return (ENAMETOOLONG);
1599 
1600 	if (jail_export) {
1601 		error = priv_check(td, PRIV_NFS_DAEMON);
1602 		if (error)
1603 			return (error);
1604 	} else if (jailed(td->td_ucred) || usermount == 0) {
1605 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1606 			return (error);
1607 	}
1608 
1609 	/*
1610 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1611 	 */
1612 	if (fsflags & MNT_EXPORTED) {
1613 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1614 		if (error)
1615 			return (error);
1616 	}
1617 	if (fsflags & MNT_SUIDDIR) {
1618 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1619 		if (error)
1620 			return (error);
1621 	}
1622 	/*
1623 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1624 	 */
1625 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1626 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1627 			fsflags |= MNT_NOSUID | MNT_USER;
1628 	}
1629 
1630 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1631 	vfsp = NULL;
1632 	if ((fsflags & MNT_UPDATE) == 0) {
1633 		/* Don't try to load KLDs if we're mounting the root. */
1634 		if (fsflags & MNT_ROOTFS) {
1635 			if ((vfsp = vfs_byname(fstype)) == NULL)
1636 				return (ENODEV);
1637 		} else {
1638 			if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1639 				return (error);
1640 		}
1641 	}
1642 
1643 	/*
1644 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1645 	 */
1646 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1647 	    UIO_SYSSPACE, fspath);
1648 	error = namei(&nd);
1649 	if (error != 0)
1650 		return (error);
1651 	vp = nd.ni_vp;
1652 	/*
1653 	 * Don't allow stacking file mounts to work around problems with the way
1654 	 * that namei sets nd.ni_dvp to vp_crossmp for these.
1655 	 */
1656 	if (vp->v_type == VREG)
1657 		fsflags |= MNT_NOCOVER;
1658 	if ((fsflags & MNT_UPDATE) == 0) {
1659 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1660 		    (fsflags & MNT_NOCOVER) != 0) {
1661 			vput(vp);
1662 			error = EBUSY;
1663 			goto out;
1664 		}
1665 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1666 		strcpy(pathbuf, fspath);
1667 		/*
1668 		 * Note: we allow any vnode type here. If the path sanity check
1669 		 * succeeds, the type will be validated in vfs_domount_first
1670 		 * above.
1671 		 */
1672 		if (vp->v_type == VDIR)
1673 			error = vn_path_to_global_path(td, vp, pathbuf,
1674 			    MNAMELEN);
1675 		else
1676 			error = vn_path_to_global_path_hardlink(td, vp,
1677 			    nd.ni_dvp, pathbuf, MNAMELEN,
1678 			    nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1679 		if (error == 0) {
1680 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1681 			    fsflags, optlist);
1682 		}
1683 		free(pathbuf, M_TEMP);
1684 	} else
1685 		error = vfs_domount_update(td, vp, fsflags, jail_export,
1686 		    optlist);
1687 
1688 out:
1689 	NDFREE_PNBUF(&nd);
1690 	vrele(nd.ni_dvp);
1691 
1692 	return (error);
1693 }
1694 
1695 /*
1696  * Unmount a filesystem.
1697  *
1698  * Note: unmount takes a path to the vnode mounted on as argument, not
1699  * special file (as before).
1700  */
1701 #ifndef _SYS_SYSPROTO_H_
1702 struct unmount_args {
1703 	char	*path;
1704 	int	flags;
1705 };
1706 #endif
1707 /* ARGSUSED */
1708 int
1709 sys_unmount(struct thread *td, struct unmount_args *uap)
1710 {
1711 
1712 	return (kern_unmount(td, uap->path, uap->flags));
1713 }
1714 
1715 int
1716 kern_unmount(struct thread *td, const char *path, int flags)
1717 {
1718 	struct nameidata nd;
1719 	struct mount *mp;
1720 	char *fsidbuf, *pathbuf;
1721 	fsid_t fsid;
1722 	int error;
1723 
1724 	AUDIT_ARG_VALUE(flags);
1725 	if (jailed(td->td_ucred) || usermount == 0) {
1726 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1727 		if (error)
1728 			return (error);
1729 	}
1730 
1731 	if (flags & MNT_BYFSID) {
1732 		fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1733 		error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1734 		if (error) {
1735 			free(fsidbuf, M_TEMP);
1736 			return (error);
1737 		}
1738 
1739 		AUDIT_ARG_TEXT(fsidbuf);
1740 		/* Decode the filesystem ID. */
1741 		if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1742 			free(fsidbuf, M_TEMP);
1743 			return (EINVAL);
1744 		}
1745 
1746 		mp = vfs_getvfs(&fsid);
1747 		free(fsidbuf, M_TEMP);
1748 		if (mp == NULL) {
1749 			return (ENOENT);
1750 		}
1751 	} else {
1752 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1753 		error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1754 		if (error) {
1755 			free(pathbuf, M_TEMP);
1756 			return (error);
1757 		}
1758 
1759 		/*
1760 		 * Try to find global path for path argument.
1761 		 */
1762 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1763 		    UIO_SYSSPACE, pathbuf);
1764 		if (namei(&nd) == 0) {
1765 			NDFREE_PNBUF(&nd);
1766 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1767 			    MNAMELEN);
1768 			if (error == 0)
1769 				vput(nd.ni_vp);
1770 		}
1771 		mtx_lock(&mountlist_mtx);
1772 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1773 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1774 				vfs_ref(mp);
1775 				break;
1776 			}
1777 		}
1778 		mtx_unlock(&mountlist_mtx);
1779 		free(pathbuf, M_TEMP);
1780 		if (mp == NULL) {
1781 			/*
1782 			 * Previously we returned ENOENT for a nonexistent path and
1783 			 * EINVAL for a non-mountpoint.  We cannot tell these apart
1784 			 * now, so in the !MNT_BYFSID case return the more likely
1785 			 * EINVAL for compatibility.
1786 			 */
1787 			return (EINVAL);
1788 		}
1789 	}
1790 
1791 	/*
1792 	 * Don't allow unmounting the root filesystem.
1793 	 */
1794 	if (mp->mnt_flag & MNT_ROOTFS) {
1795 		vfs_rel(mp);
1796 		return (EINVAL);
1797 	}
1798 	error = dounmount(mp, flags, td);
1799 	return (error);
1800 }
1801 
1802 /*
1803  * Return error if any of the vnodes, ignoring the root vnode
1804  * and the syncer vnode, have non-zero usecount.
1805  *
1806  * This function is purely advisory - it can return false positives
1807  * and negatives.
1808  */
1809 static int
1810 vfs_check_usecounts(struct mount *mp)
1811 {
1812 	struct vnode *vp, *mvp;
1813 
1814 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1815 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1816 		    vp->v_usecount != 0) {
1817 			VI_UNLOCK(vp);
1818 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1819 			return (EBUSY);
1820 		}
1821 		VI_UNLOCK(vp);
1822 	}
1823 
1824 	return (0);
1825 }
1826 
1827 static void
1828 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1829 {
1830 
1831 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1832 	mp->mnt_kern_flag &= ~mntkflags;
1833 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1834 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1835 		wakeup(mp);
1836 	}
1837 	vfs_op_exit_locked(mp);
1838 	MNT_IUNLOCK(mp);
1839 	if (coveredvp != NULL) {
1840 		VOP_UNLOCK(coveredvp);
1841 		vdrop(coveredvp);
1842 	}
1843 	vn_finished_write(mp);
1844 	vfs_rel(mp);
1845 }
1846 
1847 /*
1848  * There are various reference counters associated with the mount point.
1849  * Normally it is permitted to modify them without taking the mnt ilock,
1850  * but this behavior can be temporarily disabled if stable value is needed
1851  * or callers are expected to block (e.g. to not allow new users during
1852  * forced unmount).
1853  */
1854 void
1855 vfs_op_enter(struct mount *mp)
1856 {
1857 	struct mount_pcpu *mpcpu;
1858 	int cpu;
1859 
1860 	MNT_ILOCK(mp);
1861 	mp->mnt_vfs_ops++;
1862 	if (mp->mnt_vfs_ops > 1) {
1863 		MNT_IUNLOCK(mp);
1864 		return;
1865 	}
1866 	vfs_op_barrier_wait(mp);
1867 	CPU_FOREACH(cpu) {
1868 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1869 
1870 		mp->mnt_ref += mpcpu->mntp_ref;
1871 		mpcpu->mntp_ref = 0;
1872 
1873 		mp->mnt_lockref += mpcpu->mntp_lockref;
1874 		mpcpu->mntp_lockref = 0;
1875 
1876 		mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1877 		mpcpu->mntp_writeopcount = 0;
1878 	}
1879 	MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1880 	    mp->mnt_writeopcount >= 0, mp,
1881 	    ("invalid count(s): ref %d lockref %d writeopcount %d",
1882 	    mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1883 	MNT_IUNLOCK(mp);
1884 	vfs_assert_mount_counters(mp);
1885 }
1886 
1887 void
1888 vfs_op_exit_locked(struct mount *mp)
1889 {
1890 
1891 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1892 
1893 	MPASSERT(mp->mnt_vfs_ops > 0, mp,
1894 	    ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1895 	MPASSERT(mp->mnt_vfs_ops > 1 ||
1896 	    (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1897 	    ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1898 	mp->mnt_vfs_ops--;
1899 }
1900 
1901 void
1902 vfs_op_exit(struct mount *mp)
1903 {
1904 
1905 	MNT_ILOCK(mp);
1906 	vfs_op_exit_locked(mp);
1907 	MNT_IUNLOCK(mp);
1908 }
1909 
1910 struct vfs_op_barrier_ipi {
1911 	struct mount *mp;
1912 	struct smp_rendezvous_cpus_retry_arg srcra;
1913 };
1914 
1915 static void
1916 vfs_op_action_func(void *arg)
1917 {
1918 	struct vfs_op_barrier_ipi *vfsopipi;
1919 	struct mount *mp;
1920 
1921 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1922 	mp = vfsopipi->mp;
1923 
1924 	if (!vfs_op_thread_entered(mp))
1925 		smp_rendezvous_cpus_done(arg);
1926 }
1927 
1928 static void
1929 vfs_op_wait_func(void *arg, int cpu)
1930 {
1931 	struct vfs_op_barrier_ipi *vfsopipi;
1932 	struct mount *mp;
1933 	struct mount_pcpu *mpcpu;
1934 
1935 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1936 	mp = vfsopipi->mp;
1937 
1938 	mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1939 	while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1940 		cpu_spinwait();
1941 }
1942 
1943 void
1944 vfs_op_barrier_wait(struct mount *mp)
1945 {
1946 	struct vfs_op_barrier_ipi vfsopipi;
1947 
1948 	vfsopipi.mp = mp;
1949 
1950 	smp_rendezvous_cpus_retry(all_cpus,
1951 	    smp_no_rendezvous_barrier,
1952 	    vfs_op_action_func,
1953 	    smp_no_rendezvous_barrier,
1954 	    vfs_op_wait_func,
1955 	    &vfsopipi.srcra);
1956 }
1957 
1958 #ifdef DIAGNOSTIC
1959 void
1960 vfs_assert_mount_counters(struct mount *mp)
1961 {
1962 	struct mount_pcpu *mpcpu;
1963 	int cpu;
1964 
1965 	if (mp->mnt_vfs_ops == 0)
1966 		return;
1967 
1968 	CPU_FOREACH(cpu) {
1969 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1970 		if (mpcpu->mntp_ref != 0 ||
1971 		    mpcpu->mntp_lockref != 0 ||
1972 		    mpcpu->mntp_writeopcount != 0)
1973 			vfs_dump_mount_counters(mp);
1974 	}
1975 }
1976 
1977 void
1978 vfs_dump_mount_counters(struct mount *mp)
1979 {
1980 	struct mount_pcpu *mpcpu;
1981 	int ref, lockref, writeopcount;
1982 	int cpu;
1983 
1984 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1985 
1986 	printf("        ref : ");
1987 	ref = mp->mnt_ref;
1988 	CPU_FOREACH(cpu) {
1989 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1990 		printf("%d ", mpcpu->mntp_ref);
1991 		ref += mpcpu->mntp_ref;
1992 	}
1993 	printf("\n");
1994 	printf("    lockref : ");
1995 	lockref = mp->mnt_lockref;
1996 	CPU_FOREACH(cpu) {
1997 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1998 		printf("%d ", mpcpu->mntp_lockref);
1999 		lockref += mpcpu->mntp_lockref;
2000 	}
2001 	printf("\n");
2002 	printf("writeopcount: ");
2003 	writeopcount = mp->mnt_writeopcount;
2004 	CPU_FOREACH(cpu) {
2005 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2006 		printf("%d ", mpcpu->mntp_writeopcount);
2007 		writeopcount += mpcpu->mntp_writeopcount;
2008 	}
2009 	printf("\n");
2010 
2011 	printf("counter       struct total\n");
2012 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
2013 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
2014 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
2015 
2016 	panic("invalid counts on struct mount");
2017 }
2018 #endif
2019 
2020 int
2021 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
2022 {
2023 	struct mount_pcpu *mpcpu;
2024 	int cpu, sum;
2025 
2026 	switch (which) {
2027 	case MNT_COUNT_REF:
2028 		sum = mp->mnt_ref;
2029 		break;
2030 	case MNT_COUNT_LOCKREF:
2031 		sum = mp->mnt_lockref;
2032 		break;
2033 	case MNT_COUNT_WRITEOPCOUNT:
2034 		sum = mp->mnt_writeopcount;
2035 		break;
2036 	}
2037 
2038 	CPU_FOREACH(cpu) {
2039 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2040 		switch (which) {
2041 		case MNT_COUNT_REF:
2042 			sum += mpcpu->mntp_ref;
2043 			break;
2044 		case MNT_COUNT_LOCKREF:
2045 			sum += mpcpu->mntp_lockref;
2046 			break;
2047 		case MNT_COUNT_WRITEOPCOUNT:
2048 			sum += mpcpu->mntp_writeopcount;
2049 			break;
2050 		}
2051 	}
2052 	return (sum);
2053 }
2054 
2055 static bool
2056 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
2057     int timeout_ticks)
2058 {
2059 	bool enqueued;
2060 
2061 	enqueued = false;
2062 	mtx_lock(&deferred_unmount_lock);
2063 	if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
2064 		mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
2065 		STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
2066 		    mnt_taskqueue_link);
2067 		enqueued = true;
2068 	}
2069 	mtx_unlock(&deferred_unmount_lock);
2070 
2071 	if (enqueued) {
2072 		taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
2073 		    &deferred_unmount_task, timeout_ticks);
2074 	}
2075 
2076 	return (enqueued);
2077 }
2078 
2079 /*
2080  * Taskqueue handler for processing async/recursive unmounts
2081  */
2082 static void
2083 vfs_deferred_unmount(void *argi __unused, int pending __unused)
2084 {
2085 	STAILQ_HEAD(, mount) local_unmounts;
2086 	uint64_t flags;
2087 	struct mount *mp, *tmp;
2088 	int error;
2089 	unsigned int retries;
2090 	bool unmounted;
2091 
2092 	STAILQ_INIT(&local_unmounts);
2093 	mtx_lock(&deferred_unmount_lock);
2094 	STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
2095 	mtx_unlock(&deferred_unmount_lock);
2096 
2097 	STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
2098 		flags = mp->mnt_taskqueue_flags;
2099 		KASSERT((flags & MNT_DEFERRED) != 0,
2100 		    ("taskqueue unmount without MNT_DEFERRED"));
2101 		error = dounmount(mp, flags, curthread);
2102 		if (error != 0) {
2103 			MNT_ILOCK(mp);
2104 			unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
2105 			MNT_IUNLOCK(mp);
2106 
2107 			/*
2108 			 * The deferred unmount thread is the only thread that
2109 			 * modifies the retry counts, so locking/atomics aren't
2110 			 * needed here.
2111 			 */
2112 			retries = (mp->mnt_unmount_retries)++;
2113 			deferred_unmount_total_retries++;
2114 			if (!unmounted && retries < deferred_unmount_retry_limit) {
2115 				deferred_unmount_enqueue(mp, flags, true,
2116 				    -deferred_unmount_retry_delay_hz);
2117 			} else {
2118 				if (retries >= deferred_unmount_retry_limit) {
2119 					printf("giving up on deferred unmount "
2120 					    "of %s after %d retries, error %d\n",
2121 					    mp->mnt_stat.f_mntonname, retries, error);
2122 				}
2123 				vfs_rel(mp);
2124 			}
2125 		}
2126 	}
2127 }
2128 
2129 /*
2130  * Do the actual filesystem unmount.
2131  */
2132 int
2133 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2134 {
2135 	struct mount_upper_node *upper;
2136 	struct vnode *coveredvp, *rootvp;
2137 	int error;
2138 	uint64_t async_flag;
2139 	int mnt_gen_r;
2140 	unsigned int retries;
2141 
2142 	KASSERT((flags & MNT_DEFERRED) == 0 ||
2143 	    (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2144 	    ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2145 
2146 	/*
2147 	 * If the caller has explicitly requested the unmount to be handled by
2148 	 * the taskqueue and we're not already in taskqueue context, queue
2149 	 * up the unmount request and exit.  This is done prior to any
2150 	 * credential checks; MNT_DEFERRED should be used only for kernel-
2151 	 * initiated unmounts and will therefore be processed with the
2152 	 * (kernel) credentials of the taskqueue thread.  Still, callers
2153 	 * should be sure this is the behavior they want.
2154 	 */
2155 	if ((flags & MNT_DEFERRED) != 0 &&
2156 	    taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2157 		if (!deferred_unmount_enqueue(mp, flags, false, 0))
2158 			vfs_rel(mp);
2159 		return (EINPROGRESS);
2160 	}
2161 
2162 	/*
2163 	 * Only privileged root, or (if MNT_USER is set) the user that did the
2164 	 * original mount is permitted to unmount this filesystem.
2165 	 * This check should be made prior to queueing up any recursive
2166 	 * unmounts of upper filesystems.  Those unmounts will be executed
2167 	 * with kernel thread credentials and are expected to succeed, so
2168 	 * we must at least ensure the originating context has sufficient
2169 	 * privilege to unmount the base filesystem before proceeding with
2170 	 * the uppers.
2171 	 */
2172 	error = vfs_suser(mp, td);
2173 	if (error != 0) {
2174 		KASSERT((flags & MNT_DEFERRED) == 0,
2175 		    ("taskqueue unmount with insufficient privilege"));
2176 		vfs_rel(mp);
2177 		return (error);
2178 	}
2179 
2180 	if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2181 		flags |= MNT_RECURSE;
2182 
2183 	if ((flags & MNT_RECURSE) != 0) {
2184 		KASSERT((flags & MNT_FORCE) != 0,
2185 		    ("MNT_RECURSE requires MNT_FORCE"));
2186 
2187 		MNT_ILOCK(mp);
2188 		/*
2189 		 * Set MNTK_RECURSE to prevent new upper mounts from being
2190 		 * added, and note that an operation on the uppers list is in
2191 		 * progress.  This will ensure that unregistration from the
2192 		 * uppers list, and therefore any pending unmount of the upper
2193 		 * FS, can't complete until after we finish walking the list.
2194 		 */
2195 		mp->mnt_kern_flag |= MNTK_RECURSE;
2196 		mp->mnt_upper_pending++;
2197 		TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2198 			retries = upper->mp->mnt_unmount_retries;
2199 			if (retries > deferred_unmount_retry_limit) {
2200 				error = EBUSY;
2201 				continue;
2202 			}
2203 			MNT_IUNLOCK(mp);
2204 
2205 			vfs_ref(upper->mp);
2206 			if (!deferred_unmount_enqueue(upper->mp, flags,
2207 			    false, 0))
2208 				vfs_rel(upper->mp);
2209 			MNT_ILOCK(mp);
2210 		}
2211 		mp->mnt_upper_pending--;
2212 		if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2213 		    mp->mnt_upper_pending == 0) {
2214 			mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2215 			wakeup(&mp->mnt_uppers);
2216 		}
2217 
2218 		/*
2219 		 * If we're not on the taskqueue, wait until the uppers list
2220 		 * is drained before proceeding with unmount.  Otherwise, if
2221 		 * we are on the taskqueue and there are still pending uppers,
2222 		 * just re-enqueue on the end of the taskqueue.
2223 		 */
2224 		if ((flags & MNT_DEFERRED) == 0) {
2225 			while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2226 				mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2227 				error = msleep(&mp->mnt_taskqueue_link,
2228 				    MNT_MTX(mp), PCATCH, "umntqw", 0);
2229 			}
2230 			if (error != 0) {
2231 				MNT_REL(mp);
2232 				MNT_IUNLOCK(mp);
2233 				return (error);
2234 			}
2235 		} else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2236 			MNT_IUNLOCK(mp);
2237 			if (error == 0)
2238 				deferred_unmount_enqueue(mp, flags, true, 0);
2239 			return (error);
2240 		}
2241 		MNT_IUNLOCK(mp);
2242 		KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2243 	}
2244 
2245 	/* Allow the taskqueue to safely re-enqueue on failure */
2246 	if ((flags & MNT_DEFERRED) != 0)
2247 		vfs_ref(mp);
2248 
2249 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2250 		mnt_gen_r = mp->mnt_gen;
2251 		VI_LOCK(coveredvp);
2252 		vholdl(coveredvp);
2253 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2254 		/*
2255 		 * Check for mp being unmounted while waiting for the
2256 		 * covered vnode lock.
2257 		 */
2258 		if (coveredvp->v_mountedhere != mp ||
2259 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2260 			VOP_UNLOCK(coveredvp);
2261 			vdrop(coveredvp);
2262 			vfs_rel(mp);
2263 			return (EBUSY);
2264 		}
2265 	}
2266 
2267 	vfs_op_enter(mp);
2268 
2269 	vn_start_write(NULL, &mp, V_WAIT);
2270 	MNT_ILOCK(mp);
2271 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2272 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
2273 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
2274 		dounmount_cleanup(mp, coveredvp, 0);
2275 		return (EBUSY);
2276 	}
2277 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
2278 	rootvp = vfs_cache_root_clear(mp);
2279 	if (coveredvp != NULL)
2280 		vn_seqc_write_begin(coveredvp);
2281 	if (flags & MNT_NONBUSY) {
2282 		MNT_IUNLOCK(mp);
2283 		error = vfs_check_usecounts(mp);
2284 		MNT_ILOCK(mp);
2285 		if (error != 0) {
2286 			vn_seqc_write_end(coveredvp);
2287 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2288 			if (rootvp != NULL) {
2289 				vn_seqc_write_end(rootvp);
2290 				vrele(rootvp);
2291 			}
2292 			return (error);
2293 		}
2294 	}
2295 	/* Allow filesystems to detect that a forced unmount is in progress. */
2296 	if (flags & MNT_FORCE) {
2297 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2298 		MNT_IUNLOCK(mp);
2299 		/*
2300 		 * Must be done after setting MNTK_UNMOUNTF and before
2301 		 * waiting for mnt_lockref to become 0.
2302 		 */
2303 		VFS_PURGE(mp);
2304 		MNT_ILOCK(mp);
2305 	}
2306 	error = 0;
2307 	if (mp->mnt_lockref) {
2308 		mp->mnt_kern_flag |= MNTK_DRAINING;
2309 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2310 		    "mount drain", 0);
2311 	}
2312 	MNT_IUNLOCK(mp);
2313 	KASSERT(mp->mnt_lockref == 0,
2314 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
2315 	    __func__, __FILE__, __LINE__));
2316 	KASSERT(error == 0,
2317 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
2318 	    __func__, __FILE__, __LINE__));
2319 
2320 	/*
2321 	 * We want to keep the vnode around so that we can vn_seqc_write_end
2322 	 * after we are done with unmount. Downgrade our reference to a mere
2323 	 * hold count so that we don't interefere with anything.
2324 	 */
2325 	if (rootvp != NULL) {
2326 		vhold(rootvp);
2327 		vrele(rootvp);
2328 	}
2329 
2330 	if (mp->mnt_flag & MNT_EXPUBLIC)
2331 		vfs_setpublicfs(NULL, NULL, NULL);
2332 
2333 	vfs_periodic(mp, MNT_WAIT);
2334 	MNT_ILOCK(mp);
2335 	async_flag = mp->mnt_flag & MNT_ASYNC;
2336 	mp->mnt_flag &= ~MNT_ASYNC;
2337 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
2338 	MNT_IUNLOCK(mp);
2339 	vfs_deallocate_syncvnode(mp);
2340 	error = VFS_UNMOUNT(mp, flags);
2341 	vn_finished_write(mp);
2342 	vfs_rel(mp);
2343 	/*
2344 	 * If we failed to flush the dirty blocks for this mount point,
2345 	 * undo all the cdir/rdir and rootvnode changes we made above.
2346 	 * Unless we failed to do so because the device is reporting that
2347 	 * it doesn't exist anymore.
2348 	 */
2349 	if (error && error != ENXIO) {
2350 		MNT_ILOCK(mp);
2351 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2352 			MNT_IUNLOCK(mp);
2353 			vfs_allocate_syncvnode(mp);
2354 			MNT_ILOCK(mp);
2355 		}
2356 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2357 		mp->mnt_flag |= async_flag;
2358 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2359 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2360 			mp->mnt_kern_flag |= MNTK_ASYNC;
2361 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
2362 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
2363 			wakeup(mp);
2364 		}
2365 		vfs_op_exit_locked(mp);
2366 		MNT_IUNLOCK(mp);
2367 		if (coveredvp) {
2368 			vn_seqc_write_end(coveredvp);
2369 			VOP_UNLOCK(coveredvp);
2370 			vdrop(coveredvp);
2371 		}
2372 		if (rootvp != NULL) {
2373 			vn_seqc_write_end(rootvp);
2374 			vdrop(rootvp);
2375 		}
2376 		return (error);
2377 	}
2378 
2379 	mtx_lock(&mountlist_mtx);
2380 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
2381 	mtx_unlock(&mountlist_mtx);
2382 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2383 	if (coveredvp != NULL) {
2384 		VI_LOCK(coveredvp);
2385 		vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2386 		coveredvp->v_mountedhere = NULL;
2387 		vn_seqc_write_end_locked(coveredvp);
2388 		VI_UNLOCK(coveredvp);
2389 		VOP_UNLOCK(coveredvp);
2390 		vdrop(coveredvp);
2391 	}
2392 	mount_devctl_event("UNMOUNT", mp, false);
2393 	if (rootvp != NULL) {
2394 		vn_seqc_write_end(rootvp);
2395 		vdrop(rootvp);
2396 	}
2397 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2398 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
2399 		vrele(rootvnode);
2400 		rootvnode = NULL;
2401 	}
2402 	if (mp == rootdevmp)
2403 		rootdevmp = NULL;
2404 	if ((flags & MNT_DEFERRED) != 0)
2405 		vfs_rel(mp);
2406 	vfs_mount_destroy(mp);
2407 	return (0);
2408 }
2409 
2410 /*
2411  * Report errors during filesystem mounting.
2412  */
2413 void
2414 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2415 {
2416 	struct vfsoptlist *moptlist = mp->mnt_optnew;
2417 	va_list ap;
2418 	int error, len;
2419 	char *errmsg;
2420 
2421 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2422 	if (error || errmsg == NULL || len <= 0)
2423 		return;
2424 
2425 	va_start(ap, fmt);
2426 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2427 	va_end(ap);
2428 }
2429 
2430 void
2431 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2432 {
2433 	va_list ap;
2434 	int error, len;
2435 	char *errmsg;
2436 
2437 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2438 	if (error || errmsg == NULL || len <= 0)
2439 		return;
2440 
2441 	va_start(ap, fmt);
2442 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2443 	va_end(ap);
2444 }
2445 
2446 /*
2447  * ---------------------------------------------------------------------
2448  * Functions for querying mount options/arguments from filesystems.
2449  */
2450 
2451 /*
2452  * Check that no unknown options are given
2453  */
2454 int
2455 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2456 {
2457 	struct vfsopt *opt;
2458 	char errmsg[255];
2459 	const char **t, *p, *q;
2460 	int ret = 0;
2461 
2462 	TAILQ_FOREACH(opt, opts, link) {
2463 		p = opt->name;
2464 		q = NULL;
2465 		if (p[0] == 'n' && p[1] == 'o')
2466 			q = p + 2;
2467 		for(t = global_opts; *t != NULL; t++) {
2468 			if (strcmp(*t, p) == 0)
2469 				break;
2470 			if (q != NULL) {
2471 				if (strcmp(*t, q) == 0)
2472 					break;
2473 			}
2474 		}
2475 		if (*t != NULL)
2476 			continue;
2477 		for(t = legal; *t != NULL; t++) {
2478 			if (strcmp(*t, p) == 0)
2479 				break;
2480 			if (q != NULL) {
2481 				if (strcmp(*t, q) == 0)
2482 					break;
2483 			}
2484 		}
2485 		if (*t != NULL)
2486 			continue;
2487 		snprintf(errmsg, sizeof(errmsg),
2488 		    "mount option <%s> is unknown", p);
2489 		ret = EINVAL;
2490 	}
2491 	if (ret != 0) {
2492 		TAILQ_FOREACH(opt, opts, link) {
2493 			if (strcmp(opt->name, "errmsg") == 0) {
2494 				strncpy((char *)opt->value, errmsg, opt->len);
2495 				break;
2496 			}
2497 		}
2498 		if (opt == NULL)
2499 			printf("%s\n", errmsg);
2500 	}
2501 	return (ret);
2502 }
2503 
2504 /*
2505  * Get a mount option by its name.
2506  *
2507  * Return 0 if the option was found, ENOENT otherwise.
2508  * If len is non-NULL it will be filled with the length
2509  * of the option. If buf is non-NULL, it will be filled
2510  * with the address of the option.
2511  */
2512 int
2513 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2514 {
2515 	struct vfsopt *opt;
2516 
2517 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2518 
2519 	TAILQ_FOREACH(opt, opts, link) {
2520 		if (strcmp(name, opt->name) == 0) {
2521 			opt->seen = 1;
2522 			if (len != NULL)
2523 				*len = opt->len;
2524 			if (buf != NULL)
2525 				*buf = opt->value;
2526 			return (0);
2527 		}
2528 	}
2529 	return (ENOENT);
2530 }
2531 
2532 int
2533 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2534 {
2535 	struct vfsopt *opt;
2536 
2537 	if (opts == NULL)
2538 		return (-1);
2539 
2540 	TAILQ_FOREACH(opt, opts, link) {
2541 		if (strcmp(name, opt->name) == 0) {
2542 			opt->seen = 1;
2543 			return (opt->pos);
2544 		}
2545 	}
2546 	return (-1);
2547 }
2548 
2549 int
2550 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2551 {
2552 	char *opt_value, *vtp;
2553 	quad_t iv;
2554 	int error, opt_len;
2555 
2556 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2557 	if (error != 0)
2558 		return (error);
2559 	if (opt_len == 0 || opt_value == NULL)
2560 		return (EINVAL);
2561 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2562 		return (EINVAL);
2563 	iv = strtoq(opt_value, &vtp, 0);
2564 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2565 		return (EINVAL);
2566 	if (iv < 0)
2567 		return (EINVAL);
2568 	switch (vtp[0]) {
2569 	case 't': case 'T':
2570 		iv *= 1024;
2571 		/* FALLTHROUGH */
2572 	case 'g': case 'G':
2573 		iv *= 1024;
2574 		/* FALLTHROUGH */
2575 	case 'm': case 'M':
2576 		iv *= 1024;
2577 		/* FALLTHROUGH */
2578 	case 'k': case 'K':
2579 		iv *= 1024;
2580 	case '\0':
2581 		break;
2582 	default:
2583 		return (EINVAL);
2584 	}
2585 	*value = iv;
2586 
2587 	return (0);
2588 }
2589 
2590 char *
2591 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2592 {
2593 	struct vfsopt *opt;
2594 
2595 	*error = 0;
2596 	TAILQ_FOREACH(opt, opts, link) {
2597 		if (strcmp(name, opt->name) != 0)
2598 			continue;
2599 		opt->seen = 1;
2600 		if (opt->len == 0 ||
2601 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2602 			*error = EINVAL;
2603 			return (NULL);
2604 		}
2605 		return (opt->value);
2606 	}
2607 	*error = ENOENT;
2608 	return (NULL);
2609 }
2610 
2611 int
2612 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2613 	uint64_t val)
2614 {
2615 	struct vfsopt *opt;
2616 
2617 	TAILQ_FOREACH(opt, opts, link) {
2618 		if (strcmp(name, opt->name) == 0) {
2619 			opt->seen = 1;
2620 			if (w != NULL)
2621 				*w |= val;
2622 			return (1);
2623 		}
2624 	}
2625 	if (w != NULL)
2626 		*w &= ~val;
2627 	return (0);
2628 }
2629 
2630 int
2631 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2632 {
2633 	va_list ap;
2634 	struct vfsopt *opt;
2635 	int ret;
2636 
2637 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2638 
2639 	TAILQ_FOREACH(opt, opts, link) {
2640 		if (strcmp(name, opt->name) != 0)
2641 			continue;
2642 		opt->seen = 1;
2643 		if (opt->len == 0 || opt->value == NULL)
2644 			return (0);
2645 		if (((char *)opt->value)[opt->len - 1] != '\0')
2646 			return (0);
2647 		va_start(ap, fmt);
2648 		ret = vsscanf(opt->value, fmt, ap);
2649 		va_end(ap);
2650 		return (ret);
2651 	}
2652 	return (0);
2653 }
2654 
2655 int
2656 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2657 {
2658 	struct vfsopt *opt;
2659 
2660 	TAILQ_FOREACH(opt, opts, link) {
2661 		if (strcmp(name, opt->name) != 0)
2662 			continue;
2663 		opt->seen = 1;
2664 		if (opt->value == NULL)
2665 			opt->len = len;
2666 		else {
2667 			if (opt->len != len)
2668 				return (EINVAL);
2669 			bcopy(value, opt->value, len);
2670 		}
2671 		return (0);
2672 	}
2673 	return (ENOENT);
2674 }
2675 
2676 int
2677 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2678 {
2679 	struct vfsopt *opt;
2680 
2681 	TAILQ_FOREACH(opt, opts, link) {
2682 		if (strcmp(name, opt->name) != 0)
2683 			continue;
2684 		opt->seen = 1;
2685 		if (opt->value == NULL)
2686 			opt->len = len;
2687 		else {
2688 			if (opt->len < len)
2689 				return (EINVAL);
2690 			opt->len = len;
2691 			bcopy(value, opt->value, len);
2692 		}
2693 		return (0);
2694 	}
2695 	return (ENOENT);
2696 }
2697 
2698 int
2699 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2700 {
2701 	struct vfsopt *opt;
2702 
2703 	TAILQ_FOREACH(opt, opts, link) {
2704 		if (strcmp(name, opt->name) != 0)
2705 			continue;
2706 		opt->seen = 1;
2707 		if (opt->value == NULL)
2708 			opt->len = strlen(value) + 1;
2709 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2710 			return (EINVAL);
2711 		return (0);
2712 	}
2713 	return (ENOENT);
2714 }
2715 
2716 /*
2717  * Find and copy a mount option.
2718  *
2719  * The size of the buffer has to be specified
2720  * in len, if it is not the same length as the
2721  * mount option, EINVAL is returned.
2722  * Returns ENOENT if the option is not found.
2723  */
2724 int
2725 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2726 {
2727 	struct vfsopt *opt;
2728 
2729 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2730 
2731 	TAILQ_FOREACH(opt, opts, link) {
2732 		if (strcmp(name, opt->name) == 0) {
2733 			opt->seen = 1;
2734 			if (len != opt->len)
2735 				return (EINVAL);
2736 			bcopy(opt->value, dest, opt->len);
2737 			return (0);
2738 		}
2739 	}
2740 	return (ENOENT);
2741 }
2742 
2743 int
2744 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2745 {
2746 	/*
2747 	 * Filesystems only fill in part of the structure for updates, we
2748 	 * have to read the entirety first to get all content.
2749 	 */
2750 	if (sbp != &mp->mnt_stat)
2751 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2752 
2753 	/*
2754 	 * Set these in case the underlying filesystem fails to do so.
2755 	 */
2756 	sbp->f_version = STATFS_VERSION;
2757 	sbp->f_namemax = NAME_MAX;
2758 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2759 	sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2760 
2761 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2762 }
2763 
2764 void
2765 vfs_mountedfrom(struct mount *mp, const char *from)
2766 {
2767 
2768 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2769 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2770 	    sizeof mp->mnt_stat.f_mntfromname);
2771 }
2772 
2773 /*
2774  * ---------------------------------------------------------------------
2775  * This is the api for building mount args and mounting filesystems from
2776  * inside the kernel.
2777  *
2778  * The API works by accumulation of individual args.  First error is
2779  * latched.
2780  *
2781  * XXX: should be documented in new manpage kernel_mount(9)
2782  */
2783 
2784 /* A memory allocation which must be freed when we are done */
2785 struct mntaarg {
2786 	SLIST_ENTRY(mntaarg)	next;
2787 };
2788 
2789 /* The header for the mount arguments */
2790 struct mntarg {
2791 	struct iovec *v;
2792 	int len;
2793 	int error;
2794 	SLIST_HEAD(, mntaarg)	list;
2795 };
2796 
2797 /*
2798  * Add a boolean argument.
2799  *
2800  * flag is the boolean value.
2801  * name must start with "no".
2802  */
2803 struct mntarg *
2804 mount_argb(struct mntarg *ma, int flag, const char *name)
2805 {
2806 
2807 	KASSERT(name[0] == 'n' && name[1] == 'o',
2808 	    ("mount_argb(...,%s): name must start with 'no'", name));
2809 
2810 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2811 }
2812 
2813 /*
2814  * Add an argument printf style
2815  */
2816 struct mntarg *
2817 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2818 {
2819 	va_list ap;
2820 	struct mntaarg *maa;
2821 	struct sbuf *sb;
2822 	int len;
2823 
2824 	if (ma == NULL) {
2825 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2826 		SLIST_INIT(&ma->list);
2827 	}
2828 	if (ma->error)
2829 		return (ma);
2830 
2831 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2832 	    M_MOUNT, M_WAITOK);
2833 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2834 	ma->v[ma->len].iov_len = strlen(name) + 1;
2835 	ma->len++;
2836 
2837 	sb = sbuf_new_auto();
2838 	va_start(ap, fmt);
2839 	sbuf_vprintf(sb, fmt, ap);
2840 	va_end(ap);
2841 	sbuf_finish(sb);
2842 	len = sbuf_len(sb) + 1;
2843 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2844 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2845 	bcopy(sbuf_data(sb), maa + 1, len);
2846 	sbuf_delete(sb);
2847 
2848 	ma->v[ma->len].iov_base = maa + 1;
2849 	ma->v[ma->len].iov_len = len;
2850 	ma->len++;
2851 
2852 	return (ma);
2853 }
2854 
2855 /*
2856  * Add an argument which is a userland string.
2857  */
2858 struct mntarg *
2859 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2860 {
2861 	struct mntaarg *maa;
2862 	char *tbuf;
2863 
2864 	if (val == NULL)
2865 		return (ma);
2866 	if (ma == NULL) {
2867 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2868 		SLIST_INIT(&ma->list);
2869 	}
2870 	if (ma->error)
2871 		return (ma);
2872 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2873 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2874 	tbuf = (void *)(maa + 1);
2875 	ma->error = copyinstr(val, tbuf, len, NULL);
2876 	return (mount_arg(ma, name, tbuf, -1));
2877 }
2878 
2879 /*
2880  * Plain argument.
2881  *
2882  * If length is -1, treat value as a C string.
2883  */
2884 struct mntarg *
2885 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2886 {
2887 
2888 	if (ma == NULL) {
2889 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2890 		SLIST_INIT(&ma->list);
2891 	}
2892 	if (ma->error)
2893 		return (ma);
2894 
2895 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2896 	    M_MOUNT, M_WAITOK);
2897 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2898 	ma->v[ma->len].iov_len = strlen(name) + 1;
2899 	ma->len++;
2900 
2901 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2902 	if (len < 0)
2903 		ma->v[ma->len].iov_len = strlen(val) + 1;
2904 	else
2905 		ma->v[ma->len].iov_len = len;
2906 	ma->len++;
2907 	return (ma);
2908 }
2909 
2910 /*
2911  * Free a mntarg structure
2912  */
2913 static void
2914 free_mntarg(struct mntarg *ma)
2915 {
2916 	struct mntaarg *maa;
2917 
2918 	while (!SLIST_EMPTY(&ma->list)) {
2919 		maa = SLIST_FIRST(&ma->list);
2920 		SLIST_REMOVE_HEAD(&ma->list, next);
2921 		free(maa, M_MOUNT);
2922 	}
2923 	free(ma->v, M_MOUNT);
2924 	free(ma, M_MOUNT);
2925 }
2926 
2927 /*
2928  * Mount a filesystem
2929  */
2930 int
2931 kernel_mount(struct mntarg *ma, uint64_t flags)
2932 {
2933 	struct uio auio;
2934 	int error;
2935 
2936 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2937 	KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2938 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2939 
2940 	error = ma->error;
2941 	if (error == 0) {
2942 		auio.uio_iov = ma->v;
2943 		auio.uio_iovcnt = ma->len;
2944 		auio.uio_segflg = UIO_SYSSPACE;
2945 		error = vfs_donmount(curthread, flags, &auio);
2946 	}
2947 	free_mntarg(ma);
2948 	return (error);
2949 }
2950 
2951 /* Map from mount options to printable formats. */
2952 static struct mntoptnames optnames[] = {
2953 	MNTOPT_NAMES
2954 };
2955 
2956 #define DEVCTL_LEN 1024
2957 static void
2958 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2959 {
2960 	const uint8_t *cp;
2961 	struct mntoptnames *fp;
2962 	struct sbuf sb;
2963 	struct statfs *sfp = &mp->mnt_stat;
2964 	char *buf;
2965 
2966 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2967 	if (buf == NULL)
2968 		return;
2969 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2970 	sbuf_cpy(&sb, "mount-point=\"");
2971 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2972 	sbuf_cat(&sb, "\" mount-dev=\"");
2973 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2974 	sbuf_cat(&sb, "\" mount-type=\"");
2975 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2976 	sbuf_cat(&sb, "\" fsid=0x");
2977 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2978 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2979 		sbuf_printf(&sb, "%02x", cp[i]);
2980 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2981 	for (fp = optnames; fp->o_opt != 0; fp++) {
2982 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2983 			sbuf_cat(&sb, fp->o_name);
2984 			sbuf_putc(&sb, ';');
2985 		}
2986 	}
2987 	sbuf_putc(&sb, '"');
2988 	sbuf_finish(&sb);
2989 
2990 	/*
2991 	 * Options are not published because the form of the options depends on
2992 	 * the file system and may include binary data. In addition, they don't
2993 	 * necessarily provide enough useful information to be actionable when
2994 	 * devd processes them.
2995 	 */
2996 
2997 	if (sbuf_error(&sb) == 0)
2998 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2999 	sbuf_delete(&sb);
3000 	free(buf, M_MOUNT);
3001 }
3002 
3003 /*
3004  * Force remount specified mount point to read-only.  The argument
3005  * must be busied to avoid parallel unmount attempts.
3006  *
3007  * Intended use is to prevent further writes if some metadata
3008  * inconsistency is detected.  Note that the function still flushes
3009  * all cached metadata and data for the mount point, which might be
3010  * not always suitable.
3011  */
3012 int
3013 vfs_remount_ro(struct mount *mp)
3014 {
3015 	struct vfsoptlist *opts;
3016 	struct vfsopt *opt;
3017 	struct vnode *vp_covered, *rootvp;
3018 	int error;
3019 
3020 	vfs_op_enter(mp);
3021 	KASSERT(mp->mnt_lockref > 0,
3022 	    ("vfs_remount_ro: mp %p is not busied", mp));
3023 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
3024 	    ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
3025 
3026 	rootvp = NULL;
3027 	vp_covered = mp->mnt_vnodecovered;
3028 	error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
3029 	if (error != 0) {
3030 		vfs_op_exit(mp);
3031 		return (error);
3032 	}
3033 	VI_LOCK(vp_covered);
3034 	if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
3035 		VI_UNLOCK(vp_covered);
3036 		vput(vp_covered);
3037 		vfs_op_exit(mp);
3038 		return (EBUSY);
3039 	}
3040 	vp_covered->v_iflag |= VI_MOUNT;
3041 	VI_UNLOCK(vp_covered);
3042 	vn_seqc_write_begin(vp_covered);
3043 
3044 	MNT_ILOCK(mp);
3045 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
3046 		MNT_IUNLOCK(mp);
3047 		error = EBUSY;
3048 		goto out;
3049 	}
3050 	mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
3051 	rootvp = vfs_cache_root_clear(mp);
3052 	MNT_IUNLOCK(mp);
3053 
3054 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
3055 	TAILQ_INIT(opts);
3056 	opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
3057 	opt->name = strdup("ro", M_MOUNT);
3058 	opt->value = NULL;
3059 	TAILQ_INSERT_TAIL(opts, opt, link);
3060 	vfs_mergeopts(opts, mp->mnt_opt);
3061 	mp->mnt_optnew = opts;
3062 
3063 	error = VFS_MOUNT(mp);
3064 
3065 	if (error == 0) {
3066 		MNT_ILOCK(mp);
3067 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
3068 		MNT_IUNLOCK(mp);
3069 		vfs_deallocate_syncvnode(mp);
3070 		if (mp->mnt_opt != NULL)
3071 			vfs_freeopts(mp->mnt_opt);
3072 		mp->mnt_opt = mp->mnt_optnew;
3073 	} else {
3074 		MNT_ILOCK(mp);
3075 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
3076 		MNT_IUNLOCK(mp);
3077 		vfs_freeopts(mp->mnt_optnew);
3078 	}
3079 	mp->mnt_optnew = NULL;
3080 
3081 out:
3082 	vfs_op_exit(mp);
3083 	VI_LOCK(vp_covered);
3084 	vp_covered->v_iflag &= ~VI_MOUNT;
3085 	VI_UNLOCK(vp_covered);
3086 	vput(vp_covered);
3087 	vn_seqc_write_end(vp_covered);
3088 	if (rootvp != NULL) {
3089 		vn_seqc_write_end(rootvp);
3090 		vrele(rootvp);
3091 	}
3092 	return (error);
3093 }
3094 
3095 /*
3096  * Suspend write operations on all local writeable filesystems.  Does
3097  * full sync of them in the process.
3098  *
3099  * Iterate over the mount points in reverse order, suspending most
3100  * recently mounted filesystems first.  It handles a case where a
3101  * filesystem mounted from a md(4) vnode-backed device should be
3102  * suspended before the filesystem that owns the vnode.
3103  */
3104 void
3105 suspend_all_fs(void)
3106 {
3107 	struct mount *mp;
3108 	int error;
3109 
3110 	mtx_lock(&mountlist_mtx);
3111 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
3112 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
3113 		if (error != 0)
3114 			continue;
3115 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
3116 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
3117 			mtx_lock(&mountlist_mtx);
3118 			vfs_unbusy(mp);
3119 			continue;
3120 		}
3121 		error = vfs_write_suspend(mp, 0);
3122 		if (error == 0) {
3123 			MNT_ILOCK(mp);
3124 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
3125 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
3126 			MNT_IUNLOCK(mp);
3127 			mtx_lock(&mountlist_mtx);
3128 		} else {
3129 			printf("suspend of %s failed, error %d\n",
3130 			    mp->mnt_stat.f_mntonname, error);
3131 			mtx_lock(&mountlist_mtx);
3132 			vfs_unbusy(mp);
3133 		}
3134 	}
3135 	mtx_unlock(&mountlist_mtx);
3136 }
3137 
3138 /*
3139  * Clone the mnt_exjail field to a new mount point.
3140  */
3141 void
3142 vfs_exjail_clone(struct mount *inmp, struct mount *outmp)
3143 {
3144 	struct ucred *cr;
3145 	struct prison *pr;
3146 
3147 	MNT_ILOCK(inmp);
3148 	cr = inmp->mnt_exjail;
3149 	if (cr != NULL) {
3150 		crhold(cr);
3151 		MNT_IUNLOCK(inmp);
3152 		pr = cr->cr_prison;
3153 		sx_slock(&allprison_lock);
3154 		if (!prison_isalive(pr)) {
3155 			sx_sunlock(&allprison_lock);
3156 			crfree(cr);
3157 			return;
3158 		}
3159 		MNT_ILOCK(outmp);
3160 		if (outmp->mnt_exjail == NULL) {
3161 			outmp->mnt_exjail = cr;
3162 			atomic_add_int(&pr->pr_exportcnt, 1);
3163 			cr = NULL;
3164 		}
3165 		MNT_IUNLOCK(outmp);
3166 		sx_sunlock(&allprison_lock);
3167 		if (cr != NULL)
3168 			crfree(cr);
3169 	} else
3170 		MNT_IUNLOCK(inmp);
3171 }
3172 
3173 void
3174 resume_all_fs(void)
3175 {
3176 	struct mount *mp;
3177 
3178 	mtx_lock(&mountlist_mtx);
3179 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3180 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3181 			continue;
3182 		mtx_unlock(&mountlist_mtx);
3183 		MNT_ILOCK(mp);
3184 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3185 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3186 		MNT_IUNLOCK(mp);
3187 		vfs_write_resume(mp, 0);
3188 		mtx_lock(&mountlist_mtx);
3189 		vfs_unbusy(mp);
3190 	}
3191 	mtx_unlock(&mountlist_mtx);
3192 }
3193