xref: /freebsd/sys/kern/vfs_mount.c (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
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 	freeuio(auio);
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, mnt_union;
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_union = 0;
1399 	MNT_ILOCK(mp);
1400 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1401 		MNT_IUNLOCK(mp);
1402 		error = EBUSY;
1403 		goto end;
1404 	}
1405 	if (vfs_suser_failed) {
1406 		KASSERT((fsflags & (MNT_EXPORTED | MNT_UPDATE)) ==
1407 		    (MNT_EXPORTED | MNT_UPDATE),
1408 		    ("%s: jailed export did not set expected fsflags",
1409 		     __func__));
1410 		/*
1411 		 * For this case, only MNT_UPDATE and
1412 		 * MNT_EXPORTED have been set in fsflags
1413 		 * by the options.  Only set MNT_UPDATE,
1414 		 * since that is the one that would be set
1415 		 * when set in fsflags, below.
1416 		 */
1417 		mp->mnt_flag |= MNT_UPDATE;
1418 	} else {
1419 		mp->mnt_flag &= ~MNT_UPDATEMASK;
1420 		if ((mp->mnt_flag & MNT_UNION) == 0 &&
1421 		    (fsflags & MNT_UNION) != 0) {
1422 			fsflags &= ~MNT_UNION;
1423 			mnt_union = MNT_UNION;
1424 		}
1425 		mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1426 		    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1427 		if ((mp->mnt_flag & MNT_ASYNC) == 0)
1428 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1429 	}
1430 	rootvp = vfs_cache_root_clear(mp);
1431 	MNT_IUNLOCK(mp);
1432 	mp->mnt_optnew = *optlist;
1433 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1434 
1435 	/*
1436 	 * Mount the filesystem.
1437 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1438 	 * get.  No freeing of cn_pnbuf.
1439 	 */
1440 	/*
1441 	 * For the case of mountd(8) doing exports from within a vnet jail,
1442 	 * "from" is typically not set correctly such that VFS_MOUNT() will
1443 	 * return ENOENT. It is not obvious that VFS_MOUNT() ever needs to be
1444 	 * called when mountd is doing exports, but this check only applies to
1445 	 * the specific case where it is running inside a vnet jail, to
1446 	 * avoid any POLA violation.
1447 	 */
1448 	error = 0;
1449 	if (!jail_export)
1450 		error = VFS_MOUNT(mp);
1451 
1452 	export_error = 0;
1453 	/* Process the export option. */
1454 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1455 	    &len) == 0) {
1456 		/* Assume that there is only 1 ABI for each length. */
1457 		switch (len) {
1458 		case (sizeof(struct oexport_args)):
1459 			bzero(&o2export, sizeof(o2export));
1460 			/* FALLTHROUGH */
1461 		case (sizeof(o2export)):
1462 			bcopy(bufp, &o2export, len);
1463 			export.ex_flags = (uint64_t)o2export.ex_flags;
1464 			export.ex_root = o2export.ex_root;
1465 			export.ex_uid = o2export.ex_anon.cr_uid;
1466 			export.ex_groups = NULL;
1467 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1468 			if (export.ex_ngroups > 0) {
1469 				if (export.ex_ngroups <= XU_NGROUPS) {
1470 					export.ex_groups = malloc(
1471 					    export.ex_ngroups * sizeof(gid_t),
1472 					    M_TEMP, M_WAITOK);
1473 					for (i = 0; i < export.ex_ngroups; i++)
1474 						export.ex_groups[i] =
1475 						  o2export.ex_anon.cr_groups[i];
1476 				} else
1477 					export_error = EINVAL;
1478 			} else if (export.ex_ngroups < 0)
1479 				export_error = EINVAL;
1480 			export.ex_addr = o2export.ex_addr;
1481 			export.ex_addrlen = o2export.ex_addrlen;
1482 			export.ex_mask = o2export.ex_mask;
1483 			export.ex_masklen = o2export.ex_masklen;
1484 			export.ex_indexfile = o2export.ex_indexfile;
1485 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1486 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1487 				for (i = 0; i < export.ex_numsecflavors; i++)
1488 					export.ex_secflavors[i] =
1489 					    o2export.ex_secflavors[i];
1490 			} else
1491 				export_error = EINVAL;
1492 			if (export_error == 0)
1493 				export_error = vfs_export(mp, &export, true);
1494 			free(export.ex_groups, M_TEMP);
1495 			break;
1496 		case (sizeof(export)):
1497 			bcopy(bufp, &export, len);
1498 			grps = NULL;
1499 			if (export.ex_ngroups > 0) {
1500 				if (export.ex_ngroups <= NGROUPS_MAX) {
1501 					grps = malloc(export.ex_ngroups *
1502 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1503 					export_error = copyin(export.ex_groups,
1504 					    grps, export.ex_ngroups *
1505 					    sizeof(gid_t));
1506 					if (export_error == 0)
1507 						export.ex_groups = grps;
1508 				} else
1509 					export_error = EINVAL;
1510 			} else if (export.ex_ngroups == 0)
1511 				export.ex_groups = NULL;
1512 			else
1513 				export_error = EINVAL;
1514 			if (export_error == 0)
1515 				export_error = vfs_export(mp, &export, true);
1516 			free(grps, M_TEMP);
1517 			break;
1518 		default:
1519 			export_error = EINVAL;
1520 			break;
1521 		}
1522 	}
1523 
1524 	MNT_ILOCK(mp);
1525 	if (error == 0) {
1526 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1527 		    MNT_SNAPSHOT);
1528 		mp->mnt_flag |= mnt_union;
1529 	} else {
1530 		/*
1531 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1532 		 * because it is not part of MNT_UPDATEMASK, but it could have
1533 		 * changed in the meantime if quotactl(2) was called.
1534 		 * All in all we want current value of MNT_QUOTA, not the old
1535 		 * one.
1536 		 */
1537 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1538 	}
1539 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1540 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1541 		mp->mnt_kern_flag |= MNTK_ASYNC;
1542 	else
1543 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1544 	MNT_IUNLOCK(mp);
1545 
1546 	if (error != 0)
1547 		goto end;
1548 
1549 	mount_devctl_event("REMOUNT", mp, true);
1550 	if (mp->mnt_opt != NULL)
1551 		vfs_freeopts(mp->mnt_opt);
1552 	mp->mnt_opt = mp->mnt_optnew;
1553 	*optlist = NULL;
1554 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1555 	/*
1556 	 * Prevent external consumers of mount options from reading
1557 	 * mnt_optnew.
1558 	 */
1559 	mp->mnt_optnew = NULL;
1560 
1561 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1562 		vfs_allocate_syncvnode(mp);
1563 	else
1564 		vfs_deallocate_syncvnode(mp);
1565 end:
1566 	vfs_op_exit(mp);
1567 	if (rootvp != NULL) {
1568 		vn_seqc_write_end(rootvp);
1569 		vrele(rootvp);
1570 	}
1571 	vn_seqc_write_end(vp);
1572 	vfs_unbusy(mp);
1573 	VI_LOCK(vp);
1574 	vp->v_iflag &= ~VI_MOUNT;
1575 	VI_UNLOCK(vp);
1576 	vrele(vp);
1577 	return (error != 0 ? error : export_error);
1578 }
1579 
1580 /*
1581  * vfs_domount(): actually attempt a filesystem mount.
1582  */
1583 static int
1584 vfs_domount(
1585 	struct thread *td,		/* Calling thread. */
1586 	const char *fstype,		/* Filesystem type. */
1587 	char *fspath,			/* Mount path. */
1588 	uint64_t fsflags,		/* Flags common to all filesystems. */
1589 	bool jail_export,		/* Got export option in vnet prison. */
1590 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1591 	)
1592 {
1593 	struct vfsconf *vfsp;
1594 	struct nameidata nd;
1595 	struct vnode *vp;
1596 	char *pathbuf;
1597 	int error;
1598 
1599 	/*
1600 	 * Be ultra-paranoid about making sure the type and fspath
1601 	 * variables will fit in our mp buffers, including the
1602 	 * terminating NUL.
1603 	 */
1604 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1605 		return (ENAMETOOLONG);
1606 
1607 	if (jail_export) {
1608 		error = priv_check(td, PRIV_NFS_DAEMON);
1609 		if (error)
1610 			return (error);
1611 	} else if (jailed(td->td_ucred) || usermount == 0) {
1612 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1613 			return (error);
1614 	}
1615 
1616 	/*
1617 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1618 	 */
1619 	if (fsflags & MNT_EXPORTED) {
1620 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1621 		if (error)
1622 			return (error);
1623 	}
1624 	if (fsflags & MNT_SUIDDIR) {
1625 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1626 		if (error)
1627 			return (error);
1628 	}
1629 	/*
1630 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1631 	 */
1632 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1633 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1634 			fsflags |= MNT_NOSUID | MNT_USER;
1635 	}
1636 
1637 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1638 	vfsp = NULL;
1639 	if ((fsflags & MNT_UPDATE) == 0) {
1640 		/* Don't try to load KLDs if we're mounting the root. */
1641 		if (fsflags & MNT_ROOTFS) {
1642 			if ((vfsp = vfs_byname(fstype)) == NULL)
1643 				return (ENODEV);
1644 		} else {
1645 			if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1646 				return (error);
1647 		}
1648 	}
1649 
1650 	/*
1651 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1652 	 */
1653 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1654 	    UIO_SYSSPACE, fspath);
1655 	error = namei(&nd);
1656 	if (error != 0)
1657 		return (error);
1658 	vp = nd.ni_vp;
1659 	/*
1660 	 * Don't allow stacking file mounts to work around problems with the way
1661 	 * that namei sets nd.ni_dvp to vp_crossmp for these.
1662 	 */
1663 	if (vp->v_type == VREG)
1664 		fsflags |= MNT_NOCOVER;
1665 	if ((fsflags & MNT_UPDATE) == 0) {
1666 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1667 		    (fsflags & MNT_NOCOVER) != 0) {
1668 			vput(vp);
1669 			error = EBUSY;
1670 			goto out;
1671 		}
1672 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1673 		strcpy(pathbuf, fspath);
1674 		/*
1675 		 * Note: we allow any vnode type here. If the path sanity check
1676 		 * succeeds, the type will be validated in vfs_domount_first
1677 		 * above.
1678 		 */
1679 		if (vp->v_type == VDIR)
1680 			error = vn_path_to_global_path(td, vp, pathbuf,
1681 			    MNAMELEN);
1682 		else
1683 			error = vn_path_to_global_path_hardlink(td, vp,
1684 			    nd.ni_dvp, pathbuf, MNAMELEN,
1685 			    nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1686 		if (error == 0) {
1687 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1688 			    fsflags, optlist);
1689 		}
1690 		free(pathbuf, M_TEMP);
1691 	} else
1692 		error = vfs_domount_update(td, vp, fsflags, jail_export,
1693 		    optlist);
1694 
1695 out:
1696 	NDFREE_PNBUF(&nd);
1697 	vrele(nd.ni_dvp);
1698 
1699 	return (error);
1700 }
1701 
1702 /*
1703  * Unmount a filesystem.
1704  *
1705  * Note: unmount takes a path to the vnode mounted on as argument, not
1706  * special file (as before).
1707  */
1708 #ifndef _SYS_SYSPROTO_H_
1709 struct unmount_args {
1710 	char	*path;
1711 	int	flags;
1712 };
1713 #endif
1714 /* ARGSUSED */
1715 int
1716 sys_unmount(struct thread *td, struct unmount_args *uap)
1717 {
1718 
1719 	return (kern_unmount(td, uap->path, uap->flags));
1720 }
1721 
1722 int
1723 kern_unmount(struct thread *td, const char *path, int flags)
1724 {
1725 	struct nameidata nd;
1726 	struct mount *mp;
1727 	char *fsidbuf, *pathbuf;
1728 	fsid_t fsid;
1729 	int error;
1730 
1731 	AUDIT_ARG_VALUE(flags);
1732 	if (jailed(td->td_ucred) || usermount == 0) {
1733 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1734 		if (error)
1735 			return (error);
1736 	}
1737 
1738 	if (flags & MNT_BYFSID) {
1739 		fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1740 		error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1741 		if (error) {
1742 			free(fsidbuf, M_TEMP);
1743 			return (error);
1744 		}
1745 
1746 		AUDIT_ARG_TEXT(fsidbuf);
1747 		/* Decode the filesystem ID. */
1748 		if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1749 			free(fsidbuf, M_TEMP);
1750 			return (EINVAL);
1751 		}
1752 
1753 		mp = vfs_getvfs(&fsid);
1754 		free(fsidbuf, M_TEMP);
1755 		if (mp == NULL) {
1756 			return (ENOENT);
1757 		}
1758 	} else {
1759 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1760 		error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1761 		if (error) {
1762 			free(pathbuf, M_TEMP);
1763 			return (error);
1764 		}
1765 
1766 		/*
1767 		 * Try to find global path for path argument.
1768 		 */
1769 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1770 		    UIO_SYSSPACE, pathbuf);
1771 		if (namei(&nd) == 0) {
1772 			NDFREE_PNBUF(&nd);
1773 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1774 			    MNAMELEN);
1775 			if (error == 0)
1776 				vput(nd.ni_vp);
1777 		}
1778 		mtx_lock(&mountlist_mtx);
1779 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1780 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1781 				vfs_ref(mp);
1782 				break;
1783 			}
1784 		}
1785 		mtx_unlock(&mountlist_mtx);
1786 		free(pathbuf, M_TEMP);
1787 		if (mp == NULL) {
1788 			/*
1789 			 * Previously we returned ENOENT for a nonexistent path and
1790 			 * EINVAL for a non-mountpoint.  We cannot tell these apart
1791 			 * now, so in the !MNT_BYFSID case return the more likely
1792 			 * EINVAL for compatibility.
1793 			 */
1794 			return (EINVAL);
1795 		}
1796 	}
1797 
1798 	/*
1799 	 * Don't allow unmounting the root filesystem.
1800 	 */
1801 	if (mp->mnt_flag & MNT_ROOTFS) {
1802 		vfs_rel(mp);
1803 		return (EINVAL);
1804 	}
1805 	error = dounmount(mp, flags, td);
1806 	return (error);
1807 }
1808 
1809 /*
1810  * Return error if any of the vnodes, ignoring the root vnode
1811  * and the syncer vnode, have non-zero usecount.
1812  *
1813  * This function is purely advisory - it can return false positives
1814  * and negatives.
1815  */
1816 static int
1817 vfs_check_usecounts(struct mount *mp)
1818 {
1819 	struct vnode *vp, *mvp;
1820 
1821 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1822 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1823 		    vp->v_usecount != 0) {
1824 			VI_UNLOCK(vp);
1825 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1826 			return (EBUSY);
1827 		}
1828 		VI_UNLOCK(vp);
1829 	}
1830 
1831 	return (0);
1832 }
1833 
1834 static void
1835 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1836 {
1837 
1838 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1839 	mp->mnt_kern_flag &= ~mntkflags;
1840 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1841 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1842 		wakeup(mp);
1843 	}
1844 	vfs_op_exit_locked(mp);
1845 	MNT_IUNLOCK(mp);
1846 	if (coveredvp != NULL) {
1847 		VOP_UNLOCK(coveredvp);
1848 		vdrop(coveredvp);
1849 	}
1850 	vn_finished_write(mp);
1851 	vfs_rel(mp);
1852 }
1853 
1854 /*
1855  * There are various reference counters associated with the mount point.
1856  * Normally it is permitted to modify them without taking the mnt ilock,
1857  * but this behavior can be temporarily disabled if stable value is needed
1858  * or callers are expected to block (e.g. to not allow new users during
1859  * forced unmount).
1860  */
1861 void
1862 vfs_op_enter(struct mount *mp)
1863 {
1864 	struct mount_pcpu *mpcpu;
1865 	int cpu;
1866 
1867 	MNT_ILOCK(mp);
1868 	mp->mnt_vfs_ops++;
1869 	if (mp->mnt_vfs_ops > 1) {
1870 		MNT_IUNLOCK(mp);
1871 		return;
1872 	}
1873 	vfs_op_barrier_wait(mp);
1874 	CPU_FOREACH(cpu) {
1875 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1876 
1877 		mp->mnt_ref += mpcpu->mntp_ref;
1878 		mpcpu->mntp_ref = 0;
1879 
1880 		mp->mnt_lockref += mpcpu->mntp_lockref;
1881 		mpcpu->mntp_lockref = 0;
1882 
1883 		mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1884 		mpcpu->mntp_writeopcount = 0;
1885 	}
1886 	MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1887 	    mp->mnt_writeopcount >= 0, mp,
1888 	    ("invalid count(s): ref %d lockref %d writeopcount %d",
1889 	    mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1890 	MNT_IUNLOCK(mp);
1891 	vfs_assert_mount_counters(mp);
1892 }
1893 
1894 void
1895 vfs_op_exit_locked(struct mount *mp)
1896 {
1897 
1898 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1899 
1900 	MPASSERT(mp->mnt_vfs_ops > 0, mp,
1901 	    ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1902 	MPASSERT(mp->mnt_vfs_ops > 1 ||
1903 	    (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1904 	    ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1905 	mp->mnt_vfs_ops--;
1906 }
1907 
1908 void
1909 vfs_op_exit(struct mount *mp)
1910 {
1911 
1912 	MNT_ILOCK(mp);
1913 	vfs_op_exit_locked(mp);
1914 	MNT_IUNLOCK(mp);
1915 }
1916 
1917 struct vfs_op_barrier_ipi {
1918 	struct mount *mp;
1919 	struct smp_rendezvous_cpus_retry_arg srcra;
1920 };
1921 
1922 static void
1923 vfs_op_action_func(void *arg)
1924 {
1925 	struct vfs_op_barrier_ipi *vfsopipi;
1926 	struct mount *mp;
1927 
1928 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1929 	mp = vfsopipi->mp;
1930 
1931 	if (!vfs_op_thread_entered(mp))
1932 		smp_rendezvous_cpus_done(arg);
1933 }
1934 
1935 static void
1936 vfs_op_wait_func(void *arg, int cpu)
1937 {
1938 	struct vfs_op_barrier_ipi *vfsopipi;
1939 	struct mount *mp;
1940 	struct mount_pcpu *mpcpu;
1941 
1942 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1943 	mp = vfsopipi->mp;
1944 
1945 	mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1946 	while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1947 		cpu_spinwait();
1948 }
1949 
1950 void
1951 vfs_op_barrier_wait(struct mount *mp)
1952 {
1953 	struct vfs_op_barrier_ipi vfsopipi;
1954 
1955 	vfsopipi.mp = mp;
1956 
1957 	smp_rendezvous_cpus_retry(all_cpus,
1958 	    smp_no_rendezvous_barrier,
1959 	    vfs_op_action_func,
1960 	    smp_no_rendezvous_barrier,
1961 	    vfs_op_wait_func,
1962 	    &vfsopipi.srcra);
1963 }
1964 
1965 #ifdef DIAGNOSTIC
1966 void
1967 vfs_assert_mount_counters(struct mount *mp)
1968 {
1969 	struct mount_pcpu *mpcpu;
1970 	int cpu;
1971 
1972 	if (mp->mnt_vfs_ops == 0)
1973 		return;
1974 
1975 	CPU_FOREACH(cpu) {
1976 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1977 		if (mpcpu->mntp_ref != 0 ||
1978 		    mpcpu->mntp_lockref != 0 ||
1979 		    mpcpu->mntp_writeopcount != 0)
1980 			vfs_dump_mount_counters(mp);
1981 	}
1982 }
1983 
1984 void
1985 vfs_dump_mount_counters(struct mount *mp)
1986 {
1987 	struct mount_pcpu *mpcpu;
1988 	int ref, lockref, writeopcount;
1989 	int cpu;
1990 
1991 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1992 
1993 	printf("        ref : ");
1994 	ref = mp->mnt_ref;
1995 	CPU_FOREACH(cpu) {
1996 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1997 		printf("%d ", mpcpu->mntp_ref);
1998 		ref += mpcpu->mntp_ref;
1999 	}
2000 	printf("\n");
2001 	printf("    lockref : ");
2002 	lockref = mp->mnt_lockref;
2003 	CPU_FOREACH(cpu) {
2004 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2005 		printf("%d ", mpcpu->mntp_lockref);
2006 		lockref += mpcpu->mntp_lockref;
2007 	}
2008 	printf("\n");
2009 	printf("writeopcount: ");
2010 	writeopcount = mp->mnt_writeopcount;
2011 	CPU_FOREACH(cpu) {
2012 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2013 		printf("%d ", mpcpu->mntp_writeopcount);
2014 		writeopcount += mpcpu->mntp_writeopcount;
2015 	}
2016 	printf("\n");
2017 
2018 	printf("counter       struct total\n");
2019 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
2020 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
2021 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
2022 
2023 	panic("invalid counts on struct mount");
2024 }
2025 #endif
2026 
2027 int
2028 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
2029 {
2030 	struct mount_pcpu *mpcpu;
2031 	int cpu, sum;
2032 
2033 	switch (which) {
2034 	case MNT_COUNT_REF:
2035 		sum = mp->mnt_ref;
2036 		break;
2037 	case MNT_COUNT_LOCKREF:
2038 		sum = mp->mnt_lockref;
2039 		break;
2040 	case MNT_COUNT_WRITEOPCOUNT:
2041 		sum = mp->mnt_writeopcount;
2042 		break;
2043 	}
2044 
2045 	CPU_FOREACH(cpu) {
2046 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2047 		switch (which) {
2048 		case MNT_COUNT_REF:
2049 			sum += mpcpu->mntp_ref;
2050 			break;
2051 		case MNT_COUNT_LOCKREF:
2052 			sum += mpcpu->mntp_lockref;
2053 			break;
2054 		case MNT_COUNT_WRITEOPCOUNT:
2055 			sum += mpcpu->mntp_writeopcount;
2056 			break;
2057 		}
2058 	}
2059 	return (sum);
2060 }
2061 
2062 static bool
2063 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
2064     int timeout_ticks)
2065 {
2066 	bool enqueued;
2067 
2068 	enqueued = false;
2069 	mtx_lock(&deferred_unmount_lock);
2070 	if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
2071 		mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
2072 		STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
2073 		    mnt_taskqueue_link);
2074 		enqueued = true;
2075 	}
2076 	mtx_unlock(&deferred_unmount_lock);
2077 
2078 	if (enqueued) {
2079 		taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
2080 		    &deferred_unmount_task, timeout_ticks);
2081 	}
2082 
2083 	return (enqueued);
2084 }
2085 
2086 /*
2087  * Taskqueue handler for processing async/recursive unmounts
2088  */
2089 static void
2090 vfs_deferred_unmount(void *argi __unused, int pending __unused)
2091 {
2092 	STAILQ_HEAD(, mount) local_unmounts;
2093 	uint64_t flags;
2094 	struct mount *mp, *tmp;
2095 	int error;
2096 	unsigned int retries;
2097 	bool unmounted;
2098 
2099 	STAILQ_INIT(&local_unmounts);
2100 	mtx_lock(&deferred_unmount_lock);
2101 	STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
2102 	mtx_unlock(&deferred_unmount_lock);
2103 
2104 	STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
2105 		flags = mp->mnt_taskqueue_flags;
2106 		KASSERT((flags & MNT_DEFERRED) != 0,
2107 		    ("taskqueue unmount without MNT_DEFERRED"));
2108 		error = dounmount(mp, flags, curthread);
2109 		if (error != 0) {
2110 			MNT_ILOCK(mp);
2111 			unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
2112 			MNT_IUNLOCK(mp);
2113 
2114 			/*
2115 			 * The deferred unmount thread is the only thread that
2116 			 * modifies the retry counts, so locking/atomics aren't
2117 			 * needed here.
2118 			 */
2119 			retries = (mp->mnt_unmount_retries)++;
2120 			deferred_unmount_total_retries++;
2121 			if (!unmounted && retries < deferred_unmount_retry_limit) {
2122 				deferred_unmount_enqueue(mp, flags, true,
2123 				    -deferred_unmount_retry_delay_hz);
2124 			} else {
2125 				if (retries >= deferred_unmount_retry_limit) {
2126 					printf("giving up on deferred unmount "
2127 					    "of %s after %d retries, error %d\n",
2128 					    mp->mnt_stat.f_mntonname, retries, error);
2129 				}
2130 				vfs_rel(mp);
2131 			}
2132 		}
2133 	}
2134 }
2135 
2136 /*
2137  * Do the actual filesystem unmount.
2138  */
2139 int
2140 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2141 {
2142 	struct mount_upper_node *upper;
2143 	struct vnode *coveredvp, *rootvp;
2144 	int error;
2145 	uint64_t async_flag;
2146 	int mnt_gen_r;
2147 	unsigned int retries;
2148 
2149 	KASSERT((flags & MNT_DEFERRED) == 0 ||
2150 	    (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2151 	    ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2152 
2153 	/*
2154 	 * If the caller has explicitly requested the unmount to be handled by
2155 	 * the taskqueue and we're not already in taskqueue context, queue
2156 	 * up the unmount request and exit.  This is done prior to any
2157 	 * credential checks; MNT_DEFERRED should be used only for kernel-
2158 	 * initiated unmounts and will therefore be processed with the
2159 	 * (kernel) credentials of the taskqueue thread.  Still, callers
2160 	 * should be sure this is the behavior they want.
2161 	 */
2162 	if ((flags & MNT_DEFERRED) != 0 &&
2163 	    taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2164 		if (!deferred_unmount_enqueue(mp, flags, false, 0))
2165 			vfs_rel(mp);
2166 		return (EINPROGRESS);
2167 	}
2168 
2169 	/*
2170 	 * Only privileged root, or (if MNT_USER is set) the user that did the
2171 	 * original mount is permitted to unmount this filesystem.
2172 	 * This check should be made prior to queueing up any recursive
2173 	 * unmounts of upper filesystems.  Those unmounts will be executed
2174 	 * with kernel thread credentials and are expected to succeed, so
2175 	 * we must at least ensure the originating context has sufficient
2176 	 * privilege to unmount the base filesystem before proceeding with
2177 	 * the uppers.
2178 	 */
2179 	error = vfs_suser(mp, td);
2180 	if (error != 0) {
2181 		KASSERT((flags & MNT_DEFERRED) == 0,
2182 		    ("taskqueue unmount with insufficient privilege"));
2183 		vfs_rel(mp);
2184 		return (error);
2185 	}
2186 
2187 	if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2188 		flags |= MNT_RECURSE;
2189 
2190 	if ((flags & MNT_RECURSE) != 0) {
2191 		KASSERT((flags & MNT_FORCE) != 0,
2192 		    ("MNT_RECURSE requires MNT_FORCE"));
2193 
2194 		MNT_ILOCK(mp);
2195 		/*
2196 		 * Set MNTK_RECURSE to prevent new upper mounts from being
2197 		 * added, and note that an operation on the uppers list is in
2198 		 * progress.  This will ensure that unregistration from the
2199 		 * uppers list, and therefore any pending unmount of the upper
2200 		 * FS, can't complete until after we finish walking the list.
2201 		 */
2202 		mp->mnt_kern_flag |= MNTK_RECURSE;
2203 		mp->mnt_upper_pending++;
2204 		TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2205 			retries = upper->mp->mnt_unmount_retries;
2206 			if (retries > deferred_unmount_retry_limit) {
2207 				error = EBUSY;
2208 				continue;
2209 			}
2210 			MNT_IUNLOCK(mp);
2211 
2212 			vfs_ref(upper->mp);
2213 			if (!deferred_unmount_enqueue(upper->mp, flags,
2214 			    false, 0))
2215 				vfs_rel(upper->mp);
2216 			MNT_ILOCK(mp);
2217 		}
2218 		mp->mnt_upper_pending--;
2219 		if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2220 		    mp->mnt_upper_pending == 0) {
2221 			mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2222 			wakeup(&mp->mnt_uppers);
2223 		}
2224 
2225 		/*
2226 		 * If we're not on the taskqueue, wait until the uppers list
2227 		 * is drained before proceeding with unmount.  Otherwise, if
2228 		 * we are on the taskqueue and there are still pending uppers,
2229 		 * just re-enqueue on the end of the taskqueue.
2230 		 */
2231 		if ((flags & MNT_DEFERRED) == 0) {
2232 			while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2233 				mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2234 				error = msleep(&mp->mnt_taskqueue_link,
2235 				    MNT_MTX(mp), PCATCH, "umntqw", 0);
2236 			}
2237 			if (error != 0) {
2238 				MNT_REL(mp);
2239 				MNT_IUNLOCK(mp);
2240 				return (error);
2241 			}
2242 		} else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2243 			MNT_IUNLOCK(mp);
2244 			if (error == 0)
2245 				deferred_unmount_enqueue(mp, flags, true, 0);
2246 			return (error);
2247 		}
2248 		MNT_IUNLOCK(mp);
2249 		KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2250 	}
2251 
2252 	/* Allow the taskqueue to safely re-enqueue on failure */
2253 	if ((flags & MNT_DEFERRED) != 0)
2254 		vfs_ref(mp);
2255 
2256 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2257 		mnt_gen_r = mp->mnt_gen;
2258 		VI_LOCK(coveredvp);
2259 		vholdl(coveredvp);
2260 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2261 		/*
2262 		 * Check for mp being unmounted while waiting for the
2263 		 * covered vnode lock.
2264 		 */
2265 		if (coveredvp->v_mountedhere != mp ||
2266 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2267 			VOP_UNLOCK(coveredvp);
2268 			vdrop(coveredvp);
2269 			vfs_rel(mp);
2270 			return (EBUSY);
2271 		}
2272 	}
2273 
2274 	vfs_op_enter(mp);
2275 
2276 	vn_start_write(NULL, &mp, V_WAIT);
2277 	MNT_ILOCK(mp);
2278 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2279 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
2280 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
2281 		dounmount_cleanup(mp, coveredvp, 0);
2282 		return (EBUSY);
2283 	}
2284 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
2285 	rootvp = vfs_cache_root_clear(mp);
2286 	if (coveredvp != NULL)
2287 		vn_seqc_write_begin(coveredvp);
2288 	if (flags & MNT_NONBUSY) {
2289 		MNT_IUNLOCK(mp);
2290 		error = vfs_check_usecounts(mp);
2291 		MNT_ILOCK(mp);
2292 		if (error != 0) {
2293 			vn_seqc_write_end(coveredvp);
2294 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2295 			if (rootvp != NULL) {
2296 				vn_seqc_write_end(rootvp);
2297 				vrele(rootvp);
2298 			}
2299 			return (error);
2300 		}
2301 	}
2302 	/* Allow filesystems to detect that a forced unmount is in progress. */
2303 	if (flags & MNT_FORCE) {
2304 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2305 		MNT_IUNLOCK(mp);
2306 		/*
2307 		 * Must be done after setting MNTK_UNMOUNTF and before
2308 		 * waiting for mnt_lockref to become 0.
2309 		 */
2310 		VFS_PURGE(mp);
2311 		MNT_ILOCK(mp);
2312 	}
2313 	error = 0;
2314 	if (mp->mnt_lockref) {
2315 		mp->mnt_kern_flag |= MNTK_DRAINING;
2316 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2317 		    "mount drain", 0);
2318 	}
2319 	MNT_IUNLOCK(mp);
2320 	KASSERT(mp->mnt_lockref == 0,
2321 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
2322 	    __func__, __FILE__, __LINE__));
2323 	KASSERT(error == 0,
2324 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
2325 	    __func__, __FILE__, __LINE__));
2326 
2327 	/*
2328 	 * We want to keep the vnode around so that we can vn_seqc_write_end
2329 	 * after we are done with unmount. Downgrade our reference to a mere
2330 	 * hold count so that we don't interefere with anything.
2331 	 */
2332 	if (rootvp != NULL) {
2333 		vhold(rootvp);
2334 		vrele(rootvp);
2335 	}
2336 
2337 	if (mp->mnt_flag & MNT_EXPUBLIC)
2338 		vfs_setpublicfs(NULL, NULL, NULL);
2339 
2340 	vfs_periodic(mp, MNT_WAIT);
2341 	MNT_ILOCK(mp);
2342 	async_flag = mp->mnt_flag & MNT_ASYNC;
2343 	mp->mnt_flag &= ~MNT_ASYNC;
2344 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
2345 	MNT_IUNLOCK(mp);
2346 	vfs_deallocate_syncvnode(mp);
2347 	error = VFS_UNMOUNT(mp, flags);
2348 	vn_finished_write(mp);
2349 	vfs_rel(mp);
2350 	/*
2351 	 * If we failed to flush the dirty blocks for this mount point,
2352 	 * undo all the cdir/rdir and rootvnode changes we made above.
2353 	 * Unless we failed to do so because the device is reporting that
2354 	 * it doesn't exist anymore.
2355 	 */
2356 	if (error && error != ENXIO) {
2357 		MNT_ILOCK(mp);
2358 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2359 			MNT_IUNLOCK(mp);
2360 			vfs_allocate_syncvnode(mp);
2361 			MNT_ILOCK(mp);
2362 		}
2363 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2364 		mp->mnt_flag |= async_flag;
2365 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2366 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2367 			mp->mnt_kern_flag |= MNTK_ASYNC;
2368 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
2369 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
2370 			wakeup(mp);
2371 		}
2372 		vfs_op_exit_locked(mp);
2373 		MNT_IUNLOCK(mp);
2374 		if (coveredvp) {
2375 			vn_seqc_write_end(coveredvp);
2376 			VOP_UNLOCK(coveredvp);
2377 			vdrop(coveredvp);
2378 		}
2379 		if (rootvp != NULL) {
2380 			vn_seqc_write_end(rootvp);
2381 			vdrop(rootvp);
2382 		}
2383 		return (error);
2384 	}
2385 
2386 	mtx_lock(&mountlist_mtx);
2387 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
2388 	mtx_unlock(&mountlist_mtx);
2389 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2390 	if (coveredvp != NULL) {
2391 		VI_LOCK(coveredvp);
2392 		vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2393 		coveredvp->v_mountedhere = NULL;
2394 		vn_seqc_write_end_locked(coveredvp);
2395 		VI_UNLOCK(coveredvp);
2396 		VOP_UNLOCK(coveredvp);
2397 		vdrop(coveredvp);
2398 	}
2399 	mount_devctl_event("UNMOUNT", mp, false);
2400 	if (rootvp != NULL) {
2401 		vn_seqc_write_end(rootvp);
2402 		vdrop(rootvp);
2403 	}
2404 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2405 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
2406 		vrele(rootvnode);
2407 		rootvnode = NULL;
2408 	}
2409 	if (mp == rootdevmp)
2410 		rootdevmp = NULL;
2411 	if ((flags & MNT_DEFERRED) != 0)
2412 		vfs_rel(mp);
2413 	vfs_mount_destroy(mp);
2414 	return (0);
2415 }
2416 
2417 /*
2418  * Report errors during filesystem mounting.
2419  */
2420 void
2421 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2422 {
2423 	struct vfsoptlist *moptlist = mp->mnt_optnew;
2424 	va_list ap;
2425 	int error, len;
2426 	char *errmsg;
2427 
2428 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2429 	if (error || errmsg == NULL || len <= 0)
2430 		return;
2431 
2432 	va_start(ap, fmt);
2433 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2434 	va_end(ap);
2435 }
2436 
2437 void
2438 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2439 {
2440 	va_list ap;
2441 	int error, len;
2442 	char *errmsg;
2443 
2444 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2445 	if (error || errmsg == NULL || len <= 0)
2446 		return;
2447 
2448 	va_start(ap, fmt);
2449 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2450 	va_end(ap);
2451 }
2452 
2453 /*
2454  * ---------------------------------------------------------------------
2455  * Functions for querying mount options/arguments from filesystems.
2456  */
2457 
2458 /*
2459  * Check that no unknown options are given
2460  */
2461 int
2462 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2463 {
2464 	struct vfsopt *opt;
2465 	char errmsg[255];
2466 	const char **t, *p, *q;
2467 	int ret = 0;
2468 
2469 	TAILQ_FOREACH(opt, opts, link) {
2470 		p = opt->name;
2471 		q = NULL;
2472 		if (p[0] == 'n' && p[1] == 'o')
2473 			q = p + 2;
2474 		for(t = global_opts; *t != NULL; t++) {
2475 			if (strcmp(*t, p) == 0)
2476 				break;
2477 			if (q != NULL) {
2478 				if (strcmp(*t, q) == 0)
2479 					break;
2480 			}
2481 		}
2482 		if (*t != NULL)
2483 			continue;
2484 		for(t = legal; *t != NULL; t++) {
2485 			if (strcmp(*t, p) == 0)
2486 				break;
2487 			if (q != NULL) {
2488 				if (strcmp(*t, q) == 0)
2489 					break;
2490 			}
2491 		}
2492 		if (*t != NULL)
2493 			continue;
2494 		snprintf(errmsg, sizeof(errmsg),
2495 		    "mount option <%s> is unknown", p);
2496 		ret = EINVAL;
2497 	}
2498 	if (ret != 0) {
2499 		TAILQ_FOREACH(opt, opts, link) {
2500 			if (strcmp(opt->name, "errmsg") == 0) {
2501 				strncpy((char *)opt->value, errmsg, opt->len);
2502 				break;
2503 			}
2504 		}
2505 		if (opt == NULL)
2506 			printf("%s\n", errmsg);
2507 	}
2508 	return (ret);
2509 }
2510 
2511 /*
2512  * Get a mount option by its name.
2513  *
2514  * Return 0 if the option was found, ENOENT otherwise.
2515  * If len is non-NULL it will be filled with the length
2516  * of the option. If buf is non-NULL, it will be filled
2517  * with the address of the option.
2518  */
2519 int
2520 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2521 {
2522 	struct vfsopt *opt;
2523 
2524 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2525 
2526 	TAILQ_FOREACH(opt, opts, link) {
2527 		if (strcmp(name, opt->name) == 0) {
2528 			opt->seen = 1;
2529 			if (len != NULL)
2530 				*len = opt->len;
2531 			if (buf != NULL)
2532 				*buf = opt->value;
2533 			return (0);
2534 		}
2535 	}
2536 	return (ENOENT);
2537 }
2538 
2539 int
2540 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2541 {
2542 	struct vfsopt *opt;
2543 
2544 	if (opts == NULL)
2545 		return (-1);
2546 
2547 	TAILQ_FOREACH(opt, opts, link) {
2548 		if (strcmp(name, opt->name) == 0) {
2549 			opt->seen = 1;
2550 			return (opt->pos);
2551 		}
2552 	}
2553 	return (-1);
2554 }
2555 
2556 int
2557 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2558 {
2559 	char *opt_value, *vtp;
2560 	quad_t iv;
2561 	int error, opt_len;
2562 
2563 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2564 	if (error != 0)
2565 		return (error);
2566 	if (opt_len == 0 || opt_value == NULL)
2567 		return (EINVAL);
2568 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2569 		return (EINVAL);
2570 	iv = strtoq(opt_value, &vtp, 0);
2571 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2572 		return (EINVAL);
2573 	if (iv < 0)
2574 		return (EINVAL);
2575 	switch (vtp[0]) {
2576 	case 't': case 'T':
2577 		iv *= 1024;
2578 		/* FALLTHROUGH */
2579 	case 'g': case 'G':
2580 		iv *= 1024;
2581 		/* FALLTHROUGH */
2582 	case 'm': case 'M':
2583 		iv *= 1024;
2584 		/* FALLTHROUGH */
2585 	case 'k': case 'K':
2586 		iv *= 1024;
2587 	case '\0':
2588 		break;
2589 	default:
2590 		return (EINVAL);
2591 	}
2592 	*value = iv;
2593 
2594 	return (0);
2595 }
2596 
2597 char *
2598 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2599 {
2600 	struct vfsopt *opt;
2601 
2602 	*error = 0;
2603 	TAILQ_FOREACH(opt, opts, link) {
2604 		if (strcmp(name, opt->name) != 0)
2605 			continue;
2606 		opt->seen = 1;
2607 		if (opt->len == 0 ||
2608 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2609 			*error = EINVAL;
2610 			return (NULL);
2611 		}
2612 		return (opt->value);
2613 	}
2614 	*error = ENOENT;
2615 	return (NULL);
2616 }
2617 
2618 int
2619 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2620 	uint64_t val)
2621 {
2622 	struct vfsopt *opt;
2623 
2624 	TAILQ_FOREACH(opt, opts, link) {
2625 		if (strcmp(name, opt->name) == 0) {
2626 			opt->seen = 1;
2627 			if (w != NULL)
2628 				*w |= val;
2629 			return (1);
2630 		}
2631 	}
2632 	if (w != NULL)
2633 		*w &= ~val;
2634 	return (0);
2635 }
2636 
2637 int
2638 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2639 {
2640 	va_list ap;
2641 	struct vfsopt *opt;
2642 	int ret;
2643 
2644 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2645 
2646 	TAILQ_FOREACH(opt, opts, link) {
2647 		if (strcmp(name, opt->name) != 0)
2648 			continue;
2649 		opt->seen = 1;
2650 		if (opt->len == 0 || opt->value == NULL)
2651 			return (0);
2652 		if (((char *)opt->value)[opt->len - 1] != '\0')
2653 			return (0);
2654 		va_start(ap, fmt);
2655 		ret = vsscanf(opt->value, fmt, ap);
2656 		va_end(ap);
2657 		return (ret);
2658 	}
2659 	return (0);
2660 }
2661 
2662 int
2663 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2664 {
2665 	struct vfsopt *opt;
2666 
2667 	TAILQ_FOREACH(opt, opts, link) {
2668 		if (strcmp(name, opt->name) != 0)
2669 			continue;
2670 		opt->seen = 1;
2671 		if (opt->value == NULL)
2672 			opt->len = len;
2673 		else {
2674 			if (opt->len != len)
2675 				return (EINVAL);
2676 			bcopy(value, opt->value, len);
2677 		}
2678 		return (0);
2679 	}
2680 	return (ENOENT);
2681 }
2682 
2683 int
2684 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2685 {
2686 	struct vfsopt *opt;
2687 
2688 	TAILQ_FOREACH(opt, opts, link) {
2689 		if (strcmp(name, opt->name) != 0)
2690 			continue;
2691 		opt->seen = 1;
2692 		if (opt->value == NULL)
2693 			opt->len = len;
2694 		else {
2695 			if (opt->len < len)
2696 				return (EINVAL);
2697 			opt->len = len;
2698 			bcopy(value, opt->value, len);
2699 		}
2700 		return (0);
2701 	}
2702 	return (ENOENT);
2703 }
2704 
2705 int
2706 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2707 {
2708 	struct vfsopt *opt;
2709 
2710 	TAILQ_FOREACH(opt, opts, link) {
2711 		if (strcmp(name, opt->name) != 0)
2712 			continue;
2713 		opt->seen = 1;
2714 		if (opt->value == NULL)
2715 			opt->len = strlen(value) + 1;
2716 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2717 			return (EINVAL);
2718 		return (0);
2719 	}
2720 	return (ENOENT);
2721 }
2722 
2723 /*
2724  * Find and copy a mount option.
2725  *
2726  * The size of the buffer has to be specified
2727  * in len, if it is not the same length as the
2728  * mount option, EINVAL is returned.
2729  * Returns ENOENT if the option is not found.
2730  */
2731 int
2732 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2733 {
2734 	struct vfsopt *opt;
2735 
2736 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2737 
2738 	TAILQ_FOREACH(opt, opts, link) {
2739 		if (strcmp(name, opt->name) == 0) {
2740 			opt->seen = 1;
2741 			if (len != opt->len)
2742 				return (EINVAL);
2743 			bcopy(opt->value, dest, opt->len);
2744 			return (0);
2745 		}
2746 	}
2747 	return (ENOENT);
2748 }
2749 
2750 int
2751 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2752 {
2753 	/*
2754 	 * Filesystems only fill in part of the structure for updates, we
2755 	 * have to read the entirety first to get all content.
2756 	 */
2757 	if (sbp != &mp->mnt_stat)
2758 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2759 
2760 	/*
2761 	 * Set these in case the underlying filesystem fails to do so.
2762 	 */
2763 	sbp->f_version = STATFS_VERSION;
2764 	sbp->f_namemax = NAME_MAX;
2765 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2766 	sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2767 
2768 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2769 }
2770 
2771 void
2772 vfs_mountedfrom(struct mount *mp, const char *from)
2773 {
2774 
2775 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2776 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2777 	    sizeof mp->mnt_stat.f_mntfromname);
2778 }
2779 
2780 /*
2781  * ---------------------------------------------------------------------
2782  * This is the api for building mount args and mounting filesystems from
2783  * inside the kernel.
2784  *
2785  * The API works by accumulation of individual args.  First error is
2786  * latched.
2787  *
2788  * XXX: should be documented in new manpage kernel_mount(9)
2789  */
2790 
2791 /* A memory allocation which must be freed when we are done */
2792 struct mntaarg {
2793 	SLIST_ENTRY(mntaarg)	next;
2794 };
2795 
2796 /* The header for the mount arguments */
2797 struct mntarg {
2798 	struct iovec *v;
2799 	int len;
2800 	int error;
2801 	SLIST_HEAD(, mntaarg)	list;
2802 };
2803 
2804 /*
2805  * Add a boolean argument.
2806  *
2807  * flag is the boolean value.
2808  * name must start with "no".
2809  */
2810 struct mntarg *
2811 mount_argb(struct mntarg *ma, int flag, const char *name)
2812 {
2813 
2814 	KASSERT(name[0] == 'n' && name[1] == 'o',
2815 	    ("mount_argb(...,%s): name must start with 'no'", name));
2816 
2817 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2818 }
2819 
2820 /*
2821  * Add an argument printf style
2822  */
2823 struct mntarg *
2824 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2825 {
2826 	va_list ap;
2827 	struct mntaarg *maa;
2828 	struct sbuf *sb;
2829 	int len;
2830 
2831 	if (ma == NULL) {
2832 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2833 		SLIST_INIT(&ma->list);
2834 	}
2835 	if (ma->error)
2836 		return (ma);
2837 
2838 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2839 	    M_MOUNT, M_WAITOK);
2840 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2841 	ma->v[ma->len].iov_len = strlen(name) + 1;
2842 	ma->len++;
2843 
2844 	sb = sbuf_new_auto();
2845 	va_start(ap, fmt);
2846 	sbuf_vprintf(sb, fmt, ap);
2847 	va_end(ap);
2848 	sbuf_finish(sb);
2849 	len = sbuf_len(sb) + 1;
2850 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2851 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2852 	bcopy(sbuf_data(sb), maa + 1, len);
2853 	sbuf_delete(sb);
2854 
2855 	ma->v[ma->len].iov_base = maa + 1;
2856 	ma->v[ma->len].iov_len = len;
2857 	ma->len++;
2858 
2859 	return (ma);
2860 }
2861 
2862 /*
2863  * Add an argument which is a userland string.
2864  */
2865 struct mntarg *
2866 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2867 {
2868 	struct mntaarg *maa;
2869 	char *tbuf;
2870 
2871 	if (val == NULL)
2872 		return (ma);
2873 	if (ma == NULL) {
2874 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2875 		SLIST_INIT(&ma->list);
2876 	}
2877 	if (ma->error)
2878 		return (ma);
2879 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2880 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2881 	tbuf = (void *)(maa + 1);
2882 	ma->error = copyinstr(val, tbuf, len, NULL);
2883 	return (mount_arg(ma, name, tbuf, -1));
2884 }
2885 
2886 /*
2887  * Plain argument.
2888  *
2889  * If length is -1, treat value as a C string.
2890  */
2891 struct mntarg *
2892 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2893 {
2894 
2895 	if (ma == NULL) {
2896 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2897 		SLIST_INIT(&ma->list);
2898 	}
2899 	if (ma->error)
2900 		return (ma);
2901 
2902 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2903 	    M_MOUNT, M_WAITOK);
2904 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2905 	ma->v[ma->len].iov_len = strlen(name) + 1;
2906 	ma->len++;
2907 
2908 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2909 	if (len < 0)
2910 		ma->v[ma->len].iov_len = strlen(val) + 1;
2911 	else
2912 		ma->v[ma->len].iov_len = len;
2913 	ma->len++;
2914 	return (ma);
2915 }
2916 
2917 /*
2918  * Free a mntarg structure
2919  */
2920 static void
2921 free_mntarg(struct mntarg *ma)
2922 {
2923 	struct mntaarg *maa;
2924 
2925 	while (!SLIST_EMPTY(&ma->list)) {
2926 		maa = SLIST_FIRST(&ma->list);
2927 		SLIST_REMOVE_HEAD(&ma->list, next);
2928 		free(maa, M_MOUNT);
2929 	}
2930 	free(ma->v, M_MOUNT);
2931 	free(ma, M_MOUNT);
2932 }
2933 
2934 /*
2935  * Mount a filesystem
2936  */
2937 int
2938 kernel_mount(struct mntarg *ma, uint64_t flags)
2939 {
2940 	struct uio auio;
2941 	int error;
2942 
2943 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2944 	KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2945 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2946 
2947 	error = ma->error;
2948 	if (error == 0) {
2949 		auio.uio_iov = ma->v;
2950 		auio.uio_iovcnt = ma->len;
2951 		auio.uio_segflg = UIO_SYSSPACE;
2952 		error = vfs_donmount(curthread, flags, &auio);
2953 	}
2954 	free_mntarg(ma);
2955 	return (error);
2956 }
2957 
2958 /* Map from mount options to printable formats. */
2959 static struct mntoptnames optnames[] = {
2960 	MNTOPT_NAMES
2961 };
2962 
2963 #define DEVCTL_LEN 1024
2964 static void
2965 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2966 {
2967 	const uint8_t *cp;
2968 	struct mntoptnames *fp;
2969 	struct sbuf sb;
2970 	struct statfs *sfp = &mp->mnt_stat;
2971 	char *buf;
2972 
2973 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2974 	if (buf == NULL)
2975 		return;
2976 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2977 	sbuf_cpy(&sb, "mount-point=\"");
2978 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2979 	sbuf_cat(&sb, "\" mount-dev=\"");
2980 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2981 	sbuf_cat(&sb, "\" mount-type=\"");
2982 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2983 	sbuf_cat(&sb, "\" fsid=0x");
2984 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2985 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2986 		sbuf_printf(&sb, "%02x", cp[i]);
2987 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2988 	for (fp = optnames; fp->o_opt != 0; fp++) {
2989 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2990 			sbuf_cat(&sb, fp->o_name);
2991 			sbuf_putc(&sb, ';');
2992 		}
2993 	}
2994 	sbuf_putc(&sb, '"');
2995 	sbuf_finish(&sb);
2996 
2997 	/*
2998 	 * Options are not published because the form of the options depends on
2999 	 * the file system and may include binary data. In addition, they don't
3000 	 * necessarily provide enough useful information to be actionable when
3001 	 * devd processes them.
3002 	 */
3003 
3004 	if (sbuf_error(&sb) == 0)
3005 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
3006 	sbuf_delete(&sb);
3007 	free(buf, M_MOUNT);
3008 }
3009 
3010 /*
3011  * Force remount specified mount point to read-only.  The argument
3012  * must be busied to avoid parallel unmount attempts.
3013  *
3014  * Intended use is to prevent further writes if some metadata
3015  * inconsistency is detected.  Note that the function still flushes
3016  * all cached metadata and data for the mount point, which might be
3017  * not always suitable.
3018  */
3019 int
3020 vfs_remount_ro(struct mount *mp)
3021 {
3022 	struct vfsoptlist *opts;
3023 	struct vfsopt *opt;
3024 	struct vnode *vp_covered, *rootvp;
3025 	int error;
3026 
3027 	vfs_op_enter(mp);
3028 	KASSERT(mp->mnt_lockref > 0,
3029 	    ("vfs_remount_ro: mp %p is not busied", mp));
3030 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
3031 	    ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
3032 
3033 	rootvp = NULL;
3034 	vp_covered = mp->mnt_vnodecovered;
3035 	error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
3036 	if (error != 0) {
3037 		vfs_op_exit(mp);
3038 		return (error);
3039 	}
3040 	VI_LOCK(vp_covered);
3041 	if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
3042 		VI_UNLOCK(vp_covered);
3043 		vput(vp_covered);
3044 		vfs_op_exit(mp);
3045 		return (EBUSY);
3046 	}
3047 	vp_covered->v_iflag |= VI_MOUNT;
3048 	VI_UNLOCK(vp_covered);
3049 	vn_seqc_write_begin(vp_covered);
3050 
3051 	MNT_ILOCK(mp);
3052 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
3053 		MNT_IUNLOCK(mp);
3054 		error = EBUSY;
3055 		goto out;
3056 	}
3057 	mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
3058 	rootvp = vfs_cache_root_clear(mp);
3059 	MNT_IUNLOCK(mp);
3060 
3061 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
3062 	TAILQ_INIT(opts);
3063 	opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
3064 	opt->name = strdup("ro", M_MOUNT);
3065 	opt->value = NULL;
3066 	TAILQ_INSERT_TAIL(opts, opt, link);
3067 	vfs_mergeopts(opts, mp->mnt_opt);
3068 	mp->mnt_optnew = opts;
3069 
3070 	error = VFS_MOUNT(mp);
3071 
3072 	if (error == 0) {
3073 		MNT_ILOCK(mp);
3074 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
3075 		MNT_IUNLOCK(mp);
3076 		vfs_deallocate_syncvnode(mp);
3077 		if (mp->mnt_opt != NULL)
3078 			vfs_freeopts(mp->mnt_opt);
3079 		mp->mnt_opt = mp->mnt_optnew;
3080 	} else {
3081 		MNT_ILOCK(mp);
3082 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
3083 		MNT_IUNLOCK(mp);
3084 		vfs_freeopts(mp->mnt_optnew);
3085 	}
3086 	mp->mnt_optnew = NULL;
3087 
3088 out:
3089 	vfs_op_exit(mp);
3090 	VI_LOCK(vp_covered);
3091 	vp_covered->v_iflag &= ~VI_MOUNT;
3092 	VI_UNLOCK(vp_covered);
3093 	vput(vp_covered);
3094 	vn_seqc_write_end(vp_covered);
3095 	if (rootvp != NULL) {
3096 		vn_seqc_write_end(rootvp);
3097 		vrele(rootvp);
3098 	}
3099 	return (error);
3100 }
3101 
3102 /*
3103  * Suspend write operations on all local writeable filesystems.  Does
3104  * full sync of them in the process.
3105  *
3106  * Iterate over the mount points in reverse order, suspending most
3107  * recently mounted filesystems first.  It handles a case where a
3108  * filesystem mounted from a md(4) vnode-backed device should be
3109  * suspended before the filesystem that owns the vnode.
3110  */
3111 void
3112 suspend_all_fs(void)
3113 {
3114 	struct mount *mp;
3115 	int error;
3116 
3117 	mtx_lock(&mountlist_mtx);
3118 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
3119 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
3120 		if (error != 0)
3121 			continue;
3122 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
3123 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
3124 			mtx_lock(&mountlist_mtx);
3125 			vfs_unbusy(mp);
3126 			continue;
3127 		}
3128 		error = vfs_write_suspend(mp, 0);
3129 		if (error == 0) {
3130 			MNT_ILOCK(mp);
3131 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
3132 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
3133 			MNT_IUNLOCK(mp);
3134 			mtx_lock(&mountlist_mtx);
3135 		} else {
3136 			printf("suspend of %s failed, error %d\n",
3137 			    mp->mnt_stat.f_mntonname, error);
3138 			mtx_lock(&mountlist_mtx);
3139 			vfs_unbusy(mp);
3140 		}
3141 	}
3142 	mtx_unlock(&mountlist_mtx);
3143 }
3144 
3145 /*
3146  * Clone the mnt_exjail field to a new mount point.
3147  */
3148 void
3149 vfs_exjail_clone(struct mount *inmp, struct mount *outmp)
3150 {
3151 	struct ucred *cr;
3152 	struct prison *pr;
3153 
3154 	MNT_ILOCK(inmp);
3155 	cr = inmp->mnt_exjail;
3156 	if (cr != NULL) {
3157 		crhold(cr);
3158 		MNT_IUNLOCK(inmp);
3159 		pr = cr->cr_prison;
3160 		sx_slock(&allprison_lock);
3161 		if (!prison_isalive(pr)) {
3162 			sx_sunlock(&allprison_lock);
3163 			crfree(cr);
3164 			return;
3165 		}
3166 		MNT_ILOCK(outmp);
3167 		if (outmp->mnt_exjail == NULL) {
3168 			outmp->mnt_exjail = cr;
3169 			atomic_add_int(&pr->pr_exportcnt, 1);
3170 			cr = NULL;
3171 		}
3172 		MNT_IUNLOCK(outmp);
3173 		sx_sunlock(&allprison_lock);
3174 		if (cr != NULL)
3175 			crfree(cr);
3176 	} else
3177 		MNT_IUNLOCK(inmp);
3178 }
3179 
3180 void
3181 resume_all_fs(void)
3182 {
3183 	struct mount *mp;
3184 
3185 	mtx_lock(&mountlist_mtx);
3186 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3187 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3188 			continue;
3189 		mtx_unlock(&mountlist_mtx);
3190 		MNT_ILOCK(mp);
3191 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3192 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3193 		MNT_IUNLOCK(mp);
3194 		vfs_write_resume(mp, 0);
3195 		mtx_lock(&mountlist_mtx);
3196 		vfs_unbusy(mp);
3197 	}
3198 	mtx_unlock(&mountlist_mtx);
3199 }
3200