xref: /freebsd/sys/kern/vfs_mount.c (revision a2f733abcff64628b7771a47089628b7327a88bd)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1999-2004 Poul-Henning Kamp
5  * Copyright (c) 1999 Michael Smith
6  * Copyright (c) 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 #include <sys/param.h>
41 #include <sys/conf.h>
42 #include <sys/smp.h>
43 #include <sys/devctl.h>
44 #include <sys/eventhandler.h>
45 #include <sys/fcntl.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/libkern.h>
50 #include <sys/limits.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/filedesc.h>
58 #include <sys/reboot.h>
59 #include <sys/sbuf.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/systm.h>
65 #include <sys/taskqueue.h>
66 #include <sys/vnode.h>
67 #include <vm/uma.h>
68 
69 #include <geom/geom.h>
70 
71 #include <machine/stdarg.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
77 
78 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
79 		    uint64_t fsflags, bool jail_export,
80 		    struct vfsoptlist **optlist);
81 static void	free_mntarg(struct mntarg *ma);
82 
83 static int	usermount = 0;
84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
85     "Unprivileged users may mount and unmount file systems");
86 
87 static bool	default_autoro = false;
88 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
89     "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
90 
91 static bool	recursive_forced_unmount = false;
92 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW,
93     &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts"
94     " when a file system is forcibly unmounted");
95 
96 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount,
97     CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls");
98 
99 static unsigned int	deferred_unmount_retry_limit = 10;
100 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW,
101     &deferred_unmount_retry_limit, 0,
102     "Maximum number of retries for deferred unmount failure");
103 
104 static int	deferred_unmount_retry_delay_hz;
105 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW,
106     &deferred_unmount_retry_delay_hz, 0,
107     "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount");
108 
109 static int	deferred_unmount_total_retries = 0;
110 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD,
111     &deferred_unmount_total_retries, 0,
112     "Total number of retried deferred unmounts");
113 
114 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
115 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
116 static uma_zone_t mount_zone;
117 
118 /* List of mounted filesystems. */
119 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
120 
121 /* For any iteration/modification of mountlist */
122 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
123 
124 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
125 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
126 
127 static void vfs_deferred_unmount(void *arg, int pending);
128 static struct timeout_task deferred_unmount_task;
129 static struct mtx deferred_unmount_lock;
130 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount",
131     MTX_DEF);
132 static STAILQ_HEAD(, mount) deferred_unmount_list =
133     STAILQ_HEAD_INITIALIZER(deferred_unmount_list);
134 TASKQUEUE_DEFINE_THREAD(deferred_unmount);
135 
136 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
137 
138 /*
139  * Global opts, taken by all filesystems
140  */
141 static const char *global_opts[] = {
142 	"errmsg",
143 	"fstype",
144 	"fspath",
145 	"ro",
146 	"rw",
147 	"nosuid",
148 	"noexec",
149 	NULL
150 };
151 
152 static int
153 mount_init(void *mem, int size, int flags)
154 {
155 	struct mount *mp;
156 
157 	mp = (struct mount *)mem;
158 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
159 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
160 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
161 	mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
162 	mp->mnt_ref = 0;
163 	mp->mnt_vfs_ops = 1;
164 	mp->mnt_rootvnode = NULL;
165 	return (0);
166 }
167 
168 static void
169 mount_fini(void *mem, int size)
170 {
171 	struct mount *mp;
172 
173 	mp = (struct mount *)mem;
174 	uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
175 	lockdestroy(&mp->mnt_explock);
176 	mtx_destroy(&mp->mnt_listmtx);
177 	mtx_destroy(&mp->mnt_mtx);
178 }
179 
180 static void
181 vfs_mount_init(void *dummy __unused)
182 {
183 	TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task,
184 	    0, vfs_deferred_unmount, NULL);
185 	deferred_unmount_retry_delay_hz = hz;
186 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
187 	    NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
188 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
189 }
190 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
191 
192 /*
193  * ---------------------------------------------------------------------
194  * Functions for building and sanitizing the mount options
195  */
196 
197 /* Remove one mount option. */
198 static void
199 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
200 {
201 
202 	TAILQ_REMOVE(opts, opt, link);
203 	free(opt->name, M_MOUNT);
204 	if (opt->value != NULL)
205 		free(opt->value, M_MOUNT);
206 	free(opt, M_MOUNT);
207 }
208 
209 /* Release all resources related to the mount options. */
210 void
211 vfs_freeopts(struct vfsoptlist *opts)
212 {
213 	struct vfsopt *opt;
214 
215 	while (!TAILQ_EMPTY(opts)) {
216 		opt = TAILQ_FIRST(opts);
217 		vfs_freeopt(opts, opt);
218 	}
219 	free(opts, M_MOUNT);
220 }
221 
222 void
223 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
224 {
225 	struct vfsopt *opt, *temp;
226 
227 	if (opts == NULL)
228 		return;
229 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
230 		if (strcmp(opt->name, name) == 0)
231 			vfs_freeopt(opts, opt);
232 	}
233 }
234 
235 static int
236 vfs_isopt_ro(const char *opt)
237 {
238 
239 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
240 	    strcmp(opt, "norw") == 0)
241 		return (1);
242 	return (0);
243 }
244 
245 static int
246 vfs_isopt_rw(const char *opt)
247 {
248 
249 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
250 		return (1);
251 	return (0);
252 }
253 
254 /*
255  * Check if options are equal (with or without the "no" prefix).
256  */
257 static int
258 vfs_equalopts(const char *opt1, const char *opt2)
259 {
260 	char *p;
261 
262 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
263 	if (strcmp(opt1, opt2) == 0)
264 		return (1);
265 	/* "noopt" vs. "opt" */
266 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
267 		return (1);
268 	/* "opt" vs. "noopt" */
269 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
270 		return (1);
271 	while ((p = strchr(opt1, '.')) != NULL &&
272 	    !strncmp(opt1, opt2, ++p - opt1)) {
273 		opt2 += p - opt1;
274 		opt1 = p;
275 		/* "foo.noopt" vs. "foo.opt" */
276 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
277 			return (1);
278 		/* "foo.opt" vs. "foo.noopt" */
279 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
280 			return (1);
281 	}
282 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
283 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
284 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
285 		return (1);
286 	return (0);
287 }
288 
289 /*
290  * If a mount option is specified several times,
291  * (with or without the "no" prefix) only keep
292  * the last occurrence of it.
293  */
294 static void
295 vfs_sanitizeopts(struct vfsoptlist *opts)
296 {
297 	struct vfsopt *opt, *opt2, *tmp;
298 
299 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
300 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
301 		while (opt2 != NULL) {
302 			if (vfs_equalopts(opt->name, opt2->name)) {
303 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
304 				vfs_freeopt(opts, opt2);
305 				opt2 = tmp;
306 			} else {
307 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
308 			}
309 		}
310 	}
311 }
312 
313 /*
314  * Build a linked list of mount options from a struct uio.
315  */
316 int
317 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
318 {
319 	struct vfsoptlist *opts;
320 	struct vfsopt *opt;
321 	size_t memused, namelen, optlen;
322 	unsigned int i, iovcnt;
323 	int error;
324 
325 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
326 	TAILQ_INIT(opts);
327 	memused = 0;
328 	iovcnt = auio->uio_iovcnt;
329 	for (i = 0; i < iovcnt; i += 2) {
330 		namelen = auio->uio_iov[i].iov_len;
331 		optlen = auio->uio_iov[i + 1].iov_len;
332 		memused += sizeof(struct vfsopt) + optlen + namelen;
333 		/*
334 		 * Avoid consuming too much memory, and attempts to overflow
335 		 * memused.
336 		 */
337 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
338 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
339 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
340 			error = EINVAL;
341 			goto bad;
342 		}
343 
344 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
345 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
346 		opt->value = NULL;
347 		opt->len = 0;
348 		opt->pos = i / 2;
349 		opt->seen = 0;
350 
351 		/*
352 		 * Do this early, so jumps to "bad" will free the current
353 		 * option.
354 		 */
355 		TAILQ_INSERT_TAIL(opts, opt, link);
356 
357 		if (auio->uio_segflg == UIO_SYSSPACE) {
358 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
359 		} else {
360 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
361 			    namelen);
362 			if (error)
363 				goto bad;
364 		}
365 		/* Ensure names are null-terminated strings. */
366 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
367 			error = EINVAL;
368 			goto bad;
369 		}
370 		if (optlen != 0) {
371 			opt->len = optlen;
372 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
373 			if (auio->uio_segflg == UIO_SYSSPACE) {
374 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
375 				    optlen);
376 			} else {
377 				error = copyin(auio->uio_iov[i + 1].iov_base,
378 				    opt->value, optlen);
379 				if (error)
380 					goto bad;
381 			}
382 		}
383 	}
384 	vfs_sanitizeopts(opts);
385 	*options = opts;
386 	return (0);
387 bad:
388 	vfs_freeopts(opts);
389 	return (error);
390 }
391 
392 /*
393  * Merge the old mount options with the new ones passed
394  * in the MNT_UPDATE case.
395  *
396  * XXX: This function will keep a "nofoo" option in the new
397  * options.  E.g, if the option's canonical name is "foo",
398  * "nofoo" ends up in the mount point's active options.
399  */
400 static void
401 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
402 {
403 	struct vfsopt *opt, *new;
404 
405 	TAILQ_FOREACH(opt, oldopts, link) {
406 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
407 		new->name = strdup(opt->name, M_MOUNT);
408 		if (opt->len != 0) {
409 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
410 			bcopy(opt->value, new->value, opt->len);
411 		} else
412 			new->value = NULL;
413 		new->len = opt->len;
414 		new->seen = opt->seen;
415 		TAILQ_INSERT_HEAD(toopts, new, link);
416 	}
417 	vfs_sanitizeopts(toopts);
418 }
419 
420 /*
421  * Mount a filesystem.
422  */
423 #ifndef _SYS_SYSPROTO_H_
424 struct nmount_args {
425 	struct iovec *iovp;
426 	unsigned int iovcnt;
427 	int flags;
428 };
429 #endif
430 int
431 sys_nmount(struct thread *td, struct nmount_args *uap)
432 {
433 	struct uio *auio;
434 	int error;
435 	u_int iovcnt;
436 	uint64_t flags;
437 
438 	/*
439 	 * Mount flags are now 64-bits. On 32-bit archtectures only
440 	 * 32-bits are passed in, but from here on everything handles
441 	 * 64-bit flags correctly.
442 	 */
443 	flags = uap->flags;
444 
445 	AUDIT_ARG_FFLAGS(flags);
446 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
447 	    uap->iovp, uap->iovcnt, flags);
448 
449 	/*
450 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
451 	 * userspace to set this flag, but we must filter it out if we want
452 	 * MNT_UPDATE on the root file system to work.
453 	 * MNT_ROOTFS should only be set by the kernel when mounting its
454 	 * root file system.
455 	 */
456 	flags &= ~MNT_ROOTFS;
457 
458 	iovcnt = uap->iovcnt;
459 	/*
460 	 * Check that we have an even number of iovec's
461 	 * and that we have at least two options.
462 	 */
463 	if ((iovcnt & 1) || (iovcnt < 4)) {
464 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
465 		    uap->iovcnt);
466 		return (EINVAL);
467 	}
468 
469 	error = copyinuio(uap->iovp, iovcnt, &auio);
470 	if (error) {
471 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
472 		    __func__, error);
473 		return (error);
474 	}
475 	error = vfs_donmount(td, flags, auio);
476 
477 	free(auio, M_IOV);
478 	return (error);
479 }
480 
481 /*
482  * ---------------------------------------------------------------------
483  * Various utility functions
484  */
485 
486 /*
487  * Get a reference on a mount point from a vnode.
488  *
489  * The vnode is allowed to be passed unlocked and race against dooming. Note in
490  * such case there are no guarantees the referenced mount point will still be
491  * associated with it after the function returns.
492  */
493 struct mount *
494 vfs_ref_from_vp(struct vnode *vp)
495 {
496 	struct mount *mp;
497 	struct mount_pcpu *mpcpu;
498 
499 	mp = atomic_load_ptr(&vp->v_mount);
500 	if (__predict_false(mp == NULL)) {
501 		return (mp);
502 	}
503 	if (vfs_op_thread_enter(mp, mpcpu)) {
504 		if (__predict_true(mp == vp->v_mount)) {
505 			vfs_mp_count_add_pcpu(mpcpu, ref, 1);
506 			vfs_op_thread_exit(mp, mpcpu);
507 		} else {
508 			vfs_op_thread_exit(mp, mpcpu);
509 			mp = NULL;
510 		}
511 	} else {
512 		MNT_ILOCK(mp);
513 		if (mp == vp->v_mount) {
514 			MNT_REF(mp);
515 			MNT_IUNLOCK(mp);
516 		} else {
517 			MNT_IUNLOCK(mp);
518 			mp = NULL;
519 		}
520 	}
521 	return (mp);
522 }
523 
524 void
525 vfs_ref(struct mount *mp)
526 {
527 	struct mount_pcpu *mpcpu;
528 
529 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
530 	if (vfs_op_thread_enter(mp, mpcpu)) {
531 		vfs_mp_count_add_pcpu(mpcpu, ref, 1);
532 		vfs_op_thread_exit(mp, mpcpu);
533 		return;
534 	}
535 
536 	MNT_ILOCK(mp);
537 	MNT_REF(mp);
538 	MNT_IUNLOCK(mp);
539 }
540 
541 /*
542  * Register ump as an upper mount of the mount associated with
543  * vnode vp.  This registration will be tracked through
544  * mount_upper_node upper, which should be allocated by the
545  * caller and stored in per-mount data associated with mp.
546  *
547  * If successful, this function will return the mount associated
548  * with vp, and will ensure that it cannot be unmounted until
549  * ump has been unregistered as one of its upper mounts.
550  *
551  * Upon failure this function will return NULL.
552  */
553 struct mount *
554 vfs_register_upper_from_vp(struct vnode *vp, struct mount *ump,
555     struct mount_upper_node *upper)
556 {
557 	struct mount *mp;
558 
559 	mp = atomic_load_ptr(&vp->v_mount);
560 	if (mp == NULL)
561 		return (NULL);
562 	MNT_ILOCK(mp);
563 	if (mp != vp->v_mount ||
564 	    ((mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_RECURSE)) != 0)) {
565 		MNT_IUNLOCK(mp);
566 		return (NULL);
567 	}
568 	KASSERT(ump != mp, ("upper and lower mounts are identical"));
569 	upper->mp = ump;
570 	MNT_REF(mp);
571 	TAILQ_INSERT_TAIL(&mp->mnt_uppers, upper, mnt_upper_link);
572 	MNT_IUNLOCK(mp);
573 	return (mp);
574 }
575 
576 /*
577  * Register upper mount ump to receive vnode unlink/reclaim
578  * notifications from lower mount mp. This registration will
579  * be tracked through mount_upper_node upper, which should be
580  * allocated by the caller and stored in per-mount data
581  * associated with mp.
582  *
583  * ump must already be registered as an upper mount of mp
584  * through a call to vfs_register_upper_from_vp().
585  */
586 void
587 vfs_register_for_notification(struct mount *mp, struct mount *ump,
588     struct mount_upper_node *upper)
589 {
590 	upper->mp = ump;
591 	MNT_ILOCK(mp);
592 	TAILQ_INSERT_TAIL(&mp->mnt_notify, upper, mnt_upper_link);
593 	MNT_IUNLOCK(mp);
594 }
595 
596 static void
597 vfs_drain_upper_locked(struct mount *mp)
598 {
599 	mtx_assert(MNT_MTX(mp), MA_OWNED);
600 	while (mp->mnt_upper_pending != 0) {
601 		mp->mnt_kern_flag |= MNTK_UPPER_WAITER;
602 		msleep(&mp->mnt_uppers, MNT_MTX(mp), 0, "mntupw", 0);
603 	}
604 }
605 
606 /*
607  * Undo a previous call to vfs_register_for_notification().
608  * The mount represented by upper must be currently registered
609  * as an upper mount for mp.
610  */
611 void
612 vfs_unregister_for_notification(struct mount *mp,
613     struct mount_upper_node *upper)
614 {
615 	MNT_ILOCK(mp);
616 	vfs_drain_upper_locked(mp);
617 	TAILQ_REMOVE(&mp->mnt_notify, upper, mnt_upper_link);
618 	MNT_IUNLOCK(mp);
619 }
620 
621 /*
622  * Undo a previous call to vfs_register_upper_from_vp().
623  * This must be done before mp can be unmounted.
624  */
625 void
626 vfs_unregister_upper(struct mount *mp, struct mount_upper_node *upper)
627 {
628 	MNT_ILOCK(mp);
629 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
630 	    ("registered upper with pending unmount"));
631 	vfs_drain_upper_locked(mp);
632 	TAILQ_REMOVE(&mp->mnt_uppers, upper, mnt_upper_link);
633 	if ((mp->mnt_kern_flag & MNTK_TASKQUEUE_WAITER) != 0 &&
634 	    TAILQ_EMPTY(&mp->mnt_uppers)) {
635 		mp->mnt_kern_flag &= ~MNTK_TASKQUEUE_WAITER;
636 		wakeup(&mp->mnt_taskqueue_link);
637 	}
638 	MNT_REL(mp);
639 	MNT_IUNLOCK(mp);
640 }
641 
642 void
643 vfs_rel(struct mount *mp)
644 {
645 	struct mount_pcpu *mpcpu;
646 
647 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
648 	if (vfs_op_thread_enter(mp, mpcpu)) {
649 		vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
650 		vfs_op_thread_exit(mp, mpcpu);
651 		return;
652 	}
653 
654 	MNT_ILOCK(mp);
655 	MNT_REL(mp);
656 	MNT_IUNLOCK(mp);
657 }
658 
659 /*
660  * Allocate and initialize the mount point struct.
661  */
662 struct mount *
663 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
664     struct ucred *cred)
665 {
666 	struct mount *mp;
667 
668 	mp = uma_zalloc(mount_zone, M_WAITOK);
669 	bzero(&mp->mnt_startzero,
670 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
671 	mp->mnt_kern_flag = 0;
672 	mp->mnt_flag = 0;
673 	mp->mnt_rootvnode = NULL;
674 	mp->mnt_vnodecovered = NULL;
675 	mp->mnt_op = NULL;
676 	mp->mnt_vfc = NULL;
677 	TAILQ_INIT(&mp->mnt_nvnodelist);
678 	mp->mnt_nvnodelistsize = 0;
679 	TAILQ_INIT(&mp->mnt_lazyvnodelist);
680 	mp->mnt_lazyvnodelistsize = 0;
681 	MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 &&
682 	    mp->mnt_writeopcount == 0, mp);
683 	MPASSERT(mp->mnt_vfs_ops == 1, mp,
684 	    ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
685 	(void) vfs_busy(mp, MBF_NOWAIT);
686 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
687 	mp->mnt_op = vfsp->vfc_vfsops;
688 	mp->mnt_vfc = vfsp;
689 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
690 	mp->mnt_gen++;
691 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
692 	mp->mnt_vnodecovered = vp;
693 	mp->mnt_cred = crdup(cred);
694 	mp->mnt_stat.f_owner = cred->cr_uid;
695 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
696 	mp->mnt_iosize_max = DFLTPHYS;
697 #ifdef MAC
698 	mac_mount_init(mp);
699 	mac_mount_create(cred, mp);
700 #endif
701 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
702 	mp->mnt_upper_pending = 0;
703 	TAILQ_INIT(&mp->mnt_uppers);
704 	TAILQ_INIT(&mp->mnt_notify);
705 	mp->mnt_taskqueue_flags = 0;
706 	mp->mnt_unmount_retries = 0;
707 	return (mp);
708 }
709 
710 /*
711  * Destroy the mount struct previously allocated by vfs_mount_alloc().
712  */
713 void
714 vfs_mount_destroy(struct mount *mp)
715 {
716 
717 	MPPASS(mp->mnt_vfs_ops != 0, mp);
718 
719 	vfs_assert_mount_counters(mp);
720 
721 	MNT_ILOCK(mp);
722 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
723 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
724 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
725 		wakeup(mp);
726 	}
727 	while (mp->mnt_ref)
728 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
729 	KASSERT(mp->mnt_ref == 0,
730 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
731 	    __FILE__, __LINE__));
732 	MPPASS(mp->mnt_writeopcount == 0, mp);
733 	MPPASS(mp->mnt_secondary_writes == 0, mp);
734 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
735 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
736 		struct vnode *vp;
737 
738 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
739 			vn_printf(vp, "dangling vnode ");
740 		panic("unmount: dangling vnode");
741 	}
742 	KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending"));
743 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
744 	KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify"));
745 	MPPASS(mp->mnt_nvnodelistsize == 0, mp);
746 	MPPASS(mp->mnt_lazyvnodelistsize == 0, mp);
747 	MPPASS(mp->mnt_lockref == 0, mp);
748 	MNT_IUNLOCK(mp);
749 
750 	MPASSERT(mp->mnt_vfs_ops == 1, mp,
751 	    ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
752 
753 	MPASSERT(mp->mnt_rootvnode == NULL, mp,
754 	    ("mount point still has a root vnode %p", mp->mnt_rootvnode));
755 
756 	if (mp->mnt_vnodecovered != NULL)
757 		vrele(mp->mnt_vnodecovered);
758 #ifdef MAC
759 	mac_mount_destroy(mp);
760 #endif
761 	if (mp->mnt_opt != NULL)
762 		vfs_freeopts(mp->mnt_opt);
763 	if (mp->mnt_exjail != NULL) {
764 		atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt,
765 		    1);
766 		crfree(mp->mnt_exjail);
767 	}
768 	if (mp->mnt_export != NULL) {
769 		vfs_free_addrlist(mp->mnt_export);
770 		free(mp->mnt_export, M_MOUNT);
771 	}
772 	crfree(mp->mnt_cred);
773 	uma_zfree(mount_zone, mp);
774 }
775 
776 static bool
777 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
778 {
779 	/* This is an upgrade of an exisiting mount. */
780 	if ((fsflags & MNT_UPDATE) != 0)
781 		return (false);
782 	/* This is already an R/O mount. */
783 	if ((fsflags & MNT_RDONLY) != 0)
784 		return (false);
785 
786 	switch (error) {
787 	case ENODEV:	/* generic, geom, ... */
788 	case EACCES:	/* cam/scsi, ... */
789 	case EROFS:	/* md, mmcsd, ... */
790 		/*
791 		 * These errors can be returned by the storage layer to signal
792 		 * that the media is read-only.  No harm in the R/O mount
793 		 * attempt if the error was returned for some other reason.
794 		 */
795 		return (true);
796 	default:
797 		return (false);
798 	}
799 }
800 
801 int
802 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
803 {
804 	struct vfsoptlist *optlist;
805 	struct vfsopt *opt, *tmp_opt;
806 	char *fstype, *fspath, *errmsg;
807 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
808 	bool autoro, has_nonexport, jail_export;
809 
810 	errmsg = fspath = NULL;
811 	errmsg_len = fspathlen = 0;
812 	errmsg_pos = -1;
813 	autoro = default_autoro;
814 
815 	error = vfs_buildopts(fsoptions, &optlist);
816 	if (error)
817 		return (error);
818 
819 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
820 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
821 
822 	/*
823 	 * We need these two options before the others,
824 	 * and they are mandatory for any filesystem.
825 	 * Ensure they are NUL terminated as well.
826 	 */
827 	fstypelen = 0;
828 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
829 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
830 		error = EINVAL;
831 		if (errmsg != NULL)
832 			strncpy(errmsg, "Invalid fstype", errmsg_len);
833 		goto bail;
834 	}
835 	fspathlen = 0;
836 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
837 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
838 		error = EINVAL;
839 		if (errmsg != NULL)
840 			strncpy(errmsg, "Invalid fspath", errmsg_len);
841 		goto bail;
842 	}
843 
844 	/*
845 	 * Check to see that "export" is only used with the "update", "fstype",
846 	 * "fspath", "from" and "errmsg" options when in a vnet jail.
847 	 * These are the ones used to set/update exports by mountd(8).
848 	 * If only the above options are set in a jail that can run mountd(8),
849 	 * then the jail_export argument of vfs_domount() will be true.
850 	 * When jail_export is true, the vfs_suser() check does not cause
851 	 * failure, but limits the update to exports only.
852 	 * This allows mountd(8) running within the vnet jail
853 	 * to export file systems visible within the jail, but
854 	 * mounted outside of the jail.
855 	 */
856 	/*
857 	 * We need to see if we have the "update" option
858 	 * before we call vfs_domount(), since vfs_domount() has special
859 	 * logic based on MNT_UPDATE.  This is very important
860 	 * when we want to update the root filesystem.
861 	 */
862 	has_nonexport = false;
863 	jail_export = false;
864 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
865 		int do_freeopt = 0;
866 
867 		if (jailed(td->td_ucred) &&
868 		    strcmp(opt->name, "export") != 0 &&
869 		    strcmp(opt->name, "update") != 0 &&
870 		    strcmp(opt->name, "fstype") != 0 &&
871 		    strcmp(opt->name, "fspath") != 0 &&
872 		    strcmp(opt->name, "from") != 0 &&
873 		    strcmp(opt->name, "errmsg") != 0)
874 			has_nonexport = true;
875 		if (strcmp(opt->name, "update") == 0) {
876 			fsflags |= MNT_UPDATE;
877 			do_freeopt = 1;
878 		}
879 		else if (strcmp(opt->name, "async") == 0)
880 			fsflags |= MNT_ASYNC;
881 		else if (strcmp(opt->name, "force") == 0) {
882 			fsflags |= MNT_FORCE;
883 			do_freeopt = 1;
884 		}
885 		else if (strcmp(opt->name, "reload") == 0) {
886 			fsflags |= MNT_RELOAD;
887 			do_freeopt = 1;
888 		}
889 		else if (strcmp(opt->name, "multilabel") == 0)
890 			fsflags |= MNT_MULTILABEL;
891 		else if (strcmp(opt->name, "noasync") == 0)
892 			fsflags &= ~MNT_ASYNC;
893 		else if (strcmp(opt->name, "noatime") == 0)
894 			fsflags |= MNT_NOATIME;
895 		else if (strcmp(opt->name, "atime") == 0) {
896 			free(opt->name, M_MOUNT);
897 			opt->name = strdup("nonoatime", M_MOUNT);
898 		}
899 		else if (strcmp(opt->name, "noclusterr") == 0)
900 			fsflags |= MNT_NOCLUSTERR;
901 		else if (strcmp(opt->name, "clusterr") == 0) {
902 			free(opt->name, M_MOUNT);
903 			opt->name = strdup("nonoclusterr", M_MOUNT);
904 		}
905 		else if (strcmp(opt->name, "noclusterw") == 0)
906 			fsflags |= MNT_NOCLUSTERW;
907 		else if (strcmp(opt->name, "clusterw") == 0) {
908 			free(opt->name, M_MOUNT);
909 			opt->name = strdup("nonoclusterw", M_MOUNT);
910 		}
911 		else if (strcmp(opt->name, "noexec") == 0)
912 			fsflags |= MNT_NOEXEC;
913 		else if (strcmp(opt->name, "exec") == 0) {
914 			free(opt->name, M_MOUNT);
915 			opt->name = strdup("nonoexec", M_MOUNT);
916 		}
917 		else if (strcmp(opt->name, "nosuid") == 0)
918 			fsflags |= MNT_NOSUID;
919 		else if (strcmp(opt->name, "suid") == 0) {
920 			free(opt->name, M_MOUNT);
921 			opt->name = strdup("nonosuid", M_MOUNT);
922 		}
923 		else if (strcmp(opt->name, "nosymfollow") == 0)
924 			fsflags |= MNT_NOSYMFOLLOW;
925 		else if (strcmp(opt->name, "symfollow") == 0) {
926 			free(opt->name, M_MOUNT);
927 			opt->name = strdup("nonosymfollow", M_MOUNT);
928 		}
929 		else if (strcmp(opt->name, "noro") == 0) {
930 			fsflags &= ~MNT_RDONLY;
931 			autoro = false;
932 		}
933 		else if (strcmp(opt->name, "rw") == 0) {
934 			fsflags &= ~MNT_RDONLY;
935 			autoro = false;
936 		}
937 		else if (strcmp(opt->name, "ro") == 0) {
938 			fsflags |= MNT_RDONLY;
939 			autoro = false;
940 		}
941 		else if (strcmp(opt->name, "rdonly") == 0) {
942 			free(opt->name, M_MOUNT);
943 			opt->name = strdup("ro", M_MOUNT);
944 			fsflags |= MNT_RDONLY;
945 			autoro = false;
946 		}
947 		else if (strcmp(opt->name, "autoro") == 0) {
948 			do_freeopt = 1;
949 			autoro = true;
950 		}
951 		else if (strcmp(opt->name, "suiddir") == 0)
952 			fsflags |= MNT_SUIDDIR;
953 		else if (strcmp(opt->name, "sync") == 0)
954 			fsflags |= MNT_SYNCHRONOUS;
955 		else if (strcmp(opt->name, "union") == 0)
956 			fsflags |= MNT_UNION;
957 		else if (strcmp(opt->name, "export") == 0) {
958 			fsflags |= MNT_EXPORTED;
959 			jail_export = true;
960 		} else if (strcmp(opt->name, "automounted") == 0) {
961 			fsflags |= MNT_AUTOMOUNTED;
962 			do_freeopt = 1;
963 		} else if (strcmp(opt->name, "nocover") == 0) {
964 			fsflags |= MNT_NOCOVER;
965 			do_freeopt = 1;
966 		} else if (strcmp(opt->name, "cover") == 0) {
967 			fsflags &= ~MNT_NOCOVER;
968 			do_freeopt = 1;
969 		} else if (strcmp(opt->name, "emptydir") == 0) {
970 			fsflags |= MNT_EMPTYDIR;
971 			do_freeopt = 1;
972 		} else if (strcmp(opt->name, "noemptydir") == 0) {
973 			fsflags &= ~MNT_EMPTYDIR;
974 			do_freeopt = 1;
975 		}
976 		if (do_freeopt)
977 			vfs_freeopt(optlist, opt);
978 	}
979 
980 	/*
981 	 * Be ultra-paranoid about making sure the type and fspath
982 	 * variables will fit in our mp buffers, including the
983 	 * terminating NUL.
984 	 */
985 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
986 		error = ENAMETOOLONG;
987 		goto bail;
988 	}
989 
990 	/*
991 	 * If has_nonexport is true or the caller is not running within a
992 	 * vnet prison that can run mountd(8), set jail_export false.
993 	 */
994 	if (has_nonexport || !jailed(td->td_ucred) ||
995 	    !prison_check_nfsd(td->td_ucred))
996 		jail_export = false;
997 
998 	error = vfs_domount(td, fstype, fspath, fsflags, jail_export, &optlist);
999 	if (error == ENODEV) {
1000 		error = EINVAL;
1001 		if (errmsg != NULL)
1002 			strncpy(errmsg, "Invalid fstype", errmsg_len);
1003 		goto bail;
1004 	}
1005 
1006 	/*
1007 	 * See if we can mount in the read-only mode if the error code suggests
1008 	 * that it could be possible and the mount options allow for that.
1009 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
1010 	 * overridden by "autoro".
1011 	 */
1012 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
1013 		printf("%s: R/W mount failed, possibly R/O media,"
1014 		    " trying R/O mount\n", __func__);
1015 		fsflags |= MNT_RDONLY;
1016 		error = vfs_domount(td, fstype, fspath, fsflags, jail_export,
1017 		    &optlist);
1018 	}
1019 bail:
1020 	/* copyout the errmsg */
1021 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
1022 	    && errmsg_len > 0 && errmsg != NULL) {
1023 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
1024 			bcopy(errmsg,
1025 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1026 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1027 		} else {
1028 			copyout(errmsg,
1029 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1030 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1031 		}
1032 	}
1033 
1034 	if (optlist != NULL)
1035 		vfs_freeopts(optlist);
1036 	return (error);
1037 }
1038 
1039 /*
1040  * Old mount API.
1041  */
1042 #ifndef _SYS_SYSPROTO_H_
1043 struct mount_args {
1044 	char	*type;
1045 	char	*path;
1046 	int	flags;
1047 	caddr_t	data;
1048 };
1049 #endif
1050 /* ARGSUSED */
1051 int
1052 sys_mount(struct thread *td, struct mount_args *uap)
1053 {
1054 	char *fstype;
1055 	struct vfsconf *vfsp = NULL;
1056 	struct mntarg *ma = NULL;
1057 	uint64_t flags;
1058 	int error;
1059 
1060 	/*
1061 	 * Mount flags are now 64-bits. On 32-bit architectures only
1062 	 * 32-bits are passed in, but from here on everything handles
1063 	 * 64-bit flags correctly.
1064 	 */
1065 	flags = uap->flags;
1066 
1067 	AUDIT_ARG_FFLAGS(flags);
1068 
1069 	/*
1070 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
1071 	 * userspace to set this flag, but we must filter it out if we want
1072 	 * MNT_UPDATE on the root file system to work.
1073 	 * MNT_ROOTFS should only be set by the kernel when mounting its
1074 	 * root file system.
1075 	 */
1076 	flags &= ~MNT_ROOTFS;
1077 
1078 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1079 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1080 	if (error) {
1081 		free(fstype, M_TEMP);
1082 		return (error);
1083 	}
1084 
1085 	AUDIT_ARG_TEXT(fstype);
1086 	vfsp = vfs_byname_kld(fstype, td, &error);
1087 	free(fstype, M_TEMP);
1088 	if (vfsp == NULL)
1089 		return (EINVAL);
1090 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1091 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1092 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1093 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
1094 		return (EOPNOTSUPP);
1095 
1096 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1097 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1098 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1099 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1100 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1101 
1102 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1103 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1104 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1105 }
1106 
1107 /*
1108  * vfs_domount_first(): first file system mount (not update)
1109  */
1110 static int
1111 vfs_domount_first(
1112 	struct thread *td,		/* Calling thread. */
1113 	struct vfsconf *vfsp,		/* File system type. */
1114 	char *fspath,			/* Mount path. */
1115 	struct vnode *vp,		/* Vnode to be covered. */
1116 	uint64_t fsflags,		/* Flags common to all filesystems. */
1117 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1118 	)
1119 {
1120 	struct vattr va;
1121 	struct mount *mp;
1122 	struct vnode *newdp, *rootvp;
1123 	int error, error1;
1124 	bool unmounted;
1125 
1126 	ASSERT_VOP_ELOCKED(vp, __func__);
1127 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1128 
1129 	/*
1130 	 * If the jail of the calling thread lacks permission for this type of
1131 	 * file system, or is trying to cover its own root, deny immediately.
1132 	 */
1133 	if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1134 	    vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1135 		vput(vp);
1136 		return (EPERM);
1137 	}
1138 
1139 	/*
1140 	 * If the user is not root, ensure that they own the directory
1141 	 * onto which we are attempting to mount.
1142 	 */
1143 	error = VOP_GETATTR(vp, &va, td->td_ucred);
1144 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1145 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1146 	if (error == 0)
1147 		error = vinvalbuf(vp, V_SAVE, 0, 0);
1148 	if (vfsp->vfc_flags & VFCF_FILEMOUNT) {
1149 		if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG)
1150 			error = EINVAL;
1151 		/*
1152 		 * For file mounts, ensure that there is only one hardlink to the file.
1153 		 */
1154 		if (error == 0 && vp->v_type == VREG && va.va_nlink != 1)
1155 			error = EINVAL;
1156 	} else {
1157 		if (error == 0 && vp->v_type != VDIR)
1158 			error = ENOTDIR;
1159 	}
1160 	if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
1161 		error = vn_dir_check_empty(vp);
1162 	if (error == 0) {
1163 		VI_LOCK(vp);
1164 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1165 			vp->v_iflag |= VI_MOUNT;
1166 		else
1167 			error = EBUSY;
1168 		VI_UNLOCK(vp);
1169 	}
1170 	if (error != 0) {
1171 		vput(vp);
1172 		return (error);
1173 	}
1174 	vn_seqc_write_begin(vp);
1175 	VOP_UNLOCK(vp);
1176 
1177 	/* Allocate and initialize the filesystem. */
1178 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1179 	/* XXXMAC: pass to vfs_mount_alloc? */
1180 	mp->mnt_optnew = *optlist;
1181 	/* Set the mount level flags. */
1182 	mp->mnt_flag = (fsflags &
1183 	    (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE));
1184 
1185 	/*
1186 	 * Mount the filesystem.
1187 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1188 	 * get.  No freeing of cn_pnbuf.
1189 	 */
1190 	error1 = 0;
1191 	unmounted = true;
1192 	if ((error = VFS_MOUNT(mp)) != 0 ||
1193 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1194 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1195 		rootvp = NULL;
1196 		if (error1 != 0) {
1197 			MPASS(error == 0);
1198 			rootvp = vfs_cache_root_clear(mp);
1199 			if (rootvp != NULL) {
1200 				vhold(rootvp);
1201 				vrele(rootvp);
1202 			}
1203 			(void)vn_start_write(NULL, &mp, V_WAIT);
1204 			MNT_ILOCK(mp);
1205 			mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1206 			MNT_IUNLOCK(mp);
1207 			VFS_PURGE(mp);
1208 			error = VFS_UNMOUNT(mp, 0);
1209 			vn_finished_write(mp);
1210 			if (error != 0) {
1211 				printf(
1212 		    "failed post-mount (%d): rollback unmount returned %d\n",
1213 				    error1, error);
1214 				unmounted = false;
1215 			}
1216 			error = error1;
1217 		}
1218 		vfs_unbusy(mp);
1219 		mp->mnt_vnodecovered = NULL;
1220 		if (unmounted) {
1221 			/* XXXKIB wait for mnt_lockref drain? */
1222 			vfs_mount_destroy(mp);
1223 		}
1224 		VI_LOCK(vp);
1225 		vp->v_iflag &= ~VI_MOUNT;
1226 		VI_UNLOCK(vp);
1227 		if (rootvp != NULL) {
1228 			vn_seqc_write_end(rootvp);
1229 			vdrop(rootvp);
1230 		}
1231 		vn_seqc_write_end(vp);
1232 		vrele(vp);
1233 		return (error);
1234 	}
1235 	vn_seqc_write_begin(newdp);
1236 	VOP_UNLOCK(newdp);
1237 
1238 	if (mp->mnt_opt != NULL)
1239 		vfs_freeopts(mp->mnt_opt);
1240 	mp->mnt_opt = mp->mnt_optnew;
1241 	*optlist = NULL;
1242 
1243 	/*
1244 	 * Prevent external consumers of mount options from reading mnt_optnew.
1245 	 */
1246 	mp->mnt_optnew = NULL;
1247 
1248 	MNT_ILOCK(mp);
1249 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1250 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1251 		mp->mnt_kern_flag |= MNTK_ASYNC;
1252 	else
1253 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1254 	MNT_IUNLOCK(mp);
1255 
1256 	/*
1257 	 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the
1258 	 * vp lock to satisfy vfs_lookup() requirements.
1259 	 */
1260 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1261 	VI_LOCK(vp);
1262 	vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1263 	vp->v_mountedhere = mp;
1264 	VI_UNLOCK(vp);
1265 	VOP_UNLOCK(vp);
1266 	cache_purge(vp);
1267 
1268 	/*
1269 	 * We need to lock both vnodes.
1270 	 *
1271 	 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1272 	 * from different filesystems.
1273 	 */
1274 	vn_lock_pair(vp, false, LK_EXCLUSIVE, newdp, false, LK_EXCLUSIVE);
1275 
1276 	VI_LOCK(vp);
1277 	vp->v_iflag &= ~VI_MOUNT;
1278 	VI_UNLOCK(vp);
1279 	/* Place the new filesystem at the end of the mount list. */
1280 	mtx_lock(&mountlist_mtx);
1281 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1282 	mtx_unlock(&mountlist_mtx);
1283 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1284 	VOP_UNLOCK(vp);
1285 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1286 	VOP_UNLOCK(newdp);
1287 	mount_devctl_event("MOUNT", mp, false);
1288 	mountcheckdirs(vp, newdp);
1289 	vn_seqc_write_end(vp);
1290 	vn_seqc_write_end(newdp);
1291 	vrele(newdp);
1292 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1293 		vfs_allocate_syncvnode(mp);
1294 	vfs_op_exit(mp);
1295 	vfs_unbusy(mp);
1296 	return (0);
1297 }
1298 
1299 /*
1300  * vfs_domount_update(): update of mounted file system
1301  */
1302 static int
1303 vfs_domount_update(
1304 	struct thread *td,		/* Calling thread. */
1305 	struct vnode *vp,		/* Mount point vnode. */
1306 	uint64_t fsflags,		/* Flags common to all filesystems. */
1307 	bool jail_export,		/* Got export option in vnet prison. */
1308 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1309 	)
1310 {
1311 	struct export_args export;
1312 	struct o2export_args o2export;
1313 	struct vnode *rootvp;
1314 	void *bufp;
1315 	struct mount *mp;
1316 	int error, export_error, i, len, fsid_up_len;
1317 	uint64_t flag;
1318 	gid_t *grps;
1319 	fsid_t *fsid_up;
1320 	bool vfs_suser_failed;
1321 
1322 	ASSERT_VOP_ELOCKED(vp, __func__);
1323 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1324 	mp = vp->v_mount;
1325 
1326 	if ((vp->v_vflag & VV_ROOT) == 0) {
1327 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1328 		    == 0)
1329 			error = EXDEV;
1330 		else
1331 			error = EINVAL;
1332 		vput(vp);
1333 		return (error);
1334 	}
1335 
1336 	/*
1337 	 * We only allow the filesystem to be reloaded if it
1338 	 * is currently mounted read-only.
1339 	 */
1340 	flag = mp->mnt_flag;
1341 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1342 		vput(vp);
1343 		return (EOPNOTSUPP);	/* Needs translation */
1344 	}
1345 	/*
1346 	 * Only privileged root, or (if MNT_USER is set) the user that
1347 	 * did the original mount is permitted to update it.
1348 	 */
1349 	/*
1350 	 * For the case of mountd(8) doing exports in a jail, the vfs_suser()
1351 	 * call does not cause failure.  vfs_domount() has already checked
1352 	 * that "root" is doing this and vfs_suser() will fail when
1353 	 * the file system has been mounted outside the jail.
1354 	 * jail_export set true indicates that "export" is not mixed
1355 	 * with other options that change mount behaviour.
1356 	 */
1357 	vfs_suser_failed = false;
1358 	error = vfs_suser(mp, td);
1359 	if (jail_export && error != 0) {
1360 		error = 0;
1361 		vfs_suser_failed = true;
1362 	}
1363 	if (error != 0) {
1364 		vput(vp);
1365 		return (error);
1366 	}
1367 	if (vfs_busy(mp, MBF_NOWAIT)) {
1368 		vput(vp);
1369 		return (EBUSY);
1370 	}
1371 	VI_LOCK(vp);
1372 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1373 		VI_UNLOCK(vp);
1374 		vfs_unbusy(mp);
1375 		vput(vp);
1376 		return (EBUSY);
1377 	}
1378 	vp->v_iflag |= VI_MOUNT;
1379 	VI_UNLOCK(vp);
1380 	VOP_UNLOCK(vp);
1381 
1382 	rootvp = NULL;
1383 	vfs_op_enter(mp);
1384 	vn_seqc_write_begin(vp);
1385 
1386 	if (vfs_getopt(*optlist, "fsid", (void **)&fsid_up,
1387 	    &fsid_up_len) == 0) {
1388 		if (fsid_up_len != sizeof(*fsid_up)) {
1389 			error = EINVAL;
1390 			goto end;
1391 		}
1392 		if (fsidcmp(&fsid_up, &mp->mnt_stat.f_fsid) != 0) {
1393 			error = ENOENT;
1394 			goto end;
1395 		}
1396 		vfs_deleteopt(*optlist, "fsid");
1397 	}
1398 
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 		mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1421 		    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1422 		if ((mp->mnt_flag & MNT_ASYNC) == 0)
1423 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1424 	}
1425 	rootvp = vfs_cache_root_clear(mp);
1426 	MNT_IUNLOCK(mp);
1427 	mp->mnt_optnew = *optlist;
1428 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1429 
1430 	/*
1431 	 * Mount the filesystem.
1432 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1433 	 * get.  No freeing of cn_pnbuf.
1434 	 */
1435 	/*
1436 	 * For the case of mountd(8) doing exports from within a vnet jail,
1437 	 * "from" is typically not set correctly such that VFS_MOUNT() will
1438 	 * return ENOENT. It is not obvious that VFS_MOUNT() ever needs to be
1439 	 * called when mountd is doing exports, but this check only applies to
1440 	 * the specific case where it is running inside a vnet jail, to
1441 	 * avoid any POLA violation.
1442 	 */
1443 	error = 0;
1444 	if (!jail_export)
1445 		error = VFS_MOUNT(mp);
1446 
1447 	export_error = 0;
1448 	/* Process the export option. */
1449 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1450 	    &len) == 0) {
1451 		/* Assume that there is only 1 ABI for each length. */
1452 		switch (len) {
1453 		case (sizeof(struct oexport_args)):
1454 			bzero(&o2export, sizeof(o2export));
1455 			/* FALLTHROUGH */
1456 		case (sizeof(o2export)):
1457 			bcopy(bufp, &o2export, len);
1458 			export.ex_flags = (uint64_t)o2export.ex_flags;
1459 			export.ex_root = o2export.ex_root;
1460 			export.ex_uid = o2export.ex_anon.cr_uid;
1461 			export.ex_groups = NULL;
1462 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1463 			if (export.ex_ngroups > 0) {
1464 				if (export.ex_ngroups <= XU_NGROUPS) {
1465 					export.ex_groups = malloc(
1466 					    export.ex_ngroups * sizeof(gid_t),
1467 					    M_TEMP, M_WAITOK);
1468 					for (i = 0; i < export.ex_ngroups; i++)
1469 						export.ex_groups[i] =
1470 						  o2export.ex_anon.cr_groups[i];
1471 				} else
1472 					export_error = EINVAL;
1473 			} else if (export.ex_ngroups < 0)
1474 				export_error = EINVAL;
1475 			export.ex_addr = o2export.ex_addr;
1476 			export.ex_addrlen = o2export.ex_addrlen;
1477 			export.ex_mask = o2export.ex_mask;
1478 			export.ex_masklen = o2export.ex_masklen;
1479 			export.ex_indexfile = o2export.ex_indexfile;
1480 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1481 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1482 				for (i = 0; i < export.ex_numsecflavors; i++)
1483 					export.ex_secflavors[i] =
1484 					    o2export.ex_secflavors[i];
1485 			} else
1486 				export_error = EINVAL;
1487 			if (export_error == 0)
1488 				export_error = vfs_export(mp, &export, true);
1489 			free(export.ex_groups, M_TEMP);
1490 			break;
1491 		case (sizeof(export)):
1492 			bcopy(bufp, &export, len);
1493 			grps = NULL;
1494 			if (export.ex_ngroups > 0) {
1495 				if (export.ex_ngroups <= NGROUPS_MAX) {
1496 					grps = malloc(export.ex_ngroups *
1497 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1498 					export_error = copyin(export.ex_groups,
1499 					    grps, export.ex_ngroups *
1500 					    sizeof(gid_t));
1501 					if (export_error == 0)
1502 						export.ex_groups = grps;
1503 				} else
1504 					export_error = EINVAL;
1505 			} else if (export.ex_ngroups == 0)
1506 				export.ex_groups = NULL;
1507 			else
1508 				export_error = EINVAL;
1509 			if (export_error == 0)
1510 				export_error = vfs_export(mp, &export, true);
1511 			free(grps, M_TEMP);
1512 			break;
1513 		default:
1514 			export_error = EINVAL;
1515 			break;
1516 		}
1517 	}
1518 
1519 	MNT_ILOCK(mp);
1520 	if (error == 0) {
1521 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1522 		    MNT_SNAPSHOT);
1523 	} else {
1524 		/*
1525 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1526 		 * because it is not part of MNT_UPDATEMASK, but it could have
1527 		 * changed in the meantime if quotactl(2) was called.
1528 		 * All in all we want current value of MNT_QUOTA, not the old
1529 		 * one.
1530 		 */
1531 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1532 	}
1533 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1534 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1535 		mp->mnt_kern_flag |= MNTK_ASYNC;
1536 	else
1537 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1538 	MNT_IUNLOCK(mp);
1539 
1540 	if (error != 0)
1541 		goto end;
1542 
1543 	mount_devctl_event("REMOUNT", mp, true);
1544 	if (mp->mnt_opt != NULL)
1545 		vfs_freeopts(mp->mnt_opt);
1546 	mp->mnt_opt = mp->mnt_optnew;
1547 	*optlist = NULL;
1548 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1549 	/*
1550 	 * Prevent external consumers of mount options from reading
1551 	 * mnt_optnew.
1552 	 */
1553 	mp->mnt_optnew = NULL;
1554 
1555 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1556 		vfs_allocate_syncvnode(mp);
1557 	else
1558 		vfs_deallocate_syncvnode(mp);
1559 end:
1560 	vfs_op_exit(mp);
1561 	if (rootvp != NULL) {
1562 		vn_seqc_write_end(rootvp);
1563 		vrele(rootvp);
1564 	}
1565 	vn_seqc_write_end(vp);
1566 	vfs_unbusy(mp);
1567 	VI_LOCK(vp);
1568 	vp->v_iflag &= ~VI_MOUNT;
1569 	VI_UNLOCK(vp);
1570 	vrele(vp);
1571 	return (error != 0 ? error : export_error);
1572 }
1573 
1574 /*
1575  * vfs_domount(): actually attempt a filesystem mount.
1576  */
1577 static int
1578 vfs_domount(
1579 	struct thread *td,		/* Calling thread. */
1580 	const char *fstype,		/* Filesystem type. */
1581 	char *fspath,			/* Mount path. */
1582 	uint64_t fsflags,		/* Flags common to all filesystems. */
1583 	bool jail_export,		/* Got export option in vnet prison. */
1584 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1585 	)
1586 {
1587 	struct vfsconf *vfsp;
1588 	struct nameidata nd;
1589 	struct vnode *vp;
1590 	char *pathbuf;
1591 	int error;
1592 
1593 	/*
1594 	 * Be ultra-paranoid about making sure the type and fspath
1595 	 * variables will fit in our mp buffers, including the
1596 	 * terminating NUL.
1597 	 */
1598 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1599 		return (ENAMETOOLONG);
1600 
1601 	if (jail_export) {
1602 		error = priv_check(td, PRIV_NFS_DAEMON);
1603 		if (error)
1604 			return (error);
1605 	} else if (jailed(td->td_ucred) || usermount == 0) {
1606 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1607 			return (error);
1608 	}
1609 
1610 	/*
1611 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1612 	 */
1613 	if (fsflags & MNT_EXPORTED) {
1614 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1615 		if (error)
1616 			return (error);
1617 	}
1618 	if (fsflags & MNT_SUIDDIR) {
1619 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1620 		if (error)
1621 			return (error);
1622 	}
1623 	/*
1624 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1625 	 */
1626 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1627 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1628 			fsflags |= MNT_NOSUID | MNT_USER;
1629 	}
1630 
1631 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1632 	vfsp = NULL;
1633 	if ((fsflags & MNT_UPDATE) == 0) {
1634 		/* Don't try to load KLDs if we're mounting the root. */
1635 		if (fsflags & MNT_ROOTFS) {
1636 			if ((vfsp = vfs_byname(fstype)) == NULL)
1637 				return (ENODEV);
1638 		} else {
1639 			if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1640 				return (error);
1641 		}
1642 	}
1643 
1644 	/*
1645 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1646 	 */
1647 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1648 	    UIO_SYSSPACE, fspath);
1649 	error = namei(&nd);
1650 	if (error != 0)
1651 		return (error);
1652 	vp = nd.ni_vp;
1653 	/*
1654 	 * Don't allow stacking file mounts to work around problems with the way
1655 	 * that namei sets nd.ni_dvp to vp_crossmp for these.
1656 	 */
1657 	if (vp->v_type == VREG)
1658 		fsflags |= MNT_NOCOVER;
1659 	if ((fsflags & MNT_UPDATE) == 0) {
1660 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1661 		    (fsflags & MNT_NOCOVER) != 0) {
1662 			vput(vp);
1663 			error = EBUSY;
1664 			goto out;
1665 		}
1666 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1667 		strcpy(pathbuf, fspath);
1668 		/*
1669 		 * Note: we allow any vnode type here. If the path sanity check
1670 		 * succeeds, the type will be validated in vfs_domount_first
1671 		 * above.
1672 		 */
1673 		if (vp->v_type == VDIR)
1674 			error = vn_path_to_global_path(td, vp, pathbuf,
1675 			    MNAMELEN);
1676 		else
1677 			error = vn_path_to_global_path_hardlink(td, vp,
1678 			    nd.ni_dvp, pathbuf, MNAMELEN,
1679 			    nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1680 		if (error == 0) {
1681 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1682 			    fsflags, optlist);
1683 		}
1684 		free(pathbuf, M_TEMP);
1685 	} else
1686 		error = vfs_domount_update(td, vp, fsflags, jail_export,
1687 		    optlist);
1688 
1689 out:
1690 	NDFREE_PNBUF(&nd);
1691 	vrele(nd.ni_dvp);
1692 
1693 	return (error);
1694 }
1695 
1696 /*
1697  * Unmount a filesystem.
1698  *
1699  * Note: unmount takes a path to the vnode mounted on as argument, not
1700  * special file (as before).
1701  */
1702 #ifndef _SYS_SYSPROTO_H_
1703 struct unmount_args {
1704 	char	*path;
1705 	int	flags;
1706 };
1707 #endif
1708 /* ARGSUSED */
1709 int
1710 sys_unmount(struct thread *td, struct unmount_args *uap)
1711 {
1712 
1713 	return (kern_unmount(td, uap->path, uap->flags));
1714 }
1715 
1716 int
1717 kern_unmount(struct thread *td, const char *path, int flags)
1718 {
1719 	struct nameidata nd;
1720 	struct mount *mp;
1721 	char *fsidbuf, *pathbuf;
1722 	fsid_t fsid;
1723 	int error;
1724 
1725 	AUDIT_ARG_VALUE(flags);
1726 	if (jailed(td->td_ucred) || usermount == 0) {
1727 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1728 		if (error)
1729 			return (error);
1730 	}
1731 
1732 	if (flags & MNT_BYFSID) {
1733 		fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1734 		error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1735 		if (error) {
1736 			free(fsidbuf, M_TEMP);
1737 			return (error);
1738 		}
1739 
1740 		AUDIT_ARG_TEXT(fsidbuf);
1741 		/* Decode the filesystem ID. */
1742 		if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1743 			free(fsidbuf, M_TEMP);
1744 			return (EINVAL);
1745 		}
1746 
1747 		mp = vfs_getvfs(&fsid);
1748 		free(fsidbuf, M_TEMP);
1749 		if (mp == NULL) {
1750 			return (ENOENT);
1751 		}
1752 	} else {
1753 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1754 		error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1755 		if (error) {
1756 			free(pathbuf, M_TEMP);
1757 			return (error);
1758 		}
1759 
1760 		/*
1761 		 * Try to find global path for path argument.
1762 		 */
1763 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1764 		    UIO_SYSSPACE, pathbuf);
1765 		if (namei(&nd) == 0) {
1766 			NDFREE_PNBUF(&nd);
1767 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1768 			    MNAMELEN);
1769 			if (error == 0)
1770 				vput(nd.ni_vp);
1771 		}
1772 		mtx_lock(&mountlist_mtx);
1773 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1774 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1775 				vfs_ref(mp);
1776 				break;
1777 			}
1778 		}
1779 		mtx_unlock(&mountlist_mtx);
1780 		free(pathbuf, M_TEMP);
1781 		if (mp == NULL) {
1782 			/*
1783 			 * Previously we returned ENOENT for a nonexistent path and
1784 			 * EINVAL for a non-mountpoint.  We cannot tell these apart
1785 			 * now, so in the !MNT_BYFSID case return the more likely
1786 			 * EINVAL for compatibility.
1787 			 */
1788 			return (EINVAL);
1789 		}
1790 	}
1791 
1792 	/*
1793 	 * Don't allow unmounting the root filesystem.
1794 	 */
1795 	if (mp->mnt_flag & MNT_ROOTFS) {
1796 		vfs_rel(mp);
1797 		return (EINVAL);
1798 	}
1799 	error = dounmount(mp, flags, td);
1800 	return (error);
1801 }
1802 
1803 /*
1804  * Return error if any of the vnodes, ignoring the root vnode
1805  * and the syncer vnode, have non-zero usecount.
1806  *
1807  * This function is purely advisory - it can return false positives
1808  * and negatives.
1809  */
1810 static int
1811 vfs_check_usecounts(struct mount *mp)
1812 {
1813 	struct vnode *vp, *mvp;
1814 
1815 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1816 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1817 		    vp->v_usecount != 0) {
1818 			VI_UNLOCK(vp);
1819 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1820 			return (EBUSY);
1821 		}
1822 		VI_UNLOCK(vp);
1823 	}
1824 
1825 	return (0);
1826 }
1827 
1828 static void
1829 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1830 {
1831 
1832 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1833 	mp->mnt_kern_flag &= ~mntkflags;
1834 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1835 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1836 		wakeup(mp);
1837 	}
1838 	vfs_op_exit_locked(mp);
1839 	MNT_IUNLOCK(mp);
1840 	if (coveredvp != NULL) {
1841 		VOP_UNLOCK(coveredvp);
1842 		vdrop(coveredvp);
1843 	}
1844 	vn_finished_write(mp);
1845 	vfs_rel(mp);
1846 }
1847 
1848 /*
1849  * There are various reference counters associated with the mount point.
1850  * Normally it is permitted to modify them without taking the mnt ilock,
1851  * but this behavior can be temporarily disabled if stable value is needed
1852  * or callers are expected to block (e.g. to not allow new users during
1853  * forced unmount).
1854  */
1855 void
1856 vfs_op_enter(struct mount *mp)
1857 {
1858 	struct mount_pcpu *mpcpu;
1859 	int cpu;
1860 
1861 	MNT_ILOCK(mp);
1862 	mp->mnt_vfs_ops++;
1863 	if (mp->mnt_vfs_ops > 1) {
1864 		MNT_IUNLOCK(mp);
1865 		return;
1866 	}
1867 	vfs_op_barrier_wait(mp);
1868 	CPU_FOREACH(cpu) {
1869 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1870 
1871 		mp->mnt_ref += mpcpu->mntp_ref;
1872 		mpcpu->mntp_ref = 0;
1873 
1874 		mp->mnt_lockref += mpcpu->mntp_lockref;
1875 		mpcpu->mntp_lockref = 0;
1876 
1877 		mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1878 		mpcpu->mntp_writeopcount = 0;
1879 	}
1880 	MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1881 	    mp->mnt_writeopcount >= 0, mp,
1882 	    ("invalid count(s): ref %d lockref %d writeopcount %d",
1883 	    mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1884 	MNT_IUNLOCK(mp);
1885 	vfs_assert_mount_counters(mp);
1886 }
1887 
1888 void
1889 vfs_op_exit_locked(struct mount *mp)
1890 {
1891 
1892 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1893 
1894 	MPASSERT(mp->mnt_vfs_ops > 0, mp,
1895 	    ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1896 	MPASSERT(mp->mnt_vfs_ops > 1 ||
1897 	    (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1898 	    ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1899 	mp->mnt_vfs_ops--;
1900 }
1901 
1902 void
1903 vfs_op_exit(struct mount *mp)
1904 {
1905 
1906 	MNT_ILOCK(mp);
1907 	vfs_op_exit_locked(mp);
1908 	MNT_IUNLOCK(mp);
1909 }
1910 
1911 struct vfs_op_barrier_ipi {
1912 	struct mount *mp;
1913 	struct smp_rendezvous_cpus_retry_arg srcra;
1914 };
1915 
1916 static void
1917 vfs_op_action_func(void *arg)
1918 {
1919 	struct vfs_op_barrier_ipi *vfsopipi;
1920 	struct mount *mp;
1921 
1922 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1923 	mp = vfsopipi->mp;
1924 
1925 	if (!vfs_op_thread_entered(mp))
1926 		smp_rendezvous_cpus_done(arg);
1927 }
1928 
1929 static void
1930 vfs_op_wait_func(void *arg, int cpu)
1931 {
1932 	struct vfs_op_barrier_ipi *vfsopipi;
1933 	struct mount *mp;
1934 	struct mount_pcpu *mpcpu;
1935 
1936 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1937 	mp = vfsopipi->mp;
1938 
1939 	mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1940 	while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1941 		cpu_spinwait();
1942 }
1943 
1944 void
1945 vfs_op_barrier_wait(struct mount *mp)
1946 {
1947 	struct vfs_op_barrier_ipi vfsopipi;
1948 
1949 	vfsopipi.mp = mp;
1950 
1951 	smp_rendezvous_cpus_retry(all_cpus,
1952 	    smp_no_rendezvous_barrier,
1953 	    vfs_op_action_func,
1954 	    smp_no_rendezvous_barrier,
1955 	    vfs_op_wait_func,
1956 	    &vfsopipi.srcra);
1957 }
1958 
1959 #ifdef DIAGNOSTIC
1960 void
1961 vfs_assert_mount_counters(struct mount *mp)
1962 {
1963 	struct mount_pcpu *mpcpu;
1964 	int cpu;
1965 
1966 	if (mp->mnt_vfs_ops == 0)
1967 		return;
1968 
1969 	CPU_FOREACH(cpu) {
1970 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1971 		if (mpcpu->mntp_ref != 0 ||
1972 		    mpcpu->mntp_lockref != 0 ||
1973 		    mpcpu->mntp_writeopcount != 0)
1974 			vfs_dump_mount_counters(mp);
1975 	}
1976 }
1977 
1978 void
1979 vfs_dump_mount_counters(struct mount *mp)
1980 {
1981 	struct mount_pcpu *mpcpu;
1982 	int ref, lockref, writeopcount;
1983 	int cpu;
1984 
1985 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1986 
1987 	printf("        ref : ");
1988 	ref = mp->mnt_ref;
1989 	CPU_FOREACH(cpu) {
1990 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1991 		printf("%d ", mpcpu->mntp_ref);
1992 		ref += mpcpu->mntp_ref;
1993 	}
1994 	printf("\n");
1995 	printf("    lockref : ");
1996 	lockref = mp->mnt_lockref;
1997 	CPU_FOREACH(cpu) {
1998 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1999 		printf("%d ", mpcpu->mntp_lockref);
2000 		lockref += mpcpu->mntp_lockref;
2001 	}
2002 	printf("\n");
2003 	printf("writeopcount: ");
2004 	writeopcount = mp->mnt_writeopcount;
2005 	CPU_FOREACH(cpu) {
2006 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2007 		printf("%d ", mpcpu->mntp_writeopcount);
2008 		writeopcount += mpcpu->mntp_writeopcount;
2009 	}
2010 	printf("\n");
2011 
2012 	printf("counter       struct total\n");
2013 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
2014 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
2015 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
2016 
2017 	panic("invalid counts on struct mount");
2018 }
2019 #endif
2020 
2021 int
2022 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
2023 {
2024 	struct mount_pcpu *mpcpu;
2025 	int cpu, sum;
2026 
2027 	switch (which) {
2028 	case MNT_COUNT_REF:
2029 		sum = mp->mnt_ref;
2030 		break;
2031 	case MNT_COUNT_LOCKREF:
2032 		sum = mp->mnt_lockref;
2033 		break;
2034 	case MNT_COUNT_WRITEOPCOUNT:
2035 		sum = mp->mnt_writeopcount;
2036 		break;
2037 	}
2038 
2039 	CPU_FOREACH(cpu) {
2040 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2041 		switch (which) {
2042 		case MNT_COUNT_REF:
2043 			sum += mpcpu->mntp_ref;
2044 			break;
2045 		case MNT_COUNT_LOCKREF:
2046 			sum += mpcpu->mntp_lockref;
2047 			break;
2048 		case MNT_COUNT_WRITEOPCOUNT:
2049 			sum += mpcpu->mntp_writeopcount;
2050 			break;
2051 		}
2052 	}
2053 	return (sum);
2054 }
2055 
2056 static bool
2057 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
2058     int timeout_ticks)
2059 {
2060 	bool enqueued;
2061 
2062 	enqueued = false;
2063 	mtx_lock(&deferred_unmount_lock);
2064 	if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
2065 		mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
2066 		STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
2067 		    mnt_taskqueue_link);
2068 		enqueued = true;
2069 	}
2070 	mtx_unlock(&deferred_unmount_lock);
2071 
2072 	if (enqueued) {
2073 		taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
2074 		    &deferred_unmount_task, timeout_ticks);
2075 	}
2076 
2077 	return (enqueued);
2078 }
2079 
2080 /*
2081  * Taskqueue handler for processing async/recursive unmounts
2082  */
2083 static void
2084 vfs_deferred_unmount(void *argi __unused, int pending __unused)
2085 {
2086 	STAILQ_HEAD(, mount) local_unmounts;
2087 	uint64_t flags;
2088 	struct mount *mp, *tmp;
2089 	int error;
2090 	unsigned int retries;
2091 	bool unmounted;
2092 
2093 	STAILQ_INIT(&local_unmounts);
2094 	mtx_lock(&deferred_unmount_lock);
2095 	STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
2096 	mtx_unlock(&deferred_unmount_lock);
2097 
2098 	STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
2099 		flags = mp->mnt_taskqueue_flags;
2100 		KASSERT((flags & MNT_DEFERRED) != 0,
2101 		    ("taskqueue unmount without MNT_DEFERRED"));
2102 		error = dounmount(mp, flags, curthread);
2103 		if (error != 0) {
2104 			MNT_ILOCK(mp);
2105 			unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
2106 			MNT_IUNLOCK(mp);
2107 
2108 			/*
2109 			 * The deferred unmount thread is the only thread that
2110 			 * modifies the retry counts, so locking/atomics aren't
2111 			 * needed here.
2112 			 */
2113 			retries = (mp->mnt_unmount_retries)++;
2114 			deferred_unmount_total_retries++;
2115 			if (!unmounted && retries < deferred_unmount_retry_limit) {
2116 				deferred_unmount_enqueue(mp, flags, true,
2117 				    -deferred_unmount_retry_delay_hz);
2118 			} else {
2119 				if (retries >= deferred_unmount_retry_limit) {
2120 					printf("giving up on deferred unmount "
2121 					    "of %s after %d retries, error %d\n",
2122 					    mp->mnt_stat.f_mntonname, retries, error);
2123 				}
2124 				vfs_rel(mp);
2125 			}
2126 		}
2127 	}
2128 }
2129 
2130 /*
2131  * Do the actual filesystem unmount.
2132  */
2133 int
2134 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2135 {
2136 	struct mount_upper_node *upper;
2137 	struct vnode *coveredvp, *rootvp;
2138 	int error;
2139 	uint64_t async_flag;
2140 	int mnt_gen_r;
2141 	unsigned int retries;
2142 
2143 	KASSERT((flags & MNT_DEFERRED) == 0 ||
2144 	    (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2145 	    ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2146 
2147 	/*
2148 	 * If the caller has explicitly requested the unmount to be handled by
2149 	 * the taskqueue and we're not already in taskqueue context, queue
2150 	 * up the unmount request and exit.  This is done prior to any
2151 	 * credential checks; MNT_DEFERRED should be used only for kernel-
2152 	 * initiated unmounts and will therefore be processed with the
2153 	 * (kernel) credentials of the taskqueue thread.  Still, callers
2154 	 * should be sure this is the behavior they want.
2155 	 */
2156 	if ((flags & MNT_DEFERRED) != 0 &&
2157 	    taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2158 		if (!deferred_unmount_enqueue(mp, flags, false, 0))
2159 			vfs_rel(mp);
2160 		return (EINPROGRESS);
2161 	}
2162 
2163 	/*
2164 	 * Only privileged root, or (if MNT_USER is set) the user that did the
2165 	 * original mount is permitted to unmount this filesystem.
2166 	 * This check should be made prior to queueing up any recursive
2167 	 * unmounts of upper filesystems.  Those unmounts will be executed
2168 	 * with kernel thread credentials and are expected to succeed, so
2169 	 * we must at least ensure the originating context has sufficient
2170 	 * privilege to unmount the base filesystem before proceeding with
2171 	 * the uppers.
2172 	 */
2173 	error = vfs_suser(mp, td);
2174 	if (error != 0) {
2175 		KASSERT((flags & MNT_DEFERRED) == 0,
2176 		    ("taskqueue unmount with insufficient privilege"));
2177 		vfs_rel(mp);
2178 		return (error);
2179 	}
2180 
2181 	if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2182 		flags |= MNT_RECURSE;
2183 
2184 	if ((flags & MNT_RECURSE) != 0) {
2185 		KASSERT((flags & MNT_FORCE) != 0,
2186 		    ("MNT_RECURSE requires MNT_FORCE"));
2187 
2188 		MNT_ILOCK(mp);
2189 		/*
2190 		 * Set MNTK_RECURSE to prevent new upper mounts from being
2191 		 * added, and note that an operation on the uppers list is in
2192 		 * progress.  This will ensure that unregistration from the
2193 		 * uppers list, and therefore any pending unmount of the upper
2194 		 * FS, can't complete until after we finish walking the list.
2195 		 */
2196 		mp->mnt_kern_flag |= MNTK_RECURSE;
2197 		mp->mnt_upper_pending++;
2198 		TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2199 			retries = upper->mp->mnt_unmount_retries;
2200 			if (retries > deferred_unmount_retry_limit) {
2201 				error = EBUSY;
2202 				continue;
2203 			}
2204 			MNT_IUNLOCK(mp);
2205 
2206 			vfs_ref(upper->mp);
2207 			if (!deferred_unmount_enqueue(upper->mp, flags,
2208 			    false, 0))
2209 				vfs_rel(upper->mp);
2210 			MNT_ILOCK(mp);
2211 		}
2212 		mp->mnt_upper_pending--;
2213 		if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2214 		    mp->mnt_upper_pending == 0) {
2215 			mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2216 			wakeup(&mp->mnt_uppers);
2217 		}
2218 
2219 		/*
2220 		 * If we're not on the taskqueue, wait until the uppers list
2221 		 * is drained before proceeding with unmount.  Otherwise, if
2222 		 * we are on the taskqueue and there are still pending uppers,
2223 		 * just re-enqueue on the end of the taskqueue.
2224 		 */
2225 		if ((flags & MNT_DEFERRED) == 0) {
2226 			while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2227 				mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2228 				error = msleep(&mp->mnt_taskqueue_link,
2229 				    MNT_MTX(mp), PCATCH, "umntqw", 0);
2230 			}
2231 			if (error != 0) {
2232 				MNT_REL(mp);
2233 				MNT_IUNLOCK(mp);
2234 				return (error);
2235 			}
2236 		} else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2237 			MNT_IUNLOCK(mp);
2238 			if (error == 0)
2239 				deferred_unmount_enqueue(mp, flags, true, 0);
2240 			return (error);
2241 		}
2242 		MNT_IUNLOCK(mp);
2243 		KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2244 	}
2245 
2246 	/* Allow the taskqueue to safely re-enqueue on failure */
2247 	if ((flags & MNT_DEFERRED) != 0)
2248 		vfs_ref(mp);
2249 
2250 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2251 		mnt_gen_r = mp->mnt_gen;
2252 		VI_LOCK(coveredvp);
2253 		vholdl(coveredvp);
2254 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2255 		/*
2256 		 * Check for mp being unmounted while waiting for the
2257 		 * covered vnode lock.
2258 		 */
2259 		if (coveredvp->v_mountedhere != mp ||
2260 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2261 			VOP_UNLOCK(coveredvp);
2262 			vdrop(coveredvp);
2263 			vfs_rel(mp);
2264 			return (EBUSY);
2265 		}
2266 	}
2267 
2268 	vfs_op_enter(mp);
2269 
2270 	vn_start_write(NULL, &mp, V_WAIT);
2271 	MNT_ILOCK(mp);
2272 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2273 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
2274 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
2275 		dounmount_cleanup(mp, coveredvp, 0);
2276 		return (EBUSY);
2277 	}
2278 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
2279 	rootvp = vfs_cache_root_clear(mp);
2280 	if (coveredvp != NULL)
2281 		vn_seqc_write_begin(coveredvp);
2282 	if (flags & MNT_NONBUSY) {
2283 		MNT_IUNLOCK(mp);
2284 		error = vfs_check_usecounts(mp);
2285 		MNT_ILOCK(mp);
2286 		if (error != 0) {
2287 			vn_seqc_write_end(coveredvp);
2288 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2289 			if (rootvp != NULL) {
2290 				vn_seqc_write_end(rootvp);
2291 				vrele(rootvp);
2292 			}
2293 			return (error);
2294 		}
2295 	}
2296 	/* Allow filesystems to detect that a forced unmount is in progress. */
2297 	if (flags & MNT_FORCE) {
2298 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2299 		MNT_IUNLOCK(mp);
2300 		/*
2301 		 * Must be done after setting MNTK_UNMOUNTF and before
2302 		 * waiting for mnt_lockref to become 0.
2303 		 */
2304 		VFS_PURGE(mp);
2305 		MNT_ILOCK(mp);
2306 	}
2307 	error = 0;
2308 	if (mp->mnt_lockref) {
2309 		mp->mnt_kern_flag |= MNTK_DRAINING;
2310 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2311 		    "mount drain", 0);
2312 	}
2313 	MNT_IUNLOCK(mp);
2314 	KASSERT(mp->mnt_lockref == 0,
2315 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
2316 	    __func__, __FILE__, __LINE__));
2317 	KASSERT(error == 0,
2318 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
2319 	    __func__, __FILE__, __LINE__));
2320 
2321 	/*
2322 	 * We want to keep the vnode around so that we can vn_seqc_write_end
2323 	 * after we are done with unmount. Downgrade our reference to a mere
2324 	 * hold count so that we don't interefere with anything.
2325 	 */
2326 	if (rootvp != NULL) {
2327 		vhold(rootvp);
2328 		vrele(rootvp);
2329 	}
2330 
2331 	if (mp->mnt_flag & MNT_EXPUBLIC)
2332 		vfs_setpublicfs(NULL, NULL, NULL);
2333 
2334 	vfs_periodic(mp, MNT_WAIT);
2335 	MNT_ILOCK(mp);
2336 	async_flag = mp->mnt_flag & MNT_ASYNC;
2337 	mp->mnt_flag &= ~MNT_ASYNC;
2338 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
2339 	MNT_IUNLOCK(mp);
2340 	vfs_deallocate_syncvnode(mp);
2341 	error = VFS_UNMOUNT(mp, flags);
2342 	vn_finished_write(mp);
2343 	vfs_rel(mp);
2344 	/*
2345 	 * If we failed to flush the dirty blocks for this mount point,
2346 	 * undo all the cdir/rdir and rootvnode changes we made above.
2347 	 * Unless we failed to do so because the device is reporting that
2348 	 * it doesn't exist anymore.
2349 	 */
2350 	if (error && error != ENXIO) {
2351 		MNT_ILOCK(mp);
2352 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2353 			MNT_IUNLOCK(mp);
2354 			vfs_allocate_syncvnode(mp);
2355 			MNT_ILOCK(mp);
2356 		}
2357 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2358 		mp->mnt_flag |= async_flag;
2359 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2360 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2361 			mp->mnt_kern_flag |= MNTK_ASYNC;
2362 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
2363 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
2364 			wakeup(mp);
2365 		}
2366 		vfs_op_exit_locked(mp);
2367 		MNT_IUNLOCK(mp);
2368 		if (coveredvp) {
2369 			vn_seqc_write_end(coveredvp);
2370 			VOP_UNLOCK(coveredvp);
2371 			vdrop(coveredvp);
2372 		}
2373 		if (rootvp != NULL) {
2374 			vn_seqc_write_end(rootvp);
2375 			vdrop(rootvp);
2376 		}
2377 		return (error);
2378 	}
2379 
2380 	mtx_lock(&mountlist_mtx);
2381 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
2382 	mtx_unlock(&mountlist_mtx);
2383 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2384 	if (coveredvp != NULL) {
2385 		VI_LOCK(coveredvp);
2386 		vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2387 		coveredvp->v_mountedhere = NULL;
2388 		vn_seqc_write_end_locked(coveredvp);
2389 		VI_UNLOCK(coveredvp);
2390 		VOP_UNLOCK(coveredvp);
2391 		vdrop(coveredvp);
2392 	}
2393 	mount_devctl_event("UNMOUNT", mp, false);
2394 	if (rootvp != NULL) {
2395 		vn_seqc_write_end(rootvp);
2396 		vdrop(rootvp);
2397 	}
2398 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2399 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
2400 		vrele(rootvnode);
2401 		rootvnode = NULL;
2402 	}
2403 	if (mp == rootdevmp)
2404 		rootdevmp = NULL;
2405 	if ((flags & MNT_DEFERRED) != 0)
2406 		vfs_rel(mp);
2407 	vfs_mount_destroy(mp);
2408 	return (0);
2409 }
2410 
2411 /*
2412  * Report errors during filesystem mounting.
2413  */
2414 void
2415 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2416 {
2417 	struct vfsoptlist *moptlist = mp->mnt_optnew;
2418 	va_list ap;
2419 	int error, len;
2420 	char *errmsg;
2421 
2422 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2423 	if (error || errmsg == NULL || len <= 0)
2424 		return;
2425 
2426 	va_start(ap, fmt);
2427 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2428 	va_end(ap);
2429 }
2430 
2431 void
2432 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2433 {
2434 	va_list ap;
2435 	int error, len;
2436 	char *errmsg;
2437 
2438 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2439 	if (error || errmsg == NULL || len <= 0)
2440 		return;
2441 
2442 	va_start(ap, fmt);
2443 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2444 	va_end(ap);
2445 }
2446 
2447 /*
2448  * ---------------------------------------------------------------------
2449  * Functions for querying mount options/arguments from filesystems.
2450  */
2451 
2452 /*
2453  * Check that no unknown options are given
2454  */
2455 int
2456 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2457 {
2458 	struct vfsopt *opt;
2459 	char errmsg[255];
2460 	const char **t, *p, *q;
2461 	int ret = 0;
2462 
2463 	TAILQ_FOREACH(opt, opts, link) {
2464 		p = opt->name;
2465 		q = NULL;
2466 		if (p[0] == 'n' && p[1] == 'o')
2467 			q = p + 2;
2468 		for(t = global_opts; *t != NULL; t++) {
2469 			if (strcmp(*t, p) == 0)
2470 				break;
2471 			if (q != NULL) {
2472 				if (strcmp(*t, q) == 0)
2473 					break;
2474 			}
2475 		}
2476 		if (*t != NULL)
2477 			continue;
2478 		for(t = legal; *t != NULL; t++) {
2479 			if (strcmp(*t, p) == 0)
2480 				break;
2481 			if (q != NULL) {
2482 				if (strcmp(*t, q) == 0)
2483 					break;
2484 			}
2485 		}
2486 		if (*t != NULL)
2487 			continue;
2488 		snprintf(errmsg, sizeof(errmsg),
2489 		    "mount option <%s> is unknown", p);
2490 		ret = EINVAL;
2491 	}
2492 	if (ret != 0) {
2493 		TAILQ_FOREACH(opt, opts, link) {
2494 			if (strcmp(opt->name, "errmsg") == 0) {
2495 				strncpy((char *)opt->value, errmsg, opt->len);
2496 				break;
2497 			}
2498 		}
2499 		if (opt == NULL)
2500 			printf("%s\n", errmsg);
2501 	}
2502 	return (ret);
2503 }
2504 
2505 /*
2506  * Get a mount option by its name.
2507  *
2508  * Return 0 if the option was found, ENOENT otherwise.
2509  * If len is non-NULL it will be filled with the length
2510  * of the option. If buf is non-NULL, it will be filled
2511  * with the address of the option.
2512  */
2513 int
2514 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2515 {
2516 	struct vfsopt *opt;
2517 
2518 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2519 
2520 	TAILQ_FOREACH(opt, opts, link) {
2521 		if (strcmp(name, opt->name) == 0) {
2522 			opt->seen = 1;
2523 			if (len != NULL)
2524 				*len = opt->len;
2525 			if (buf != NULL)
2526 				*buf = opt->value;
2527 			return (0);
2528 		}
2529 	}
2530 	return (ENOENT);
2531 }
2532 
2533 int
2534 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2535 {
2536 	struct vfsopt *opt;
2537 
2538 	if (opts == NULL)
2539 		return (-1);
2540 
2541 	TAILQ_FOREACH(opt, opts, link) {
2542 		if (strcmp(name, opt->name) == 0) {
2543 			opt->seen = 1;
2544 			return (opt->pos);
2545 		}
2546 	}
2547 	return (-1);
2548 }
2549 
2550 int
2551 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2552 {
2553 	char *opt_value, *vtp;
2554 	quad_t iv;
2555 	int error, opt_len;
2556 
2557 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2558 	if (error != 0)
2559 		return (error);
2560 	if (opt_len == 0 || opt_value == NULL)
2561 		return (EINVAL);
2562 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2563 		return (EINVAL);
2564 	iv = strtoq(opt_value, &vtp, 0);
2565 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2566 		return (EINVAL);
2567 	if (iv < 0)
2568 		return (EINVAL);
2569 	switch (vtp[0]) {
2570 	case 't': case 'T':
2571 		iv *= 1024;
2572 		/* FALLTHROUGH */
2573 	case 'g': case 'G':
2574 		iv *= 1024;
2575 		/* FALLTHROUGH */
2576 	case 'm': case 'M':
2577 		iv *= 1024;
2578 		/* FALLTHROUGH */
2579 	case 'k': case 'K':
2580 		iv *= 1024;
2581 	case '\0':
2582 		break;
2583 	default:
2584 		return (EINVAL);
2585 	}
2586 	*value = iv;
2587 
2588 	return (0);
2589 }
2590 
2591 char *
2592 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2593 {
2594 	struct vfsopt *opt;
2595 
2596 	*error = 0;
2597 	TAILQ_FOREACH(opt, opts, link) {
2598 		if (strcmp(name, opt->name) != 0)
2599 			continue;
2600 		opt->seen = 1;
2601 		if (opt->len == 0 ||
2602 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2603 			*error = EINVAL;
2604 			return (NULL);
2605 		}
2606 		return (opt->value);
2607 	}
2608 	*error = ENOENT;
2609 	return (NULL);
2610 }
2611 
2612 int
2613 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2614 	uint64_t val)
2615 {
2616 	struct vfsopt *opt;
2617 
2618 	TAILQ_FOREACH(opt, opts, link) {
2619 		if (strcmp(name, opt->name) == 0) {
2620 			opt->seen = 1;
2621 			if (w != NULL)
2622 				*w |= val;
2623 			return (1);
2624 		}
2625 	}
2626 	if (w != NULL)
2627 		*w &= ~val;
2628 	return (0);
2629 }
2630 
2631 int
2632 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2633 {
2634 	va_list ap;
2635 	struct vfsopt *opt;
2636 	int ret;
2637 
2638 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2639 
2640 	TAILQ_FOREACH(opt, opts, link) {
2641 		if (strcmp(name, opt->name) != 0)
2642 			continue;
2643 		opt->seen = 1;
2644 		if (opt->len == 0 || opt->value == NULL)
2645 			return (0);
2646 		if (((char *)opt->value)[opt->len - 1] != '\0')
2647 			return (0);
2648 		va_start(ap, fmt);
2649 		ret = vsscanf(opt->value, fmt, ap);
2650 		va_end(ap);
2651 		return (ret);
2652 	}
2653 	return (0);
2654 }
2655 
2656 int
2657 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2658 {
2659 	struct vfsopt *opt;
2660 
2661 	TAILQ_FOREACH(opt, opts, link) {
2662 		if (strcmp(name, opt->name) != 0)
2663 			continue;
2664 		opt->seen = 1;
2665 		if (opt->value == NULL)
2666 			opt->len = len;
2667 		else {
2668 			if (opt->len != len)
2669 				return (EINVAL);
2670 			bcopy(value, opt->value, len);
2671 		}
2672 		return (0);
2673 	}
2674 	return (ENOENT);
2675 }
2676 
2677 int
2678 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2679 {
2680 	struct vfsopt *opt;
2681 
2682 	TAILQ_FOREACH(opt, opts, link) {
2683 		if (strcmp(name, opt->name) != 0)
2684 			continue;
2685 		opt->seen = 1;
2686 		if (opt->value == NULL)
2687 			opt->len = len;
2688 		else {
2689 			if (opt->len < len)
2690 				return (EINVAL);
2691 			opt->len = len;
2692 			bcopy(value, opt->value, len);
2693 		}
2694 		return (0);
2695 	}
2696 	return (ENOENT);
2697 }
2698 
2699 int
2700 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2701 {
2702 	struct vfsopt *opt;
2703 
2704 	TAILQ_FOREACH(opt, opts, link) {
2705 		if (strcmp(name, opt->name) != 0)
2706 			continue;
2707 		opt->seen = 1;
2708 		if (opt->value == NULL)
2709 			opt->len = strlen(value) + 1;
2710 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2711 			return (EINVAL);
2712 		return (0);
2713 	}
2714 	return (ENOENT);
2715 }
2716 
2717 /*
2718  * Find and copy a mount option.
2719  *
2720  * The size of the buffer has to be specified
2721  * in len, if it is not the same length as the
2722  * mount option, EINVAL is returned.
2723  * Returns ENOENT if the option is not found.
2724  */
2725 int
2726 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2727 {
2728 	struct vfsopt *opt;
2729 
2730 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2731 
2732 	TAILQ_FOREACH(opt, opts, link) {
2733 		if (strcmp(name, opt->name) == 0) {
2734 			opt->seen = 1;
2735 			if (len != opt->len)
2736 				return (EINVAL);
2737 			bcopy(opt->value, dest, opt->len);
2738 			return (0);
2739 		}
2740 	}
2741 	return (ENOENT);
2742 }
2743 
2744 int
2745 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2746 {
2747 	/*
2748 	 * Filesystems only fill in part of the structure for updates, we
2749 	 * have to read the entirety first to get all content.
2750 	 */
2751 	if (sbp != &mp->mnt_stat)
2752 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2753 
2754 	/*
2755 	 * Set these in case the underlying filesystem fails to do so.
2756 	 */
2757 	sbp->f_version = STATFS_VERSION;
2758 	sbp->f_namemax = NAME_MAX;
2759 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2760 	sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2761 
2762 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2763 }
2764 
2765 void
2766 vfs_mountedfrom(struct mount *mp, const char *from)
2767 {
2768 
2769 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2770 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2771 	    sizeof mp->mnt_stat.f_mntfromname);
2772 }
2773 
2774 /*
2775  * ---------------------------------------------------------------------
2776  * This is the api for building mount args and mounting filesystems from
2777  * inside the kernel.
2778  *
2779  * The API works by accumulation of individual args.  First error is
2780  * latched.
2781  *
2782  * XXX: should be documented in new manpage kernel_mount(9)
2783  */
2784 
2785 /* A memory allocation which must be freed when we are done */
2786 struct mntaarg {
2787 	SLIST_ENTRY(mntaarg)	next;
2788 };
2789 
2790 /* The header for the mount arguments */
2791 struct mntarg {
2792 	struct iovec *v;
2793 	int len;
2794 	int error;
2795 	SLIST_HEAD(, mntaarg)	list;
2796 };
2797 
2798 /*
2799  * Add a boolean argument.
2800  *
2801  * flag is the boolean value.
2802  * name must start with "no".
2803  */
2804 struct mntarg *
2805 mount_argb(struct mntarg *ma, int flag, const char *name)
2806 {
2807 
2808 	KASSERT(name[0] == 'n' && name[1] == 'o',
2809 	    ("mount_argb(...,%s): name must start with 'no'", name));
2810 
2811 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2812 }
2813 
2814 /*
2815  * Add an argument printf style
2816  */
2817 struct mntarg *
2818 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2819 {
2820 	va_list ap;
2821 	struct mntaarg *maa;
2822 	struct sbuf *sb;
2823 	int len;
2824 
2825 	if (ma == NULL) {
2826 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2827 		SLIST_INIT(&ma->list);
2828 	}
2829 	if (ma->error)
2830 		return (ma);
2831 
2832 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2833 	    M_MOUNT, M_WAITOK);
2834 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2835 	ma->v[ma->len].iov_len = strlen(name) + 1;
2836 	ma->len++;
2837 
2838 	sb = sbuf_new_auto();
2839 	va_start(ap, fmt);
2840 	sbuf_vprintf(sb, fmt, ap);
2841 	va_end(ap);
2842 	sbuf_finish(sb);
2843 	len = sbuf_len(sb) + 1;
2844 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2845 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2846 	bcopy(sbuf_data(sb), maa + 1, len);
2847 	sbuf_delete(sb);
2848 
2849 	ma->v[ma->len].iov_base = maa + 1;
2850 	ma->v[ma->len].iov_len = len;
2851 	ma->len++;
2852 
2853 	return (ma);
2854 }
2855 
2856 /*
2857  * Add an argument which is a userland string.
2858  */
2859 struct mntarg *
2860 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2861 {
2862 	struct mntaarg *maa;
2863 	char *tbuf;
2864 
2865 	if (val == NULL)
2866 		return (ma);
2867 	if (ma == NULL) {
2868 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2869 		SLIST_INIT(&ma->list);
2870 	}
2871 	if (ma->error)
2872 		return (ma);
2873 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2874 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2875 	tbuf = (void *)(maa + 1);
2876 	ma->error = copyinstr(val, tbuf, len, NULL);
2877 	return (mount_arg(ma, name, tbuf, -1));
2878 }
2879 
2880 /*
2881  * Plain argument.
2882  *
2883  * If length is -1, treat value as a C string.
2884  */
2885 struct mntarg *
2886 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2887 {
2888 
2889 	if (ma == NULL) {
2890 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2891 		SLIST_INIT(&ma->list);
2892 	}
2893 	if (ma->error)
2894 		return (ma);
2895 
2896 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2897 	    M_MOUNT, M_WAITOK);
2898 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2899 	ma->v[ma->len].iov_len = strlen(name) + 1;
2900 	ma->len++;
2901 
2902 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2903 	if (len < 0)
2904 		ma->v[ma->len].iov_len = strlen(val) + 1;
2905 	else
2906 		ma->v[ma->len].iov_len = len;
2907 	ma->len++;
2908 	return (ma);
2909 }
2910 
2911 /*
2912  * Free a mntarg structure
2913  */
2914 static void
2915 free_mntarg(struct mntarg *ma)
2916 {
2917 	struct mntaarg *maa;
2918 
2919 	while (!SLIST_EMPTY(&ma->list)) {
2920 		maa = SLIST_FIRST(&ma->list);
2921 		SLIST_REMOVE_HEAD(&ma->list, next);
2922 		free(maa, M_MOUNT);
2923 	}
2924 	free(ma->v, M_MOUNT);
2925 	free(ma, M_MOUNT);
2926 }
2927 
2928 /*
2929  * Mount a filesystem
2930  */
2931 int
2932 kernel_mount(struct mntarg *ma, uint64_t flags)
2933 {
2934 	struct uio auio;
2935 	int error;
2936 
2937 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2938 	KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2939 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2940 
2941 	error = ma->error;
2942 	if (error == 0) {
2943 		auio.uio_iov = ma->v;
2944 		auio.uio_iovcnt = ma->len;
2945 		auio.uio_segflg = UIO_SYSSPACE;
2946 		error = vfs_donmount(curthread, flags, &auio);
2947 	}
2948 	free_mntarg(ma);
2949 	return (error);
2950 }
2951 
2952 /* Map from mount options to printable formats. */
2953 static struct mntoptnames optnames[] = {
2954 	MNTOPT_NAMES
2955 };
2956 
2957 #define DEVCTL_LEN 1024
2958 static void
2959 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2960 {
2961 	const uint8_t *cp;
2962 	struct mntoptnames *fp;
2963 	struct sbuf sb;
2964 	struct statfs *sfp = &mp->mnt_stat;
2965 	char *buf;
2966 
2967 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2968 	if (buf == NULL)
2969 		return;
2970 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2971 	sbuf_cpy(&sb, "mount-point=\"");
2972 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2973 	sbuf_cat(&sb, "\" mount-dev=\"");
2974 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2975 	sbuf_cat(&sb, "\" mount-type=\"");
2976 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2977 	sbuf_cat(&sb, "\" fsid=0x");
2978 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2979 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2980 		sbuf_printf(&sb, "%02x", cp[i]);
2981 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2982 	for (fp = optnames; fp->o_opt != 0; fp++) {
2983 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2984 			sbuf_cat(&sb, fp->o_name);
2985 			sbuf_putc(&sb, ';');
2986 		}
2987 	}
2988 	sbuf_putc(&sb, '"');
2989 	sbuf_finish(&sb);
2990 
2991 	/*
2992 	 * Options are not published because the form of the options depends on
2993 	 * the file system and may include binary data. In addition, they don't
2994 	 * necessarily provide enough useful information to be actionable when
2995 	 * devd processes them.
2996 	 */
2997 
2998 	if (sbuf_error(&sb) == 0)
2999 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
3000 	sbuf_delete(&sb);
3001 	free(buf, M_MOUNT);
3002 }
3003 
3004 /*
3005  * Force remount specified mount point to read-only.  The argument
3006  * must be busied to avoid parallel unmount attempts.
3007  *
3008  * Intended use is to prevent further writes if some metadata
3009  * inconsistency is detected.  Note that the function still flushes
3010  * all cached metadata and data for the mount point, which might be
3011  * not always suitable.
3012  */
3013 int
3014 vfs_remount_ro(struct mount *mp)
3015 {
3016 	struct vfsoptlist *opts;
3017 	struct vfsopt *opt;
3018 	struct vnode *vp_covered, *rootvp;
3019 	int error;
3020 
3021 	vfs_op_enter(mp);
3022 	KASSERT(mp->mnt_lockref > 0,
3023 	    ("vfs_remount_ro: mp %p is not busied", mp));
3024 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
3025 	    ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
3026 
3027 	rootvp = NULL;
3028 	vp_covered = mp->mnt_vnodecovered;
3029 	error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
3030 	if (error != 0) {
3031 		vfs_op_exit(mp);
3032 		return (error);
3033 	}
3034 	VI_LOCK(vp_covered);
3035 	if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
3036 		VI_UNLOCK(vp_covered);
3037 		vput(vp_covered);
3038 		vfs_op_exit(mp);
3039 		return (EBUSY);
3040 	}
3041 	vp_covered->v_iflag |= VI_MOUNT;
3042 	VI_UNLOCK(vp_covered);
3043 	vn_seqc_write_begin(vp_covered);
3044 
3045 	MNT_ILOCK(mp);
3046 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
3047 		MNT_IUNLOCK(mp);
3048 		error = EBUSY;
3049 		goto out;
3050 	}
3051 	mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
3052 	rootvp = vfs_cache_root_clear(mp);
3053 	MNT_IUNLOCK(mp);
3054 
3055 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
3056 	TAILQ_INIT(opts);
3057 	opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
3058 	opt->name = strdup("ro", M_MOUNT);
3059 	opt->value = NULL;
3060 	TAILQ_INSERT_TAIL(opts, opt, link);
3061 	vfs_mergeopts(opts, mp->mnt_opt);
3062 	mp->mnt_optnew = opts;
3063 
3064 	error = VFS_MOUNT(mp);
3065 
3066 	if (error == 0) {
3067 		MNT_ILOCK(mp);
3068 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
3069 		MNT_IUNLOCK(mp);
3070 		vfs_deallocate_syncvnode(mp);
3071 		if (mp->mnt_opt != NULL)
3072 			vfs_freeopts(mp->mnt_opt);
3073 		mp->mnt_opt = mp->mnt_optnew;
3074 	} else {
3075 		MNT_ILOCK(mp);
3076 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
3077 		MNT_IUNLOCK(mp);
3078 		vfs_freeopts(mp->mnt_optnew);
3079 	}
3080 	mp->mnt_optnew = NULL;
3081 
3082 out:
3083 	vfs_op_exit(mp);
3084 	VI_LOCK(vp_covered);
3085 	vp_covered->v_iflag &= ~VI_MOUNT;
3086 	VI_UNLOCK(vp_covered);
3087 	vput(vp_covered);
3088 	vn_seqc_write_end(vp_covered);
3089 	if (rootvp != NULL) {
3090 		vn_seqc_write_end(rootvp);
3091 		vrele(rootvp);
3092 	}
3093 	return (error);
3094 }
3095 
3096 /*
3097  * Suspend write operations on all local writeable filesystems.  Does
3098  * full sync of them in the process.
3099  *
3100  * Iterate over the mount points in reverse order, suspending most
3101  * recently mounted filesystems first.  It handles a case where a
3102  * filesystem mounted from a md(4) vnode-backed device should be
3103  * suspended before the filesystem that owns the vnode.
3104  */
3105 void
3106 suspend_all_fs(void)
3107 {
3108 	struct mount *mp;
3109 	int error;
3110 
3111 	mtx_lock(&mountlist_mtx);
3112 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
3113 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
3114 		if (error != 0)
3115 			continue;
3116 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
3117 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
3118 			mtx_lock(&mountlist_mtx);
3119 			vfs_unbusy(mp);
3120 			continue;
3121 		}
3122 		error = vfs_write_suspend(mp, 0);
3123 		if (error == 0) {
3124 			MNT_ILOCK(mp);
3125 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
3126 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
3127 			MNT_IUNLOCK(mp);
3128 			mtx_lock(&mountlist_mtx);
3129 		} else {
3130 			printf("suspend of %s failed, error %d\n",
3131 			    mp->mnt_stat.f_mntonname, error);
3132 			mtx_lock(&mountlist_mtx);
3133 			vfs_unbusy(mp);
3134 		}
3135 	}
3136 	mtx_unlock(&mountlist_mtx);
3137 }
3138 
3139 /*
3140  * Clone the mnt_exjail field to a new mount point.
3141  */
3142 void
3143 vfs_exjail_clone(struct mount *inmp, struct mount *outmp)
3144 {
3145 	struct ucred *cr;
3146 	struct prison *pr;
3147 
3148 	MNT_ILOCK(inmp);
3149 	cr = inmp->mnt_exjail;
3150 	if (cr != NULL) {
3151 		crhold(cr);
3152 		MNT_IUNLOCK(inmp);
3153 		pr = cr->cr_prison;
3154 		sx_slock(&allprison_lock);
3155 		if (!prison_isalive(pr)) {
3156 			sx_sunlock(&allprison_lock);
3157 			crfree(cr);
3158 			return;
3159 		}
3160 		MNT_ILOCK(outmp);
3161 		if (outmp->mnt_exjail == NULL) {
3162 			outmp->mnt_exjail = cr;
3163 			atomic_add_int(&pr->pr_exportcnt, 1);
3164 			cr = NULL;
3165 		}
3166 		MNT_IUNLOCK(outmp);
3167 		sx_sunlock(&allprison_lock);
3168 		if (cr != NULL)
3169 			crfree(cr);
3170 	} else
3171 		MNT_IUNLOCK(inmp);
3172 }
3173 
3174 void
3175 resume_all_fs(void)
3176 {
3177 	struct mount *mp;
3178 
3179 	mtx_lock(&mountlist_mtx);
3180 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3181 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3182 			continue;
3183 		mtx_unlock(&mountlist_mtx);
3184 		MNT_ILOCK(mp);
3185 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3186 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3187 		MNT_IUNLOCK(mp);
3188 		vfs_write_resume(mp, 0);
3189 		mtx_lock(&mountlist_mtx);
3190 		vfs_unbusy(mp);
3191 	}
3192 	mtx_unlock(&mountlist_mtx);
3193 }
3194