xref: /freebsd/sys/kern/vfs_mount.c (revision d5b0e70f7e04d971691517ce1304d86a1e367e2e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1999-2004 Poul-Henning Kamp
5  * Copyright (c) 1999 Michael Smith
6  * Copyright (c) 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/conf.h>
44 #include <sys/smp.h>
45 #include <sys/devctl.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
48 #include <sys/jail.h>
49 #include <sys/kernel.h>
50 #include <sys/ktr.h>
51 #include <sys/libkern.h>
52 #include <sys/limits.h>
53 #include <sys/malloc.h>
54 #include <sys/mount.h>
55 #include <sys/mutex.h>
56 #include <sys/namei.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/filedesc.h>
60 #include <sys/reboot.h>
61 #include <sys/sbuf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysproto.h>
64 #include <sys/sx.h>
65 #include <sys/sysctl.h>
66 #include <sys/systm.h>
67 #include <sys/taskqueue.h>
68 #include <sys/vnode.h>
69 #include <vm/uma.h>
70 
71 #include <geom/geom.h>
72 
73 #include <machine/stdarg.h>
74 
75 #include <security/audit/audit.h>
76 #include <security/mac/mac_framework.h>
77 
78 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
79 
80 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
81 		    uint64_t fsflags, struct vfsoptlist **optlist);
82 static void	free_mntarg(struct mntarg *ma);
83 
84 static int	usermount = 0;
85 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
86     "Unprivileged users may mount and unmount file systems");
87 
88 static bool	default_autoro = false;
89 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
90     "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
91 
92 static bool	recursive_forced_unmount = false;
93 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW,
94     &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts"
95     " when a file system is forcibly unmounted");
96 
97 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount,
98     CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls");
99 
100 static unsigned int	deferred_unmount_retry_limit = 10;
101 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW,
102     &deferred_unmount_retry_limit, 0,
103     "Maximum number of retries for deferred unmount failure");
104 
105 static int	deferred_unmount_retry_delay_hz;
106 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW,
107     &deferred_unmount_retry_delay_hz, 0,
108     "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount");
109 
110 static int	deferred_unmount_total_retries = 0;
111 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD,
112     &deferred_unmount_total_retries, 0,
113     "Total number of retried deferred unmounts");
114 
115 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
116 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
117 static uma_zone_t mount_zone;
118 
119 /* List of mounted filesystems. */
120 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
121 
122 /* For any iteration/modification of mountlist */
123 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
124 
125 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
126 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
127 
128 static void vfs_deferred_unmount(void *arg, int pending);
129 static struct timeout_task deferred_unmount_task;
130 static struct mtx deferred_unmount_lock;
131 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount",
132     MTX_DEF);
133 static STAILQ_HEAD(, mount) deferred_unmount_list =
134     STAILQ_HEAD_INITIALIZER(deferred_unmount_list);
135 TASKQUEUE_DEFINE_THREAD(deferred_unmount);
136 
137 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
138 
139 /*
140  * Global opts, taken by all filesystems
141  */
142 static const char *global_opts[] = {
143 	"errmsg",
144 	"fstype",
145 	"fspath",
146 	"ro",
147 	"rw",
148 	"nosuid",
149 	"noexec",
150 	NULL
151 };
152 
153 static int
154 mount_init(void *mem, int size, int flags)
155 {
156 	struct mount *mp;
157 
158 	mp = (struct mount *)mem;
159 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
160 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
161 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
162 	mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
163 	mp->mnt_ref = 0;
164 	mp->mnt_vfs_ops = 1;
165 	mp->mnt_rootvnode = NULL;
166 	return (0);
167 }
168 
169 static void
170 mount_fini(void *mem, int size)
171 {
172 	struct mount *mp;
173 
174 	mp = (struct mount *)mem;
175 	uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
176 	lockdestroy(&mp->mnt_explock);
177 	mtx_destroy(&mp->mnt_listmtx);
178 	mtx_destroy(&mp->mnt_mtx);
179 }
180 
181 static void
182 vfs_mount_init(void *dummy __unused)
183 {
184 	TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task,
185 	    0, vfs_deferred_unmount, NULL);
186 	deferred_unmount_retry_delay_hz = hz;
187 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
188 	    NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
189 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
190 }
191 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
192 
193 /*
194  * ---------------------------------------------------------------------
195  * Functions for building and sanitizing the mount options
196  */
197 
198 /* Remove one mount option. */
199 static void
200 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
201 {
202 
203 	TAILQ_REMOVE(opts, opt, link);
204 	free(opt->name, M_MOUNT);
205 	if (opt->value != NULL)
206 		free(opt->value, M_MOUNT);
207 	free(opt, M_MOUNT);
208 }
209 
210 /* Release all resources related to the mount options. */
211 void
212 vfs_freeopts(struct vfsoptlist *opts)
213 {
214 	struct vfsopt *opt;
215 
216 	while (!TAILQ_EMPTY(opts)) {
217 		opt = TAILQ_FIRST(opts);
218 		vfs_freeopt(opts, opt);
219 	}
220 	free(opts, M_MOUNT);
221 }
222 
223 void
224 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
225 {
226 	struct vfsopt *opt, *temp;
227 
228 	if (opts == NULL)
229 		return;
230 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
231 		if (strcmp(opt->name, name) == 0)
232 			vfs_freeopt(opts, opt);
233 	}
234 }
235 
236 static int
237 vfs_isopt_ro(const char *opt)
238 {
239 
240 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
241 	    strcmp(opt, "norw") == 0)
242 		return (1);
243 	return (0);
244 }
245 
246 static int
247 vfs_isopt_rw(const char *opt)
248 {
249 
250 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
251 		return (1);
252 	return (0);
253 }
254 
255 /*
256  * Check if options are equal (with or without the "no" prefix).
257  */
258 static int
259 vfs_equalopts(const char *opt1, const char *opt2)
260 {
261 	char *p;
262 
263 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
264 	if (strcmp(opt1, opt2) == 0)
265 		return (1);
266 	/* "noopt" vs. "opt" */
267 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
268 		return (1);
269 	/* "opt" vs. "noopt" */
270 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
271 		return (1);
272 	while ((p = strchr(opt1, '.')) != NULL &&
273 	    !strncmp(opt1, opt2, ++p - opt1)) {
274 		opt2 += p - opt1;
275 		opt1 = p;
276 		/* "foo.noopt" vs. "foo.opt" */
277 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
278 			return (1);
279 		/* "foo.opt" vs. "foo.noopt" */
280 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
281 			return (1);
282 	}
283 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
284 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
285 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
286 		return (1);
287 	return (0);
288 }
289 
290 /*
291  * If a mount option is specified several times,
292  * (with or without the "no" prefix) only keep
293  * the last occurrence of it.
294  */
295 static void
296 vfs_sanitizeopts(struct vfsoptlist *opts)
297 {
298 	struct vfsopt *opt, *opt2, *tmp;
299 
300 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
301 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
302 		while (opt2 != NULL) {
303 			if (vfs_equalopts(opt->name, opt2->name)) {
304 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
305 				vfs_freeopt(opts, opt2);
306 				opt2 = tmp;
307 			} else {
308 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
309 			}
310 		}
311 	}
312 }
313 
314 /*
315  * Build a linked list of mount options from a struct uio.
316  */
317 int
318 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
319 {
320 	struct vfsoptlist *opts;
321 	struct vfsopt *opt;
322 	size_t memused, namelen, optlen;
323 	unsigned int i, iovcnt;
324 	int error;
325 
326 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
327 	TAILQ_INIT(opts);
328 	memused = 0;
329 	iovcnt = auio->uio_iovcnt;
330 	for (i = 0; i < iovcnt; i += 2) {
331 		namelen = auio->uio_iov[i].iov_len;
332 		optlen = auio->uio_iov[i + 1].iov_len;
333 		memused += sizeof(struct vfsopt) + optlen + namelen;
334 		/*
335 		 * Avoid consuming too much memory, and attempts to overflow
336 		 * memused.
337 		 */
338 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
339 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
340 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
341 			error = EINVAL;
342 			goto bad;
343 		}
344 
345 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
346 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
347 		opt->value = NULL;
348 		opt->len = 0;
349 		opt->pos = i / 2;
350 		opt->seen = 0;
351 
352 		/*
353 		 * Do this early, so jumps to "bad" will free the current
354 		 * option.
355 		 */
356 		TAILQ_INSERT_TAIL(opts, opt, link);
357 
358 		if (auio->uio_segflg == UIO_SYSSPACE) {
359 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
360 		} else {
361 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
362 			    namelen);
363 			if (error)
364 				goto bad;
365 		}
366 		/* Ensure names are null-terminated strings. */
367 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
368 			error = EINVAL;
369 			goto bad;
370 		}
371 		if (optlen != 0) {
372 			opt->len = optlen;
373 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
374 			if (auio->uio_segflg == UIO_SYSSPACE) {
375 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
376 				    optlen);
377 			} else {
378 				error = copyin(auio->uio_iov[i + 1].iov_base,
379 				    opt->value, optlen);
380 				if (error)
381 					goto bad;
382 			}
383 		}
384 	}
385 	vfs_sanitizeopts(opts);
386 	*options = opts;
387 	return (0);
388 bad:
389 	vfs_freeopts(opts);
390 	return (error);
391 }
392 
393 /*
394  * Merge the old mount options with the new ones passed
395  * in the MNT_UPDATE case.
396  *
397  * XXX: This function will keep a "nofoo" option in the new
398  * options.  E.g, if the option's canonical name is "foo",
399  * "nofoo" ends up in the mount point's active options.
400  */
401 static void
402 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
403 {
404 	struct vfsopt *opt, *new;
405 
406 	TAILQ_FOREACH(opt, oldopts, link) {
407 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
408 		new->name = strdup(opt->name, M_MOUNT);
409 		if (opt->len != 0) {
410 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
411 			bcopy(opt->value, new->value, opt->len);
412 		} else
413 			new->value = NULL;
414 		new->len = opt->len;
415 		new->seen = opt->seen;
416 		TAILQ_INSERT_HEAD(toopts, new, link);
417 	}
418 	vfs_sanitizeopts(toopts);
419 }
420 
421 /*
422  * Mount a filesystem.
423  */
424 #ifndef _SYS_SYSPROTO_H_
425 struct nmount_args {
426 	struct iovec *iovp;
427 	unsigned int iovcnt;
428 	int flags;
429 };
430 #endif
431 int
432 sys_nmount(struct thread *td, struct nmount_args *uap)
433 {
434 	struct uio *auio;
435 	int error;
436 	u_int iovcnt;
437 	uint64_t flags;
438 
439 	/*
440 	 * Mount flags are now 64-bits. On 32-bit archtectures only
441 	 * 32-bits are passed in, but from here on everything handles
442 	 * 64-bit flags correctly.
443 	 */
444 	flags = uap->flags;
445 
446 	AUDIT_ARG_FFLAGS(flags);
447 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
448 	    uap->iovp, uap->iovcnt, flags);
449 
450 	/*
451 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
452 	 * userspace to set this flag, but we must filter it out if we want
453 	 * MNT_UPDATE on the root file system to work.
454 	 * MNT_ROOTFS should only be set by the kernel when mounting its
455 	 * root file system.
456 	 */
457 	flags &= ~MNT_ROOTFS;
458 
459 	iovcnt = uap->iovcnt;
460 	/*
461 	 * Check that we have an even number of iovec's
462 	 * and that we have at least two options.
463 	 */
464 	if ((iovcnt & 1) || (iovcnt < 4)) {
465 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
466 		    uap->iovcnt);
467 		return (EINVAL);
468 	}
469 
470 	error = copyinuio(uap->iovp, iovcnt, &auio);
471 	if (error) {
472 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
473 		    __func__, error);
474 		return (error);
475 	}
476 	error = vfs_donmount(td, flags, auio);
477 
478 	free(auio, M_IOV);
479 	return (error);
480 }
481 
482 /*
483  * ---------------------------------------------------------------------
484  * Various utility functions
485  */
486 
487 /*
488  * Get a reference on a mount point from a vnode.
489  *
490  * The vnode is allowed to be passed unlocked and race against dooming. Note in
491  * such case there are no guarantees the referenced mount point will still be
492  * associated with it after the function returns.
493  */
494 struct mount *
495 vfs_ref_from_vp(struct vnode *vp)
496 {
497 	struct mount *mp;
498 	struct mount_pcpu *mpcpu;
499 
500 	mp = atomic_load_ptr(&vp->v_mount);
501 	if (__predict_false(mp == NULL)) {
502 		return (mp);
503 	}
504 	if (vfs_op_thread_enter(mp, mpcpu)) {
505 		if (__predict_true(mp == vp->v_mount)) {
506 			vfs_mp_count_add_pcpu(mpcpu, ref, 1);
507 			vfs_op_thread_exit(mp, mpcpu);
508 		} else {
509 			vfs_op_thread_exit(mp, mpcpu);
510 			mp = NULL;
511 		}
512 	} else {
513 		MNT_ILOCK(mp);
514 		if (mp == vp->v_mount) {
515 			MNT_REF(mp);
516 			MNT_IUNLOCK(mp);
517 		} else {
518 			MNT_IUNLOCK(mp);
519 			mp = NULL;
520 		}
521 	}
522 	return (mp);
523 }
524 
525 void
526 vfs_ref(struct mount *mp)
527 {
528 	struct mount_pcpu *mpcpu;
529 
530 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
531 	if (vfs_op_thread_enter(mp, mpcpu)) {
532 		vfs_mp_count_add_pcpu(mpcpu, ref, 1);
533 		vfs_op_thread_exit(mp, mpcpu);
534 		return;
535 	}
536 
537 	MNT_ILOCK(mp);
538 	MNT_REF(mp);
539 	MNT_IUNLOCK(mp);
540 }
541 
542 /*
543  * Register ump as an upper mount of the mount associated with
544  * vnode vp.  This registration will be tracked through
545  * mount_upper_node upper, which should be allocated by the
546  * caller and stored in per-mount data associated with mp.
547  *
548  * If successful, this function will return the mount associated
549  * with vp, and will ensure that it cannot be unmounted until
550  * ump has been unregistered as one of its upper mounts.
551  *
552  * Upon failure this function will return NULL.
553  */
554 struct mount *
555 vfs_register_upper_from_vp(struct vnode *vp, struct mount *ump,
556     struct mount_upper_node *upper)
557 {
558 	struct mount *mp;
559 
560 	mp = atomic_load_ptr(&vp->v_mount);
561 	if (mp == NULL)
562 		return (NULL);
563 	MNT_ILOCK(mp);
564 	if (mp != vp->v_mount ||
565 	    ((mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_RECURSE)) != 0)) {
566 		MNT_IUNLOCK(mp);
567 		return (NULL);
568 	}
569 	KASSERT(ump != mp, ("upper and lower mounts are identical"));
570 	upper->mp = ump;
571 	MNT_REF(mp);
572 	TAILQ_INSERT_TAIL(&mp->mnt_uppers, upper, mnt_upper_link);
573 	MNT_IUNLOCK(mp);
574 	return (mp);
575 }
576 
577 /*
578  * Register upper mount ump to receive vnode unlink/reclaim
579  * notifications from lower mount mp. This registration will
580  * be tracked through mount_upper_node upper, which should be
581  * allocated by the caller and stored in per-mount data
582  * associated with mp.
583  *
584  * ump must already be registered as an upper mount of mp
585  * through a call to vfs_register_upper_from_vp().
586  */
587 void
588 vfs_register_for_notification(struct mount *mp, struct mount *ump,
589     struct mount_upper_node *upper)
590 {
591 	upper->mp = ump;
592 	MNT_ILOCK(mp);
593 	TAILQ_INSERT_TAIL(&mp->mnt_notify, upper, mnt_upper_link);
594 	MNT_IUNLOCK(mp);
595 }
596 
597 static void
598 vfs_drain_upper_locked(struct mount *mp)
599 {
600 	mtx_assert(MNT_MTX(mp), MA_OWNED);
601 	while (mp->mnt_upper_pending != 0) {
602 		mp->mnt_kern_flag |= MNTK_UPPER_WAITER;
603 		msleep(&mp->mnt_uppers, MNT_MTX(mp), 0, "mntupw", 0);
604 	}
605 }
606 
607 /*
608  * Undo a previous call to vfs_register_for_notification().
609  * The mount represented by upper must be currently registered
610  * as an upper mount for mp.
611  */
612 void
613 vfs_unregister_for_notification(struct mount *mp,
614     struct mount_upper_node *upper)
615 {
616 	MNT_ILOCK(mp);
617 	vfs_drain_upper_locked(mp);
618 	TAILQ_REMOVE(&mp->mnt_notify, upper, mnt_upper_link);
619 	MNT_IUNLOCK(mp);
620 }
621 
622 /*
623  * Undo a previous call to vfs_register_upper_from_vp().
624  * This must be done before mp can be unmounted.
625  */
626 void
627 vfs_unregister_upper(struct mount *mp, struct mount_upper_node *upper)
628 {
629 	MNT_ILOCK(mp);
630 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
631 	    ("registered upper with pending unmount"));
632 	vfs_drain_upper_locked(mp);
633 	TAILQ_REMOVE(&mp->mnt_uppers, upper, mnt_upper_link);
634 	if ((mp->mnt_kern_flag & MNTK_TASKQUEUE_WAITER) != 0 &&
635 	    TAILQ_EMPTY(&mp->mnt_uppers)) {
636 		mp->mnt_kern_flag &= ~MNTK_TASKQUEUE_WAITER;
637 		wakeup(&mp->mnt_taskqueue_link);
638 	}
639 	MNT_REL(mp);
640 	MNT_IUNLOCK(mp);
641 }
642 
643 void
644 vfs_rel(struct mount *mp)
645 {
646 	struct mount_pcpu *mpcpu;
647 
648 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
649 	if (vfs_op_thread_enter(mp, mpcpu)) {
650 		vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
651 		vfs_op_thread_exit(mp, mpcpu);
652 		return;
653 	}
654 
655 	MNT_ILOCK(mp);
656 	MNT_REL(mp);
657 	MNT_IUNLOCK(mp);
658 }
659 
660 /*
661  * Allocate and initialize the mount point struct.
662  */
663 struct mount *
664 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
665     struct ucred *cred)
666 {
667 	struct mount *mp;
668 
669 	mp = uma_zalloc(mount_zone, M_WAITOK);
670 	bzero(&mp->mnt_startzero,
671 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
672 	mp->mnt_kern_flag = 0;
673 	mp->mnt_flag = 0;
674 	mp->mnt_rootvnode = NULL;
675 	mp->mnt_vnodecovered = NULL;
676 	mp->mnt_op = NULL;
677 	mp->mnt_vfc = NULL;
678 	TAILQ_INIT(&mp->mnt_nvnodelist);
679 	mp->mnt_nvnodelistsize = 0;
680 	TAILQ_INIT(&mp->mnt_lazyvnodelist);
681 	mp->mnt_lazyvnodelistsize = 0;
682 	MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 &&
683 	    mp->mnt_writeopcount == 0, mp);
684 	MPASSERT(mp->mnt_vfs_ops == 1, mp,
685 	    ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
686 	(void) vfs_busy(mp, MBF_NOWAIT);
687 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
688 	mp->mnt_op = vfsp->vfc_vfsops;
689 	mp->mnt_vfc = vfsp;
690 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
691 	mp->mnt_gen++;
692 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
693 	mp->mnt_vnodecovered = vp;
694 	mp->mnt_cred = crdup(cred);
695 	mp->mnt_stat.f_owner = cred->cr_uid;
696 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
697 	mp->mnt_iosize_max = DFLTPHYS;
698 #ifdef MAC
699 	mac_mount_init(mp);
700 	mac_mount_create(cred, mp);
701 #endif
702 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
703 	mp->mnt_upper_pending = 0;
704 	TAILQ_INIT(&mp->mnt_uppers);
705 	TAILQ_INIT(&mp->mnt_notify);
706 	mp->mnt_taskqueue_flags = 0;
707 	mp->mnt_unmount_retries = 0;
708 	return (mp);
709 }
710 
711 /*
712  * Destroy the mount struct previously allocated by vfs_mount_alloc().
713  */
714 void
715 vfs_mount_destroy(struct mount *mp)
716 {
717 
718 	MPPASS(mp->mnt_vfs_ops != 0, mp);
719 
720 	vfs_assert_mount_counters(mp);
721 
722 	MNT_ILOCK(mp);
723 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
724 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
725 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
726 		wakeup(mp);
727 	}
728 	while (mp->mnt_ref)
729 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
730 	KASSERT(mp->mnt_ref == 0,
731 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
732 	    __FILE__, __LINE__));
733 	MPPASS(mp->mnt_writeopcount == 0, mp);
734 	MPPASS(mp->mnt_secondary_writes == 0, mp);
735 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
736 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
737 		struct vnode *vp;
738 
739 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
740 			vn_printf(vp, "dangling vnode ");
741 		panic("unmount: dangling vnode");
742 	}
743 	KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending"));
744 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
745 	KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify"));
746 	MPPASS(mp->mnt_nvnodelistsize == 0, mp);
747 	MPPASS(mp->mnt_lazyvnodelistsize == 0, mp);
748 	MPPASS(mp->mnt_lockref == 0, mp);
749 	MNT_IUNLOCK(mp);
750 
751 	MPASSERT(mp->mnt_vfs_ops == 1, mp,
752 	    ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
753 
754 	MPASSERT(mp->mnt_rootvnode == NULL, mp,
755 	    ("mount point still has a root vnode %p", mp->mnt_rootvnode));
756 
757 	if (mp->mnt_vnodecovered != NULL)
758 		vrele(mp->mnt_vnodecovered);
759 #ifdef MAC
760 	mac_mount_destroy(mp);
761 #endif
762 	if (mp->mnt_opt != NULL)
763 		vfs_freeopts(mp->mnt_opt);
764 	crfree(mp->mnt_cred);
765 	uma_zfree(mount_zone, mp);
766 }
767 
768 static bool
769 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
770 {
771 	/* This is an upgrade of an exisiting mount. */
772 	if ((fsflags & MNT_UPDATE) != 0)
773 		return (false);
774 	/* This is already an R/O mount. */
775 	if ((fsflags & MNT_RDONLY) != 0)
776 		return (false);
777 
778 	switch (error) {
779 	case ENODEV:	/* generic, geom, ... */
780 	case EACCES:	/* cam/scsi, ... */
781 	case EROFS:	/* md, mmcsd, ... */
782 		/*
783 		 * These errors can be returned by the storage layer to signal
784 		 * that the media is read-only.  No harm in the R/O mount
785 		 * attempt if the error was returned for some other reason.
786 		 */
787 		return (true);
788 	default:
789 		return (false);
790 	}
791 }
792 
793 int
794 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
795 {
796 	struct vfsoptlist *optlist;
797 	struct vfsopt *opt, *tmp_opt;
798 	char *fstype, *fspath, *errmsg;
799 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
800 	bool autoro;
801 
802 	errmsg = fspath = NULL;
803 	errmsg_len = fspathlen = 0;
804 	errmsg_pos = -1;
805 	autoro = default_autoro;
806 
807 	error = vfs_buildopts(fsoptions, &optlist);
808 	if (error)
809 		return (error);
810 
811 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
812 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
813 
814 	/*
815 	 * We need these two options before the others,
816 	 * and they are mandatory for any filesystem.
817 	 * Ensure they are NUL terminated as well.
818 	 */
819 	fstypelen = 0;
820 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
821 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
822 		error = EINVAL;
823 		if (errmsg != NULL)
824 			strncpy(errmsg, "Invalid fstype", errmsg_len);
825 		goto bail;
826 	}
827 	fspathlen = 0;
828 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
829 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
830 		error = EINVAL;
831 		if (errmsg != NULL)
832 			strncpy(errmsg, "Invalid fspath", errmsg_len);
833 		goto bail;
834 	}
835 
836 	/*
837 	 * We need to see if we have the "update" option
838 	 * before we call vfs_domount(), since vfs_domount() has special
839 	 * logic based on MNT_UPDATE.  This is very important
840 	 * when we want to update the root filesystem.
841 	 */
842 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
843 		int do_freeopt = 0;
844 
845 		if (strcmp(opt->name, "update") == 0) {
846 			fsflags |= MNT_UPDATE;
847 			do_freeopt = 1;
848 		}
849 		else if (strcmp(opt->name, "async") == 0)
850 			fsflags |= MNT_ASYNC;
851 		else if (strcmp(opt->name, "force") == 0) {
852 			fsflags |= MNT_FORCE;
853 			do_freeopt = 1;
854 		}
855 		else if (strcmp(opt->name, "reload") == 0) {
856 			fsflags |= MNT_RELOAD;
857 			do_freeopt = 1;
858 		}
859 		else if (strcmp(opt->name, "multilabel") == 0)
860 			fsflags |= MNT_MULTILABEL;
861 		else if (strcmp(opt->name, "noasync") == 0)
862 			fsflags &= ~MNT_ASYNC;
863 		else if (strcmp(opt->name, "noatime") == 0)
864 			fsflags |= MNT_NOATIME;
865 		else if (strcmp(opt->name, "atime") == 0) {
866 			free(opt->name, M_MOUNT);
867 			opt->name = strdup("nonoatime", M_MOUNT);
868 		}
869 		else if (strcmp(opt->name, "noclusterr") == 0)
870 			fsflags |= MNT_NOCLUSTERR;
871 		else if (strcmp(opt->name, "clusterr") == 0) {
872 			free(opt->name, M_MOUNT);
873 			opt->name = strdup("nonoclusterr", M_MOUNT);
874 		}
875 		else if (strcmp(opt->name, "noclusterw") == 0)
876 			fsflags |= MNT_NOCLUSTERW;
877 		else if (strcmp(opt->name, "clusterw") == 0) {
878 			free(opt->name, M_MOUNT);
879 			opt->name = strdup("nonoclusterw", M_MOUNT);
880 		}
881 		else if (strcmp(opt->name, "noexec") == 0)
882 			fsflags |= MNT_NOEXEC;
883 		else if (strcmp(opt->name, "exec") == 0) {
884 			free(opt->name, M_MOUNT);
885 			opt->name = strdup("nonoexec", M_MOUNT);
886 		}
887 		else if (strcmp(opt->name, "nosuid") == 0)
888 			fsflags |= MNT_NOSUID;
889 		else if (strcmp(opt->name, "suid") == 0) {
890 			free(opt->name, M_MOUNT);
891 			opt->name = strdup("nonosuid", M_MOUNT);
892 		}
893 		else if (strcmp(opt->name, "nosymfollow") == 0)
894 			fsflags |= MNT_NOSYMFOLLOW;
895 		else if (strcmp(opt->name, "symfollow") == 0) {
896 			free(opt->name, M_MOUNT);
897 			opt->name = strdup("nonosymfollow", M_MOUNT);
898 		}
899 		else if (strcmp(opt->name, "noro") == 0) {
900 			fsflags &= ~MNT_RDONLY;
901 			autoro = false;
902 		}
903 		else if (strcmp(opt->name, "rw") == 0) {
904 			fsflags &= ~MNT_RDONLY;
905 			autoro = false;
906 		}
907 		else if (strcmp(opt->name, "ro") == 0) {
908 			fsflags |= MNT_RDONLY;
909 			autoro = false;
910 		}
911 		else if (strcmp(opt->name, "rdonly") == 0) {
912 			free(opt->name, M_MOUNT);
913 			opt->name = strdup("ro", M_MOUNT);
914 			fsflags |= MNT_RDONLY;
915 			autoro = false;
916 		}
917 		else if (strcmp(opt->name, "autoro") == 0) {
918 			do_freeopt = 1;
919 			autoro = true;
920 		}
921 		else if (strcmp(opt->name, "suiddir") == 0)
922 			fsflags |= MNT_SUIDDIR;
923 		else if (strcmp(opt->name, "sync") == 0)
924 			fsflags |= MNT_SYNCHRONOUS;
925 		else if (strcmp(opt->name, "union") == 0)
926 			fsflags |= MNT_UNION;
927 		else if (strcmp(opt->name, "automounted") == 0) {
928 			fsflags |= MNT_AUTOMOUNTED;
929 			do_freeopt = 1;
930 		} else if (strcmp(opt->name, "nocover") == 0) {
931 			fsflags |= MNT_NOCOVER;
932 			do_freeopt = 1;
933 		} else if (strcmp(opt->name, "cover") == 0) {
934 			fsflags &= ~MNT_NOCOVER;
935 			do_freeopt = 1;
936 		} else if (strcmp(opt->name, "emptydir") == 0) {
937 			fsflags |= MNT_EMPTYDIR;
938 			do_freeopt = 1;
939 		} else if (strcmp(opt->name, "noemptydir") == 0) {
940 			fsflags &= ~MNT_EMPTYDIR;
941 			do_freeopt = 1;
942 		}
943 		if (do_freeopt)
944 			vfs_freeopt(optlist, opt);
945 	}
946 
947 	/*
948 	 * Be ultra-paranoid about making sure the type and fspath
949 	 * variables will fit in our mp buffers, including the
950 	 * terminating NUL.
951 	 */
952 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
953 		error = ENAMETOOLONG;
954 		goto bail;
955 	}
956 
957 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
958 	if (error == ENOENT) {
959 		error = EINVAL;
960 		if (errmsg != NULL)
961 			strncpy(errmsg, "Invalid fstype", errmsg_len);
962 		goto bail;
963 	}
964 
965 	/*
966 	 * See if we can mount in the read-only mode if the error code suggests
967 	 * that it could be possible and the mount options allow for that.
968 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
969 	 * overridden by "autoro".
970 	 */
971 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
972 		printf("%s: R/W mount failed, possibly R/O media,"
973 		    " trying R/O mount\n", __func__);
974 		fsflags |= MNT_RDONLY;
975 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
976 	}
977 bail:
978 	/* copyout the errmsg */
979 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
980 	    && errmsg_len > 0 && errmsg != NULL) {
981 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
982 			bcopy(errmsg,
983 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
984 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
985 		} else {
986 			copyout(errmsg,
987 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
988 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
989 		}
990 	}
991 
992 	if (optlist != NULL)
993 		vfs_freeopts(optlist);
994 	return (error);
995 }
996 
997 /*
998  * Old mount API.
999  */
1000 #ifndef _SYS_SYSPROTO_H_
1001 struct mount_args {
1002 	char	*type;
1003 	char	*path;
1004 	int	flags;
1005 	caddr_t	data;
1006 };
1007 #endif
1008 /* ARGSUSED */
1009 int
1010 sys_mount(struct thread *td, struct mount_args *uap)
1011 {
1012 	char *fstype;
1013 	struct vfsconf *vfsp = NULL;
1014 	struct mntarg *ma = NULL;
1015 	uint64_t flags;
1016 	int error;
1017 
1018 	/*
1019 	 * Mount flags are now 64-bits. On 32-bit architectures only
1020 	 * 32-bits are passed in, but from here on everything handles
1021 	 * 64-bit flags correctly.
1022 	 */
1023 	flags = uap->flags;
1024 
1025 	AUDIT_ARG_FFLAGS(flags);
1026 
1027 	/*
1028 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
1029 	 * userspace to set this flag, but we must filter it out if we want
1030 	 * MNT_UPDATE on the root file system to work.
1031 	 * MNT_ROOTFS should only be set by the kernel when mounting its
1032 	 * root file system.
1033 	 */
1034 	flags &= ~MNT_ROOTFS;
1035 
1036 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1037 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1038 	if (error) {
1039 		free(fstype, M_TEMP);
1040 		return (error);
1041 	}
1042 
1043 	AUDIT_ARG_TEXT(fstype);
1044 	vfsp = vfs_byname_kld(fstype, td, &error);
1045 	free(fstype, M_TEMP);
1046 	if (vfsp == NULL)
1047 		return (ENOENT);
1048 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1049 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1050 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1051 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
1052 		return (EOPNOTSUPP);
1053 
1054 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1055 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1056 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1057 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1058 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1059 
1060 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1061 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1062 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1063 }
1064 
1065 /*
1066  * vfs_domount_first(): first file system mount (not update)
1067  */
1068 static int
1069 vfs_domount_first(
1070 	struct thread *td,		/* Calling thread. */
1071 	struct vfsconf *vfsp,		/* File system type. */
1072 	char *fspath,			/* Mount path. */
1073 	struct vnode *vp,		/* Vnode to be covered. */
1074 	uint64_t fsflags,		/* Flags common to all filesystems. */
1075 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1076 	)
1077 {
1078 	struct vattr va;
1079 	struct mount *mp;
1080 	struct vnode *newdp, *rootvp;
1081 	int error, error1;
1082 	bool unmounted;
1083 
1084 	ASSERT_VOP_ELOCKED(vp, __func__);
1085 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1086 
1087 	/*
1088 	 * If the jail of the calling thread lacks permission for this type of
1089 	 * file system, or is trying to cover its own root, deny immediately.
1090 	 */
1091 	if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1092 	    vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1093 		vput(vp);
1094 		return (EPERM);
1095 	}
1096 
1097 	/*
1098 	 * If the user is not root, ensure that they own the directory
1099 	 * onto which we are attempting to mount.
1100 	 */
1101 	error = VOP_GETATTR(vp, &va, td->td_ucred);
1102 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1103 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1104 	if (error == 0)
1105 		error = vinvalbuf(vp, V_SAVE, 0, 0);
1106 	if (error == 0 && vp->v_type != VDIR)
1107 		error = ENOTDIR;
1108 	if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
1109 		error = vfs_emptydir(vp);
1110 	if (error == 0) {
1111 		VI_LOCK(vp);
1112 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1113 			vp->v_iflag |= VI_MOUNT;
1114 		else
1115 			error = EBUSY;
1116 		VI_UNLOCK(vp);
1117 	}
1118 	if (error != 0) {
1119 		vput(vp);
1120 		return (error);
1121 	}
1122 	vn_seqc_write_begin(vp);
1123 	VOP_UNLOCK(vp);
1124 
1125 	/* Allocate and initialize the filesystem. */
1126 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1127 	/* XXXMAC: pass to vfs_mount_alloc? */
1128 	mp->mnt_optnew = *optlist;
1129 	/* Set the mount level flags. */
1130 	mp->mnt_flag = (fsflags &
1131 	    (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE));
1132 
1133 	/*
1134 	 * Mount the filesystem.
1135 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1136 	 * get.  No freeing of cn_pnbuf.
1137 	 */
1138 	error1 = 0;
1139 	unmounted = true;
1140 	if ((error = VFS_MOUNT(mp)) != 0 ||
1141 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1142 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1143 		rootvp = NULL;
1144 		if (error1 != 0) {
1145 			MPASS(error == 0);
1146 			rootvp = vfs_cache_root_clear(mp);
1147 			if (rootvp != NULL) {
1148 				vhold(rootvp);
1149 				vrele(rootvp);
1150 			}
1151 			(void)vn_start_write(NULL, &mp, V_WAIT);
1152 			MNT_ILOCK(mp);
1153 			mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1154 			MNT_IUNLOCK(mp);
1155 			VFS_PURGE(mp);
1156 			error = VFS_UNMOUNT(mp, 0);
1157 			vn_finished_write(mp);
1158 			if (error != 0) {
1159 				printf(
1160 		    "failed post-mount (%d): rollback unmount returned %d\n",
1161 				    error1, error);
1162 				unmounted = false;
1163 			}
1164 			error = error1;
1165 		}
1166 		vfs_unbusy(mp);
1167 		mp->mnt_vnodecovered = NULL;
1168 		if (unmounted) {
1169 			/* XXXKIB wait for mnt_lockref drain? */
1170 			vfs_mount_destroy(mp);
1171 		}
1172 		VI_LOCK(vp);
1173 		vp->v_iflag &= ~VI_MOUNT;
1174 		VI_UNLOCK(vp);
1175 		if (rootvp != NULL) {
1176 			vn_seqc_write_end(rootvp);
1177 			vdrop(rootvp);
1178 		}
1179 		vn_seqc_write_end(vp);
1180 		vrele(vp);
1181 		return (error);
1182 	}
1183 	vn_seqc_write_begin(newdp);
1184 	VOP_UNLOCK(newdp);
1185 
1186 	if (mp->mnt_opt != NULL)
1187 		vfs_freeopts(mp->mnt_opt);
1188 	mp->mnt_opt = mp->mnt_optnew;
1189 	*optlist = NULL;
1190 
1191 	/*
1192 	 * Prevent external consumers of mount options from reading mnt_optnew.
1193 	 */
1194 	mp->mnt_optnew = NULL;
1195 
1196 	MNT_ILOCK(mp);
1197 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1198 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1199 		mp->mnt_kern_flag |= MNTK_ASYNC;
1200 	else
1201 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1202 	MNT_IUNLOCK(mp);
1203 
1204 	/*
1205 	 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the
1206 	 * vp lock to satisfy vfs_lookup() requirements.
1207 	 */
1208 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1209 	VI_LOCK(vp);
1210 	vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1211 	vp->v_mountedhere = mp;
1212 	VI_UNLOCK(vp);
1213 	VOP_UNLOCK(vp);
1214 	cache_purge(vp);
1215 
1216 	/*
1217 	 * We need to lock both vnodes.
1218 	 *
1219 	 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1220 	 * from different filesystems.
1221 	 */
1222 	vn_lock_pair(vp, false, newdp, false);
1223 
1224 	VI_LOCK(vp);
1225 	vp->v_iflag &= ~VI_MOUNT;
1226 	VI_UNLOCK(vp);
1227 	/* Place the new filesystem at the end of the mount list. */
1228 	mtx_lock(&mountlist_mtx);
1229 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1230 	mtx_unlock(&mountlist_mtx);
1231 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1232 	VOP_UNLOCK(vp);
1233 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1234 	VOP_UNLOCK(newdp);
1235 	mount_devctl_event("MOUNT", mp, false);
1236 	mountcheckdirs(vp, newdp);
1237 	vn_seqc_write_end(vp);
1238 	vn_seqc_write_end(newdp);
1239 	vrele(newdp);
1240 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1241 		vfs_allocate_syncvnode(mp);
1242 	vfs_op_exit(mp);
1243 	vfs_unbusy(mp);
1244 	return (0);
1245 }
1246 
1247 /*
1248  * vfs_domount_update(): update of mounted file system
1249  */
1250 static int
1251 vfs_domount_update(
1252 	struct thread *td,		/* Calling thread. */
1253 	struct vnode *vp,		/* Mount point vnode. */
1254 	uint64_t fsflags,		/* Flags common to all filesystems. */
1255 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1256 	)
1257 {
1258 	struct export_args export;
1259 	struct o2export_args o2export;
1260 	struct vnode *rootvp;
1261 	void *bufp;
1262 	struct mount *mp;
1263 	int error, export_error, i, len;
1264 	uint64_t flag;
1265 	gid_t *grps;
1266 
1267 	ASSERT_VOP_ELOCKED(vp, __func__);
1268 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1269 	mp = vp->v_mount;
1270 
1271 	if ((vp->v_vflag & VV_ROOT) == 0) {
1272 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1273 		    == 0)
1274 			error = EXDEV;
1275 		else
1276 			error = EINVAL;
1277 		vput(vp);
1278 		return (error);
1279 	}
1280 
1281 	/*
1282 	 * We only allow the filesystem to be reloaded if it
1283 	 * is currently mounted read-only.
1284 	 */
1285 	flag = mp->mnt_flag;
1286 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1287 		vput(vp);
1288 		return (EOPNOTSUPP);	/* Needs translation */
1289 	}
1290 	/*
1291 	 * Only privileged root, or (if MNT_USER is set) the user that
1292 	 * did the original mount is permitted to update it.
1293 	 */
1294 	error = vfs_suser(mp, td);
1295 	if (error != 0) {
1296 		vput(vp);
1297 		return (error);
1298 	}
1299 	if (vfs_busy(mp, MBF_NOWAIT)) {
1300 		vput(vp);
1301 		return (EBUSY);
1302 	}
1303 	VI_LOCK(vp);
1304 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1305 		VI_UNLOCK(vp);
1306 		vfs_unbusy(mp);
1307 		vput(vp);
1308 		return (EBUSY);
1309 	}
1310 	vp->v_iflag |= VI_MOUNT;
1311 	VI_UNLOCK(vp);
1312 	VOP_UNLOCK(vp);
1313 
1314 	vfs_op_enter(mp);
1315 	vn_seqc_write_begin(vp);
1316 
1317 	rootvp = NULL;
1318 	MNT_ILOCK(mp);
1319 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1320 		MNT_IUNLOCK(mp);
1321 		error = EBUSY;
1322 		goto end;
1323 	}
1324 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1325 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1326 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1327 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1328 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1329 	rootvp = vfs_cache_root_clear(mp);
1330 	MNT_IUNLOCK(mp);
1331 	mp->mnt_optnew = *optlist;
1332 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1333 
1334 	/*
1335 	 * Mount the filesystem.
1336 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1337 	 * get.  No freeing of cn_pnbuf.
1338 	 */
1339 	error = VFS_MOUNT(mp);
1340 
1341 	export_error = 0;
1342 	/* Process the export option. */
1343 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1344 	    &len) == 0) {
1345 		/* Assume that there is only 1 ABI for each length. */
1346 		switch (len) {
1347 		case (sizeof(struct oexport_args)):
1348 			bzero(&o2export, sizeof(o2export));
1349 			/* FALLTHROUGH */
1350 		case (sizeof(o2export)):
1351 			bcopy(bufp, &o2export, len);
1352 			export.ex_flags = (uint64_t)o2export.ex_flags;
1353 			export.ex_root = o2export.ex_root;
1354 			export.ex_uid = o2export.ex_anon.cr_uid;
1355 			export.ex_groups = NULL;
1356 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1357 			if (export.ex_ngroups > 0) {
1358 				if (export.ex_ngroups <= XU_NGROUPS) {
1359 					export.ex_groups = malloc(
1360 					    export.ex_ngroups * sizeof(gid_t),
1361 					    M_TEMP, M_WAITOK);
1362 					for (i = 0; i < export.ex_ngroups; i++)
1363 						export.ex_groups[i] =
1364 						  o2export.ex_anon.cr_groups[i];
1365 				} else
1366 					export_error = EINVAL;
1367 			} else if (export.ex_ngroups < 0)
1368 				export_error = EINVAL;
1369 			export.ex_addr = o2export.ex_addr;
1370 			export.ex_addrlen = o2export.ex_addrlen;
1371 			export.ex_mask = o2export.ex_mask;
1372 			export.ex_masklen = o2export.ex_masklen;
1373 			export.ex_indexfile = o2export.ex_indexfile;
1374 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1375 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1376 				for (i = 0; i < export.ex_numsecflavors; i++)
1377 					export.ex_secflavors[i] =
1378 					    o2export.ex_secflavors[i];
1379 			} else
1380 				export_error = EINVAL;
1381 			if (export_error == 0)
1382 				export_error = vfs_export(mp, &export);
1383 			free(export.ex_groups, M_TEMP);
1384 			break;
1385 		case (sizeof(export)):
1386 			bcopy(bufp, &export, len);
1387 			grps = NULL;
1388 			if (export.ex_ngroups > 0) {
1389 				if (export.ex_ngroups <= NGROUPS_MAX) {
1390 					grps = malloc(export.ex_ngroups *
1391 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1392 					export_error = copyin(export.ex_groups,
1393 					    grps, export.ex_ngroups *
1394 					    sizeof(gid_t));
1395 					if (export_error == 0)
1396 						export.ex_groups = grps;
1397 				} else
1398 					export_error = EINVAL;
1399 			} else if (export.ex_ngroups == 0)
1400 				export.ex_groups = NULL;
1401 			else
1402 				export_error = EINVAL;
1403 			if (export_error == 0)
1404 				export_error = vfs_export(mp, &export);
1405 			free(grps, M_TEMP);
1406 			break;
1407 		default:
1408 			export_error = EINVAL;
1409 			break;
1410 		}
1411 	}
1412 
1413 	MNT_ILOCK(mp);
1414 	if (error == 0) {
1415 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1416 		    MNT_SNAPSHOT);
1417 	} else {
1418 		/*
1419 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1420 		 * because it is not part of MNT_UPDATEMASK, but it could have
1421 		 * changed in the meantime if quotactl(2) was called.
1422 		 * All in all we want current value of MNT_QUOTA, not the old
1423 		 * one.
1424 		 */
1425 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1426 	}
1427 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1428 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1429 		mp->mnt_kern_flag |= MNTK_ASYNC;
1430 	else
1431 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1432 	MNT_IUNLOCK(mp);
1433 
1434 	if (error != 0)
1435 		goto end;
1436 
1437 	mount_devctl_event("REMOUNT", mp, true);
1438 	if (mp->mnt_opt != NULL)
1439 		vfs_freeopts(mp->mnt_opt);
1440 	mp->mnt_opt = mp->mnt_optnew;
1441 	*optlist = NULL;
1442 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1443 	/*
1444 	 * Prevent external consumers of mount options from reading
1445 	 * mnt_optnew.
1446 	 */
1447 	mp->mnt_optnew = NULL;
1448 
1449 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1450 		vfs_allocate_syncvnode(mp);
1451 	else
1452 		vfs_deallocate_syncvnode(mp);
1453 end:
1454 	vfs_op_exit(mp);
1455 	if (rootvp != NULL) {
1456 		vn_seqc_write_end(rootvp);
1457 		vrele(rootvp);
1458 	}
1459 	vn_seqc_write_end(vp);
1460 	vfs_unbusy(mp);
1461 	VI_LOCK(vp);
1462 	vp->v_iflag &= ~VI_MOUNT;
1463 	VI_UNLOCK(vp);
1464 	vrele(vp);
1465 	return (error != 0 ? error : export_error);
1466 }
1467 
1468 /*
1469  * vfs_domount(): actually attempt a filesystem mount.
1470  */
1471 static int
1472 vfs_domount(
1473 	struct thread *td,		/* Calling thread. */
1474 	const char *fstype,		/* Filesystem type. */
1475 	char *fspath,			/* Mount path. */
1476 	uint64_t fsflags,		/* Flags common to all filesystems. */
1477 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1478 	)
1479 {
1480 	struct vfsconf *vfsp;
1481 	struct nameidata nd;
1482 	struct vnode *vp;
1483 	char *pathbuf;
1484 	int error;
1485 
1486 	/*
1487 	 * Be ultra-paranoid about making sure the type and fspath
1488 	 * variables will fit in our mp buffers, including the
1489 	 * terminating NUL.
1490 	 */
1491 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1492 		return (ENAMETOOLONG);
1493 
1494 	if (jailed(td->td_ucred) || usermount == 0) {
1495 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1496 			return (error);
1497 	}
1498 
1499 	/*
1500 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1501 	 */
1502 	if (fsflags & MNT_EXPORTED) {
1503 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1504 		if (error)
1505 			return (error);
1506 	}
1507 	if (fsflags & MNT_SUIDDIR) {
1508 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1509 		if (error)
1510 			return (error);
1511 	}
1512 	/*
1513 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1514 	 */
1515 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1516 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1517 			fsflags |= MNT_NOSUID | MNT_USER;
1518 	}
1519 
1520 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1521 	vfsp = NULL;
1522 	if ((fsflags & MNT_UPDATE) == 0) {
1523 		/* Don't try to load KLDs if we're mounting the root. */
1524 		if (fsflags & MNT_ROOTFS) {
1525 			if ((vfsp = vfs_byname(fstype)) == NULL)
1526 				return (ENODEV);
1527 		} else {
1528 			if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1529 				return (error);
1530 		}
1531 	}
1532 
1533 	/*
1534 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1535 	 */
1536 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
1537 	    fspath);
1538 	error = namei(&nd);
1539 	if (error != 0)
1540 		return (error);
1541 	NDFREE_PNBUF(&nd);
1542 	vp = nd.ni_vp;
1543 	if ((fsflags & MNT_UPDATE) == 0) {
1544 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1545 		    (fsflags & MNT_NOCOVER) != 0) {
1546 			vput(vp);
1547 			return (EBUSY);
1548 		}
1549 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1550 		strcpy(pathbuf, fspath);
1551 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1552 		if (error == 0) {
1553 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1554 			    fsflags, optlist);
1555 		}
1556 		free(pathbuf, M_TEMP);
1557 	} else
1558 		error = vfs_domount_update(td, vp, fsflags, optlist);
1559 
1560 	return (error);
1561 }
1562 
1563 /*
1564  * Unmount a filesystem.
1565  *
1566  * Note: unmount takes a path to the vnode mounted on as argument, not
1567  * special file (as before).
1568  */
1569 #ifndef _SYS_SYSPROTO_H_
1570 struct unmount_args {
1571 	char	*path;
1572 	int	flags;
1573 };
1574 #endif
1575 /* ARGSUSED */
1576 int
1577 sys_unmount(struct thread *td, struct unmount_args *uap)
1578 {
1579 
1580 	return (kern_unmount(td, uap->path, uap->flags));
1581 }
1582 
1583 int
1584 kern_unmount(struct thread *td, const char *path, int flags)
1585 {
1586 	struct nameidata nd;
1587 	struct mount *mp;
1588 	char *fsidbuf, *pathbuf;
1589 	fsid_t fsid;
1590 	int error;
1591 
1592 	AUDIT_ARG_VALUE(flags);
1593 	if (jailed(td->td_ucred) || usermount == 0) {
1594 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1595 		if (error)
1596 			return (error);
1597 	}
1598 
1599 	if (flags & MNT_BYFSID) {
1600 		fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1601 		error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1602 		if (error) {
1603 			free(fsidbuf, M_TEMP);
1604 			return (error);
1605 		}
1606 
1607 		AUDIT_ARG_TEXT(fsidbuf);
1608 		/* Decode the filesystem ID. */
1609 		if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1610 			free(fsidbuf, M_TEMP);
1611 			return (EINVAL);
1612 		}
1613 
1614 		mp = vfs_getvfs(&fsid);
1615 		free(fsidbuf, M_TEMP);
1616 		if (mp == NULL) {
1617 			return (ENOENT);
1618 		}
1619 	} else {
1620 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1621 		error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1622 		if (error) {
1623 			free(pathbuf, M_TEMP);
1624 			return (error);
1625 		}
1626 
1627 		/*
1628 		 * Try to find global path for path argument.
1629 		 */
1630 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1631 		    UIO_SYSSPACE, pathbuf);
1632 		if (namei(&nd) == 0) {
1633 			NDFREE_PNBUF(&nd);
1634 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1635 			    MNAMELEN);
1636 			if (error == 0)
1637 				vput(nd.ni_vp);
1638 		}
1639 		mtx_lock(&mountlist_mtx);
1640 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1641 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1642 				vfs_ref(mp);
1643 				break;
1644 			}
1645 		}
1646 		mtx_unlock(&mountlist_mtx);
1647 		free(pathbuf, M_TEMP);
1648 		if (mp == NULL) {
1649 			/*
1650 			 * Previously we returned ENOENT for a nonexistent path and
1651 			 * EINVAL for a non-mountpoint.  We cannot tell these apart
1652 			 * now, so in the !MNT_BYFSID case return the more likely
1653 			 * EINVAL for compatibility.
1654 			 */
1655 			return (EINVAL);
1656 		}
1657 	}
1658 
1659 	/*
1660 	 * Don't allow unmounting the root filesystem.
1661 	 */
1662 	if (mp->mnt_flag & MNT_ROOTFS) {
1663 		vfs_rel(mp);
1664 		return (EINVAL);
1665 	}
1666 	error = dounmount(mp, flags, td);
1667 	return (error);
1668 }
1669 
1670 /*
1671  * Return error if any of the vnodes, ignoring the root vnode
1672  * and the syncer vnode, have non-zero usecount.
1673  *
1674  * This function is purely advisory - it can return false positives
1675  * and negatives.
1676  */
1677 static int
1678 vfs_check_usecounts(struct mount *mp)
1679 {
1680 	struct vnode *vp, *mvp;
1681 
1682 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1683 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1684 		    vp->v_usecount != 0) {
1685 			VI_UNLOCK(vp);
1686 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1687 			return (EBUSY);
1688 		}
1689 		VI_UNLOCK(vp);
1690 	}
1691 
1692 	return (0);
1693 }
1694 
1695 static void
1696 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1697 {
1698 
1699 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1700 	mp->mnt_kern_flag &= ~mntkflags;
1701 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1702 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1703 		wakeup(mp);
1704 	}
1705 	vfs_op_exit_locked(mp);
1706 	MNT_IUNLOCK(mp);
1707 	if (coveredvp != NULL) {
1708 		VOP_UNLOCK(coveredvp);
1709 		vdrop(coveredvp);
1710 	}
1711 	vn_finished_write(mp);
1712 	vfs_rel(mp);
1713 }
1714 
1715 /*
1716  * There are various reference counters associated with the mount point.
1717  * Normally it is permitted to modify them without taking the mnt ilock,
1718  * but this behavior can be temporarily disabled if stable value is needed
1719  * or callers are expected to block (e.g. to not allow new users during
1720  * forced unmount).
1721  */
1722 void
1723 vfs_op_enter(struct mount *mp)
1724 {
1725 	struct mount_pcpu *mpcpu;
1726 	int cpu;
1727 
1728 	MNT_ILOCK(mp);
1729 	mp->mnt_vfs_ops++;
1730 	if (mp->mnt_vfs_ops > 1) {
1731 		MNT_IUNLOCK(mp);
1732 		return;
1733 	}
1734 	vfs_op_barrier_wait(mp);
1735 	CPU_FOREACH(cpu) {
1736 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1737 
1738 		mp->mnt_ref += mpcpu->mntp_ref;
1739 		mpcpu->mntp_ref = 0;
1740 
1741 		mp->mnt_lockref += mpcpu->mntp_lockref;
1742 		mpcpu->mntp_lockref = 0;
1743 
1744 		mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1745 		mpcpu->mntp_writeopcount = 0;
1746 	}
1747 	MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1748 	    mp->mnt_writeopcount >= 0, mp,
1749 	    ("invalid count(s): ref %d lockref %d writeopcount %d",
1750 	    mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1751 	MNT_IUNLOCK(mp);
1752 	vfs_assert_mount_counters(mp);
1753 }
1754 
1755 void
1756 vfs_op_exit_locked(struct mount *mp)
1757 {
1758 
1759 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1760 
1761 	MPASSERT(mp->mnt_vfs_ops > 0, mp,
1762 	    ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1763 	MPASSERT(mp->mnt_vfs_ops > 1 ||
1764 	    (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1765 	    ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1766 	mp->mnt_vfs_ops--;
1767 }
1768 
1769 void
1770 vfs_op_exit(struct mount *mp)
1771 {
1772 
1773 	MNT_ILOCK(mp);
1774 	vfs_op_exit_locked(mp);
1775 	MNT_IUNLOCK(mp);
1776 }
1777 
1778 struct vfs_op_barrier_ipi {
1779 	struct mount *mp;
1780 	struct smp_rendezvous_cpus_retry_arg srcra;
1781 };
1782 
1783 static void
1784 vfs_op_action_func(void *arg)
1785 {
1786 	struct vfs_op_barrier_ipi *vfsopipi;
1787 	struct mount *mp;
1788 
1789 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1790 	mp = vfsopipi->mp;
1791 
1792 	if (!vfs_op_thread_entered(mp))
1793 		smp_rendezvous_cpus_done(arg);
1794 }
1795 
1796 static void
1797 vfs_op_wait_func(void *arg, int cpu)
1798 {
1799 	struct vfs_op_barrier_ipi *vfsopipi;
1800 	struct mount *mp;
1801 	struct mount_pcpu *mpcpu;
1802 
1803 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1804 	mp = vfsopipi->mp;
1805 
1806 	mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1807 	while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1808 		cpu_spinwait();
1809 }
1810 
1811 void
1812 vfs_op_barrier_wait(struct mount *mp)
1813 {
1814 	struct vfs_op_barrier_ipi vfsopipi;
1815 
1816 	vfsopipi.mp = mp;
1817 
1818 	smp_rendezvous_cpus_retry(all_cpus,
1819 	    smp_no_rendezvous_barrier,
1820 	    vfs_op_action_func,
1821 	    smp_no_rendezvous_barrier,
1822 	    vfs_op_wait_func,
1823 	    &vfsopipi.srcra);
1824 }
1825 
1826 #ifdef DIAGNOSTIC
1827 void
1828 vfs_assert_mount_counters(struct mount *mp)
1829 {
1830 	struct mount_pcpu *mpcpu;
1831 	int cpu;
1832 
1833 	if (mp->mnt_vfs_ops == 0)
1834 		return;
1835 
1836 	CPU_FOREACH(cpu) {
1837 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1838 		if (mpcpu->mntp_ref != 0 ||
1839 		    mpcpu->mntp_lockref != 0 ||
1840 		    mpcpu->mntp_writeopcount != 0)
1841 			vfs_dump_mount_counters(mp);
1842 	}
1843 }
1844 
1845 void
1846 vfs_dump_mount_counters(struct mount *mp)
1847 {
1848 	struct mount_pcpu *mpcpu;
1849 	int ref, lockref, writeopcount;
1850 	int cpu;
1851 
1852 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1853 
1854 	printf("        ref : ");
1855 	ref = mp->mnt_ref;
1856 	CPU_FOREACH(cpu) {
1857 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1858 		printf("%d ", mpcpu->mntp_ref);
1859 		ref += mpcpu->mntp_ref;
1860 	}
1861 	printf("\n");
1862 	printf("    lockref : ");
1863 	lockref = mp->mnt_lockref;
1864 	CPU_FOREACH(cpu) {
1865 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1866 		printf("%d ", mpcpu->mntp_lockref);
1867 		lockref += mpcpu->mntp_lockref;
1868 	}
1869 	printf("\n");
1870 	printf("writeopcount: ");
1871 	writeopcount = mp->mnt_writeopcount;
1872 	CPU_FOREACH(cpu) {
1873 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1874 		printf("%d ", mpcpu->mntp_writeopcount);
1875 		writeopcount += mpcpu->mntp_writeopcount;
1876 	}
1877 	printf("\n");
1878 
1879 	printf("counter       struct total\n");
1880 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1881 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1882 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1883 
1884 	panic("invalid counts on struct mount");
1885 }
1886 #endif
1887 
1888 int
1889 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1890 {
1891 	struct mount_pcpu *mpcpu;
1892 	int cpu, sum;
1893 
1894 	switch (which) {
1895 	case MNT_COUNT_REF:
1896 		sum = mp->mnt_ref;
1897 		break;
1898 	case MNT_COUNT_LOCKREF:
1899 		sum = mp->mnt_lockref;
1900 		break;
1901 	case MNT_COUNT_WRITEOPCOUNT:
1902 		sum = mp->mnt_writeopcount;
1903 		break;
1904 	}
1905 
1906 	CPU_FOREACH(cpu) {
1907 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1908 		switch (which) {
1909 		case MNT_COUNT_REF:
1910 			sum += mpcpu->mntp_ref;
1911 			break;
1912 		case MNT_COUNT_LOCKREF:
1913 			sum += mpcpu->mntp_lockref;
1914 			break;
1915 		case MNT_COUNT_WRITEOPCOUNT:
1916 			sum += mpcpu->mntp_writeopcount;
1917 			break;
1918 		}
1919 	}
1920 	return (sum);
1921 }
1922 
1923 static bool
1924 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
1925     int timeout_ticks)
1926 {
1927 	bool enqueued;
1928 
1929 	enqueued = false;
1930 	mtx_lock(&deferred_unmount_lock);
1931 	if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
1932 		mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
1933 		STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
1934 		    mnt_taskqueue_link);
1935 		enqueued = true;
1936 	}
1937 	mtx_unlock(&deferred_unmount_lock);
1938 
1939 	if (enqueued) {
1940 		taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
1941 		    &deferred_unmount_task, timeout_ticks);
1942 	}
1943 
1944 	return (enqueued);
1945 }
1946 
1947 /*
1948  * Taskqueue handler for processing async/recursive unmounts
1949  */
1950 static void
1951 vfs_deferred_unmount(void *argi __unused, int pending __unused)
1952 {
1953 	STAILQ_HEAD(, mount) local_unmounts;
1954 	uint64_t flags;
1955 	struct mount *mp, *tmp;
1956 	int error;
1957 	unsigned int retries;
1958 	bool unmounted;
1959 
1960 	STAILQ_INIT(&local_unmounts);
1961 	mtx_lock(&deferred_unmount_lock);
1962 	STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
1963 	mtx_unlock(&deferred_unmount_lock);
1964 
1965 	STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
1966 		flags = mp->mnt_taskqueue_flags;
1967 		KASSERT((flags & MNT_DEFERRED) != 0,
1968 		    ("taskqueue unmount without MNT_DEFERRED"));
1969 		error = dounmount(mp, flags, curthread);
1970 		if (error != 0) {
1971 			MNT_ILOCK(mp);
1972 			unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
1973 			MNT_IUNLOCK(mp);
1974 
1975 			/*
1976 			 * The deferred unmount thread is the only thread that
1977 			 * modifies the retry counts, so locking/atomics aren't
1978 			 * needed here.
1979 			 */
1980 			retries = (mp->mnt_unmount_retries)++;
1981 			deferred_unmount_total_retries++;
1982 			if (!unmounted && retries < deferred_unmount_retry_limit) {
1983 				deferred_unmount_enqueue(mp, flags, true,
1984 				    -deferred_unmount_retry_delay_hz);
1985 			} else {
1986 				if (retries >= deferred_unmount_retry_limit) {
1987 					printf("giving up on deferred unmount "
1988 					    "of %s after %d retries, error %d\n",
1989 					    mp->mnt_stat.f_mntonname, retries, error);
1990 				}
1991 				vfs_rel(mp);
1992 			}
1993 		}
1994 	}
1995 }
1996 
1997 /*
1998  * Do the actual filesystem unmount.
1999  */
2000 int
2001 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2002 {
2003 	struct mount_upper_node *upper;
2004 	struct vnode *coveredvp, *rootvp;
2005 	int error;
2006 	uint64_t async_flag;
2007 	int mnt_gen_r;
2008 	unsigned int retries;
2009 
2010 	KASSERT((flags & MNT_DEFERRED) == 0 ||
2011 	    (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2012 	    ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2013 
2014 	/*
2015 	 * If the caller has explicitly requested the unmount to be handled by
2016 	 * the taskqueue and we're not already in taskqueue context, queue
2017 	 * up the unmount request and exit.  This is done prior to any
2018 	 * credential checks; MNT_DEFERRED should be used only for kernel-
2019 	 * initiated unmounts and will therefore be processed with the
2020 	 * (kernel) credentials of the taskqueue thread.  Still, callers
2021 	 * should be sure this is the behavior they want.
2022 	 */
2023 	if ((flags & MNT_DEFERRED) != 0 &&
2024 	    taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2025 		if (!deferred_unmount_enqueue(mp, flags, false, 0))
2026 			vfs_rel(mp);
2027 		return (EINPROGRESS);
2028 	}
2029 
2030 	/*
2031 	 * Only privileged root, or (if MNT_USER is set) the user that did the
2032 	 * original mount is permitted to unmount this filesystem.
2033 	 * This check should be made prior to queueing up any recursive
2034 	 * unmounts of upper filesystems.  Those unmounts will be executed
2035 	 * with kernel thread credentials and are expected to succeed, so
2036 	 * we must at least ensure the originating context has sufficient
2037 	 * privilege to unmount the base filesystem before proceeding with
2038 	 * the uppers.
2039 	 */
2040 	error = vfs_suser(mp, td);
2041 	if (error != 0) {
2042 		KASSERT((flags & MNT_DEFERRED) == 0,
2043 		    ("taskqueue unmount with insufficient privilege"));
2044 		vfs_rel(mp);
2045 		return (error);
2046 	}
2047 
2048 	if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2049 		flags |= MNT_RECURSE;
2050 
2051 	if ((flags & MNT_RECURSE) != 0) {
2052 		KASSERT((flags & MNT_FORCE) != 0,
2053 		    ("MNT_RECURSE requires MNT_FORCE"));
2054 
2055 		MNT_ILOCK(mp);
2056 		/*
2057 		 * Set MNTK_RECURSE to prevent new upper mounts from being
2058 		 * added, and note that an operation on the uppers list is in
2059 		 * progress.  This will ensure that unregistration from the
2060 		 * uppers list, and therefore any pending unmount of the upper
2061 		 * FS, can't complete until after we finish walking the list.
2062 		 */
2063 		mp->mnt_kern_flag |= MNTK_RECURSE;
2064 		mp->mnt_upper_pending++;
2065 		TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2066 			retries = upper->mp->mnt_unmount_retries;
2067 			if (retries > deferred_unmount_retry_limit) {
2068 				error = EBUSY;
2069 				continue;
2070 			}
2071 			MNT_IUNLOCK(mp);
2072 
2073 			vfs_ref(upper->mp);
2074 			if (!deferred_unmount_enqueue(upper->mp, flags,
2075 			    false, 0))
2076 				vfs_rel(upper->mp);
2077 			MNT_ILOCK(mp);
2078 		}
2079 		mp->mnt_upper_pending--;
2080 		if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2081 		    mp->mnt_upper_pending == 0) {
2082 			mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2083 			wakeup(&mp->mnt_uppers);
2084 		}
2085 
2086 		/*
2087 		 * If we're not on the taskqueue, wait until the uppers list
2088 		 * is drained before proceeding with unmount.  Otherwise, if
2089 		 * we are on the taskqueue and there are still pending uppers,
2090 		 * just re-enqueue on the end of the taskqueue.
2091 		 */
2092 		if ((flags & MNT_DEFERRED) == 0) {
2093 			while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2094 				mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2095 				error = msleep(&mp->mnt_taskqueue_link,
2096 				    MNT_MTX(mp), PCATCH, "umntqw", 0);
2097 			}
2098 			if (error != 0) {
2099 				MNT_REL(mp);
2100 				MNT_IUNLOCK(mp);
2101 				return (error);
2102 			}
2103 		} else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2104 			MNT_IUNLOCK(mp);
2105 			if (error == 0)
2106 				deferred_unmount_enqueue(mp, flags, true, 0);
2107 			return (error);
2108 		}
2109 		MNT_IUNLOCK(mp);
2110 		KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2111 	}
2112 
2113 	/* Allow the taskqueue to safely re-enqueue on failure */
2114 	if ((flags & MNT_DEFERRED) != 0)
2115 		vfs_ref(mp);
2116 
2117 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2118 		mnt_gen_r = mp->mnt_gen;
2119 		VI_LOCK(coveredvp);
2120 		vholdl(coveredvp);
2121 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2122 		/*
2123 		 * Check for mp being unmounted while waiting for the
2124 		 * covered vnode lock.
2125 		 */
2126 		if (coveredvp->v_mountedhere != mp ||
2127 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2128 			VOP_UNLOCK(coveredvp);
2129 			vdrop(coveredvp);
2130 			vfs_rel(mp);
2131 			return (EBUSY);
2132 		}
2133 	}
2134 
2135 	vfs_op_enter(mp);
2136 
2137 	vn_start_write(NULL, &mp, V_WAIT);
2138 	MNT_ILOCK(mp);
2139 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2140 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
2141 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
2142 		dounmount_cleanup(mp, coveredvp, 0);
2143 		return (EBUSY);
2144 	}
2145 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
2146 	rootvp = vfs_cache_root_clear(mp);
2147 	if (coveredvp != NULL)
2148 		vn_seqc_write_begin(coveredvp);
2149 	if (flags & MNT_NONBUSY) {
2150 		MNT_IUNLOCK(mp);
2151 		error = vfs_check_usecounts(mp);
2152 		MNT_ILOCK(mp);
2153 		if (error != 0) {
2154 			vn_seqc_write_end(coveredvp);
2155 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2156 			if (rootvp != NULL) {
2157 				vn_seqc_write_end(rootvp);
2158 				vrele(rootvp);
2159 			}
2160 			return (error);
2161 		}
2162 	}
2163 	/* Allow filesystems to detect that a forced unmount is in progress. */
2164 	if (flags & MNT_FORCE) {
2165 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2166 		MNT_IUNLOCK(mp);
2167 		/*
2168 		 * Must be done after setting MNTK_UNMOUNTF and before
2169 		 * waiting for mnt_lockref to become 0.
2170 		 */
2171 		VFS_PURGE(mp);
2172 		MNT_ILOCK(mp);
2173 	}
2174 	error = 0;
2175 	if (mp->mnt_lockref) {
2176 		mp->mnt_kern_flag |= MNTK_DRAINING;
2177 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2178 		    "mount drain", 0);
2179 	}
2180 	MNT_IUNLOCK(mp);
2181 	KASSERT(mp->mnt_lockref == 0,
2182 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
2183 	    __func__, __FILE__, __LINE__));
2184 	KASSERT(error == 0,
2185 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
2186 	    __func__, __FILE__, __LINE__));
2187 
2188 	/*
2189 	 * We want to keep the vnode around so that we can vn_seqc_write_end
2190 	 * after we are done with unmount. Downgrade our reference to a mere
2191 	 * hold count so that we don't interefere with anything.
2192 	 */
2193 	if (rootvp != NULL) {
2194 		vhold(rootvp);
2195 		vrele(rootvp);
2196 	}
2197 
2198 	if (mp->mnt_flag & MNT_EXPUBLIC)
2199 		vfs_setpublicfs(NULL, NULL, NULL);
2200 
2201 	vfs_periodic(mp, MNT_WAIT);
2202 	MNT_ILOCK(mp);
2203 	async_flag = mp->mnt_flag & MNT_ASYNC;
2204 	mp->mnt_flag &= ~MNT_ASYNC;
2205 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
2206 	MNT_IUNLOCK(mp);
2207 	vfs_deallocate_syncvnode(mp);
2208 	error = VFS_UNMOUNT(mp, flags);
2209 	vn_finished_write(mp);
2210 	vfs_rel(mp);
2211 	/*
2212 	 * If we failed to flush the dirty blocks for this mount point,
2213 	 * undo all the cdir/rdir and rootvnode changes we made above.
2214 	 * Unless we failed to do so because the device is reporting that
2215 	 * it doesn't exist anymore.
2216 	 */
2217 	if (error && error != ENXIO) {
2218 		MNT_ILOCK(mp);
2219 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2220 			MNT_IUNLOCK(mp);
2221 			vfs_allocate_syncvnode(mp);
2222 			MNT_ILOCK(mp);
2223 		}
2224 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2225 		mp->mnt_flag |= async_flag;
2226 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2227 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2228 			mp->mnt_kern_flag |= MNTK_ASYNC;
2229 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
2230 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
2231 			wakeup(mp);
2232 		}
2233 		vfs_op_exit_locked(mp);
2234 		MNT_IUNLOCK(mp);
2235 		if (coveredvp) {
2236 			vn_seqc_write_end(coveredvp);
2237 			VOP_UNLOCK(coveredvp);
2238 			vdrop(coveredvp);
2239 		}
2240 		if (rootvp != NULL) {
2241 			vn_seqc_write_end(rootvp);
2242 			vdrop(rootvp);
2243 		}
2244 		return (error);
2245 	}
2246 
2247 	mtx_lock(&mountlist_mtx);
2248 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
2249 	mtx_unlock(&mountlist_mtx);
2250 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2251 	if (coveredvp != NULL) {
2252 		VI_LOCK(coveredvp);
2253 		vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2254 		coveredvp->v_mountedhere = NULL;
2255 		vn_seqc_write_end_locked(coveredvp);
2256 		VI_UNLOCK(coveredvp);
2257 		VOP_UNLOCK(coveredvp);
2258 		vdrop(coveredvp);
2259 	}
2260 	mount_devctl_event("UNMOUNT", mp, false);
2261 	if (rootvp != NULL) {
2262 		vn_seqc_write_end(rootvp);
2263 		vdrop(rootvp);
2264 	}
2265 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2266 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
2267 		vrele(rootvnode);
2268 		rootvnode = NULL;
2269 	}
2270 	if (mp == rootdevmp)
2271 		rootdevmp = NULL;
2272 	if ((flags & MNT_DEFERRED) != 0)
2273 		vfs_rel(mp);
2274 	vfs_mount_destroy(mp);
2275 	return (0);
2276 }
2277 
2278 /*
2279  * Report errors during filesystem mounting.
2280  */
2281 void
2282 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2283 {
2284 	struct vfsoptlist *moptlist = mp->mnt_optnew;
2285 	va_list ap;
2286 	int error, len;
2287 	char *errmsg;
2288 
2289 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2290 	if (error || errmsg == NULL || len <= 0)
2291 		return;
2292 
2293 	va_start(ap, fmt);
2294 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2295 	va_end(ap);
2296 }
2297 
2298 void
2299 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2300 {
2301 	va_list ap;
2302 	int error, len;
2303 	char *errmsg;
2304 
2305 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2306 	if (error || errmsg == NULL || len <= 0)
2307 		return;
2308 
2309 	va_start(ap, fmt);
2310 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2311 	va_end(ap);
2312 }
2313 
2314 /*
2315  * ---------------------------------------------------------------------
2316  * Functions for querying mount options/arguments from filesystems.
2317  */
2318 
2319 /*
2320  * Check that no unknown options are given
2321  */
2322 int
2323 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2324 {
2325 	struct vfsopt *opt;
2326 	char errmsg[255];
2327 	const char **t, *p, *q;
2328 	int ret = 0;
2329 
2330 	TAILQ_FOREACH(opt, opts, link) {
2331 		p = opt->name;
2332 		q = NULL;
2333 		if (p[0] == 'n' && p[1] == 'o')
2334 			q = p + 2;
2335 		for(t = global_opts; *t != NULL; t++) {
2336 			if (strcmp(*t, p) == 0)
2337 				break;
2338 			if (q != NULL) {
2339 				if (strcmp(*t, q) == 0)
2340 					break;
2341 			}
2342 		}
2343 		if (*t != NULL)
2344 			continue;
2345 		for(t = legal; *t != NULL; t++) {
2346 			if (strcmp(*t, p) == 0)
2347 				break;
2348 			if (q != NULL) {
2349 				if (strcmp(*t, q) == 0)
2350 					break;
2351 			}
2352 		}
2353 		if (*t != NULL)
2354 			continue;
2355 		snprintf(errmsg, sizeof(errmsg),
2356 		    "mount option <%s> is unknown", p);
2357 		ret = EINVAL;
2358 	}
2359 	if (ret != 0) {
2360 		TAILQ_FOREACH(opt, opts, link) {
2361 			if (strcmp(opt->name, "errmsg") == 0) {
2362 				strncpy((char *)opt->value, errmsg, opt->len);
2363 				break;
2364 			}
2365 		}
2366 		if (opt == NULL)
2367 			printf("%s\n", errmsg);
2368 	}
2369 	return (ret);
2370 }
2371 
2372 /*
2373  * Get a mount option by its name.
2374  *
2375  * Return 0 if the option was found, ENOENT otherwise.
2376  * If len is non-NULL it will be filled with the length
2377  * of the option. If buf is non-NULL, it will be filled
2378  * with the address of the option.
2379  */
2380 int
2381 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2382 {
2383 	struct vfsopt *opt;
2384 
2385 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2386 
2387 	TAILQ_FOREACH(opt, opts, link) {
2388 		if (strcmp(name, opt->name) == 0) {
2389 			opt->seen = 1;
2390 			if (len != NULL)
2391 				*len = opt->len;
2392 			if (buf != NULL)
2393 				*buf = opt->value;
2394 			return (0);
2395 		}
2396 	}
2397 	return (ENOENT);
2398 }
2399 
2400 int
2401 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2402 {
2403 	struct vfsopt *opt;
2404 
2405 	if (opts == NULL)
2406 		return (-1);
2407 
2408 	TAILQ_FOREACH(opt, opts, link) {
2409 		if (strcmp(name, opt->name) == 0) {
2410 			opt->seen = 1;
2411 			return (opt->pos);
2412 		}
2413 	}
2414 	return (-1);
2415 }
2416 
2417 int
2418 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2419 {
2420 	char *opt_value, *vtp;
2421 	quad_t iv;
2422 	int error, opt_len;
2423 
2424 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2425 	if (error != 0)
2426 		return (error);
2427 	if (opt_len == 0 || opt_value == NULL)
2428 		return (EINVAL);
2429 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2430 		return (EINVAL);
2431 	iv = strtoq(opt_value, &vtp, 0);
2432 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2433 		return (EINVAL);
2434 	if (iv < 0)
2435 		return (EINVAL);
2436 	switch (vtp[0]) {
2437 	case 't': case 'T':
2438 		iv *= 1024;
2439 		/* FALLTHROUGH */
2440 	case 'g': case 'G':
2441 		iv *= 1024;
2442 		/* FALLTHROUGH */
2443 	case 'm': case 'M':
2444 		iv *= 1024;
2445 		/* FALLTHROUGH */
2446 	case 'k': case 'K':
2447 		iv *= 1024;
2448 	case '\0':
2449 		break;
2450 	default:
2451 		return (EINVAL);
2452 	}
2453 	*value = iv;
2454 
2455 	return (0);
2456 }
2457 
2458 char *
2459 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2460 {
2461 	struct vfsopt *opt;
2462 
2463 	*error = 0;
2464 	TAILQ_FOREACH(opt, opts, link) {
2465 		if (strcmp(name, opt->name) != 0)
2466 			continue;
2467 		opt->seen = 1;
2468 		if (opt->len == 0 ||
2469 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2470 			*error = EINVAL;
2471 			return (NULL);
2472 		}
2473 		return (opt->value);
2474 	}
2475 	*error = ENOENT;
2476 	return (NULL);
2477 }
2478 
2479 int
2480 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2481 	uint64_t val)
2482 {
2483 	struct vfsopt *opt;
2484 
2485 	TAILQ_FOREACH(opt, opts, link) {
2486 		if (strcmp(name, opt->name) == 0) {
2487 			opt->seen = 1;
2488 			if (w != NULL)
2489 				*w |= val;
2490 			return (1);
2491 		}
2492 	}
2493 	if (w != NULL)
2494 		*w &= ~val;
2495 	return (0);
2496 }
2497 
2498 int
2499 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2500 {
2501 	va_list ap;
2502 	struct vfsopt *opt;
2503 	int ret;
2504 
2505 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2506 
2507 	TAILQ_FOREACH(opt, opts, link) {
2508 		if (strcmp(name, opt->name) != 0)
2509 			continue;
2510 		opt->seen = 1;
2511 		if (opt->len == 0 || opt->value == NULL)
2512 			return (0);
2513 		if (((char *)opt->value)[opt->len - 1] != '\0')
2514 			return (0);
2515 		va_start(ap, fmt);
2516 		ret = vsscanf(opt->value, fmt, ap);
2517 		va_end(ap);
2518 		return (ret);
2519 	}
2520 	return (0);
2521 }
2522 
2523 int
2524 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2525 {
2526 	struct vfsopt *opt;
2527 
2528 	TAILQ_FOREACH(opt, opts, link) {
2529 		if (strcmp(name, opt->name) != 0)
2530 			continue;
2531 		opt->seen = 1;
2532 		if (opt->value == NULL)
2533 			opt->len = len;
2534 		else {
2535 			if (opt->len != len)
2536 				return (EINVAL);
2537 			bcopy(value, opt->value, len);
2538 		}
2539 		return (0);
2540 	}
2541 	return (ENOENT);
2542 }
2543 
2544 int
2545 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2546 {
2547 	struct vfsopt *opt;
2548 
2549 	TAILQ_FOREACH(opt, opts, link) {
2550 		if (strcmp(name, opt->name) != 0)
2551 			continue;
2552 		opt->seen = 1;
2553 		if (opt->value == NULL)
2554 			opt->len = len;
2555 		else {
2556 			if (opt->len < len)
2557 				return (EINVAL);
2558 			opt->len = len;
2559 			bcopy(value, opt->value, len);
2560 		}
2561 		return (0);
2562 	}
2563 	return (ENOENT);
2564 }
2565 
2566 int
2567 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2568 {
2569 	struct vfsopt *opt;
2570 
2571 	TAILQ_FOREACH(opt, opts, link) {
2572 		if (strcmp(name, opt->name) != 0)
2573 			continue;
2574 		opt->seen = 1;
2575 		if (opt->value == NULL)
2576 			opt->len = strlen(value) + 1;
2577 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2578 			return (EINVAL);
2579 		return (0);
2580 	}
2581 	return (ENOENT);
2582 }
2583 
2584 /*
2585  * Find and copy a mount option.
2586  *
2587  * The size of the buffer has to be specified
2588  * in len, if it is not the same length as the
2589  * mount option, EINVAL is returned.
2590  * Returns ENOENT if the option is not found.
2591  */
2592 int
2593 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2594 {
2595 	struct vfsopt *opt;
2596 
2597 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2598 
2599 	TAILQ_FOREACH(opt, opts, link) {
2600 		if (strcmp(name, opt->name) == 0) {
2601 			opt->seen = 1;
2602 			if (len != opt->len)
2603 				return (EINVAL);
2604 			bcopy(opt->value, dest, opt->len);
2605 			return (0);
2606 		}
2607 	}
2608 	return (ENOENT);
2609 }
2610 
2611 int
2612 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2613 {
2614 	/*
2615 	 * Filesystems only fill in part of the structure for updates, we
2616 	 * have to read the entirety first to get all content.
2617 	 */
2618 	if (sbp != &mp->mnt_stat)
2619 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2620 
2621 	/*
2622 	 * Set these in case the underlying filesystem fails to do so.
2623 	 */
2624 	sbp->f_version = STATFS_VERSION;
2625 	sbp->f_namemax = NAME_MAX;
2626 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2627 	sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2628 
2629 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2630 }
2631 
2632 void
2633 vfs_mountedfrom(struct mount *mp, const char *from)
2634 {
2635 
2636 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2637 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2638 	    sizeof mp->mnt_stat.f_mntfromname);
2639 }
2640 
2641 /*
2642  * ---------------------------------------------------------------------
2643  * This is the api for building mount args and mounting filesystems from
2644  * inside the kernel.
2645  *
2646  * The API works by accumulation of individual args.  First error is
2647  * latched.
2648  *
2649  * XXX: should be documented in new manpage kernel_mount(9)
2650  */
2651 
2652 /* A memory allocation which must be freed when we are done */
2653 struct mntaarg {
2654 	SLIST_ENTRY(mntaarg)	next;
2655 };
2656 
2657 /* The header for the mount arguments */
2658 struct mntarg {
2659 	struct iovec *v;
2660 	int len;
2661 	int error;
2662 	SLIST_HEAD(, mntaarg)	list;
2663 };
2664 
2665 /*
2666  * Add a boolean argument.
2667  *
2668  * flag is the boolean value.
2669  * name must start with "no".
2670  */
2671 struct mntarg *
2672 mount_argb(struct mntarg *ma, int flag, const char *name)
2673 {
2674 
2675 	KASSERT(name[0] == 'n' && name[1] == 'o',
2676 	    ("mount_argb(...,%s): name must start with 'no'", name));
2677 
2678 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2679 }
2680 
2681 /*
2682  * Add an argument printf style
2683  */
2684 struct mntarg *
2685 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2686 {
2687 	va_list ap;
2688 	struct mntaarg *maa;
2689 	struct sbuf *sb;
2690 	int len;
2691 
2692 	if (ma == NULL) {
2693 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2694 		SLIST_INIT(&ma->list);
2695 	}
2696 	if (ma->error)
2697 		return (ma);
2698 
2699 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2700 	    M_MOUNT, M_WAITOK);
2701 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2702 	ma->v[ma->len].iov_len = strlen(name) + 1;
2703 	ma->len++;
2704 
2705 	sb = sbuf_new_auto();
2706 	va_start(ap, fmt);
2707 	sbuf_vprintf(sb, fmt, ap);
2708 	va_end(ap);
2709 	sbuf_finish(sb);
2710 	len = sbuf_len(sb) + 1;
2711 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2712 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2713 	bcopy(sbuf_data(sb), maa + 1, len);
2714 	sbuf_delete(sb);
2715 
2716 	ma->v[ma->len].iov_base = maa + 1;
2717 	ma->v[ma->len].iov_len = len;
2718 	ma->len++;
2719 
2720 	return (ma);
2721 }
2722 
2723 /*
2724  * Add an argument which is a userland string.
2725  */
2726 struct mntarg *
2727 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2728 {
2729 	struct mntaarg *maa;
2730 	char *tbuf;
2731 
2732 	if (val == NULL)
2733 		return (ma);
2734 	if (ma == NULL) {
2735 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2736 		SLIST_INIT(&ma->list);
2737 	}
2738 	if (ma->error)
2739 		return (ma);
2740 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2741 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2742 	tbuf = (void *)(maa + 1);
2743 	ma->error = copyinstr(val, tbuf, len, NULL);
2744 	return (mount_arg(ma, name, tbuf, -1));
2745 }
2746 
2747 /*
2748  * Plain argument.
2749  *
2750  * If length is -1, treat value as a C string.
2751  */
2752 struct mntarg *
2753 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2754 {
2755 
2756 	if (ma == NULL) {
2757 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2758 		SLIST_INIT(&ma->list);
2759 	}
2760 	if (ma->error)
2761 		return (ma);
2762 
2763 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2764 	    M_MOUNT, M_WAITOK);
2765 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2766 	ma->v[ma->len].iov_len = strlen(name) + 1;
2767 	ma->len++;
2768 
2769 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2770 	if (len < 0)
2771 		ma->v[ma->len].iov_len = strlen(val) + 1;
2772 	else
2773 		ma->v[ma->len].iov_len = len;
2774 	ma->len++;
2775 	return (ma);
2776 }
2777 
2778 /*
2779  * Free a mntarg structure
2780  */
2781 static void
2782 free_mntarg(struct mntarg *ma)
2783 {
2784 	struct mntaarg *maa;
2785 
2786 	while (!SLIST_EMPTY(&ma->list)) {
2787 		maa = SLIST_FIRST(&ma->list);
2788 		SLIST_REMOVE_HEAD(&ma->list, next);
2789 		free(maa, M_MOUNT);
2790 	}
2791 	free(ma->v, M_MOUNT);
2792 	free(ma, M_MOUNT);
2793 }
2794 
2795 /*
2796  * Mount a filesystem
2797  */
2798 int
2799 kernel_mount(struct mntarg *ma, uint64_t flags)
2800 {
2801 	struct uio auio;
2802 	int error;
2803 
2804 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2805 	KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2806 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2807 
2808 	error = ma->error;
2809 	if (error == 0) {
2810 		auio.uio_iov = ma->v;
2811 		auio.uio_iovcnt = ma->len;
2812 		auio.uio_segflg = UIO_SYSSPACE;
2813 		error = vfs_donmount(curthread, flags, &auio);
2814 	}
2815 	free_mntarg(ma);
2816 	return (error);
2817 }
2818 
2819 /* Map from mount options to printable formats. */
2820 static struct mntoptnames optnames[] = {
2821 	MNTOPT_NAMES
2822 };
2823 
2824 #define DEVCTL_LEN 1024
2825 static void
2826 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2827 {
2828 	const uint8_t *cp;
2829 	struct mntoptnames *fp;
2830 	struct sbuf sb;
2831 	struct statfs *sfp = &mp->mnt_stat;
2832 	char *buf;
2833 
2834 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2835 	if (buf == NULL)
2836 		return;
2837 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2838 	sbuf_cpy(&sb, "mount-point=\"");
2839 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2840 	sbuf_cat(&sb, "\" mount-dev=\"");
2841 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2842 	sbuf_cat(&sb, "\" mount-type=\"");
2843 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2844 	sbuf_cat(&sb, "\" fsid=0x");
2845 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2846 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2847 		sbuf_printf(&sb, "%02x", cp[i]);
2848 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2849 	for (fp = optnames; fp->o_opt != 0; fp++) {
2850 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2851 			sbuf_cat(&sb, fp->o_name);
2852 			sbuf_putc(&sb, ';');
2853 		}
2854 	}
2855 	sbuf_putc(&sb, '"');
2856 	sbuf_finish(&sb);
2857 
2858 	/*
2859 	 * Options are not published because the form of the options depends on
2860 	 * the file system and may include binary data. In addition, they don't
2861 	 * necessarily provide enough useful information to be actionable when
2862 	 * devd processes them.
2863 	 */
2864 
2865 	if (sbuf_error(&sb) == 0)
2866 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2867 	sbuf_delete(&sb);
2868 	free(buf, M_MOUNT);
2869 }
2870 
2871 /*
2872  * Force remount specified mount point to read-only.  The argument
2873  * must be busied to avoid parallel unmount attempts.
2874  *
2875  * Intended use is to prevent further writes if some metadata
2876  * inconsistency is detected.  Note that the function still flushes
2877  * all cached metadata and data for the mount point, which might be
2878  * not always suitable.
2879  */
2880 int
2881 vfs_remount_ro(struct mount *mp)
2882 {
2883 	struct vfsoptlist *opts;
2884 	struct vfsopt *opt;
2885 	struct vnode *vp_covered, *rootvp;
2886 	int error;
2887 
2888 	KASSERT(mp->mnt_lockref > 0,
2889 	    ("vfs_remount_ro: mp %p is not busied", mp));
2890 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
2891 	    ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
2892 
2893 	rootvp = NULL;
2894 	vp_covered = mp->mnt_vnodecovered;
2895 	error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
2896 	if (error != 0)
2897 		return (error);
2898 	VI_LOCK(vp_covered);
2899 	if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
2900 		VI_UNLOCK(vp_covered);
2901 		vput(vp_covered);
2902 		return (EBUSY);
2903 	}
2904 	vp_covered->v_iflag |= VI_MOUNT;
2905 	VI_UNLOCK(vp_covered);
2906 	vfs_op_enter(mp);
2907 	vn_seqc_write_begin(vp_covered);
2908 
2909 	MNT_ILOCK(mp);
2910 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
2911 		MNT_IUNLOCK(mp);
2912 		error = EBUSY;
2913 		goto out;
2914 	}
2915 	mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
2916 	rootvp = vfs_cache_root_clear(mp);
2917 	MNT_IUNLOCK(mp);
2918 
2919 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
2920 	TAILQ_INIT(opts);
2921 	opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
2922 	opt->name = strdup("ro", M_MOUNT);
2923 	opt->value = NULL;
2924 	TAILQ_INSERT_TAIL(opts, opt, link);
2925 	vfs_mergeopts(opts, mp->mnt_opt);
2926 	mp->mnt_optnew = opts;
2927 
2928 	error = VFS_MOUNT(mp);
2929 
2930 	if (error == 0) {
2931 		MNT_ILOCK(mp);
2932 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
2933 		MNT_IUNLOCK(mp);
2934 		vfs_deallocate_syncvnode(mp);
2935 		if (mp->mnt_opt != NULL)
2936 			vfs_freeopts(mp->mnt_opt);
2937 		mp->mnt_opt = mp->mnt_optnew;
2938 	} else {
2939 		MNT_ILOCK(mp);
2940 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
2941 		MNT_IUNLOCK(mp);
2942 		vfs_freeopts(mp->mnt_optnew);
2943 	}
2944 	mp->mnt_optnew = NULL;
2945 
2946 out:
2947 	vfs_op_exit(mp);
2948 	VI_LOCK(vp_covered);
2949 	vp_covered->v_iflag &= ~VI_MOUNT;
2950 	VI_UNLOCK(vp_covered);
2951 	vput(vp_covered);
2952 	vn_seqc_write_end(vp_covered);
2953 	if (rootvp != NULL) {
2954 		vn_seqc_write_end(rootvp);
2955 		vrele(rootvp);
2956 	}
2957 	return (error);
2958 }
2959 
2960 /*
2961  * Suspend write operations on all local writeable filesystems.  Does
2962  * full sync of them in the process.
2963  *
2964  * Iterate over the mount points in reverse order, suspending most
2965  * recently mounted filesystems first.  It handles a case where a
2966  * filesystem mounted from a md(4) vnode-backed device should be
2967  * suspended before the filesystem that owns the vnode.
2968  */
2969 void
2970 suspend_all_fs(void)
2971 {
2972 	struct mount *mp;
2973 	int error;
2974 
2975 	mtx_lock(&mountlist_mtx);
2976 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2977 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2978 		if (error != 0)
2979 			continue;
2980 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2981 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2982 			mtx_lock(&mountlist_mtx);
2983 			vfs_unbusy(mp);
2984 			continue;
2985 		}
2986 		error = vfs_write_suspend(mp, 0);
2987 		if (error == 0) {
2988 			MNT_ILOCK(mp);
2989 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2990 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2991 			MNT_IUNLOCK(mp);
2992 			mtx_lock(&mountlist_mtx);
2993 		} else {
2994 			printf("suspend of %s failed, error %d\n",
2995 			    mp->mnt_stat.f_mntonname, error);
2996 			mtx_lock(&mountlist_mtx);
2997 			vfs_unbusy(mp);
2998 		}
2999 	}
3000 	mtx_unlock(&mountlist_mtx);
3001 }
3002 
3003 void
3004 resume_all_fs(void)
3005 {
3006 	struct mount *mp;
3007 
3008 	mtx_lock(&mountlist_mtx);
3009 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3010 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3011 			continue;
3012 		mtx_unlock(&mountlist_mtx);
3013 		MNT_ILOCK(mp);
3014 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3015 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3016 		MNT_IUNLOCK(mp);
3017 		vfs_write_resume(mp, 0);
3018 		mtx_lock(&mountlist_mtx);
3019 		vfs_unbusy(mp);
3020 	}
3021 	mtx_unlock(&mountlist_mtx);
3022 }
3023