xref: /freebsd/sys/kern/vfs_mount.c (revision ec965063070e5753c166cf592c9336444b74720a)
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, "export") == 0)
928 			fsflags |= MNT_EXPORTED;
929 		else if (strcmp(opt->name, "automounted") == 0) {
930 			fsflags |= MNT_AUTOMOUNTED;
931 			do_freeopt = 1;
932 		} else if (strcmp(opt->name, "nocover") == 0) {
933 			fsflags |= MNT_NOCOVER;
934 			do_freeopt = 1;
935 		} else if (strcmp(opt->name, "cover") == 0) {
936 			fsflags &= ~MNT_NOCOVER;
937 			do_freeopt = 1;
938 		} else if (strcmp(opt->name, "emptydir") == 0) {
939 			fsflags |= MNT_EMPTYDIR;
940 			do_freeopt = 1;
941 		} else if (strcmp(opt->name, "noemptydir") == 0) {
942 			fsflags &= ~MNT_EMPTYDIR;
943 			do_freeopt = 1;
944 		}
945 		if (do_freeopt)
946 			vfs_freeopt(optlist, opt);
947 	}
948 
949 	/*
950 	 * Be ultra-paranoid about making sure the type and fspath
951 	 * variables will fit in our mp buffers, including the
952 	 * terminating NUL.
953 	 */
954 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
955 		error = ENAMETOOLONG;
956 		goto bail;
957 	}
958 
959 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
960 	if (error == ENOENT) {
961 		error = EINVAL;
962 		if (errmsg != NULL)
963 			strncpy(errmsg, "Invalid fstype", errmsg_len);
964 		goto bail;
965 	}
966 
967 	/*
968 	 * See if we can mount in the read-only mode if the error code suggests
969 	 * that it could be possible and the mount options allow for that.
970 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
971 	 * overridden by "autoro".
972 	 */
973 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
974 		printf("%s: R/W mount failed, possibly R/O media,"
975 		    " trying R/O mount\n", __func__);
976 		fsflags |= MNT_RDONLY;
977 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
978 	}
979 bail:
980 	/* copyout the errmsg */
981 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
982 	    && errmsg_len > 0 && errmsg != NULL) {
983 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
984 			bcopy(errmsg,
985 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
986 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
987 		} else {
988 			copyout(errmsg,
989 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
990 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
991 		}
992 	}
993 
994 	if (optlist != NULL)
995 		vfs_freeopts(optlist);
996 	return (error);
997 }
998 
999 /*
1000  * Old mount API.
1001  */
1002 #ifndef _SYS_SYSPROTO_H_
1003 struct mount_args {
1004 	char	*type;
1005 	char	*path;
1006 	int	flags;
1007 	caddr_t	data;
1008 };
1009 #endif
1010 /* ARGSUSED */
1011 int
1012 sys_mount(struct thread *td, struct mount_args *uap)
1013 {
1014 	char *fstype;
1015 	struct vfsconf *vfsp = NULL;
1016 	struct mntarg *ma = NULL;
1017 	uint64_t flags;
1018 	int error;
1019 
1020 	/*
1021 	 * Mount flags are now 64-bits. On 32-bit architectures only
1022 	 * 32-bits are passed in, but from here on everything handles
1023 	 * 64-bit flags correctly.
1024 	 */
1025 	flags = uap->flags;
1026 
1027 	AUDIT_ARG_FFLAGS(flags);
1028 
1029 	/*
1030 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
1031 	 * userspace to set this flag, but we must filter it out if we want
1032 	 * MNT_UPDATE on the root file system to work.
1033 	 * MNT_ROOTFS should only be set by the kernel when mounting its
1034 	 * root file system.
1035 	 */
1036 	flags &= ~MNT_ROOTFS;
1037 
1038 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1039 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1040 	if (error) {
1041 		free(fstype, M_TEMP);
1042 		return (error);
1043 	}
1044 
1045 	AUDIT_ARG_TEXT(fstype);
1046 	vfsp = vfs_byname_kld(fstype, td, &error);
1047 	free(fstype, M_TEMP);
1048 	if (vfsp == NULL)
1049 		return (ENOENT);
1050 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1051 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1052 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1053 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
1054 		return (EOPNOTSUPP);
1055 
1056 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1057 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1058 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1059 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1060 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1061 
1062 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1063 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1064 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1065 }
1066 
1067 /*
1068  * vfs_domount_first(): first file system mount (not update)
1069  */
1070 static int
1071 vfs_domount_first(
1072 	struct thread *td,		/* Calling thread. */
1073 	struct vfsconf *vfsp,		/* File system type. */
1074 	char *fspath,			/* Mount path. */
1075 	struct vnode *vp,		/* Vnode to be covered. */
1076 	uint64_t fsflags,		/* Flags common to all filesystems. */
1077 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1078 	)
1079 {
1080 	struct vattr va;
1081 	struct mount *mp;
1082 	struct vnode *newdp, *rootvp;
1083 	int error, error1;
1084 	bool unmounted;
1085 
1086 	ASSERT_VOP_ELOCKED(vp, __func__);
1087 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1088 
1089 	/*
1090 	 * If the jail of the calling thread lacks permission for this type of
1091 	 * file system, or is trying to cover its own root, deny immediately.
1092 	 */
1093 	if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1094 	    vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1095 		vput(vp);
1096 		return (EPERM);
1097 	}
1098 
1099 	/*
1100 	 * If the user is not root, ensure that they own the directory
1101 	 * onto which we are attempting to mount.
1102 	 */
1103 	error = VOP_GETATTR(vp, &va, td->td_ucred);
1104 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1105 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1106 	if (error == 0)
1107 		error = vinvalbuf(vp, V_SAVE, 0, 0);
1108 	if (vfsp->vfc_flags & VFCF_FILEMOUNT) {
1109 		if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG)
1110 			error = EINVAL;
1111 		/*
1112 		 * For file mounts, ensure that there is only one hardlink to the file.
1113 		 */
1114 		if (error == 0 && vp->v_type == VREG && va.va_nlink != 1)
1115 			error = EINVAL;
1116 	} else {
1117 		if (error == 0 && vp->v_type != VDIR)
1118 			error = ENOTDIR;
1119 	}
1120 	if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
1121 		error = vfs_emptydir(vp);
1122 	if (error == 0) {
1123 		VI_LOCK(vp);
1124 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1125 			vp->v_iflag |= VI_MOUNT;
1126 		else
1127 			error = EBUSY;
1128 		VI_UNLOCK(vp);
1129 	}
1130 	if (error != 0) {
1131 		vput(vp);
1132 		return (error);
1133 	}
1134 	vn_seqc_write_begin(vp);
1135 	VOP_UNLOCK(vp);
1136 
1137 	/* Allocate and initialize the filesystem. */
1138 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1139 	/* XXXMAC: pass to vfs_mount_alloc? */
1140 	mp->mnt_optnew = *optlist;
1141 	/* Set the mount level flags. */
1142 	mp->mnt_flag = (fsflags &
1143 	    (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE));
1144 
1145 	/*
1146 	 * Mount the filesystem.
1147 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1148 	 * get.  No freeing of cn_pnbuf.
1149 	 */
1150 	error1 = 0;
1151 	unmounted = true;
1152 	if ((error = VFS_MOUNT(mp)) != 0 ||
1153 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1154 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1155 		rootvp = NULL;
1156 		if (error1 != 0) {
1157 			MPASS(error == 0);
1158 			rootvp = vfs_cache_root_clear(mp);
1159 			if (rootvp != NULL) {
1160 				vhold(rootvp);
1161 				vrele(rootvp);
1162 			}
1163 			(void)vn_start_write(NULL, &mp, V_WAIT);
1164 			MNT_ILOCK(mp);
1165 			mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1166 			MNT_IUNLOCK(mp);
1167 			VFS_PURGE(mp);
1168 			error = VFS_UNMOUNT(mp, 0);
1169 			vn_finished_write(mp);
1170 			if (error != 0) {
1171 				printf(
1172 		    "failed post-mount (%d): rollback unmount returned %d\n",
1173 				    error1, error);
1174 				unmounted = false;
1175 			}
1176 			error = error1;
1177 		}
1178 		vfs_unbusy(mp);
1179 		mp->mnt_vnodecovered = NULL;
1180 		if (unmounted) {
1181 			/* XXXKIB wait for mnt_lockref drain? */
1182 			vfs_mount_destroy(mp);
1183 		}
1184 		VI_LOCK(vp);
1185 		vp->v_iflag &= ~VI_MOUNT;
1186 		VI_UNLOCK(vp);
1187 		if (rootvp != NULL) {
1188 			vn_seqc_write_end(rootvp);
1189 			vdrop(rootvp);
1190 		}
1191 		vn_seqc_write_end(vp);
1192 		vrele(vp);
1193 		return (error);
1194 	}
1195 	vn_seqc_write_begin(newdp);
1196 	VOP_UNLOCK(newdp);
1197 
1198 	if (mp->mnt_opt != NULL)
1199 		vfs_freeopts(mp->mnt_opt);
1200 	mp->mnt_opt = mp->mnt_optnew;
1201 	*optlist = NULL;
1202 
1203 	/*
1204 	 * Prevent external consumers of mount options from reading mnt_optnew.
1205 	 */
1206 	mp->mnt_optnew = NULL;
1207 
1208 	MNT_ILOCK(mp);
1209 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1210 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1211 		mp->mnt_kern_flag |= MNTK_ASYNC;
1212 	else
1213 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1214 	MNT_IUNLOCK(mp);
1215 
1216 	/*
1217 	 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the
1218 	 * vp lock to satisfy vfs_lookup() requirements.
1219 	 */
1220 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1221 	VI_LOCK(vp);
1222 	vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1223 	vp->v_mountedhere = mp;
1224 	VI_UNLOCK(vp);
1225 	VOP_UNLOCK(vp);
1226 	cache_purge(vp);
1227 
1228 	/*
1229 	 * We need to lock both vnodes.
1230 	 *
1231 	 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1232 	 * from different filesystems.
1233 	 */
1234 	vn_lock_pair(vp, false, newdp, false);
1235 
1236 	VI_LOCK(vp);
1237 	vp->v_iflag &= ~VI_MOUNT;
1238 	VI_UNLOCK(vp);
1239 	/* Place the new filesystem at the end of the mount list. */
1240 	mtx_lock(&mountlist_mtx);
1241 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1242 	mtx_unlock(&mountlist_mtx);
1243 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1244 	VOP_UNLOCK(vp);
1245 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1246 	VOP_UNLOCK(newdp);
1247 	mount_devctl_event("MOUNT", mp, false);
1248 	mountcheckdirs(vp, newdp);
1249 	vn_seqc_write_end(vp);
1250 	vn_seqc_write_end(newdp);
1251 	vrele(newdp);
1252 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1253 		vfs_allocate_syncvnode(mp);
1254 	vfs_op_exit(mp);
1255 	vfs_unbusy(mp);
1256 	return (0);
1257 }
1258 
1259 /*
1260  * vfs_domount_update(): update of mounted file system
1261  */
1262 static int
1263 vfs_domount_update(
1264 	struct thread *td,		/* Calling thread. */
1265 	struct vnode *vp,		/* Mount point vnode. */
1266 	uint64_t fsflags,		/* Flags common to all filesystems. */
1267 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1268 	)
1269 {
1270 	struct export_args export;
1271 	struct o2export_args o2export;
1272 	struct vnode *rootvp;
1273 	void *bufp;
1274 	struct mount *mp;
1275 	int error, export_error, i, len;
1276 	uint64_t flag;
1277 	gid_t *grps;
1278 
1279 	ASSERT_VOP_ELOCKED(vp, __func__);
1280 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1281 	mp = vp->v_mount;
1282 
1283 	if ((vp->v_vflag & VV_ROOT) == 0) {
1284 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1285 		    == 0)
1286 			error = EXDEV;
1287 		else
1288 			error = EINVAL;
1289 		vput(vp);
1290 		return (error);
1291 	}
1292 
1293 	/*
1294 	 * We only allow the filesystem to be reloaded if it
1295 	 * is currently mounted read-only.
1296 	 */
1297 	flag = mp->mnt_flag;
1298 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1299 		vput(vp);
1300 		return (EOPNOTSUPP);	/* Needs translation */
1301 	}
1302 	/*
1303 	 * Only privileged root, or (if MNT_USER is set) the user that
1304 	 * did the original mount is permitted to update it.
1305 	 */
1306 	error = vfs_suser(mp, td);
1307 	if (error != 0) {
1308 		vput(vp);
1309 		return (error);
1310 	}
1311 	if (vfs_busy(mp, MBF_NOWAIT)) {
1312 		vput(vp);
1313 		return (EBUSY);
1314 	}
1315 	VI_LOCK(vp);
1316 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1317 		VI_UNLOCK(vp);
1318 		vfs_unbusy(mp);
1319 		vput(vp);
1320 		return (EBUSY);
1321 	}
1322 	vp->v_iflag |= VI_MOUNT;
1323 	VI_UNLOCK(vp);
1324 	VOP_UNLOCK(vp);
1325 
1326 	vfs_op_enter(mp);
1327 	vn_seqc_write_begin(vp);
1328 
1329 	rootvp = NULL;
1330 	MNT_ILOCK(mp);
1331 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1332 		MNT_IUNLOCK(mp);
1333 		error = EBUSY;
1334 		goto end;
1335 	}
1336 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1337 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1338 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1339 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1340 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1341 	rootvp = vfs_cache_root_clear(mp);
1342 	MNT_IUNLOCK(mp);
1343 	mp->mnt_optnew = *optlist;
1344 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1345 
1346 	/*
1347 	 * Mount the filesystem.
1348 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1349 	 * get.  No freeing of cn_pnbuf.
1350 	 */
1351 	error = VFS_MOUNT(mp);
1352 
1353 	export_error = 0;
1354 	/* Process the export option. */
1355 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1356 	    &len) == 0) {
1357 		/* Assume that there is only 1 ABI for each length. */
1358 		switch (len) {
1359 		case (sizeof(struct oexport_args)):
1360 			bzero(&o2export, sizeof(o2export));
1361 			/* FALLTHROUGH */
1362 		case (sizeof(o2export)):
1363 			bcopy(bufp, &o2export, len);
1364 			export.ex_flags = (uint64_t)o2export.ex_flags;
1365 			export.ex_root = o2export.ex_root;
1366 			export.ex_uid = o2export.ex_anon.cr_uid;
1367 			export.ex_groups = NULL;
1368 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1369 			if (export.ex_ngroups > 0) {
1370 				if (export.ex_ngroups <= XU_NGROUPS) {
1371 					export.ex_groups = malloc(
1372 					    export.ex_ngroups * sizeof(gid_t),
1373 					    M_TEMP, M_WAITOK);
1374 					for (i = 0; i < export.ex_ngroups; i++)
1375 						export.ex_groups[i] =
1376 						  o2export.ex_anon.cr_groups[i];
1377 				} else
1378 					export_error = EINVAL;
1379 			} else if (export.ex_ngroups < 0)
1380 				export_error = EINVAL;
1381 			export.ex_addr = o2export.ex_addr;
1382 			export.ex_addrlen = o2export.ex_addrlen;
1383 			export.ex_mask = o2export.ex_mask;
1384 			export.ex_masklen = o2export.ex_masklen;
1385 			export.ex_indexfile = o2export.ex_indexfile;
1386 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1387 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1388 				for (i = 0; i < export.ex_numsecflavors; i++)
1389 					export.ex_secflavors[i] =
1390 					    o2export.ex_secflavors[i];
1391 			} else
1392 				export_error = EINVAL;
1393 			if (export_error == 0)
1394 				export_error = vfs_export(mp, &export);
1395 			free(export.ex_groups, M_TEMP);
1396 			break;
1397 		case (sizeof(export)):
1398 			bcopy(bufp, &export, len);
1399 			grps = NULL;
1400 			if (export.ex_ngroups > 0) {
1401 				if (export.ex_ngroups <= NGROUPS_MAX) {
1402 					grps = malloc(export.ex_ngroups *
1403 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1404 					export_error = copyin(export.ex_groups,
1405 					    grps, export.ex_ngroups *
1406 					    sizeof(gid_t));
1407 					if (export_error == 0)
1408 						export.ex_groups = grps;
1409 				} else
1410 					export_error = EINVAL;
1411 			} else if (export.ex_ngroups == 0)
1412 				export.ex_groups = NULL;
1413 			else
1414 				export_error = EINVAL;
1415 			if (export_error == 0)
1416 				export_error = vfs_export(mp, &export);
1417 			free(grps, M_TEMP);
1418 			break;
1419 		default:
1420 			export_error = EINVAL;
1421 			break;
1422 		}
1423 	}
1424 
1425 	MNT_ILOCK(mp);
1426 	if (error == 0) {
1427 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1428 		    MNT_SNAPSHOT);
1429 	} else {
1430 		/*
1431 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1432 		 * because it is not part of MNT_UPDATEMASK, but it could have
1433 		 * changed in the meantime if quotactl(2) was called.
1434 		 * All in all we want current value of MNT_QUOTA, not the old
1435 		 * one.
1436 		 */
1437 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1438 	}
1439 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1440 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1441 		mp->mnt_kern_flag |= MNTK_ASYNC;
1442 	else
1443 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1444 	MNT_IUNLOCK(mp);
1445 
1446 	if (error != 0)
1447 		goto end;
1448 
1449 	mount_devctl_event("REMOUNT", mp, true);
1450 	if (mp->mnt_opt != NULL)
1451 		vfs_freeopts(mp->mnt_opt);
1452 	mp->mnt_opt = mp->mnt_optnew;
1453 	*optlist = NULL;
1454 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1455 	/*
1456 	 * Prevent external consumers of mount options from reading
1457 	 * mnt_optnew.
1458 	 */
1459 	mp->mnt_optnew = NULL;
1460 
1461 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1462 		vfs_allocate_syncvnode(mp);
1463 	else
1464 		vfs_deallocate_syncvnode(mp);
1465 end:
1466 	vfs_op_exit(mp);
1467 	if (rootvp != NULL) {
1468 		vn_seqc_write_end(rootvp);
1469 		vrele(rootvp);
1470 	}
1471 	vn_seqc_write_end(vp);
1472 	vfs_unbusy(mp);
1473 	VI_LOCK(vp);
1474 	vp->v_iflag &= ~VI_MOUNT;
1475 	VI_UNLOCK(vp);
1476 	vrele(vp);
1477 	return (error != 0 ? error : export_error);
1478 }
1479 
1480 /*
1481  * vfs_domount(): actually attempt a filesystem mount.
1482  */
1483 static int
1484 vfs_domount(
1485 	struct thread *td,		/* Calling thread. */
1486 	const char *fstype,		/* Filesystem type. */
1487 	char *fspath,			/* Mount path. */
1488 	uint64_t fsflags,		/* Flags common to all filesystems. */
1489 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1490 	)
1491 {
1492 	struct vfsconf *vfsp;
1493 	struct nameidata nd;
1494 	struct vnode *vp;
1495 	char *pathbuf;
1496 	int error;
1497 
1498 	/*
1499 	 * Be ultra-paranoid about making sure the type and fspath
1500 	 * variables will fit in our mp buffers, including the
1501 	 * terminating NUL.
1502 	 */
1503 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1504 		return (ENAMETOOLONG);
1505 
1506 	if (jailed(td->td_ucred) || usermount == 0) {
1507 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1508 			return (error);
1509 	}
1510 
1511 	/*
1512 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1513 	 */
1514 	if (fsflags & MNT_EXPORTED) {
1515 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1516 		if (error)
1517 			return (error);
1518 	}
1519 	if (fsflags & MNT_SUIDDIR) {
1520 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1521 		if (error)
1522 			return (error);
1523 	}
1524 	/*
1525 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1526 	 */
1527 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1528 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1529 			fsflags |= MNT_NOSUID | MNT_USER;
1530 	}
1531 
1532 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1533 	vfsp = NULL;
1534 	if ((fsflags & MNT_UPDATE) == 0) {
1535 		/* Don't try to load KLDs if we're mounting the root. */
1536 		if (fsflags & MNT_ROOTFS) {
1537 			if ((vfsp = vfs_byname(fstype)) == NULL)
1538 				return (ENODEV);
1539 		} else {
1540 			if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1541 				return (error);
1542 		}
1543 	}
1544 
1545 	/*
1546 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1547 	 */
1548 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1549 	    UIO_SYSSPACE, fspath);
1550 	error = namei(&nd);
1551 	if (error != 0)
1552 		return (error);
1553 	vp = nd.ni_vp;
1554 	/*
1555 	 * Don't allow stacking file mounts to work around problems with the way
1556 	 * that namei sets nd.ni_dvp to vp_crossmp for these.
1557 	 */
1558 	if (vp->v_type == VREG)
1559 		fsflags |= MNT_NOCOVER;
1560 	if ((fsflags & MNT_UPDATE) == 0) {
1561 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1562 		    (fsflags & MNT_NOCOVER) != 0) {
1563 			vput(vp);
1564 			error = EBUSY;
1565 			goto out;
1566 		}
1567 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1568 		strcpy(pathbuf, fspath);
1569 		/*
1570 		 * Note: we allow any vnode type here. If the path sanity check
1571 		 * succeeds, the type will be validated in vfs_domount_first
1572 		 * above.
1573 		 */
1574 		if (vp->v_type == VDIR)
1575 			error = vn_path_to_global_path(td, vp, pathbuf,
1576 			    MNAMELEN);
1577 		else
1578 			error = vn_path_to_global_path_hardlink(td, vp,
1579 			    nd.ni_dvp, pathbuf, MNAMELEN,
1580 			    nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1581 		if (error == 0) {
1582 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1583 			    fsflags, optlist);
1584 		}
1585 		free(pathbuf, M_TEMP);
1586 	} else
1587 		error = vfs_domount_update(td, vp, fsflags, optlist);
1588 
1589 out:
1590 	NDFREE_PNBUF(&nd);
1591 	vrele(nd.ni_dvp);
1592 
1593 	return (error);
1594 }
1595 
1596 /*
1597  * Unmount a filesystem.
1598  *
1599  * Note: unmount takes a path to the vnode mounted on as argument, not
1600  * special file (as before).
1601  */
1602 #ifndef _SYS_SYSPROTO_H_
1603 struct unmount_args {
1604 	char	*path;
1605 	int	flags;
1606 };
1607 #endif
1608 /* ARGSUSED */
1609 int
1610 sys_unmount(struct thread *td, struct unmount_args *uap)
1611 {
1612 
1613 	return (kern_unmount(td, uap->path, uap->flags));
1614 }
1615 
1616 int
1617 kern_unmount(struct thread *td, const char *path, int flags)
1618 {
1619 	struct nameidata nd;
1620 	struct mount *mp;
1621 	char *fsidbuf, *pathbuf;
1622 	fsid_t fsid;
1623 	int error;
1624 
1625 	AUDIT_ARG_VALUE(flags);
1626 	if (jailed(td->td_ucred) || usermount == 0) {
1627 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1628 		if (error)
1629 			return (error);
1630 	}
1631 
1632 	if (flags & MNT_BYFSID) {
1633 		fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1634 		error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1635 		if (error) {
1636 			free(fsidbuf, M_TEMP);
1637 			return (error);
1638 		}
1639 
1640 		AUDIT_ARG_TEXT(fsidbuf);
1641 		/* Decode the filesystem ID. */
1642 		if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1643 			free(fsidbuf, M_TEMP);
1644 			return (EINVAL);
1645 		}
1646 
1647 		mp = vfs_getvfs(&fsid);
1648 		free(fsidbuf, M_TEMP);
1649 		if (mp == NULL) {
1650 			return (ENOENT);
1651 		}
1652 	} else {
1653 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1654 		error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1655 		if (error) {
1656 			free(pathbuf, M_TEMP);
1657 			return (error);
1658 		}
1659 
1660 		/*
1661 		 * Try to find global path for path argument.
1662 		 */
1663 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1664 		    UIO_SYSSPACE, pathbuf);
1665 		if (namei(&nd) == 0) {
1666 			NDFREE_PNBUF(&nd);
1667 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1668 			    MNAMELEN);
1669 			if (error == 0)
1670 				vput(nd.ni_vp);
1671 		}
1672 		mtx_lock(&mountlist_mtx);
1673 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1674 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1675 				vfs_ref(mp);
1676 				break;
1677 			}
1678 		}
1679 		mtx_unlock(&mountlist_mtx);
1680 		free(pathbuf, M_TEMP);
1681 		if (mp == NULL) {
1682 			/*
1683 			 * Previously we returned ENOENT for a nonexistent path and
1684 			 * EINVAL for a non-mountpoint.  We cannot tell these apart
1685 			 * now, so in the !MNT_BYFSID case return the more likely
1686 			 * EINVAL for compatibility.
1687 			 */
1688 			return (EINVAL);
1689 		}
1690 	}
1691 
1692 	/*
1693 	 * Don't allow unmounting the root filesystem.
1694 	 */
1695 	if (mp->mnt_flag & MNT_ROOTFS) {
1696 		vfs_rel(mp);
1697 		return (EINVAL);
1698 	}
1699 	error = dounmount(mp, flags, td);
1700 	return (error);
1701 }
1702 
1703 /*
1704  * Return error if any of the vnodes, ignoring the root vnode
1705  * and the syncer vnode, have non-zero usecount.
1706  *
1707  * This function is purely advisory - it can return false positives
1708  * and negatives.
1709  */
1710 static int
1711 vfs_check_usecounts(struct mount *mp)
1712 {
1713 	struct vnode *vp, *mvp;
1714 
1715 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1716 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1717 		    vp->v_usecount != 0) {
1718 			VI_UNLOCK(vp);
1719 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1720 			return (EBUSY);
1721 		}
1722 		VI_UNLOCK(vp);
1723 	}
1724 
1725 	return (0);
1726 }
1727 
1728 static void
1729 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1730 {
1731 
1732 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1733 	mp->mnt_kern_flag &= ~mntkflags;
1734 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1735 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1736 		wakeup(mp);
1737 	}
1738 	vfs_op_exit_locked(mp);
1739 	MNT_IUNLOCK(mp);
1740 	if (coveredvp != NULL) {
1741 		VOP_UNLOCK(coveredvp);
1742 		vdrop(coveredvp);
1743 	}
1744 	vn_finished_write(mp);
1745 	vfs_rel(mp);
1746 }
1747 
1748 /*
1749  * There are various reference counters associated with the mount point.
1750  * Normally it is permitted to modify them without taking the mnt ilock,
1751  * but this behavior can be temporarily disabled if stable value is needed
1752  * or callers are expected to block (e.g. to not allow new users during
1753  * forced unmount).
1754  */
1755 void
1756 vfs_op_enter(struct mount *mp)
1757 {
1758 	struct mount_pcpu *mpcpu;
1759 	int cpu;
1760 
1761 	MNT_ILOCK(mp);
1762 	mp->mnt_vfs_ops++;
1763 	if (mp->mnt_vfs_ops > 1) {
1764 		MNT_IUNLOCK(mp);
1765 		return;
1766 	}
1767 	vfs_op_barrier_wait(mp);
1768 	CPU_FOREACH(cpu) {
1769 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1770 
1771 		mp->mnt_ref += mpcpu->mntp_ref;
1772 		mpcpu->mntp_ref = 0;
1773 
1774 		mp->mnt_lockref += mpcpu->mntp_lockref;
1775 		mpcpu->mntp_lockref = 0;
1776 
1777 		mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1778 		mpcpu->mntp_writeopcount = 0;
1779 	}
1780 	MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1781 	    mp->mnt_writeopcount >= 0, mp,
1782 	    ("invalid count(s): ref %d lockref %d writeopcount %d",
1783 	    mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1784 	MNT_IUNLOCK(mp);
1785 	vfs_assert_mount_counters(mp);
1786 }
1787 
1788 void
1789 vfs_op_exit_locked(struct mount *mp)
1790 {
1791 
1792 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1793 
1794 	MPASSERT(mp->mnt_vfs_ops > 0, mp,
1795 	    ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1796 	MPASSERT(mp->mnt_vfs_ops > 1 ||
1797 	    (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1798 	    ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1799 	mp->mnt_vfs_ops--;
1800 }
1801 
1802 void
1803 vfs_op_exit(struct mount *mp)
1804 {
1805 
1806 	MNT_ILOCK(mp);
1807 	vfs_op_exit_locked(mp);
1808 	MNT_IUNLOCK(mp);
1809 }
1810 
1811 struct vfs_op_barrier_ipi {
1812 	struct mount *mp;
1813 	struct smp_rendezvous_cpus_retry_arg srcra;
1814 };
1815 
1816 static void
1817 vfs_op_action_func(void *arg)
1818 {
1819 	struct vfs_op_barrier_ipi *vfsopipi;
1820 	struct mount *mp;
1821 
1822 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1823 	mp = vfsopipi->mp;
1824 
1825 	if (!vfs_op_thread_entered(mp))
1826 		smp_rendezvous_cpus_done(arg);
1827 }
1828 
1829 static void
1830 vfs_op_wait_func(void *arg, int cpu)
1831 {
1832 	struct vfs_op_barrier_ipi *vfsopipi;
1833 	struct mount *mp;
1834 	struct mount_pcpu *mpcpu;
1835 
1836 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1837 	mp = vfsopipi->mp;
1838 
1839 	mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1840 	while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1841 		cpu_spinwait();
1842 }
1843 
1844 void
1845 vfs_op_barrier_wait(struct mount *mp)
1846 {
1847 	struct vfs_op_barrier_ipi vfsopipi;
1848 
1849 	vfsopipi.mp = mp;
1850 
1851 	smp_rendezvous_cpus_retry(all_cpus,
1852 	    smp_no_rendezvous_barrier,
1853 	    vfs_op_action_func,
1854 	    smp_no_rendezvous_barrier,
1855 	    vfs_op_wait_func,
1856 	    &vfsopipi.srcra);
1857 }
1858 
1859 #ifdef DIAGNOSTIC
1860 void
1861 vfs_assert_mount_counters(struct mount *mp)
1862 {
1863 	struct mount_pcpu *mpcpu;
1864 	int cpu;
1865 
1866 	if (mp->mnt_vfs_ops == 0)
1867 		return;
1868 
1869 	CPU_FOREACH(cpu) {
1870 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1871 		if (mpcpu->mntp_ref != 0 ||
1872 		    mpcpu->mntp_lockref != 0 ||
1873 		    mpcpu->mntp_writeopcount != 0)
1874 			vfs_dump_mount_counters(mp);
1875 	}
1876 }
1877 
1878 void
1879 vfs_dump_mount_counters(struct mount *mp)
1880 {
1881 	struct mount_pcpu *mpcpu;
1882 	int ref, lockref, writeopcount;
1883 	int cpu;
1884 
1885 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1886 
1887 	printf("        ref : ");
1888 	ref = mp->mnt_ref;
1889 	CPU_FOREACH(cpu) {
1890 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1891 		printf("%d ", mpcpu->mntp_ref);
1892 		ref += mpcpu->mntp_ref;
1893 	}
1894 	printf("\n");
1895 	printf("    lockref : ");
1896 	lockref = mp->mnt_lockref;
1897 	CPU_FOREACH(cpu) {
1898 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1899 		printf("%d ", mpcpu->mntp_lockref);
1900 		lockref += mpcpu->mntp_lockref;
1901 	}
1902 	printf("\n");
1903 	printf("writeopcount: ");
1904 	writeopcount = mp->mnt_writeopcount;
1905 	CPU_FOREACH(cpu) {
1906 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1907 		printf("%d ", mpcpu->mntp_writeopcount);
1908 		writeopcount += mpcpu->mntp_writeopcount;
1909 	}
1910 	printf("\n");
1911 
1912 	printf("counter       struct total\n");
1913 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1914 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1915 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1916 
1917 	panic("invalid counts on struct mount");
1918 }
1919 #endif
1920 
1921 int
1922 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1923 {
1924 	struct mount_pcpu *mpcpu;
1925 	int cpu, sum;
1926 
1927 	switch (which) {
1928 	case MNT_COUNT_REF:
1929 		sum = mp->mnt_ref;
1930 		break;
1931 	case MNT_COUNT_LOCKREF:
1932 		sum = mp->mnt_lockref;
1933 		break;
1934 	case MNT_COUNT_WRITEOPCOUNT:
1935 		sum = mp->mnt_writeopcount;
1936 		break;
1937 	}
1938 
1939 	CPU_FOREACH(cpu) {
1940 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1941 		switch (which) {
1942 		case MNT_COUNT_REF:
1943 			sum += mpcpu->mntp_ref;
1944 			break;
1945 		case MNT_COUNT_LOCKREF:
1946 			sum += mpcpu->mntp_lockref;
1947 			break;
1948 		case MNT_COUNT_WRITEOPCOUNT:
1949 			sum += mpcpu->mntp_writeopcount;
1950 			break;
1951 		}
1952 	}
1953 	return (sum);
1954 }
1955 
1956 static bool
1957 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
1958     int timeout_ticks)
1959 {
1960 	bool enqueued;
1961 
1962 	enqueued = false;
1963 	mtx_lock(&deferred_unmount_lock);
1964 	if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
1965 		mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
1966 		STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
1967 		    mnt_taskqueue_link);
1968 		enqueued = true;
1969 	}
1970 	mtx_unlock(&deferred_unmount_lock);
1971 
1972 	if (enqueued) {
1973 		taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
1974 		    &deferred_unmount_task, timeout_ticks);
1975 	}
1976 
1977 	return (enqueued);
1978 }
1979 
1980 /*
1981  * Taskqueue handler for processing async/recursive unmounts
1982  */
1983 static void
1984 vfs_deferred_unmount(void *argi __unused, int pending __unused)
1985 {
1986 	STAILQ_HEAD(, mount) local_unmounts;
1987 	uint64_t flags;
1988 	struct mount *mp, *tmp;
1989 	int error;
1990 	unsigned int retries;
1991 	bool unmounted;
1992 
1993 	STAILQ_INIT(&local_unmounts);
1994 	mtx_lock(&deferred_unmount_lock);
1995 	STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
1996 	mtx_unlock(&deferred_unmount_lock);
1997 
1998 	STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
1999 		flags = mp->mnt_taskqueue_flags;
2000 		KASSERT((flags & MNT_DEFERRED) != 0,
2001 		    ("taskqueue unmount without MNT_DEFERRED"));
2002 		error = dounmount(mp, flags, curthread);
2003 		if (error != 0) {
2004 			MNT_ILOCK(mp);
2005 			unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
2006 			MNT_IUNLOCK(mp);
2007 
2008 			/*
2009 			 * The deferred unmount thread is the only thread that
2010 			 * modifies the retry counts, so locking/atomics aren't
2011 			 * needed here.
2012 			 */
2013 			retries = (mp->mnt_unmount_retries)++;
2014 			deferred_unmount_total_retries++;
2015 			if (!unmounted && retries < deferred_unmount_retry_limit) {
2016 				deferred_unmount_enqueue(mp, flags, true,
2017 				    -deferred_unmount_retry_delay_hz);
2018 			} else {
2019 				if (retries >= deferred_unmount_retry_limit) {
2020 					printf("giving up on deferred unmount "
2021 					    "of %s after %d retries, error %d\n",
2022 					    mp->mnt_stat.f_mntonname, retries, error);
2023 				}
2024 				vfs_rel(mp);
2025 			}
2026 		}
2027 	}
2028 }
2029 
2030 /*
2031  * Do the actual filesystem unmount.
2032  */
2033 int
2034 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2035 {
2036 	struct mount_upper_node *upper;
2037 	struct vnode *coveredvp, *rootvp;
2038 	int error;
2039 	uint64_t async_flag;
2040 	int mnt_gen_r;
2041 	unsigned int retries;
2042 
2043 	KASSERT((flags & MNT_DEFERRED) == 0 ||
2044 	    (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2045 	    ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2046 
2047 	/*
2048 	 * If the caller has explicitly requested the unmount to be handled by
2049 	 * the taskqueue and we're not already in taskqueue context, queue
2050 	 * up the unmount request and exit.  This is done prior to any
2051 	 * credential checks; MNT_DEFERRED should be used only for kernel-
2052 	 * initiated unmounts and will therefore be processed with the
2053 	 * (kernel) credentials of the taskqueue thread.  Still, callers
2054 	 * should be sure this is the behavior they want.
2055 	 */
2056 	if ((flags & MNT_DEFERRED) != 0 &&
2057 	    taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2058 		if (!deferred_unmount_enqueue(mp, flags, false, 0))
2059 			vfs_rel(mp);
2060 		return (EINPROGRESS);
2061 	}
2062 
2063 	/*
2064 	 * Only privileged root, or (if MNT_USER is set) the user that did the
2065 	 * original mount is permitted to unmount this filesystem.
2066 	 * This check should be made prior to queueing up any recursive
2067 	 * unmounts of upper filesystems.  Those unmounts will be executed
2068 	 * with kernel thread credentials and are expected to succeed, so
2069 	 * we must at least ensure the originating context has sufficient
2070 	 * privilege to unmount the base filesystem before proceeding with
2071 	 * the uppers.
2072 	 */
2073 	error = vfs_suser(mp, td);
2074 	if (error != 0) {
2075 		KASSERT((flags & MNT_DEFERRED) == 0,
2076 		    ("taskqueue unmount with insufficient privilege"));
2077 		vfs_rel(mp);
2078 		return (error);
2079 	}
2080 
2081 	if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2082 		flags |= MNT_RECURSE;
2083 
2084 	if ((flags & MNT_RECURSE) != 0) {
2085 		KASSERT((flags & MNT_FORCE) != 0,
2086 		    ("MNT_RECURSE requires MNT_FORCE"));
2087 
2088 		MNT_ILOCK(mp);
2089 		/*
2090 		 * Set MNTK_RECURSE to prevent new upper mounts from being
2091 		 * added, and note that an operation on the uppers list is in
2092 		 * progress.  This will ensure that unregistration from the
2093 		 * uppers list, and therefore any pending unmount of the upper
2094 		 * FS, can't complete until after we finish walking the list.
2095 		 */
2096 		mp->mnt_kern_flag |= MNTK_RECURSE;
2097 		mp->mnt_upper_pending++;
2098 		TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2099 			retries = upper->mp->mnt_unmount_retries;
2100 			if (retries > deferred_unmount_retry_limit) {
2101 				error = EBUSY;
2102 				continue;
2103 			}
2104 			MNT_IUNLOCK(mp);
2105 
2106 			vfs_ref(upper->mp);
2107 			if (!deferred_unmount_enqueue(upper->mp, flags,
2108 			    false, 0))
2109 				vfs_rel(upper->mp);
2110 			MNT_ILOCK(mp);
2111 		}
2112 		mp->mnt_upper_pending--;
2113 		if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2114 		    mp->mnt_upper_pending == 0) {
2115 			mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2116 			wakeup(&mp->mnt_uppers);
2117 		}
2118 
2119 		/*
2120 		 * If we're not on the taskqueue, wait until the uppers list
2121 		 * is drained before proceeding with unmount.  Otherwise, if
2122 		 * we are on the taskqueue and there are still pending uppers,
2123 		 * just re-enqueue on the end of the taskqueue.
2124 		 */
2125 		if ((flags & MNT_DEFERRED) == 0) {
2126 			while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2127 				mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2128 				error = msleep(&mp->mnt_taskqueue_link,
2129 				    MNT_MTX(mp), PCATCH, "umntqw", 0);
2130 			}
2131 			if (error != 0) {
2132 				MNT_REL(mp);
2133 				MNT_IUNLOCK(mp);
2134 				return (error);
2135 			}
2136 		} else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2137 			MNT_IUNLOCK(mp);
2138 			if (error == 0)
2139 				deferred_unmount_enqueue(mp, flags, true, 0);
2140 			return (error);
2141 		}
2142 		MNT_IUNLOCK(mp);
2143 		KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2144 	}
2145 
2146 	/* Allow the taskqueue to safely re-enqueue on failure */
2147 	if ((flags & MNT_DEFERRED) != 0)
2148 		vfs_ref(mp);
2149 
2150 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2151 		mnt_gen_r = mp->mnt_gen;
2152 		VI_LOCK(coveredvp);
2153 		vholdl(coveredvp);
2154 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2155 		/*
2156 		 * Check for mp being unmounted while waiting for the
2157 		 * covered vnode lock.
2158 		 */
2159 		if (coveredvp->v_mountedhere != mp ||
2160 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2161 			VOP_UNLOCK(coveredvp);
2162 			vdrop(coveredvp);
2163 			vfs_rel(mp);
2164 			return (EBUSY);
2165 		}
2166 	}
2167 
2168 	vfs_op_enter(mp);
2169 
2170 	vn_start_write(NULL, &mp, V_WAIT);
2171 	MNT_ILOCK(mp);
2172 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2173 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
2174 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
2175 		dounmount_cleanup(mp, coveredvp, 0);
2176 		return (EBUSY);
2177 	}
2178 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
2179 	rootvp = vfs_cache_root_clear(mp);
2180 	if (coveredvp != NULL)
2181 		vn_seqc_write_begin(coveredvp);
2182 	if (flags & MNT_NONBUSY) {
2183 		MNT_IUNLOCK(mp);
2184 		error = vfs_check_usecounts(mp);
2185 		MNT_ILOCK(mp);
2186 		if (error != 0) {
2187 			vn_seqc_write_end(coveredvp);
2188 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2189 			if (rootvp != NULL) {
2190 				vn_seqc_write_end(rootvp);
2191 				vrele(rootvp);
2192 			}
2193 			return (error);
2194 		}
2195 	}
2196 	/* Allow filesystems to detect that a forced unmount is in progress. */
2197 	if (flags & MNT_FORCE) {
2198 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2199 		MNT_IUNLOCK(mp);
2200 		/*
2201 		 * Must be done after setting MNTK_UNMOUNTF and before
2202 		 * waiting for mnt_lockref to become 0.
2203 		 */
2204 		VFS_PURGE(mp);
2205 		MNT_ILOCK(mp);
2206 	}
2207 	error = 0;
2208 	if (mp->mnt_lockref) {
2209 		mp->mnt_kern_flag |= MNTK_DRAINING;
2210 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2211 		    "mount drain", 0);
2212 	}
2213 	MNT_IUNLOCK(mp);
2214 	KASSERT(mp->mnt_lockref == 0,
2215 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
2216 	    __func__, __FILE__, __LINE__));
2217 	KASSERT(error == 0,
2218 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
2219 	    __func__, __FILE__, __LINE__));
2220 
2221 	/*
2222 	 * We want to keep the vnode around so that we can vn_seqc_write_end
2223 	 * after we are done with unmount. Downgrade our reference to a mere
2224 	 * hold count so that we don't interefere with anything.
2225 	 */
2226 	if (rootvp != NULL) {
2227 		vhold(rootvp);
2228 		vrele(rootvp);
2229 	}
2230 
2231 	if (mp->mnt_flag & MNT_EXPUBLIC)
2232 		vfs_setpublicfs(NULL, NULL, NULL);
2233 
2234 	vfs_periodic(mp, MNT_WAIT);
2235 	MNT_ILOCK(mp);
2236 	async_flag = mp->mnt_flag & MNT_ASYNC;
2237 	mp->mnt_flag &= ~MNT_ASYNC;
2238 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
2239 	MNT_IUNLOCK(mp);
2240 	vfs_deallocate_syncvnode(mp);
2241 	error = VFS_UNMOUNT(mp, flags);
2242 	vn_finished_write(mp);
2243 	vfs_rel(mp);
2244 	/*
2245 	 * If we failed to flush the dirty blocks for this mount point,
2246 	 * undo all the cdir/rdir and rootvnode changes we made above.
2247 	 * Unless we failed to do so because the device is reporting that
2248 	 * it doesn't exist anymore.
2249 	 */
2250 	if (error && error != ENXIO) {
2251 		MNT_ILOCK(mp);
2252 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2253 			MNT_IUNLOCK(mp);
2254 			vfs_allocate_syncvnode(mp);
2255 			MNT_ILOCK(mp);
2256 		}
2257 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2258 		mp->mnt_flag |= async_flag;
2259 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2260 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2261 			mp->mnt_kern_flag |= MNTK_ASYNC;
2262 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
2263 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
2264 			wakeup(mp);
2265 		}
2266 		vfs_op_exit_locked(mp);
2267 		MNT_IUNLOCK(mp);
2268 		if (coveredvp) {
2269 			vn_seqc_write_end(coveredvp);
2270 			VOP_UNLOCK(coveredvp);
2271 			vdrop(coveredvp);
2272 		}
2273 		if (rootvp != NULL) {
2274 			vn_seqc_write_end(rootvp);
2275 			vdrop(rootvp);
2276 		}
2277 		return (error);
2278 	}
2279 
2280 	mtx_lock(&mountlist_mtx);
2281 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
2282 	mtx_unlock(&mountlist_mtx);
2283 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2284 	if (coveredvp != NULL) {
2285 		VI_LOCK(coveredvp);
2286 		vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2287 		coveredvp->v_mountedhere = NULL;
2288 		vn_seqc_write_end_locked(coveredvp);
2289 		VI_UNLOCK(coveredvp);
2290 		VOP_UNLOCK(coveredvp);
2291 		vdrop(coveredvp);
2292 	}
2293 	mount_devctl_event("UNMOUNT", mp, false);
2294 	if (rootvp != NULL) {
2295 		vn_seqc_write_end(rootvp);
2296 		vdrop(rootvp);
2297 	}
2298 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2299 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
2300 		vrele(rootvnode);
2301 		rootvnode = NULL;
2302 	}
2303 	if (mp == rootdevmp)
2304 		rootdevmp = NULL;
2305 	if ((flags & MNT_DEFERRED) != 0)
2306 		vfs_rel(mp);
2307 	vfs_mount_destroy(mp);
2308 	return (0);
2309 }
2310 
2311 /*
2312  * Report errors during filesystem mounting.
2313  */
2314 void
2315 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2316 {
2317 	struct vfsoptlist *moptlist = mp->mnt_optnew;
2318 	va_list ap;
2319 	int error, len;
2320 	char *errmsg;
2321 
2322 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2323 	if (error || errmsg == NULL || len <= 0)
2324 		return;
2325 
2326 	va_start(ap, fmt);
2327 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2328 	va_end(ap);
2329 }
2330 
2331 void
2332 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2333 {
2334 	va_list ap;
2335 	int error, len;
2336 	char *errmsg;
2337 
2338 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2339 	if (error || errmsg == NULL || len <= 0)
2340 		return;
2341 
2342 	va_start(ap, fmt);
2343 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2344 	va_end(ap);
2345 }
2346 
2347 /*
2348  * ---------------------------------------------------------------------
2349  * Functions for querying mount options/arguments from filesystems.
2350  */
2351 
2352 /*
2353  * Check that no unknown options are given
2354  */
2355 int
2356 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2357 {
2358 	struct vfsopt *opt;
2359 	char errmsg[255];
2360 	const char **t, *p, *q;
2361 	int ret = 0;
2362 
2363 	TAILQ_FOREACH(opt, opts, link) {
2364 		p = opt->name;
2365 		q = NULL;
2366 		if (p[0] == 'n' && p[1] == 'o')
2367 			q = p + 2;
2368 		for(t = global_opts; *t != NULL; t++) {
2369 			if (strcmp(*t, p) == 0)
2370 				break;
2371 			if (q != NULL) {
2372 				if (strcmp(*t, q) == 0)
2373 					break;
2374 			}
2375 		}
2376 		if (*t != NULL)
2377 			continue;
2378 		for(t = legal; *t != NULL; t++) {
2379 			if (strcmp(*t, p) == 0)
2380 				break;
2381 			if (q != NULL) {
2382 				if (strcmp(*t, q) == 0)
2383 					break;
2384 			}
2385 		}
2386 		if (*t != NULL)
2387 			continue;
2388 		snprintf(errmsg, sizeof(errmsg),
2389 		    "mount option <%s> is unknown", p);
2390 		ret = EINVAL;
2391 	}
2392 	if (ret != 0) {
2393 		TAILQ_FOREACH(opt, opts, link) {
2394 			if (strcmp(opt->name, "errmsg") == 0) {
2395 				strncpy((char *)opt->value, errmsg, opt->len);
2396 				break;
2397 			}
2398 		}
2399 		if (opt == NULL)
2400 			printf("%s\n", errmsg);
2401 	}
2402 	return (ret);
2403 }
2404 
2405 /*
2406  * Get a mount option by its name.
2407  *
2408  * Return 0 if the option was found, ENOENT otherwise.
2409  * If len is non-NULL it will be filled with the length
2410  * of the option. If buf is non-NULL, it will be filled
2411  * with the address of the option.
2412  */
2413 int
2414 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2415 {
2416 	struct vfsopt *opt;
2417 
2418 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2419 
2420 	TAILQ_FOREACH(opt, opts, link) {
2421 		if (strcmp(name, opt->name) == 0) {
2422 			opt->seen = 1;
2423 			if (len != NULL)
2424 				*len = opt->len;
2425 			if (buf != NULL)
2426 				*buf = opt->value;
2427 			return (0);
2428 		}
2429 	}
2430 	return (ENOENT);
2431 }
2432 
2433 int
2434 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2435 {
2436 	struct vfsopt *opt;
2437 
2438 	if (opts == NULL)
2439 		return (-1);
2440 
2441 	TAILQ_FOREACH(opt, opts, link) {
2442 		if (strcmp(name, opt->name) == 0) {
2443 			opt->seen = 1;
2444 			return (opt->pos);
2445 		}
2446 	}
2447 	return (-1);
2448 }
2449 
2450 int
2451 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2452 {
2453 	char *opt_value, *vtp;
2454 	quad_t iv;
2455 	int error, opt_len;
2456 
2457 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2458 	if (error != 0)
2459 		return (error);
2460 	if (opt_len == 0 || opt_value == NULL)
2461 		return (EINVAL);
2462 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2463 		return (EINVAL);
2464 	iv = strtoq(opt_value, &vtp, 0);
2465 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2466 		return (EINVAL);
2467 	if (iv < 0)
2468 		return (EINVAL);
2469 	switch (vtp[0]) {
2470 	case 't': case 'T':
2471 		iv *= 1024;
2472 		/* FALLTHROUGH */
2473 	case 'g': case 'G':
2474 		iv *= 1024;
2475 		/* FALLTHROUGH */
2476 	case 'm': case 'M':
2477 		iv *= 1024;
2478 		/* FALLTHROUGH */
2479 	case 'k': case 'K':
2480 		iv *= 1024;
2481 	case '\0':
2482 		break;
2483 	default:
2484 		return (EINVAL);
2485 	}
2486 	*value = iv;
2487 
2488 	return (0);
2489 }
2490 
2491 char *
2492 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2493 {
2494 	struct vfsopt *opt;
2495 
2496 	*error = 0;
2497 	TAILQ_FOREACH(opt, opts, link) {
2498 		if (strcmp(name, opt->name) != 0)
2499 			continue;
2500 		opt->seen = 1;
2501 		if (opt->len == 0 ||
2502 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2503 			*error = EINVAL;
2504 			return (NULL);
2505 		}
2506 		return (opt->value);
2507 	}
2508 	*error = ENOENT;
2509 	return (NULL);
2510 }
2511 
2512 int
2513 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2514 	uint64_t val)
2515 {
2516 	struct vfsopt *opt;
2517 
2518 	TAILQ_FOREACH(opt, opts, link) {
2519 		if (strcmp(name, opt->name) == 0) {
2520 			opt->seen = 1;
2521 			if (w != NULL)
2522 				*w |= val;
2523 			return (1);
2524 		}
2525 	}
2526 	if (w != NULL)
2527 		*w &= ~val;
2528 	return (0);
2529 }
2530 
2531 int
2532 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2533 {
2534 	va_list ap;
2535 	struct vfsopt *opt;
2536 	int ret;
2537 
2538 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2539 
2540 	TAILQ_FOREACH(opt, opts, link) {
2541 		if (strcmp(name, opt->name) != 0)
2542 			continue;
2543 		opt->seen = 1;
2544 		if (opt->len == 0 || opt->value == NULL)
2545 			return (0);
2546 		if (((char *)opt->value)[opt->len - 1] != '\0')
2547 			return (0);
2548 		va_start(ap, fmt);
2549 		ret = vsscanf(opt->value, fmt, ap);
2550 		va_end(ap);
2551 		return (ret);
2552 	}
2553 	return (0);
2554 }
2555 
2556 int
2557 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2558 {
2559 	struct vfsopt *opt;
2560 
2561 	TAILQ_FOREACH(opt, opts, link) {
2562 		if (strcmp(name, opt->name) != 0)
2563 			continue;
2564 		opt->seen = 1;
2565 		if (opt->value == NULL)
2566 			opt->len = len;
2567 		else {
2568 			if (opt->len != len)
2569 				return (EINVAL);
2570 			bcopy(value, opt->value, len);
2571 		}
2572 		return (0);
2573 	}
2574 	return (ENOENT);
2575 }
2576 
2577 int
2578 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2579 {
2580 	struct vfsopt *opt;
2581 
2582 	TAILQ_FOREACH(opt, opts, link) {
2583 		if (strcmp(name, opt->name) != 0)
2584 			continue;
2585 		opt->seen = 1;
2586 		if (opt->value == NULL)
2587 			opt->len = len;
2588 		else {
2589 			if (opt->len < len)
2590 				return (EINVAL);
2591 			opt->len = len;
2592 			bcopy(value, opt->value, len);
2593 		}
2594 		return (0);
2595 	}
2596 	return (ENOENT);
2597 }
2598 
2599 int
2600 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2601 {
2602 	struct vfsopt *opt;
2603 
2604 	TAILQ_FOREACH(opt, opts, link) {
2605 		if (strcmp(name, opt->name) != 0)
2606 			continue;
2607 		opt->seen = 1;
2608 		if (opt->value == NULL)
2609 			opt->len = strlen(value) + 1;
2610 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2611 			return (EINVAL);
2612 		return (0);
2613 	}
2614 	return (ENOENT);
2615 }
2616 
2617 /*
2618  * Find and copy a mount option.
2619  *
2620  * The size of the buffer has to be specified
2621  * in len, if it is not the same length as the
2622  * mount option, EINVAL is returned.
2623  * Returns ENOENT if the option is not found.
2624  */
2625 int
2626 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2627 {
2628 	struct vfsopt *opt;
2629 
2630 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2631 
2632 	TAILQ_FOREACH(opt, opts, link) {
2633 		if (strcmp(name, opt->name) == 0) {
2634 			opt->seen = 1;
2635 			if (len != opt->len)
2636 				return (EINVAL);
2637 			bcopy(opt->value, dest, opt->len);
2638 			return (0);
2639 		}
2640 	}
2641 	return (ENOENT);
2642 }
2643 
2644 int
2645 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2646 {
2647 	/*
2648 	 * Filesystems only fill in part of the structure for updates, we
2649 	 * have to read the entirety first to get all content.
2650 	 */
2651 	if (sbp != &mp->mnt_stat)
2652 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2653 
2654 	/*
2655 	 * Set these in case the underlying filesystem fails to do so.
2656 	 */
2657 	sbp->f_version = STATFS_VERSION;
2658 	sbp->f_namemax = NAME_MAX;
2659 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2660 	sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2661 
2662 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2663 }
2664 
2665 void
2666 vfs_mountedfrom(struct mount *mp, const char *from)
2667 {
2668 
2669 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2670 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2671 	    sizeof mp->mnt_stat.f_mntfromname);
2672 }
2673 
2674 /*
2675  * ---------------------------------------------------------------------
2676  * This is the api for building mount args and mounting filesystems from
2677  * inside the kernel.
2678  *
2679  * The API works by accumulation of individual args.  First error is
2680  * latched.
2681  *
2682  * XXX: should be documented in new manpage kernel_mount(9)
2683  */
2684 
2685 /* A memory allocation which must be freed when we are done */
2686 struct mntaarg {
2687 	SLIST_ENTRY(mntaarg)	next;
2688 };
2689 
2690 /* The header for the mount arguments */
2691 struct mntarg {
2692 	struct iovec *v;
2693 	int len;
2694 	int error;
2695 	SLIST_HEAD(, mntaarg)	list;
2696 };
2697 
2698 /*
2699  * Add a boolean argument.
2700  *
2701  * flag is the boolean value.
2702  * name must start with "no".
2703  */
2704 struct mntarg *
2705 mount_argb(struct mntarg *ma, int flag, const char *name)
2706 {
2707 
2708 	KASSERT(name[0] == 'n' && name[1] == 'o',
2709 	    ("mount_argb(...,%s): name must start with 'no'", name));
2710 
2711 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2712 }
2713 
2714 /*
2715  * Add an argument printf style
2716  */
2717 struct mntarg *
2718 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2719 {
2720 	va_list ap;
2721 	struct mntaarg *maa;
2722 	struct sbuf *sb;
2723 	int len;
2724 
2725 	if (ma == NULL) {
2726 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2727 		SLIST_INIT(&ma->list);
2728 	}
2729 	if (ma->error)
2730 		return (ma);
2731 
2732 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2733 	    M_MOUNT, M_WAITOK);
2734 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2735 	ma->v[ma->len].iov_len = strlen(name) + 1;
2736 	ma->len++;
2737 
2738 	sb = sbuf_new_auto();
2739 	va_start(ap, fmt);
2740 	sbuf_vprintf(sb, fmt, ap);
2741 	va_end(ap);
2742 	sbuf_finish(sb);
2743 	len = sbuf_len(sb) + 1;
2744 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2745 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2746 	bcopy(sbuf_data(sb), maa + 1, len);
2747 	sbuf_delete(sb);
2748 
2749 	ma->v[ma->len].iov_base = maa + 1;
2750 	ma->v[ma->len].iov_len = len;
2751 	ma->len++;
2752 
2753 	return (ma);
2754 }
2755 
2756 /*
2757  * Add an argument which is a userland string.
2758  */
2759 struct mntarg *
2760 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2761 {
2762 	struct mntaarg *maa;
2763 	char *tbuf;
2764 
2765 	if (val == NULL)
2766 		return (ma);
2767 	if (ma == NULL) {
2768 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2769 		SLIST_INIT(&ma->list);
2770 	}
2771 	if (ma->error)
2772 		return (ma);
2773 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2774 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2775 	tbuf = (void *)(maa + 1);
2776 	ma->error = copyinstr(val, tbuf, len, NULL);
2777 	return (mount_arg(ma, name, tbuf, -1));
2778 }
2779 
2780 /*
2781  * Plain argument.
2782  *
2783  * If length is -1, treat value as a C string.
2784  */
2785 struct mntarg *
2786 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2787 {
2788 
2789 	if (ma == NULL) {
2790 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2791 		SLIST_INIT(&ma->list);
2792 	}
2793 	if (ma->error)
2794 		return (ma);
2795 
2796 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2797 	    M_MOUNT, M_WAITOK);
2798 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2799 	ma->v[ma->len].iov_len = strlen(name) + 1;
2800 	ma->len++;
2801 
2802 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2803 	if (len < 0)
2804 		ma->v[ma->len].iov_len = strlen(val) + 1;
2805 	else
2806 		ma->v[ma->len].iov_len = len;
2807 	ma->len++;
2808 	return (ma);
2809 }
2810 
2811 /*
2812  * Free a mntarg structure
2813  */
2814 static void
2815 free_mntarg(struct mntarg *ma)
2816 {
2817 	struct mntaarg *maa;
2818 
2819 	while (!SLIST_EMPTY(&ma->list)) {
2820 		maa = SLIST_FIRST(&ma->list);
2821 		SLIST_REMOVE_HEAD(&ma->list, next);
2822 		free(maa, M_MOUNT);
2823 	}
2824 	free(ma->v, M_MOUNT);
2825 	free(ma, M_MOUNT);
2826 }
2827 
2828 /*
2829  * Mount a filesystem
2830  */
2831 int
2832 kernel_mount(struct mntarg *ma, uint64_t flags)
2833 {
2834 	struct uio auio;
2835 	int error;
2836 
2837 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2838 	KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2839 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2840 
2841 	error = ma->error;
2842 	if (error == 0) {
2843 		auio.uio_iov = ma->v;
2844 		auio.uio_iovcnt = ma->len;
2845 		auio.uio_segflg = UIO_SYSSPACE;
2846 		error = vfs_donmount(curthread, flags, &auio);
2847 	}
2848 	free_mntarg(ma);
2849 	return (error);
2850 }
2851 
2852 /* Map from mount options to printable formats. */
2853 static struct mntoptnames optnames[] = {
2854 	MNTOPT_NAMES
2855 };
2856 
2857 #define DEVCTL_LEN 1024
2858 static void
2859 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2860 {
2861 	const uint8_t *cp;
2862 	struct mntoptnames *fp;
2863 	struct sbuf sb;
2864 	struct statfs *sfp = &mp->mnt_stat;
2865 	char *buf;
2866 
2867 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2868 	if (buf == NULL)
2869 		return;
2870 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2871 	sbuf_cpy(&sb, "mount-point=\"");
2872 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2873 	sbuf_cat(&sb, "\" mount-dev=\"");
2874 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2875 	sbuf_cat(&sb, "\" mount-type=\"");
2876 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2877 	sbuf_cat(&sb, "\" fsid=0x");
2878 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2879 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2880 		sbuf_printf(&sb, "%02x", cp[i]);
2881 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2882 	for (fp = optnames; fp->o_opt != 0; fp++) {
2883 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2884 			sbuf_cat(&sb, fp->o_name);
2885 			sbuf_putc(&sb, ';');
2886 		}
2887 	}
2888 	sbuf_putc(&sb, '"');
2889 	sbuf_finish(&sb);
2890 
2891 	/*
2892 	 * Options are not published because the form of the options depends on
2893 	 * the file system and may include binary data. In addition, they don't
2894 	 * necessarily provide enough useful information to be actionable when
2895 	 * devd processes them.
2896 	 */
2897 
2898 	if (sbuf_error(&sb) == 0)
2899 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2900 	sbuf_delete(&sb);
2901 	free(buf, M_MOUNT);
2902 }
2903 
2904 /*
2905  * Force remount specified mount point to read-only.  The argument
2906  * must be busied to avoid parallel unmount attempts.
2907  *
2908  * Intended use is to prevent further writes if some metadata
2909  * inconsistency is detected.  Note that the function still flushes
2910  * all cached metadata and data for the mount point, which might be
2911  * not always suitable.
2912  */
2913 int
2914 vfs_remount_ro(struct mount *mp)
2915 {
2916 	struct vfsoptlist *opts;
2917 	struct vfsopt *opt;
2918 	struct vnode *vp_covered, *rootvp;
2919 	int error;
2920 
2921 	KASSERT(mp->mnt_lockref > 0,
2922 	    ("vfs_remount_ro: mp %p is not busied", mp));
2923 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
2924 	    ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
2925 
2926 	rootvp = NULL;
2927 	vp_covered = mp->mnt_vnodecovered;
2928 	error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
2929 	if (error != 0)
2930 		return (error);
2931 	VI_LOCK(vp_covered);
2932 	if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
2933 		VI_UNLOCK(vp_covered);
2934 		vput(vp_covered);
2935 		return (EBUSY);
2936 	}
2937 	vp_covered->v_iflag |= VI_MOUNT;
2938 	VI_UNLOCK(vp_covered);
2939 	vfs_op_enter(mp);
2940 	vn_seqc_write_begin(vp_covered);
2941 
2942 	MNT_ILOCK(mp);
2943 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
2944 		MNT_IUNLOCK(mp);
2945 		error = EBUSY;
2946 		goto out;
2947 	}
2948 	mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
2949 	rootvp = vfs_cache_root_clear(mp);
2950 	MNT_IUNLOCK(mp);
2951 
2952 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
2953 	TAILQ_INIT(opts);
2954 	opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
2955 	opt->name = strdup("ro", M_MOUNT);
2956 	opt->value = NULL;
2957 	TAILQ_INSERT_TAIL(opts, opt, link);
2958 	vfs_mergeopts(opts, mp->mnt_opt);
2959 	mp->mnt_optnew = opts;
2960 
2961 	error = VFS_MOUNT(mp);
2962 
2963 	if (error == 0) {
2964 		MNT_ILOCK(mp);
2965 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
2966 		MNT_IUNLOCK(mp);
2967 		vfs_deallocate_syncvnode(mp);
2968 		if (mp->mnt_opt != NULL)
2969 			vfs_freeopts(mp->mnt_opt);
2970 		mp->mnt_opt = mp->mnt_optnew;
2971 	} else {
2972 		MNT_ILOCK(mp);
2973 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
2974 		MNT_IUNLOCK(mp);
2975 		vfs_freeopts(mp->mnt_optnew);
2976 	}
2977 	mp->mnt_optnew = NULL;
2978 
2979 out:
2980 	vfs_op_exit(mp);
2981 	VI_LOCK(vp_covered);
2982 	vp_covered->v_iflag &= ~VI_MOUNT;
2983 	VI_UNLOCK(vp_covered);
2984 	vput(vp_covered);
2985 	vn_seqc_write_end(vp_covered);
2986 	if (rootvp != NULL) {
2987 		vn_seqc_write_end(rootvp);
2988 		vrele(rootvp);
2989 	}
2990 	return (error);
2991 }
2992 
2993 /*
2994  * Suspend write operations on all local writeable filesystems.  Does
2995  * full sync of them in the process.
2996  *
2997  * Iterate over the mount points in reverse order, suspending most
2998  * recently mounted filesystems first.  It handles a case where a
2999  * filesystem mounted from a md(4) vnode-backed device should be
3000  * suspended before the filesystem that owns the vnode.
3001  */
3002 void
3003 suspend_all_fs(void)
3004 {
3005 	struct mount *mp;
3006 	int error;
3007 
3008 	mtx_lock(&mountlist_mtx);
3009 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
3010 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
3011 		if (error != 0)
3012 			continue;
3013 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
3014 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
3015 			mtx_lock(&mountlist_mtx);
3016 			vfs_unbusy(mp);
3017 			continue;
3018 		}
3019 		error = vfs_write_suspend(mp, 0);
3020 		if (error == 0) {
3021 			MNT_ILOCK(mp);
3022 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
3023 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
3024 			MNT_IUNLOCK(mp);
3025 			mtx_lock(&mountlist_mtx);
3026 		} else {
3027 			printf("suspend of %s failed, error %d\n",
3028 			    mp->mnt_stat.f_mntonname, error);
3029 			mtx_lock(&mountlist_mtx);
3030 			vfs_unbusy(mp);
3031 		}
3032 	}
3033 	mtx_unlock(&mountlist_mtx);
3034 }
3035 
3036 void
3037 resume_all_fs(void)
3038 {
3039 	struct mount *mp;
3040 
3041 	mtx_lock(&mountlist_mtx);
3042 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3043 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3044 			continue;
3045 		mtx_unlock(&mountlist_mtx);
3046 		MNT_ILOCK(mp);
3047 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3048 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3049 		MNT_IUNLOCK(mp);
3050 		vfs_write_resume(mp, 0);
3051 		mtx_lock(&mountlist_mtx);
3052 		vfs_unbusy(mp);
3053 	}
3054 	mtx_unlock(&mountlist_mtx);
3055 }
3056