xref: /freebsd/sys/kern/vfs_mount.c (revision 0c428864495af9dc7d2af4d0a5ae21732af9c739)
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 	if (mp->mnt_exjail != NULL) {
765 		atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt,
766 		    1);
767 		crfree(mp->mnt_exjail);
768 	}
769 	if (mp->mnt_export != NULL) {
770 		vfs_free_addrlist(mp->mnt_export);
771 		free(mp->mnt_export, M_MOUNT);
772 	}
773 	crfree(mp->mnt_cred);
774 	uma_zfree(mount_zone, mp);
775 }
776 
777 static bool
778 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
779 {
780 	/* This is an upgrade of an exisiting mount. */
781 	if ((fsflags & MNT_UPDATE) != 0)
782 		return (false);
783 	/* This is already an R/O mount. */
784 	if ((fsflags & MNT_RDONLY) != 0)
785 		return (false);
786 
787 	switch (error) {
788 	case ENODEV:	/* generic, geom, ... */
789 	case EACCES:	/* cam/scsi, ... */
790 	case EROFS:	/* md, mmcsd, ... */
791 		/*
792 		 * These errors can be returned by the storage layer to signal
793 		 * that the media is read-only.  No harm in the R/O mount
794 		 * attempt if the error was returned for some other reason.
795 		 */
796 		return (true);
797 	default:
798 		return (false);
799 	}
800 }
801 
802 int
803 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
804 {
805 	struct vfsoptlist *optlist;
806 	struct vfsopt *opt, *tmp_opt;
807 	char *fstype, *fspath, *errmsg;
808 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
809 	bool autoro;
810 
811 	errmsg = fspath = NULL;
812 	errmsg_len = fspathlen = 0;
813 	errmsg_pos = -1;
814 	autoro = default_autoro;
815 
816 	error = vfs_buildopts(fsoptions, &optlist);
817 	if (error)
818 		return (error);
819 
820 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
821 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
822 
823 	/*
824 	 * We need these two options before the others,
825 	 * and they are mandatory for any filesystem.
826 	 * Ensure they are NUL terminated as well.
827 	 */
828 	fstypelen = 0;
829 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
830 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
831 		error = EINVAL;
832 		if (errmsg != NULL)
833 			strncpy(errmsg, "Invalid fstype", errmsg_len);
834 		goto bail;
835 	}
836 	fspathlen = 0;
837 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
838 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
839 		error = EINVAL;
840 		if (errmsg != NULL)
841 			strncpy(errmsg, "Invalid fspath", errmsg_len);
842 		goto bail;
843 	}
844 
845 	/*
846 	 * We need to see if we have the "update" option
847 	 * before we call vfs_domount(), since vfs_domount() has special
848 	 * logic based on MNT_UPDATE.  This is very important
849 	 * when we want to update the root filesystem.
850 	 */
851 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
852 		int do_freeopt = 0;
853 
854 		if (strcmp(opt->name, "update") == 0) {
855 			fsflags |= MNT_UPDATE;
856 			do_freeopt = 1;
857 		}
858 		else if (strcmp(opt->name, "async") == 0)
859 			fsflags |= MNT_ASYNC;
860 		else if (strcmp(opt->name, "force") == 0) {
861 			fsflags |= MNT_FORCE;
862 			do_freeopt = 1;
863 		}
864 		else if (strcmp(opt->name, "reload") == 0) {
865 			fsflags |= MNT_RELOAD;
866 			do_freeopt = 1;
867 		}
868 		else if (strcmp(opt->name, "multilabel") == 0)
869 			fsflags |= MNT_MULTILABEL;
870 		else if (strcmp(opt->name, "noasync") == 0)
871 			fsflags &= ~MNT_ASYNC;
872 		else if (strcmp(opt->name, "noatime") == 0)
873 			fsflags |= MNT_NOATIME;
874 		else if (strcmp(opt->name, "atime") == 0) {
875 			free(opt->name, M_MOUNT);
876 			opt->name = strdup("nonoatime", M_MOUNT);
877 		}
878 		else if (strcmp(opt->name, "noclusterr") == 0)
879 			fsflags |= MNT_NOCLUSTERR;
880 		else if (strcmp(opt->name, "clusterr") == 0) {
881 			free(opt->name, M_MOUNT);
882 			opt->name = strdup("nonoclusterr", M_MOUNT);
883 		}
884 		else if (strcmp(opt->name, "noclusterw") == 0)
885 			fsflags |= MNT_NOCLUSTERW;
886 		else if (strcmp(opt->name, "clusterw") == 0) {
887 			free(opt->name, M_MOUNT);
888 			opt->name = strdup("nonoclusterw", M_MOUNT);
889 		}
890 		else if (strcmp(opt->name, "noexec") == 0)
891 			fsflags |= MNT_NOEXEC;
892 		else if (strcmp(opt->name, "exec") == 0) {
893 			free(opt->name, M_MOUNT);
894 			opt->name = strdup("nonoexec", M_MOUNT);
895 		}
896 		else if (strcmp(opt->name, "nosuid") == 0)
897 			fsflags |= MNT_NOSUID;
898 		else if (strcmp(opt->name, "suid") == 0) {
899 			free(opt->name, M_MOUNT);
900 			opt->name = strdup("nonosuid", M_MOUNT);
901 		}
902 		else if (strcmp(opt->name, "nosymfollow") == 0)
903 			fsflags |= MNT_NOSYMFOLLOW;
904 		else if (strcmp(opt->name, "symfollow") == 0) {
905 			free(opt->name, M_MOUNT);
906 			opt->name = strdup("nonosymfollow", M_MOUNT);
907 		}
908 		else if (strcmp(opt->name, "noro") == 0) {
909 			fsflags &= ~MNT_RDONLY;
910 			autoro = false;
911 		}
912 		else if (strcmp(opt->name, "rw") == 0) {
913 			fsflags &= ~MNT_RDONLY;
914 			autoro = false;
915 		}
916 		else if (strcmp(opt->name, "ro") == 0) {
917 			fsflags |= MNT_RDONLY;
918 			autoro = false;
919 		}
920 		else if (strcmp(opt->name, "rdonly") == 0) {
921 			free(opt->name, M_MOUNT);
922 			opt->name = strdup("ro", M_MOUNT);
923 			fsflags |= MNT_RDONLY;
924 			autoro = false;
925 		}
926 		else if (strcmp(opt->name, "autoro") == 0) {
927 			do_freeopt = 1;
928 			autoro = true;
929 		}
930 		else if (strcmp(opt->name, "suiddir") == 0)
931 			fsflags |= MNT_SUIDDIR;
932 		else if (strcmp(opt->name, "sync") == 0)
933 			fsflags |= MNT_SYNCHRONOUS;
934 		else if (strcmp(opt->name, "union") == 0)
935 			fsflags |= MNT_UNION;
936 		else if (strcmp(opt->name, "export") == 0)
937 			fsflags |= MNT_EXPORTED;
938 		else if (strcmp(opt->name, "automounted") == 0) {
939 			fsflags |= MNT_AUTOMOUNTED;
940 			do_freeopt = 1;
941 		} else if (strcmp(opt->name, "nocover") == 0) {
942 			fsflags |= MNT_NOCOVER;
943 			do_freeopt = 1;
944 		} else if (strcmp(opt->name, "cover") == 0) {
945 			fsflags &= ~MNT_NOCOVER;
946 			do_freeopt = 1;
947 		} else if (strcmp(opt->name, "emptydir") == 0) {
948 			fsflags |= MNT_EMPTYDIR;
949 			do_freeopt = 1;
950 		} else if (strcmp(opt->name, "noemptydir") == 0) {
951 			fsflags &= ~MNT_EMPTYDIR;
952 			do_freeopt = 1;
953 		}
954 		if (do_freeopt)
955 			vfs_freeopt(optlist, opt);
956 	}
957 
958 	/*
959 	 * Be ultra-paranoid about making sure the type and fspath
960 	 * variables will fit in our mp buffers, including the
961 	 * terminating NUL.
962 	 */
963 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
964 		error = ENAMETOOLONG;
965 		goto bail;
966 	}
967 
968 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
969 	if (error == ENOENT) {
970 		error = EINVAL;
971 		if (errmsg != NULL)
972 			strncpy(errmsg, "Invalid fstype", errmsg_len);
973 		goto bail;
974 	}
975 
976 	/*
977 	 * See if we can mount in the read-only mode if the error code suggests
978 	 * that it could be possible and the mount options allow for that.
979 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
980 	 * overridden by "autoro".
981 	 */
982 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
983 		printf("%s: R/W mount failed, possibly R/O media,"
984 		    " trying R/O mount\n", __func__);
985 		fsflags |= MNT_RDONLY;
986 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
987 	}
988 bail:
989 	/* copyout the errmsg */
990 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
991 	    && errmsg_len > 0 && errmsg != NULL) {
992 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
993 			bcopy(errmsg,
994 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
995 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
996 		} else {
997 			copyout(errmsg,
998 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
999 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1000 		}
1001 	}
1002 
1003 	if (optlist != NULL)
1004 		vfs_freeopts(optlist);
1005 	return (error);
1006 }
1007 
1008 /*
1009  * Old mount API.
1010  */
1011 #ifndef _SYS_SYSPROTO_H_
1012 struct mount_args {
1013 	char	*type;
1014 	char	*path;
1015 	int	flags;
1016 	caddr_t	data;
1017 };
1018 #endif
1019 /* ARGSUSED */
1020 int
1021 sys_mount(struct thread *td, struct mount_args *uap)
1022 {
1023 	char *fstype;
1024 	struct vfsconf *vfsp = NULL;
1025 	struct mntarg *ma = NULL;
1026 	uint64_t flags;
1027 	int error;
1028 
1029 	/*
1030 	 * Mount flags are now 64-bits. On 32-bit architectures only
1031 	 * 32-bits are passed in, but from here on everything handles
1032 	 * 64-bit flags correctly.
1033 	 */
1034 	flags = uap->flags;
1035 
1036 	AUDIT_ARG_FFLAGS(flags);
1037 
1038 	/*
1039 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
1040 	 * userspace to set this flag, but we must filter it out if we want
1041 	 * MNT_UPDATE on the root file system to work.
1042 	 * MNT_ROOTFS should only be set by the kernel when mounting its
1043 	 * root file system.
1044 	 */
1045 	flags &= ~MNT_ROOTFS;
1046 
1047 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1048 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1049 	if (error) {
1050 		free(fstype, M_TEMP);
1051 		return (error);
1052 	}
1053 
1054 	AUDIT_ARG_TEXT(fstype);
1055 	vfsp = vfs_byname_kld(fstype, td, &error);
1056 	free(fstype, M_TEMP);
1057 	if (vfsp == NULL)
1058 		return (ENOENT);
1059 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1060 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1061 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1062 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
1063 		return (EOPNOTSUPP);
1064 
1065 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1066 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1067 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1068 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1069 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1070 
1071 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1072 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1073 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1074 }
1075 
1076 /*
1077  * vfs_domount_first(): first file system mount (not update)
1078  */
1079 static int
1080 vfs_domount_first(
1081 	struct thread *td,		/* Calling thread. */
1082 	struct vfsconf *vfsp,		/* File system type. */
1083 	char *fspath,			/* Mount path. */
1084 	struct vnode *vp,		/* Vnode to be covered. */
1085 	uint64_t fsflags,		/* Flags common to all filesystems. */
1086 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1087 	)
1088 {
1089 	struct vattr va;
1090 	struct mount *mp;
1091 	struct vnode *newdp, *rootvp;
1092 	int error, error1;
1093 	bool unmounted;
1094 
1095 	ASSERT_VOP_ELOCKED(vp, __func__);
1096 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1097 
1098 	/*
1099 	 * If the jail of the calling thread lacks permission for this type of
1100 	 * file system, or is trying to cover its own root, deny immediately.
1101 	 */
1102 	if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1103 	    vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1104 		vput(vp);
1105 		return (EPERM);
1106 	}
1107 
1108 	/*
1109 	 * If the user is not root, ensure that they own the directory
1110 	 * onto which we are attempting to mount.
1111 	 */
1112 	error = VOP_GETATTR(vp, &va, td->td_ucred);
1113 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1114 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1115 	if (error == 0)
1116 		error = vinvalbuf(vp, V_SAVE, 0, 0);
1117 	if (vfsp->vfc_flags & VFCF_FILEMOUNT) {
1118 		if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG)
1119 			error = EINVAL;
1120 		/*
1121 		 * For file mounts, ensure that there is only one hardlink to the file.
1122 		 */
1123 		if (error == 0 && vp->v_type == VREG && va.va_nlink != 1)
1124 			error = EINVAL;
1125 	} else {
1126 		if (error == 0 && vp->v_type != VDIR)
1127 			error = ENOTDIR;
1128 	}
1129 	if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
1130 		error = vfs_emptydir(vp);
1131 	if (error == 0) {
1132 		VI_LOCK(vp);
1133 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1134 			vp->v_iflag |= VI_MOUNT;
1135 		else
1136 			error = EBUSY;
1137 		VI_UNLOCK(vp);
1138 	}
1139 	if (error != 0) {
1140 		vput(vp);
1141 		return (error);
1142 	}
1143 	vn_seqc_write_begin(vp);
1144 	VOP_UNLOCK(vp);
1145 
1146 	/* Allocate and initialize the filesystem. */
1147 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1148 	/* XXXMAC: pass to vfs_mount_alloc? */
1149 	mp->mnt_optnew = *optlist;
1150 	/* Set the mount level flags. */
1151 	mp->mnt_flag = (fsflags &
1152 	    (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE));
1153 
1154 	/*
1155 	 * Mount the filesystem.
1156 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1157 	 * get.  No freeing of cn_pnbuf.
1158 	 */
1159 	error1 = 0;
1160 	unmounted = true;
1161 	if ((error = VFS_MOUNT(mp)) != 0 ||
1162 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1163 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1164 		rootvp = NULL;
1165 		if (error1 != 0) {
1166 			MPASS(error == 0);
1167 			rootvp = vfs_cache_root_clear(mp);
1168 			if (rootvp != NULL) {
1169 				vhold(rootvp);
1170 				vrele(rootvp);
1171 			}
1172 			(void)vn_start_write(NULL, &mp, V_WAIT);
1173 			MNT_ILOCK(mp);
1174 			mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1175 			MNT_IUNLOCK(mp);
1176 			VFS_PURGE(mp);
1177 			error = VFS_UNMOUNT(mp, 0);
1178 			vn_finished_write(mp);
1179 			if (error != 0) {
1180 				printf(
1181 		    "failed post-mount (%d): rollback unmount returned %d\n",
1182 				    error1, error);
1183 				unmounted = false;
1184 			}
1185 			error = error1;
1186 		}
1187 		vfs_unbusy(mp);
1188 		mp->mnt_vnodecovered = NULL;
1189 		if (unmounted) {
1190 			/* XXXKIB wait for mnt_lockref drain? */
1191 			vfs_mount_destroy(mp);
1192 		}
1193 		VI_LOCK(vp);
1194 		vp->v_iflag &= ~VI_MOUNT;
1195 		VI_UNLOCK(vp);
1196 		if (rootvp != NULL) {
1197 			vn_seqc_write_end(rootvp);
1198 			vdrop(rootvp);
1199 		}
1200 		vn_seqc_write_end(vp);
1201 		vrele(vp);
1202 		return (error);
1203 	}
1204 	vn_seqc_write_begin(newdp);
1205 	VOP_UNLOCK(newdp);
1206 
1207 	if (mp->mnt_opt != NULL)
1208 		vfs_freeopts(mp->mnt_opt);
1209 	mp->mnt_opt = mp->mnt_optnew;
1210 	*optlist = NULL;
1211 
1212 	/*
1213 	 * Prevent external consumers of mount options from reading mnt_optnew.
1214 	 */
1215 	mp->mnt_optnew = NULL;
1216 
1217 	MNT_ILOCK(mp);
1218 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1219 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1220 		mp->mnt_kern_flag |= MNTK_ASYNC;
1221 	else
1222 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1223 	MNT_IUNLOCK(mp);
1224 
1225 	/*
1226 	 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the
1227 	 * vp lock to satisfy vfs_lookup() requirements.
1228 	 */
1229 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1230 	VI_LOCK(vp);
1231 	vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1232 	vp->v_mountedhere = mp;
1233 	VI_UNLOCK(vp);
1234 	VOP_UNLOCK(vp);
1235 	cache_purge(vp);
1236 
1237 	/*
1238 	 * We need to lock both vnodes.
1239 	 *
1240 	 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1241 	 * from different filesystems.
1242 	 */
1243 	vn_lock_pair(vp, false, newdp, false);
1244 
1245 	VI_LOCK(vp);
1246 	vp->v_iflag &= ~VI_MOUNT;
1247 	VI_UNLOCK(vp);
1248 	/* Place the new filesystem at the end of the mount list. */
1249 	mtx_lock(&mountlist_mtx);
1250 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1251 	mtx_unlock(&mountlist_mtx);
1252 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1253 	VOP_UNLOCK(vp);
1254 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1255 	VOP_UNLOCK(newdp);
1256 	mount_devctl_event("MOUNT", mp, false);
1257 	mountcheckdirs(vp, newdp);
1258 	vn_seqc_write_end(vp);
1259 	vn_seqc_write_end(newdp);
1260 	vrele(newdp);
1261 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1262 		vfs_allocate_syncvnode(mp);
1263 	vfs_op_exit(mp);
1264 	vfs_unbusy(mp);
1265 	return (0);
1266 }
1267 
1268 /*
1269  * vfs_domount_update(): update of mounted file system
1270  */
1271 static int
1272 vfs_domount_update(
1273 	struct thread *td,		/* Calling thread. */
1274 	struct vnode *vp,		/* Mount point vnode. */
1275 	uint64_t fsflags,		/* Flags common to all filesystems. */
1276 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1277 	)
1278 {
1279 	struct export_args export;
1280 	struct o2export_args o2export;
1281 	struct vnode *rootvp;
1282 	void *bufp;
1283 	struct mount *mp;
1284 	int error, export_error, i, len;
1285 	uint64_t flag;
1286 	gid_t *grps;
1287 
1288 	ASSERT_VOP_ELOCKED(vp, __func__);
1289 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1290 	mp = vp->v_mount;
1291 
1292 	if ((vp->v_vflag & VV_ROOT) == 0) {
1293 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1294 		    == 0)
1295 			error = EXDEV;
1296 		else
1297 			error = EINVAL;
1298 		vput(vp);
1299 		return (error);
1300 	}
1301 
1302 	/*
1303 	 * We only allow the filesystem to be reloaded if it
1304 	 * is currently mounted read-only.
1305 	 */
1306 	flag = mp->mnt_flag;
1307 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1308 		vput(vp);
1309 		return (EOPNOTSUPP);	/* Needs translation */
1310 	}
1311 	/*
1312 	 * Only privileged root, or (if MNT_USER is set) the user that
1313 	 * did the original mount is permitted to update it.
1314 	 */
1315 	error = vfs_suser(mp, td);
1316 	if (error != 0) {
1317 		vput(vp);
1318 		return (error);
1319 	}
1320 	if (vfs_busy(mp, MBF_NOWAIT)) {
1321 		vput(vp);
1322 		return (EBUSY);
1323 	}
1324 	VI_LOCK(vp);
1325 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1326 		VI_UNLOCK(vp);
1327 		vfs_unbusy(mp);
1328 		vput(vp);
1329 		return (EBUSY);
1330 	}
1331 	vp->v_iflag |= VI_MOUNT;
1332 	VI_UNLOCK(vp);
1333 	VOP_UNLOCK(vp);
1334 
1335 	vfs_op_enter(mp);
1336 	vn_seqc_write_begin(vp);
1337 
1338 	rootvp = NULL;
1339 	MNT_ILOCK(mp);
1340 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1341 		MNT_IUNLOCK(mp);
1342 		error = EBUSY;
1343 		goto end;
1344 	}
1345 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1346 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1347 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1348 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1349 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1350 	rootvp = vfs_cache_root_clear(mp);
1351 	MNT_IUNLOCK(mp);
1352 	mp->mnt_optnew = *optlist;
1353 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1354 
1355 	/*
1356 	 * Mount the filesystem.
1357 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1358 	 * get.  No freeing of cn_pnbuf.
1359 	 */
1360 	error = VFS_MOUNT(mp);
1361 
1362 	export_error = 0;
1363 	/* Process the export option. */
1364 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1365 	    &len) == 0) {
1366 		/* Assume that there is only 1 ABI for each length. */
1367 		switch (len) {
1368 		case (sizeof(struct oexport_args)):
1369 			bzero(&o2export, sizeof(o2export));
1370 			/* FALLTHROUGH */
1371 		case (sizeof(o2export)):
1372 			bcopy(bufp, &o2export, len);
1373 			export.ex_flags = (uint64_t)o2export.ex_flags;
1374 			export.ex_root = o2export.ex_root;
1375 			export.ex_uid = o2export.ex_anon.cr_uid;
1376 			export.ex_groups = NULL;
1377 			export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1378 			if (export.ex_ngroups > 0) {
1379 				if (export.ex_ngroups <= XU_NGROUPS) {
1380 					export.ex_groups = malloc(
1381 					    export.ex_ngroups * sizeof(gid_t),
1382 					    M_TEMP, M_WAITOK);
1383 					for (i = 0; i < export.ex_ngroups; i++)
1384 						export.ex_groups[i] =
1385 						  o2export.ex_anon.cr_groups[i];
1386 				} else
1387 					export_error = EINVAL;
1388 			} else if (export.ex_ngroups < 0)
1389 				export_error = EINVAL;
1390 			export.ex_addr = o2export.ex_addr;
1391 			export.ex_addrlen = o2export.ex_addrlen;
1392 			export.ex_mask = o2export.ex_mask;
1393 			export.ex_masklen = o2export.ex_masklen;
1394 			export.ex_indexfile = o2export.ex_indexfile;
1395 			export.ex_numsecflavors = o2export.ex_numsecflavors;
1396 			if (export.ex_numsecflavors < MAXSECFLAVORS) {
1397 				for (i = 0; i < export.ex_numsecflavors; i++)
1398 					export.ex_secflavors[i] =
1399 					    o2export.ex_secflavors[i];
1400 			} else
1401 				export_error = EINVAL;
1402 			if (export_error == 0)
1403 				export_error = vfs_export(mp, &export, true);
1404 			free(export.ex_groups, M_TEMP);
1405 			break;
1406 		case (sizeof(export)):
1407 			bcopy(bufp, &export, len);
1408 			grps = NULL;
1409 			if (export.ex_ngroups > 0) {
1410 				if (export.ex_ngroups <= NGROUPS_MAX) {
1411 					grps = malloc(export.ex_ngroups *
1412 					    sizeof(gid_t), M_TEMP, M_WAITOK);
1413 					export_error = copyin(export.ex_groups,
1414 					    grps, export.ex_ngroups *
1415 					    sizeof(gid_t));
1416 					if (export_error == 0)
1417 						export.ex_groups = grps;
1418 				} else
1419 					export_error = EINVAL;
1420 			} else if (export.ex_ngroups == 0)
1421 				export.ex_groups = NULL;
1422 			else
1423 				export_error = EINVAL;
1424 			if (export_error == 0)
1425 				export_error = vfs_export(mp, &export, true);
1426 			free(grps, M_TEMP);
1427 			break;
1428 		default:
1429 			export_error = EINVAL;
1430 			break;
1431 		}
1432 	}
1433 
1434 	MNT_ILOCK(mp);
1435 	if (error == 0) {
1436 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1437 		    MNT_SNAPSHOT);
1438 	} else {
1439 		/*
1440 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1441 		 * because it is not part of MNT_UPDATEMASK, but it could have
1442 		 * changed in the meantime if quotactl(2) was called.
1443 		 * All in all we want current value of MNT_QUOTA, not the old
1444 		 * one.
1445 		 */
1446 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1447 	}
1448 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1449 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1450 		mp->mnt_kern_flag |= MNTK_ASYNC;
1451 	else
1452 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1453 	MNT_IUNLOCK(mp);
1454 
1455 	if (error != 0)
1456 		goto end;
1457 
1458 	mount_devctl_event("REMOUNT", mp, true);
1459 	if (mp->mnt_opt != NULL)
1460 		vfs_freeopts(mp->mnt_opt);
1461 	mp->mnt_opt = mp->mnt_optnew;
1462 	*optlist = NULL;
1463 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1464 	/*
1465 	 * Prevent external consumers of mount options from reading
1466 	 * mnt_optnew.
1467 	 */
1468 	mp->mnt_optnew = NULL;
1469 
1470 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1471 		vfs_allocate_syncvnode(mp);
1472 	else
1473 		vfs_deallocate_syncvnode(mp);
1474 end:
1475 	vfs_op_exit(mp);
1476 	if (rootvp != NULL) {
1477 		vn_seqc_write_end(rootvp);
1478 		vrele(rootvp);
1479 	}
1480 	vn_seqc_write_end(vp);
1481 	vfs_unbusy(mp);
1482 	VI_LOCK(vp);
1483 	vp->v_iflag &= ~VI_MOUNT;
1484 	VI_UNLOCK(vp);
1485 	vrele(vp);
1486 	return (error != 0 ? error : export_error);
1487 }
1488 
1489 /*
1490  * vfs_domount(): actually attempt a filesystem mount.
1491  */
1492 static int
1493 vfs_domount(
1494 	struct thread *td,		/* Calling thread. */
1495 	const char *fstype,		/* Filesystem type. */
1496 	char *fspath,			/* Mount path. */
1497 	uint64_t fsflags,		/* Flags common to all filesystems. */
1498 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1499 	)
1500 {
1501 	struct vfsconf *vfsp;
1502 	struct nameidata nd;
1503 	struct vnode *vp;
1504 	char *pathbuf;
1505 	int error;
1506 
1507 	/*
1508 	 * Be ultra-paranoid about making sure the type and fspath
1509 	 * variables will fit in our mp buffers, including the
1510 	 * terminating NUL.
1511 	 */
1512 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1513 		return (ENAMETOOLONG);
1514 
1515 	if (jailed(td->td_ucred) || usermount == 0) {
1516 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1517 			return (error);
1518 	}
1519 
1520 	/*
1521 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1522 	 */
1523 	if (fsflags & MNT_EXPORTED) {
1524 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1525 		if (error)
1526 			return (error);
1527 	}
1528 	if (fsflags & MNT_SUIDDIR) {
1529 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1530 		if (error)
1531 			return (error);
1532 	}
1533 	/*
1534 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1535 	 */
1536 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1537 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1538 			fsflags |= MNT_NOSUID | MNT_USER;
1539 	}
1540 
1541 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1542 	vfsp = NULL;
1543 	if ((fsflags & MNT_UPDATE) == 0) {
1544 		/* Don't try to load KLDs if we're mounting the root. */
1545 		if (fsflags & MNT_ROOTFS) {
1546 			if ((vfsp = vfs_byname(fstype)) == NULL)
1547 				return (ENODEV);
1548 		} else {
1549 			if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1550 				return (error);
1551 		}
1552 	}
1553 
1554 	/*
1555 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1556 	 */
1557 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1558 	    UIO_SYSSPACE, fspath);
1559 	error = namei(&nd);
1560 	if (error != 0)
1561 		return (error);
1562 	vp = nd.ni_vp;
1563 	/*
1564 	 * Don't allow stacking file mounts to work around problems with the way
1565 	 * that namei sets nd.ni_dvp to vp_crossmp for these.
1566 	 */
1567 	if (vp->v_type == VREG)
1568 		fsflags |= MNT_NOCOVER;
1569 	if ((fsflags & MNT_UPDATE) == 0) {
1570 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1571 		    (fsflags & MNT_NOCOVER) != 0) {
1572 			vput(vp);
1573 			error = EBUSY;
1574 			goto out;
1575 		}
1576 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1577 		strcpy(pathbuf, fspath);
1578 		/*
1579 		 * Note: we allow any vnode type here. If the path sanity check
1580 		 * succeeds, the type will be validated in vfs_domount_first
1581 		 * above.
1582 		 */
1583 		if (vp->v_type == VDIR)
1584 			error = vn_path_to_global_path(td, vp, pathbuf,
1585 			    MNAMELEN);
1586 		else
1587 			error = vn_path_to_global_path_hardlink(td, vp,
1588 			    nd.ni_dvp, pathbuf, MNAMELEN,
1589 			    nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1590 		if (error == 0) {
1591 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1592 			    fsflags, optlist);
1593 		}
1594 		free(pathbuf, M_TEMP);
1595 	} else
1596 		error = vfs_domount_update(td, vp, fsflags, optlist);
1597 
1598 out:
1599 	NDFREE_PNBUF(&nd);
1600 	vrele(nd.ni_dvp);
1601 
1602 	return (error);
1603 }
1604 
1605 /*
1606  * Unmount a filesystem.
1607  *
1608  * Note: unmount takes a path to the vnode mounted on as argument, not
1609  * special file (as before).
1610  */
1611 #ifndef _SYS_SYSPROTO_H_
1612 struct unmount_args {
1613 	char	*path;
1614 	int	flags;
1615 };
1616 #endif
1617 /* ARGSUSED */
1618 int
1619 sys_unmount(struct thread *td, struct unmount_args *uap)
1620 {
1621 
1622 	return (kern_unmount(td, uap->path, uap->flags));
1623 }
1624 
1625 int
1626 kern_unmount(struct thread *td, const char *path, int flags)
1627 {
1628 	struct nameidata nd;
1629 	struct mount *mp;
1630 	char *fsidbuf, *pathbuf;
1631 	fsid_t fsid;
1632 	int error;
1633 
1634 	AUDIT_ARG_VALUE(flags);
1635 	if (jailed(td->td_ucred) || usermount == 0) {
1636 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1637 		if (error)
1638 			return (error);
1639 	}
1640 
1641 	if (flags & MNT_BYFSID) {
1642 		fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1643 		error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1644 		if (error) {
1645 			free(fsidbuf, M_TEMP);
1646 			return (error);
1647 		}
1648 
1649 		AUDIT_ARG_TEXT(fsidbuf);
1650 		/* Decode the filesystem ID. */
1651 		if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1652 			free(fsidbuf, M_TEMP);
1653 			return (EINVAL);
1654 		}
1655 
1656 		mp = vfs_getvfs(&fsid);
1657 		free(fsidbuf, M_TEMP);
1658 		if (mp == NULL) {
1659 			return (ENOENT);
1660 		}
1661 	} else {
1662 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1663 		error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1664 		if (error) {
1665 			free(pathbuf, M_TEMP);
1666 			return (error);
1667 		}
1668 
1669 		/*
1670 		 * Try to find global path for path argument.
1671 		 */
1672 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1673 		    UIO_SYSSPACE, pathbuf);
1674 		if (namei(&nd) == 0) {
1675 			NDFREE_PNBUF(&nd);
1676 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1677 			    MNAMELEN);
1678 			if (error == 0)
1679 				vput(nd.ni_vp);
1680 		}
1681 		mtx_lock(&mountlist_mtx);
1682 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1683 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1684 				vfs_ref(mp);
1685 				break;
1686 			}
1687 		}
1688 		mtx_unlock(&mountlist_mtx);
1689 		free(pathbuf, M_TEMP);
1690 		if (mp == NULL) {
1691 			/*
1692 			 * Previously we returned ENOENT for a nonexistent path and
1693 			 * EINVAL for a non-mountpoint.  We cannot tell these apart
1694 			 * now, so in the !MNT_BYFSID case return the more likely
1695 			 * EINVAL for compatibility.
1696 			 */
1697 			return (EINVAL);
1698 		}
1699 	}
1700 
1701 	/*
1702 	 * Don't allow unmounting the root filesystem.
1703 	 */
1704 	if (mp->mnt_flag & MNT_ROOTFS) {
1705 		vfs_rel(mp);
1706 		return (EINVAL);
1707 	}
1708 	error = dounmount(mp, flags, td);
1709 	return (error);
1710 }
1711 
1712 /*
1713  * Return error if any of the vnodes, ignoring the root vnode
1714  * and the syncer vnode, have non-zero usecount.
1715  *
1716  * This function is purely advisory - it can return false positives
1717  * and negatives.
1718  */
1719 static int
1720 vfs_check_usecounts(struct mount *mp)
1721 {
1722 	struct vnode *vp, *mvp;
1723 
1724 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1725 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1726 		    vp->v_usecount != 0) {
1727 			VI_UNLOCK(vp);
1728 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1729 			return (EBUSY);
1730 		}
1731 		VI_UNLOCK(vp);
1732 	}
1733 
1734 	return (0);
1735 }
1736 
1737 static void
1738 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1739 {
1740 
1741 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1742 	mp->mnt_kern_flag &= ~mntkflags;
1743 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1744 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1745 		wakeup(mp);
1746 	}
1747 	vfs_op_exit_locked(mp);
1748 	MNT_IUNLOCK(mp);
1749 	if (coveredvp != NULL) {
1750 		VOP_UNLOCK(coveredvp);
1751 		vdrop(coveredvp);
1752 	}
1753 	vn_finished_write(mp);
1754 	vfs_rel(mp);
1755 }
1756 
1757 /*
1758  * There are various reference counters associated with the mount point.
1759  * Normally it is permitted to modify them without taking the mnt ilock,
1760  * but this behavior can be temporarily disabled if stable value is needed
1761  * or callers are expected to block (e.g. to not allow new users during
1762  * forced unmount).
1763  */
1764 void
1765 vfs_op_enter(struct mount *mp)
1766 {
1767 	struct mount_pcpu *mpcpu;
1768 	int cpu;
1769 
1770 	MNT_ILOCK(mp);
1771 	mp->mnt_vfs_ops++;
1772 	if (mp->mnt_vfs_ops > 1) {
1773 		MNT_IUNLOCK(mp);
1774 		return;
1775 	}
1776 	vfs_op_barrier_wait(mp);
1777 	CPU_FOREACH(cpu) {
1778 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1779 
1780 		mp->mnt_ref += mpcpu->mntp_ref;
1781 		mpcpu->mntp_ref = 0;
1782 
1783 		mp->mnt_lockref += mpcpu->mntp_lockref;
1784 		mpcpu->mntp_lockref = 0;
1785 
1786 		mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1787 		mpcpu->mntp_writeopcount = 0;
1788 	}
1789 	MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1790 	    mp->mnt_writeopcount >= 0, mp,
1791 	    ("invalid count(s): ref %d lockref %d writeopcount %d",
1792 	    mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1793 	MNT_IUNLOCK(mp);
1794 	vfs_assert_mount_counters(mp);
1795 }
1796 
1797 void
1798 vfs_op_exit_locked(struct mount *mp)
1799 {
1800 
1801 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1802 
1803 	MPASSERT(mp->mnt_vfs_ops > 0, mp,
1804 	    ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1805 	MPASSERT(mp->mnt_vfs_ops > 1 ||
1806 	    (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1807 	    ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1808 	mp->mnt_vfs_ops--;
1809 }
1810 
1811 void
1812 vfs_op_exit(struct mount *mp)
1813 {
1814 
1815 	MNT_ILOCK(mp);
1816 	vfs_op_exit_locked(mp);
1817 	MNT_IUNLOCK(mp);
1818 }
1819 
1820 struct vfs_op_barrier_ipi {
1821 	struct mount *mp;
1822 	struct smp_rendezvous_cpus_retry_arg srcra;
1823 };
1824 
1825 static void
1826 vfs_op_action_func(void *arg)
1827 {
1828 	struct vfs_op_barrier_ipi *vfsopipi;
1829 	struct mount *mp;
1830 
1831 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1832 	mp = vfsopipi->mp;
1833 
1834 	if (!vfs_op_thread_entered(mp))
1835 		smp_rendezvous_cpus_done(arg);
1836 }
1837 
1838 static void
1839 vfs_op_wait_func(void *arg, int cpu)
1840 {
1841 	struct vfs_op_barrier_ipi *vfsopipi;
1842 	struct mount *mp;
1843 	struct mount_pcpu *mpcpu;
1844 
1845 	vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1846 	mp = vfsopipi->mp;
1847 
1848 	mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1849 	while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1850 		cpu_spinwait();
1851 }
1852 
1853 void
1854 vfs_op_barrier_wait(struct mount *mp)
1855 {
1856 	struct vfs_op_barrier_ipi vfsopipi;
1857 
1858 	vfsopipi.mp = mp;
1859 
1860 	smp_rendezvous_cpus_retry(all_cpus,
1861 	    smp_no_rendezvous_barrier,
1862 	    vfs_op_action_func,
1863 	    smp_no_rendezvous_barrier,
1864 	    vfs_op_wait_func,
1865 	    &vfsopipi.srcra);
1866 }
1867 
1868 #ifdef DIAGNOSTIC
1869 void
1870 vfs_assert_mount_counters(struct mount *mp)
1871 {
1872 	struct mount_pcpu *mpcpu;
1873 	int cpu;
1874 
1875 	if (mp->mnt_vfs_ops == 0)
1876 		return;
1877 
1878 	CPU_FOREACH(cpu) {
1879 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1880 		if (mpcpu->mntp_ref != 0 ||
1881 		    mpcpu->mntp_lockref != 0 ||
1882 		    mpcpu->mntp_writeopcount != 0)
1883 			vfs_dump_mount_counters(mp);
1884 	}
1885 }
1886 
1887 void
1888 vfs_dump_mount_counters(struct mount *mp)
1889 {
1890 	struct mount_pcpu *mpcpu;
1891 	int ref, lockref, writeopcount;
1892 	int cpu;
1893 
1894 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1895 
1896 	printf("        ref : ");
1897 	ref = mp->mnt_ref;
1898 	CPU_FOREACH(cpu) {
1899 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1900 		printf("%d ", mpcpu->mntp_ref);
1901 		ref += mpcpu->mntp_ref;
1902 	}
1903 	printf("\n");
1904 	printf("    lockref : ");
1905 	lockref = mp->mnt_lockref;
1906 	CPU_FOREACH(cpu) {
1907 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1908 		printf("%d ", mpcpu->mntp_lockref);
1909 		lockref += mpcpu->mntp_lockref;
1910 	}
1911 	printf("\n");
1912 	printf("writeopcount: ");
1913 	writeopcount = mp->mnt_writeopcount;
1914 	CPU_FOREACH(cpu) {
1915 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1916 		printf("%d ", mpcpu->mntp_writeopcount);
1917 		writeopcount += mpcpu->mntp_writeopcount;
1918 	}
1919 	printf("\n");
1920 
1921 	printf("counter       struct total\n");
1922 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1923 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1924 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1925 
1926 	panic("invalid counts on struct mount");
1927 }
1928 #endif
1929 
1930 int
1931 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1932 {
1933 	struct mount_pcpu *mpcpu;
1934 	int cpu, sum;
1935 
1936 	switch (which) {
1937 	case MNT_COUNT_REF:
1938 		sum = mp->mnt_ref;
1939 		break;
1940 	case MNT_COUNT_LOCKREF:
1941 		sum = mp->mnt_lockref;
1942 		break;
1943 	case MNT_COUNT_WRITEOPCOUNT:
1944 		sum = mp->mnt_writeopcount;
1945 		break;
1946 	}
1947 
1948 	CPU_FOREACH(cpu) {
1949 		mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1950 		switch (which) {
1951 		case MNT_COUNT_REF:
1952 			sum += mpcpu->mntp_ref;
1953 			break;
1954 		case MNT_COUNT_LOCKREF:
1955 			sum += mpcpu->mntp_lockref;
1956 			break;
1957 		case MNT_COUNT_WRITEOPCOUNT:
1958 			sum += mpcpu->mntp_writeopcount;
1959 			break;
1960 		}
1961 	}
1962 	return (sum);
1963 }
1964 
1965 static bool
1966 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
1967     int timeout_ticks)
1968 {
1969 	bool enqueued;
1970 
1971 	enqueued = false;
1972 	mtx_lock(&deferred_unmount_lock);
1973 	if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
1974 		mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
1975 		STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
1976 		    mnt_taskqueue_link);
1977 		enqueued = true;
1978 	}
1979 	mtx_unlock(&deferred_unmount_lock);
1980 
1981 	if (enqueued) {
1982 		taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
1983 		    &deferred_unmount_task, timeout_ticks);
1984 	}
1985 
1986 	return (enqueued);
1987 }
1988 
1989 /*
1990  * Taskqueue handler for processing async/recursive unmounts
1991  */
1992 static void
1993 vfs_deferred_unmount(void *argi __unused, int pending __unused)
1994 {
1995 	STAILQ_HEAD(, mount) local_unmounts;
1996 	uint64_t flags;
1997 	struct mount *mp, *tmp;
1998 	int error;
1999 	unsigned int retries;
2000 	bool unmounted;
2001 
2002 	STAILQ_INIT(&local_unmounts);
2003 	mtx_lock(&deferred_unmount_lock);
2004 	STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
2005 	mtx_unlock(&deferred_unmount_lock);
2006 
2007 	STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
2008 		flags = mp->mnt_taskqueue_flags;
2009 		KASSERT((flags & MNT_DEFERRED) != 0,
2010 		    ("taskqueue unmount without MNT_DEFERRED"));
2011 		error = dounmount(mp, flags, curthread);
2012 		if (error != 0) {
2013 			MNT_ILOCK(mp);
2014 			unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
2015 			MNT_IUNLOCK(mp);
2016 
2017 			/*
2018 			 * The deferred unmount thread is the only thread that
2019 			 * modifies the retry counts, so locking/atomics aren't
2020 			 * needed here.
2021 			 */
2022 			retries = (mp->mnt_unmount_retries)++;
2023 			deferred_unmount_total_retries++;
2024 			if (!unmounted && retries < deferred_unmount_retry_limit) {
2025 				deferred_unmount_enqueue(mp, flags, true,
2026 				    -deferred_unmount_retry_delay_hz);
2027 			} else {
2028 				if (retries >= deferred_unmount_retry_limit) {
2029 					printf("giving up on deferred unmount "
2030 					    "of %s after %d retries, error %d\n",
2031 					    mp->mnt_stat.f_mntonname, retries, error);
2032 				}
2033 				vfs_rel(mp);
2034 			}
2035 		}
2036 	}
2037 }
2038 
2039 /*
2040  * Do the actual filesystem unmount.
2041  */
2042 int
2043 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2044 {
2045 	struct mount_upper_node *upper;
2046 	struct vnode *coveredvp, *rootvp;
2047 	int error;
2048 	uint64_t async_flag;
2049 	int mnt_gen_r;
2050 	unsigned int retries;
2051 
2052 	KASSERT((flags & MNT_DEFERRED) == 0 ||
2053 	    (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2054 	    ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2055 
2056 	/*
2057 	 * If the caller has explicitly requested the unmount to be handled by
2058 	 * the taskqueue and we're not already in taskqueue context, queue
2059 	 * up the unmount request and exit.  This is done prior to any
2060 	 * credential checks; MNT_DEFERRED should be used only for kernel-
2061 	 * initiated unmounts and will therefore be processed with the
2062 	 * (kernel) credentials of the taskqueue thread.  Still, callers
2063 	 * should be sure this is the behavior they want.
2064 	 */
2065 	if ((flags & MNT_DEFERRED) != 0 &&
2066 	    taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2067 		if (!deferred_unmount_enqueue(mp, flags, false, 0))
2068 			vfs_rel(mp);
2069 		return (EINPROGRESS);
2070 	}
2071 
2072 	/*
2073 	 * Only privileged root, or (if MNT_USER is set) the user that did the
2074 	 * original mount is permitted to unmount this filesystem.
2075 	 * This check should be made prior to queueing up any recursive
2076 	 * unmounts of upper filesystems.  Those unmounts will be executed
2077 	 * with kernel thread credentials and are expected to succeed, so
2078 	 * we must at least ensure the originating context has sufficient
2079 	 * privilege to unmount the base filesystem before proceeding with
2080 	 * the uppers.
2081 	 */
2082 	error = vfs_suser(mp, td);
2083 	if (error != 0) {
2084 		KASSERT((flags & MNT_DEFERRED) == 0,
2085 		    ("taskqueue unmount with insufficient privilege"));
2086 		vfs_rel(mp);
2087 		return (error);
2088 	}
2089 
2090 	if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2091 		flags |= MNT_RECURSE;
2092 
2093 	if ((flags & MNT_RECURSE) != 0) {
2094 		KASSERT((flags & MNT_FORCE) != 0,
2095 		    ("MNT_RECURSE requires MNT_FORCE"));
2096 
2097 		MNT_ILOCK(mp);
2098 		/*
2099 		 * Set MNTK_RECURSE to prevent new upper mounts from being
2100 		 * added, and note that an operation on the uppers list is in
2101 		 * progress.  This will ensure that unregistration from the
2102 		 * uppers list, and therefore any pending unmount of the upper
2103 		 * FS, can't complete until after we finish walking the list.
2104 		 */
2105 		mp->mnt_kern_flag |= MNTK_RECURSE;
2106 		mp->mnt_upper_pending++;
2107 		TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2108 			retries = upper->mp->mnt_unmount_retries;
2109 			if (retries > deferred_unmount_retry_limit) {
2110 				error = EBUSY;
2111 				continue;
2112 			}
2113 			MNT_IUNLOCK(mp);
2114 
2115 			vfs_ref(upper->mp);
2116 			if (!deferred_unmount_enqueue(upper->mp, flags,
2117 			    false, 0))
2118 				vfs_rel(upper->mp);
2119 			MNT_ILOCK(mp);
2120 		}
2121 		mp->mnt_upper_pending--;
2122 		if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2123 		    mp->mnt_upper_pending == 0) {
2124 			mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2125 			wakeup(&mp->mnt_uppers);
2126 		}
2127 
2128 		/*
2129 		 * If we're not on the taskqueue, wait until the uppers list
2130 		 * is drained before proceeding with unmount.  Otherwise, if
2131 		 * we are on the taskqueue and there are still pending uppers,
2132 		 * just re-enqueue on the end of the taskqueue.
2133 		 */
2134 		if ((flags & MNT_DEFERRED) == 0) {
2135 			while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2136 				mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2137 				error = msleep(&mp->mnt_taskqueue_link,
2138 				    MNT_MTX(mp), PCATCH, "umntqw", 0);
2139 			}
2140 			if (error != 0) {
2141 				MNT_REL(mp);
2142 				MNT_IUNLOCK(mp);
2143 				return (error);
2144 			}
2145 		} else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2146 			MNT_IUNLOCK(mp);
2147 			if (error == 0)
2148 				deferred_unmount_enqueue(mp, flags, true, 0);
2149 			return (error);
2150 		}
2151 		MNT_IUNLOCK(mp);
2152 		KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2153 	}
2154 
2155 	/* Allow the taskqueue to safely re-enqueue on failure */
2156 	if ((flags & MNT_DEFERRED) != 0)
2157 		vfs_ref(mp);
2158 
2159 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2160 		mnt_gen_r = mp->mnt_gen;
2161 		VI_LOCK(coveredvp);
2162 		vholdl(coveredvp);
2163 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2164 		/*
2165 		 * Check for mp being unmounted while waiting for the
2166 		 * covered vnode lock.
2167 		 */
2168 		if (coveredvp->v_mountedhere != mp ||
2169 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2170 			VOP_UNLOCK(coveredvp);
2171 			vdrop(coveredvp);
2172 			vfs_rel(mp);
2173 			return (EBUSY);
2174 		}
2175 	}
2176 
2177 	vfs_op_enter(mp);
2178 
2179 	vn_start_write(NULL, &mp, V_WAIT);
2180 	MNT_ILOCK(mp);
2181 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2182 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
2183 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
2184 		dounmount_cleanup(mp, coveredvp, 0);
2185 		return (EBUSY);
2186 	}
2187 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
2188 	rootvp = vfs_cache_root_clear(mp);
2189 	if (coveredvp != NULL)
2190 		vn_seqc_write_begin(coveredvp);
2191 	if (flags & MNT_NONBUSY) {
2192 		MNT_IUNLOCK(mp);
2193 		error = vfs_check_usecounts(mp);
2194 		MNT_ILOCK(mp);
2195 		if (error != 0) {
2196 			vn_seqc_write_end(coveredvp);
2197 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2198 			if (rootvp != NULL) {
2199 				vn_seqc_write_end(rootvp);
2200 				vrele(rootvp);
2201 			}
2202 			return (error);
2203 		}
2204 	}
2205 	/* Allow filesystems to detect that a forced unmount is in progress. */
2206 	if (flags & MNT_FORCE) {
2207 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2208 		MNT_IUNLOCK(mp);
2209 		/*
2210 		 * Must be done after setting MNTK_UNMOUNTF and before
2211 		 * waiting for mnt_lockref to become 0.
2212 		 */
2213 		VFS_PURGE(mp);
2214 		MNT_ILOCK(mp);
2215 	}
2216 	error = 0;
2217 	if (mp->mnt_lockref) {
2218 		mp->mnt_kern_flag |= MNTK_DRAINING;
2219 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2220 		    "mount drain", 0);
2221 	}
2222 	MNT_IUNLOCK(mp);
2223 	KASSERT(mp->mnt_lockref == 0,
2224 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
2225 	    __func__, __FILE__, __LINE__));
2226 	KASSERT(error == 0,
2227 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
2228 	    __func__, __FILE__, __LINE__));
2229 
2230 	/*
2231 	 * We want to keep the vnode around so that we can vn_seqc_write_end
2232 	 * after we are done with unmount. Downgrade our reference to a mere
2233 	 * hold count so that we don't interefere with anything.
2234 	 */
2235 	if (rootvp != NULL) {
2236 		vhold(rootvp);
2237 		vrele(rootvp);
2238 	}
2239 
2240 	if (mp->mnt_flag & MNT_EXPUBLIC)
2241 		vfs_setpublicfs(NULL, NULL, NULL);
2242 
2243 	vfs_periodic(mp, MNT_WAIT);
2244 	MNT_ILOCK(mp);
2245 	async_flag = mp->mnt_flag & MNT_ASYNC;
2246 	mp->mnt_flag &= ~MNT_ASYNC;
2247 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
2248 	MNT_IUNLOCK(mp);
2249 	vfs_deallocate_syncvnode(mp);
2250 	error = VFS_UNMOUNT(mp, flags);
2251 	vn_finished_write(mp);
2252 	vfs_rel(mp);
2253 	/*
2254 	 * If we failed to flush the dirty blocks for this mount point,
2255 	 * undo all the cdir/rdir and rootvnode changes we made above.
2256 	 * Unless we failed to do so because the device is reporting that
2257 	 * it doesn't exist anymore.
2258 	 */
2259 	if (error && error != ENXIO) {
2260 		MNT_ILOCK(mp);
2261 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2262 			MNT_IUNLOCK(mp);
2263 			vfs_allocate_syncvnode(mp);
2264 			MNT_ILOCK(mp);
2265 		}
2266 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2267 		mp->mnt_flag |= async_flag;
2268 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2269 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2270 			mp->mnt_kern_flag |= MNTK_ASYNC;
2271 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
2272 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
2273 			wakeup(mp);
2274 		}
2275 		vfs_op_exit_locked(mp);
2276 		MNT_IUNLOCK(mp);
2277 		if (coveredvp) {
2278 			vn_seqc_write_end(coveredvp);
2279 			VOP_UNLOCK(coveredvp);
2280 			vdrop(coveredvp);
2281 		}
2282 		if (rootvp != NULL) {
2283 			vn_seqc_write_end(rootvp);
2284 			vdrop(rootvp);
2285 		}
2286 		return (error);
2287 	}
2288 
2289 	mtx_lock(&mountlist_mtx);
2290 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
2291 	mtx_unlock(&mountlist_mtx);
2292 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2293 	if (coveredvp != NULL) {
2294 		VI_LOCK(coveredvp);
2295 		vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2296 		coveredvp->v_mountedhere = NULL;
2297 		vn_seqc_write_end_locked(coveredvp);
2298 		VI_UNLOCK(coveredvp);
2299 		VOP_UNLOCK(coveredvp);
2300 		vdrop(coveredvp);
2301 	}
2302 	mount_devctl_event("UNMOUNT", mp, false);
2303 	if (rootvp != NULL) {
2304 		vn_seqc_write_end(rootvp);
2305 		vdrop(rootvp);
2306 	}
2307 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2308 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
2309 		vrele(rootvnode);
2310 		rootvnode = NULL;
2311 	}
2312 	if (mp == rootdevmp)
2313 		rootdevmp = NULL;
2314 	if ((flags & MNT_DEFERRED) != 0)
2315 		vfs_rel(mp);
2316 	vfs_mount_destroy(mp);
2317 	return (0);
2318 }
2319 
2320 /*
2321  * Report errors during filesystem mounting.
2322  */
2323 void
2324 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2325 {
2326 	struct vfsoptlist *moptlist = mp->mnt_optnew;
2327 	va_list ap;
2328 	int error, len;
2329 	char *errmsg;
2330 
2331 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2332 	if (error || errmsg == NULL || len <= 0)
2333 		return;
2334 
2335 	va_start(ap, fmt);
2336 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2337 	va_end(ap);
2338 }
2339 
2340 void
2341 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2342 {
2343 	va_list ap;
2344 	int error, len;
2345 	char *errmsg;
2346 
2347 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2348 	if (error || errmsg == NULL || len <= 0)
2349 		return;
2350 
2351 	va_start(ap, fmt);
2352 	vsnprintf(errmsg, (size_t)len, fmt, ap);
2353 	va_end(ap);
2354 }
2355 
2356 /*
2357  * ---------------------------------------------------------------------
2358  * Functions for querying mount options/arguments from filesystems.
2359  */
2360 
2361 /*
2362  * Check that no unknown options are given
2363  */
2364 int
2365 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2366 {
2367 	struct vfsopt *opt;
2368 	char errmsg[255];
2369 	const char **t, *p, *q;
2370 	int ret = 0;
2371 
2372 	TAILQ_FOREACH(opt, opts, link) {
2373 		p = opt->name;
2374 		q = NULL;
2375 		if (p[0] == 'n' && p[1] == 'o')
2376 			q = p + 2;
2377 		for(t = global_opts; *t != NULL; t++) {
2378 			if (strcmp(*t, p) == 0)
2379 				break;
2380 			if (q != NULL) {
2381 				if (strcmp(*t, q) == 0)
2382 					break;
2383 			}
2384 		}
2385 		if (*t != NULL)
2386 			continue;
2387 		for(t = legal; *t != NULL; t++) {
2388 			if (strcmp(*t, p) == 0)
2389 				break;
2390 			if (q != NULL) {
2391 				if (strcmp(*t, q) == 0)
2392 					break;
2393 			}
2394 		}
2395 		if (*t != NULL)
2396 			continue;
2397 		snprintf(errmsg, sizeof(errmsg),
2398 		    "mount option <%s> is unknown", p);
2399 		ret = EINVAL;
2400 	}
2401 	if (ret != 0) {
2402 		TAILQ_FOREACH(opt, opts, link) {
2403 			if (strcmp(opt->name, "errmsg") == 0) {
2404 				strncpy((char *)opt->value, errmsg, opt->len);
2405 				break;
2406 			}
2407 		}
2408 		if (opt == NULL)
2409 			printf("%s\n", errmsg);
2410 	}
2411 	return (ret);
2412 }
2413 
2414 /*
2415  * Get a mount option by its name.
2416  *
2417  * Return 0 if the option was found, ENOENT otherwise.
2418  * If len is non-NULL it will be filled with the length
2419  * of the option. If buf is non-NULL, it will be filled
2420  * with the address of the option.
2421  */
2422 int
2423 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2424 {
2425 	struct vfsopt *opt;
2426 
2427 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2428 
2429 	TAILQ_FOREACH(opt, opts, link) {
2430 		if (strcmp(name, opt->name) == 0) {
2431 			opt->seen = 1;
2432 			if (len != NULL)
2433 				*len = opt->len;
2434 			if (buf != NULL)
2435 				*buf = opt->value;
2436 			return (0);
2437 		}
2438 	}
2439 	return (ENOENT);
2440 }
2441 
2442 int
2443 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2444 {
2445 	struct vfsopt *opt;
2446 
2447 	if (opts == NULL)
2448 		return (-1);
2449 
2450 	TAILQ_FOREACH(opt, opts, link) {
2451 		if (strcmp(name, opt->name) == 0) {
2452 			opt->seen = 1;
2453 			return (opt->pos);
2454 		}
2455 	}
2456 	return (-1);
2457 }
2458 
2459 int
2460 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2461 {
2462 	char *opt_value, *vtp;
2463 	quad_t iv;
2464 	int error, opt_len;
2465 
2466 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2467 	if (error != 0)
2468 		return (error);
2469 	if (opt_len == 0 || opt_value == NULL)
2470 		return (EINVAL);
2471 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2472 		return (EINVAL);
2473 	iv = strtoq(opt_value, &vtp, 0);
2474 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2475 		return (EINVAL);
2476 	if (iv < 0)
2477 		return (EINVAL);
2478 	switch (vtp[0]) {
2479 	case 't': case 'T':
2480 		iv *= 1024;
2481 		/* FALLTHROUGH */
2482 	case 'g': case 'G':
2483 		iv *= 1024;
2484 		/* FALLTHROUGH */
2485 	case 'm': case 'M':
2486 		iv *= 1024;
2487 		/* FALLTHROUGH */
2488 	case 'k': case 'K':
2489 		iv *= 1024;
2490 	case '\0':
2491 		break;
2492 	default:
2493 		return (EINVAL);
2494 	}
2495 	*value = iv;
2496 
2497 	return (0);
2498 }
2499 
2500 char *
2501 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2502 {
2503 	struct vfsopt *opt;
2504 
2505 	*error = 0;
2506 	TAILQ_FOREACH(opt, opts, link) {
2507 		if (strcmp(name, opt->name) != 0)
2508 			continue;
2509 		opt->seen = 1;
2510 		if (opt->len == 0 ||
2511 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2512 			*error = EINVAL;
2513 			return (NULL);
2514 		}
2515 		return (opt->value);
2516 	}
2517 	*error = ENOENT;
2518 	return (NULL);
2519 }
2520 
2521 int
2522 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2523 	uint64_t val)
2524 {
2525 	struct vfsopt *opt;
2526 
2527 	TAILQ_FOREACH(opt, opts, link) {
2528 		if (strcmp(name, opt->name) == 0) {
2529 			opt->seen = 1;
2530 			if (w != NULL)
2531 				*w |= val;
2532 			return (1);
2533 		}
2534 	}
2535 	if (w != NULL)
2536 		*w &= ~val;
2537 	return (0);
2538 }
2539 
2540 int
2541 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2542 {
2543 	va_list ap;
2544 	struct vfsopt *opt;
2545 	int ret;
2546 
2547 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2548 
2549 	TAILQ_FOREACH(opt, opts, link) {
2550 		if (strcmp(name, opt->name) != 0)
2551 			continue;
2552 		opt->seen = 1;
2553 		if (opt->len == 0 || opt->value == NULL)
2554 			return (0);
2555 		if (((char *)opt->value)[opt->len - 1] != '\0')
2556 			return (0);
2557 		va_start(ap, fmt);
2558 		ret = vsscanf(opt->value, fmt, ap);
2559 		va_end(ap);
2560 		return (ret);
2561 	}
2562 	return (0);
2563 }
2564 
2565 int
2566 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2567 {
2568 	struct vfsopt *opt;
2569 
2570 	TAILQ_FOREACH(opt, opts, link) {
2571 		if (strcmp(name, opt->name) != 0)
2572 			continue;
2573 		opt->seen = 1;
2574 		if (opt->value == NULL)
2575 			opt->len = len;
2576 		else {
2577 			if (opt->len != len)
2578 				return (EINVAL);
2579 			bcopy(value, opt->value, len);
2580 		}
2581 		return (0);
2582 	}
2583 	return (ENOENT);
2584 }
2585 
2586 int
2587 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2588 {
2589 	struct vfsopt *opt;
2590 
2591 	TAILQ_FOREACH(opt, opts, link) {
2592 		if (strcmp(name, opt->name) != 0)
2593 			continue;
2594 		opt->seen = 1;
2595 		if (opt->value == NULL)
2596 			opt->len = len;
2597 		else {
2598 			if (opt->len < len)
2599 				return (EINVAL);
2600 			opt->len = len;
2601 			bcopy(value, opt->value, len);
2602 		}
2603 		return (0);
2604 	}
2605 	return (ENOENT);
2606 }
2607 
2608 int
2609 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2610 {
2611 	struct vfsopt *opt;
2612 
2613 	TAILQ_FOREACH(opt, opts, link) {
2614 		if (strcmp(name, opt->name) != 0)
2615 			continue;
2616 		opt->seen = 1;
2617 		if (opt->value == NULL)
2618 			opt->len = strlen(value) + 1;
2619 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2620 			return (EINVAL);
2621 		return (0);
2622 	}
2623 	return (ENOENT);
2624 }
2625 
2626 /*
2627  * Find and copy a mount option.
2628  *
2629  * The size of the buffer has to be specified
2630  * in len, if it is not the same length as the
2631  * mount option, EINVAL is returned.
2632  * Returns ENOENT if the option is not found.
2633  */
2634 int
2635 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2636 {
2637 	struct vfsopt *opt;
2638 
2639 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2640 
2641 	TAILQ_FOREACH(opt, opts, link) {
2642 		if (strcmp(name, opt->name) == 0) {
2643 			opt->seen = 1;
2644 			if (len != opt->len)
2645 				return (EINVAL);
2646 			bcopy(opt->value, dest, opt->len);
2647 			return (0);
2648 		}
2649 	}
2650 	return (ENOENT);
2651 }
2652 
2653 int
2654 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2655 {
2656 	/*
2657 	 * Filesystems only fill in part of the structure for updates, we
2658 	 * have to read the entirety first to get all content.
2659 	 */
2660 	if (sbp != &mp->mnt_stat)
2661 		memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2662 
2663 	/*
2664 	 * Set these in case the underlying filesystem fails to do so.
2665 	 */
2666 	sbp->f_version = STATFS_VERSION;
2667 	sbp->f_namemax = NAME_MAX;
2668 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2669 	sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2670 
2671 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2672 }
2673 
2674 void
2675 vfs_mountedfrom(struct mount *mp, const char *from)
2676 {
2677 
2678 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2679 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2680 	    sizeof mp->mnt_stat.f_mntfromname);
2681 }
2682 
2683 /*
2684  * ---------------------------------------------------------------------
2685  * This is the api for building mount args and mounting filesystems from
2686  * inside the kernel.
2687  *
2688  * The API works by accumulation of individual args.  First error is
2689  * latched.
2690  *
2691  * XXX: should be documented in new manpage kernel_mount(9)
2692  */
2693 
2694 /* A memory allocation which must be freed when we are done */
2695 struct mntaarg {
2696 	SLIST_ENTRY(mntaarg)	next;
2697 };
2698 
2699 /* The header for the mount arguments */
2700 struct mntarg {
2701 	struct iovec *v;
2702 	int len;
2703 	int error;
2704 	SLIST_HEAD(, mntaarg)	list;
2705 };
2706 
2707 /*
2708  * Add a boolean argument.
2709  *
2710  * flag is the boolean value.
2711  * name must start with "no".
2712  */
2713 struct mntarg *
2714 mount_argb(struct mntarg *ma, int flag, const char *name)
2715 {
2716 
2717 	KASSERT(name[0] == 'n' && name[1] == 'o',
2718 	    ("mount_argb(...,%s): name must start with 'no'", name));
2719 
2720 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2721 }
2722 
2723 /*
2724  * Add an argument printf style
2725  */
2726 struct mntarg *
2727 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2728 {
2729 	va_list ap;
2730 	struct mntaarg *maa;
2731 	struct sbuf *sb;
2732 	int len;
2733 
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 
2741 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2742 	    M_MOUNT, M_WAITOK);
2743 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2744 	ma->v[ma->len].iov_len = strlen(name) + 1;
2745 	ma->len++;
2746 
2747 	sb = sbuf_new_auto();
2748 	va_start(ap, fmt);
2749 	sbuf_vprintf(sb, fmt, ap);
2750 	va_end(ap);
2751 	sbuf_finish(sb);
2752 	len = sbuf_len(sb) + 1;
2753 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2754 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2755 	bcopy(sbuf_data(sb), maa + 1, len);
2756 	sbuf_delete(sb);
2757 
2758 	ma->v[ma->len].iov_base = maa + 1;
2759 	ma->v[ma->len].iov_len = len;
2760 	ma->len++;
2761 
2762 	return (ma);
2763 }
2764 
2765 /*
2766  * Add an argument which is a userland string.
2767  */
2768 struct mntarg *
2769 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2770 {
2771 	struct mntaarg *maa;
2772 	char *tbuf;
2773 
2774 	if (val == NULL)
2775 		return (ma);
2776 	if (ma == NULL) {
2777 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2778 		SLIST_INIT(&ma->list);
2779 	}
2780 	if (ma->error)
2781 		return (ma);
2782 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2783 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2784 	tbuf = (void *)(maa + 1);
2785 	ma->error = copyinstr(val, tbuf, len, NULL);
2786 	return (mount_arg(ma, name, tbuf, -1));
2787 }
2788 
2789 /*
2790  * Plain argument.
2791  *
2792  * If length is -1, treat value as a C string.
2793  */
2794 struct mntarg *
2795 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2796 {
2797 
2798 	if (ma == NULL) {
2799 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2800 		SLIST_INIT(&ma->list);
2801 	}
2802 	if (ma->error)
2803 		return (ma);
2804 
2805 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2806 	    M_MOUNT, M_WAITOK);
2807 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2808 	ma->v[ma->len].iov_len = strlen(name) + 1;
2809 	ma->len++;
2810 
2811 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2812 	if (len < 0)
2813 		ma->v[ma->len].iov_len = strlen(val) + 1;
2814 	else
2815 		ma->v[ma->len].iov_len = len;
2816 	ma->len++;
2817 	return (ma);
2818 }
2819 
2820 /*
2821  * Free a mntarg structure
2822  */
2823 static void
2824 free_mntarg(struct mntarg *ma)
2825 {
2826 	struct mntaarg *maa;
2827 
2828 	while (!SLIST_EMPTY(&ma->list)) {
2829 		maa = SLIST_FIRST(&ma->list);
2830 		SLIST_REMOVE_HEAD(&ma->list, next);
2831 		free(maa, M_MOUNT);
2832 	}
2833 	free(ma->v, M_MOUNT);
2834 	free(ma, M_MOUNT);
2835 }
2836 
2837 /*
2838  * Mount a filesystem
2839  */
2840 int
2841 kernel_mount(struct mntarg *ma, uint64_t flags)
2842 {
2843 	struct uio auio;
2844 	int error;
2845 
2846 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2847 	KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2848 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2849 
2850 	error = ma->error;
2851 	if (error == 0) {
2852 		auio.uio_iov = ma->v;
2853 		auio.uio_iovcnt = ma->len;
2854 		auio.uio_segflg = UIO_SYSSPACE;
2855 		error = vfs_donmount(curthread, flags, &auio);
2856 	}
2857 	free_mntarg(ma);
2858 	return (error);
2859 }
2860 
2861 /* Map from mount options to printable formats. */
2862 static struct mntoptnames optnames[] = {
2863 	MNTOPT_NAMES
2864 };
2865 
2866 #define DEVCTL_LEN 1024
2867 static void
2868 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2869 {
2870 	const uint8_t *cp;
2871 	struct mntoptnames *fp;
2872 	struct sbuf sb;
2873 	struct statfs *sfp = &mp->mnt_stat;
2874 	char *buf;
2875 
2876 	buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2877 	if (buf == NULL)
2878 		return;
2879 	sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2880 	sbuf_cpy(&sb, "mount-point=\"");
2881 	devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2882 	sbuf_cat(&sb, "\" mount-dev=\"");
2883 	devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2884 	sbuf_cat(&sb, "\" mount-type=\"");
2885 	devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2886 	sbuf_cat(&sb, "\" fsid=0x");
2887 	cp = (const uint8_t *)&sfp->f_fsid.val[0];
2888 	for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2889 		sbuf_printf(&sb, "%02x", cp[i]);
2890 	sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2891 	for (fp = optnames; fp->o_opt != 0; fp++) {
2892 		if ((mp->mnt_flag & fp->o_opt) != 0) {
2893 			sbuf_cat(&sb, fp->o_name);
2894 			sbuf_putc(&sb, ';');
2895 		}
2896 	}
2897 	sbuf_putc(&sb, '"');
2898 	sbuf_finish(&sb);
2899 
2900 	/*
2901 	 * Options are not published because the form of the options depends on
2902 	 * the file system and may include binary data. In addition, they don't
2903 	 * necessarily provide enough useful information to be actionable when
2904 	 * devd processes them.
2905 	 */
2906 
2907 	if (sbuf_error(&sb) == 0)
2908 		devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2909 	sbuf_delete(&sb);
2910 	free(buf, M_MOUNT);
2911 }
2912 
2913 /*
2914  * Force remount specified mount point to read-only.  The argument
2915  * must be busied to avoid parallel unmount attempts.
2916  *
2917  * Intended use is to prevent further writes if some metadata
2918  * inconsistency is detected.  Note that the function still flushes
2919  * all cached metadata and data for the mount point, which might be
2920  * not always suitable.
2921  */
2922 int
2923 vfs_remount_ro(struct mount *mp)
2924 {
2925 	struct vfsoptlist *opts;
2926 	struct vfsopt *opt;
2927 	struct vnode *vp_covered, *rootvp;
2928 	int error;
2929 
2930 	KASSERT(mp->mnt_lockref > 0,
2931 	    ("vfs_remount_ro: mp %p is not busied", mp));
2932 	KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
2933 	    ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
2934 
2935 	rootvp = NULL;
2936 	vp_covered = mp->mnt_vnodecovered;
2937 	error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
2938 	if (error != 0)
2939 		return (error);
2940 	VI_LOCK(vp_covered);
2941 	if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
2942 		VI_UNLOCK(vp_covered);
2943 		vput(vp_covered);
2944 		return (EBUSY);
2945 	}
2946 	vp_covered->v_iflag |= VI_MOUNT;
2947 	VI_UNLOCK(vp_covered);
2948 	vfs_op_enter(mp);
2949 	vn_seqc_write_begin(vp_covered);
2950 
2951 	MNT_ILOCK(mp);
2952 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
2953 		MNT_IUNLOCK(mp);
2954 		error = EBUSY;
2955 		goto out;
2956 	}
2957 	mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
2958 	rootvp = vfs_cache_root_clear(mp);
2959 	MNT_IUNLOCK(mp);
2960 
2961 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
2962 	TAILQ_INIT(opts);
2963 	opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
2964 	opt->name = strdup("ro", M_MOUNT);
2965 	opt->value = NULL;
2966 	TAILQ_INSERT_TAIL(opts, opt, link);
2967 	vfs_mergeopts(opts, mp->mnt_opt);
2968 	mp->mnt_optnew = opts;
2969 
2970 	error = VFS_MOUNT(mp);
2971 
2972 	if (error == 0) {
2973 		MNT_ILOCK(mp);
2974 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
2975 		MNT_IUNLOCK(mp);
2976 		vfs_deallocate_syncvnode(mp);
2977 		if (mp->mnt_opt != NULL)
2978 			vfs_freeopts(mp->mnt_opt);
2979 		mp->mnt_opt = mp->mnt_optnew;
2980 	} else {
2981 		MNT_ILOCK(mp);
2982 		mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
2983 		MNT_IUNLOCK(mp);
2984 		vfs_freeopts(mp->mnt_optnew);
2985 	}
2986 	mp->mnt_optnew = NULL;
2987 
2988 out:
2989 	vfs_op_exit(mp);
2990 	VI_LOCK(vp_covered);
2991 	vp_covered->v_iflag &= ~VI_MOUNT;
2992 	VI_UNLOCK(vp_covered);
2993 	vput(vp_covered);
2994 	vn_seqc_write_end(vp_covered);
2995 	if (rootvp != NULL) {
2996 		vn_seqc_write_end(rootvp);
2997 		vrele(rootvp);
2998 	}
2999 	return (error);
3000 }
3001 
3002 /*
3003  * Suspend write operations on all local writeable filesystems.  Does
3004  * full sync of them in the process.
3005  *
3006  * Iterate over the mount points in reverse order, suspending most
3007  * recently mounted filesystems first.  It handles a case where a
3008  * filesystem mounted from a md(4) vnode-backed device should be
3009  * suspended before the filesystem that owns the vnode.
3010  */
3011 void
3012 suspend_all_fs(void)
3013 {
3014 	struct mount *mp;
3015 	int error;
3016 
3017 	mtx_lock(&mountlist_mtx);
3018 	TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
3019 		error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
3020 		if (error != 0)
3021 			continue;
3022 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
3023 		    (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
3024 			mtx_lock(&mountlist_mtx);
3025 			vfs_unbusy(mp);
3026 			continue;
3027 		}
3028 		error = vfs_write_suspend(mp, 0);
3029 		if (error == 0) {
3030 			MNT_ILOCK(mp);
3031 			MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
3032 			mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
3033 			MNT_IUNLOCK(mp);
3034 			mtx_lock(&mountlist_mtx);
3035 		} else {
3036 			printf("suspend of %s failed, error %d\n",
3037 			    mp->mnt_stat.f_mntonname, error);
3038 			mtx_lock(&mountlist_mtx);
3039 			vfs_unbusy(mp);
3040 		}
3041 	}
3042 	mtx_unlock(&mountlist_mtx);
3043 }
3044 
3045 void
3046 resume_all_fs(void)
3047 {
3048 	struct mount *mp;
3049 
3050 	mtx_lock(&mountlist_mtx);
3051 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3052 		if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3053 			continue;
3054 		mtx_unlock(&mountlist_mtx);
3055 		MNT_ILOCK(mp);
3056 		MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3057 		mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3058 		MNT_IUNLOCK(mp);
3059 		vfs_write_resume(mp, 0);
3060 		mtx_lock(&mountlist_mtx);
3061 		vfs_unbusy(mp);
3062 	}
3063 	mtx_unlock(&mountlist_mtx);
3064 }
3065