xref: /freebsd/sys/kern/vfs_mount.c (revision 0705bb9de50b128540428cb7f0e0b0ba8be77d55)
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/eventhandler.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/libkern.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/filedesc.h>
58 #include <sys/reboot.h>
59 #include <sys/sbuf.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/systm.h>
66 #include <sys/vnode.h>
67 #include <vm/uma.h>
68 
69 #include <geom/geom.h>
70 
71 #include <machine/stdarg.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
77 
78 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
79 		    uint64_t fsflags, struct vfsoptlist **optlist);
80 static void	free_mntarg(struct mntarg *ma);
81 
82 static int	usermount = 0;
83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
84     "Unprivileged users may mount and unmount file systems");
85 
86 static bool	default_autoro = false;
87 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
88     "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
89 
90 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
91 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
92 static uma_zone_t mount_zone;
93 
94 /* List of mounted filesystems. */
95 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
96 
97 /* For any iteration/modification of mountlist */
98 struct mtx mountlist_mtx;
99 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
100 
101 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
102 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
103 
104 /*
105  * Global opts, taken by all filesystems
106  */
107 static const char *global_opts[] = {
108 	"errmsg",
109 	"fstype",
110 	"fspath",
111 	"ro",
112 	"rw",
113 	"nosuid",
114 	"noexec",
115 	NULL
116 };
117 
118 static int
119 mount_init(void *mem, int size, int flags)
120 {
121 	struct mount *mp;
122 
123 	mp = (struct mount *)mem;
124 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
125 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
126 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
127 	mp->mnt_thread_in_ops_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
128 	    M_WAITOK | M_ZERO);
129 	mp->mnt_ref_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
130 	    M_WAITOK | M_ZERO);
131 	mp->mnt_lockref_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
132 	    M_WAITOK | M_ZERO);
133 	mp->mnt_writeopcount_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
134 	    M_WAITOK | M_ZERO);
135 	mp->mnt_ref = 0;
136 	mp->mnt_vfs_ops = 1;
137 	return (0);
138 }
139 
140 static void
141 mount_fini(void *mem, int size)
142 {
143 	struct mount *mp;
144 
145 	mp = (struct mount *)mem;
146 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_writeopcount_pcpu);
147 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_lockref_pcpu);
148 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_ref_pcpu);
149 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_thread_in_ops_pcpu);
150 	lockdestroy(&mp->mnt_explock);
151 	mtx_destroy(&mp->mnt_listmtx);
152 	mtx_destroy(&mp->mnt_mtx);
153 }
154 
155 static void
156 vfs_mount_init(void *dummy __unused)
157 {
158 
159 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
160 	    NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
161 }
162 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
163 
164 /*
165  * ---------------------------------------------------------------------
166  * Functions for building and sanitizing the mount options
167  */
168 
169 /* Remove one mount option. */
170 static void
171 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
172 {
173 
174 	TAILQ_REMOVE(opts, opt, link);
175 	free(opt->name, M_MOUNT);
176 	if (opt->value != NULL)
177 		free(opt->value, M_MOUNT);
178 	free(opt, M_MOUNT);
179 }
180 
181 /* Release all resources related to the mount options. */
182 void
183 vfs_freeopts(struct vfsoptlist *opts)
184 {
185 	struct vfsopt *opt;
186 
187 	while (!TAILQ_EMPTY(opts)) {
188 		opt = TAILQ_FIRST(opts);
189 		vfs_freeopt(opts, opt);
190 	}
191 	free(opts, M_MOUNT);
192 }
193 
194 void
195 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
196 {
197 	struct vfsopt *opt, *temp;
198 
199 	if (opts == NULL)
200 		return;
201 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
202 		if (strcmp(opt->name, name) == 0)
203 			vfs_freeopt(opts, opt);
204 	}
205 }
206 
207 static int
208 vfs_isopt_ro(const char *opt)
209 {
210 
211 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
212 	    strcmp(opt, "norw") == 0)
213 		return (1);
214 	return (0);
215 }
216 
217 static int
218 vfs_isopt_rw(const char *opt)
219 {
220 
221 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
222 		return (1);
223 	return (0);
224 }
225 
226 /*
227  * Check if options are equal (with or without the "no" prefix).
228  */
229 static int
230 vfs_equalopts(const char *opt1, const char *opt2)
231 {
232 	char *p;
233 
234 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
235 	if (strcmp(opt1, opt2) == 0)
236 		return (1);
237 	/* "noopt" vs. "opt" */
238 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
239 		return (1);
240 	/* "opt" vs. "noopt" */
241 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
242 		return (1);
243 	while ((p = strchr(opt1, '.')) != NULL &&
244 	    !strncmp(opt1, opt2, ++p - opt1)) {
245 		opt2 += p - opt1;
246 		opt1 = p;
247 		/* "foo.noopt" vs. "foo.opt" */
248 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
249 			return (1);
250 		/* "foo.opt" vs. "foo.noopt" */
251 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
252 			return (1);
253 	}
254 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
255 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
256 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
257 		return (1);
258 	return (0);
259 }
260 
261 /*
262  * If a mount option is specified several times,
263  * (with or without the "no" prefix) only keep
264  * the last occurrence of it.
265  */
266 static void
267 vfs_sanitizeopts(struct vfsoptlist *opts)
268 {
269 	struct vfsopt *opt, *opt2, *tmp;
270 
271 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
272 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
273 		while (opt2 != NULL) {
274 			if (vfs_equalopts(opt->name, opt2->name)) {
275 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
276 				vfs_freeopt(opts, opt2);
277 				opt2 = tmp;
278 			} else {
279 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
280 			}
281 		}
282 	}
283 }
284 
285 /*
286  * Build a linked list of mount options from a struct uio.
287  */
288 int
289 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
290 {
291 	struct vfsoptlist *opts;
292 	struct vfsopt *opt;
293 	size_t memused, namelen, optlen;
294 	unsigned int i, iovcnt;
295 	int error;
296 
297 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
298 	TAILQ_INIT(opts);
299 	memused = 0;
300 	iovcnt = auio->uio_iovcnt;
301 	for (i = 0; i < iovcnt; i += 2) {
302 		namelen = auio->uio_iov[i].iov_len;
303 		optlen = auio->uio_iov[i + 1].iov_len;
304 		memused += sizeof(struct vfsopt) + optlen + namelen;
305 		/*
306 		 * Avoid consuming too much memory, and attempts to overflow
307 		 * memused.
308 		 */
309 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
310 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
311 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
312 			error = EINVAL;
313 			goto bad;
314 		}
315 
316 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
317 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
318 		opt->value = NULL;
319 		opt->len = 0;
320 		opt->pos = i / 2;
321 		opt->seen = 0;
322 
323 		/*
324 		 * Do this early, so jumps to "bad" will free the current
325 		 * option.
326 		 */
327 		TAILQ_INSERT_TAIL(opts, opt, link);
328 
329 		if (auio->uio_segflg == UIO_SYSSPACE) {
330 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
331 		} else {
332 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
333 			    namelen);
334 			if (error)
335 				goto bad;
336 		}
337 		/* Ensure names are null-terminated strings. */
338 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
339 			error = EINVAL;
340 			goto bad;
341 		}
342 		if (optlen != 0) {
343 			opt->len = optlen;
344 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
345 			if (auio->uio_segflg == UIO_SYSSPACE) {
346 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
347 				    optlen);
348 			} else {
349 				error = copyin(auio->uio_iov[i + 1].iov_base,
350 				    opt->value, optlen);
351 				if (error)
352 					goto bad;
353 			}
354 		}
355 	}
356 	vfs_sanitizeopts(opts);
357 	*options = opts;
358 	return (0);
359 bad:
360 	vfs_freeopts(opts);
361 	return (error);
362 }
363 
364 /*
365  * Merge the old mount options with the new ones passed
366  * in the MNT_UPDATE case.
367  *
368  * XXX: This function will keep a "nofoo" option in the new
369  * options.  E.g, if the option's canonical name is "foo",
370  * "nofoo" ends up in the mount point's active options.
371  */
372 static void
373 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
374 {
375 	struct vfsopt *opt, *new;
376 
377 	TAILQ_FOREACH(opt, oldopts, link) {
378 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
379 		new->name = strdup(opt->name, M_MOUNT);
380 		if (opt->len != 0) {
381 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
382 			bcopy(opt->value, new->value, opt->len);
383 		} else
384 			new->value = NULL;
385 		new->len = opt->len;
386 		new->seen = opt->seen;
387 		TAILQ_INSERT_HEAD(toopts, new, link);
388 	}
389 	vfs_sanitizeopts(toopts);
390 }
391 
392 /*
393  * Mount a filesystem.
394  */
395 #ifndef _SYS_SYSPROTO_H_
396 struct nmount_args {
397 	struct iovec *iovp;
398 	unsigned int iovcnt;
399 	int flags;
400 };
401 #endif
402 int
403 sys_nmount(struct thread *td, struct nmount_args *uap)
404 {
405 	struct uio *auio;
406 	int error;
407 	u_int iovcnt;
408 	uint64_t flags;
409 
410 	/*
411 	 * Mount flags are now 64-bits. On 32-bit archtectures only
412 	 * 32-bits are passed in, but from here on everything handles
413 	 * 64-bit flags correctly.
414 	 */
415 	flags = uap->flags;
416 
417 	AUDIT_ARG_FFLAGS(flags);
418 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
419 	    uap->iovp, uap->iovcnt, flags);
420 
421 	/*
422 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
423 	 * userspace to set this flag, but we must filter it out if we want
424 	 * MNT_UPDATE on the root file system to work.
425 	 * MNT_ROOTFS should only be set by the kernel when mounting its
426 	 * root file system.
427 	 */
428 	flags &= ~MNT_ROOTFS;
429 
430 	iovcnt = uap->iovcnt;
431 	/*
432 	 * Check that we have an even number of iovec's
433 	 * and that we have at least two options.
434 	 */
435 	if ((iovcnt & 1) || (iovcnt < 4)) {
436 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
437 		    uap->iovcnt);
438 		return (EINVAL);
439 	}
440 
441 	error = copyinuio(uap->iovp, iovcnt, &auio);
442 	if (error) {
443 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
444 		    __func__, error);
445 		return (error);
446 	}
447 	error = vfs_donmount(td, flags, auio);
448 
449 	free(auio, M_IOV);
450 	return (error);
451 }
452 
453 /*
454  * ---------------------------------------------------------------------
455  * Various utility functions
456  */
457 
458 void
459 vfs_ref(struct mount *mp)
460 {
461 
462 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
463 	if (vfs_op_thread_enter(mp)) {
464 		vfs_mp_count_add_pcpu(mp, ref, 1);
465 		vfs_op_thread_exit(mp);
466 		return;
467 	}
468 
469 	MNT_ILOCK(mp);
470 	MNT_REF(mp);
471 	MNT_IUNLOCK(mp);
472 }
473 
474 void
475 vfs_rel(struct mount *mp)
476 {
477 
478 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
479 	if (vfs_op_thread_enter(mp)) {
480 		vfs_mp_count_sub_pcpu(mp, ref, 1);
481 		vfs_op_thread_exit(mp);
482 		return;
483 	}
484 
485 	MNT_ILOCK(mp);
486 	MNT_REL(mp);
487 	MNT_IUNLOCK(mp);
488 }
489 
490 /*
491  * Allocate and initialize the mount point struct.
492  */
493 struct mount *
494 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
495     struct ucred *cred)
496 {
497 	struct mount *mp;
498 
499 	mp = uma_zalloc(mount_zone, M_WAITOK);
500 	bzero(&mp->mnt_startzero,
501 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
502 	TAILQ_INIT(&mp->mnt_nvnodelist);
503 	mp->mnt_nvnodelistsize = 0;
504 	TAILQ_INIT(&mp->mnt_activevnodelist);
505 	mp->mnt_activevnodelistsize = 0;
506 	TAILQ_INIT(&mp->mnt_tmpfreevnodelist);
507 	mp->mnt_tmpfreevnodelistsize = 0;
508 	if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
509 	    mp->mnt_writeopcount != 0)
510 		panic("%s: non-zero counters on new mp %p\n", __func__, mp);
511 	if (mp->mnt_vfs_ops != 1)
512 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
513 		    mp->mnt_vfs_ops);
514 	(void) vfs_busy(mp, MBF_NOWAIT);
515 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
516 	mp->mnt_op = vfsp->vfc_vfsops;
517 	mp->mnt_vfc = vfsp;
518 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
519 	mp->mnt_gen++;
520 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
521 	mp->mnt_vnodecovered = vp;
522 	mp->mnt_cred = crdup(cred);
523 	mp->mnt_stat.f_owner = cred->cr_uid;
524 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
525 	mp->mnt_iosize_max = DFLTPHYS;
526 #ifdef MAC
527 	mac_mount_init(mp);
528 	mac_mount_create(cred, mp);
529 #endif
530 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
531 	TAILQ_INIT(&mp->mnt_uppers);
532 	return (mp);
533 }
534 
535 /*
536  * Destroy the mount struct previously allocated by vfs_mount_alloc().
537  */
538 void
539 vfs_mount_destroy(struct mount *mp)
540 {
541 
542 	if (mp->mnt_vfs_ops == 0)
543 		panic("%s: entered with zero vfs_ops\n", __func__);
544 
545 	vfs_assert_mount_counters(mp);
546 
547 	MNT_ILOCK(mp);
548 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
549 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
550 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
551 		wakeup(mp);
552 	}
553 	while (mp->mnt_ref)
554 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
555 	KASSERT(mp->mnt_ref == 0,
556 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
557 	    __FILE__, __LINE__));
558 	if (mp->mnt_writeopcount != 0)
559 		panic("vfs_mount_destroy: nonzero writeopcount");
560 	if (mp->mnt_secondary_writes != 0)
561 		panic("vfs_mount_destroy: nonzero secondary_writes");
562 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
563 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
564 		struct vnode *vp;
565 
566 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
567 			vn_printf(vp, "dangling vnode ");
568 		panic("unmount: dangling vnode");
569 	}
570 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
571 	if (mp->mnt_nvnodelistsize != 0)
572 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
573 	if (mp->mnt_activevnodelistsize != 0)
574 		panic("vfs_mount_destroy: nonzero activevnodelistsize");
575 	if (mp->mnt_tmpfreevnodelistsize != 0)
576 		panic("vfs_mount_destroy: nonzero tmpfreevnodelistsize");
577 	if (mp->mnt_lockref != 0)
578 		panic("vfs_mount_destroy: nonzero lock refcount");
579 	MNT_IUNLOCK(mp);
580 
581 	if (mp->mnt_vfs_ops != 1)
582 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
583 		    mp->mnt_vfs_ops);
584 
585 	if (mp->mnt_vnodecovered != NULL)
586 		vrele(mp->mnt_vnodecovered);
587 #ifdef MAC
588 	mac_mount_destroy(mp);
589 #endif
590 	if (mp->mnt_opt != NULL)
591 		vfs_freeopts(mp->mnt_opt);
592 	crfree(mp->mnt_cred);
593 	uma_zfree(mount_zone, mp);
594 }
595 
596 static bool
597 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
598 {
599 	/* This is an upgrade of an exisiting mount. */
600 	if ((fsflags & MNT_UPDATE) != 0)
601 		return (false);
602 	/* This is already an R/O mount. */
603 	if ((fsflags & MNT_RDONLY) != 0)
604 		return (false);
605 
606 	switch (error) {
607 	case ENODEV:	/* generic, geom, ... */
608 	case EACCES:	/* cam/scsi, ... */
609 	case EROFS:	/* md, mmcsd, ... */
610 		/*
611 		 * These errors can be returned by the storage layer to signal
612 		 * that the media is read-only.  No harm in the R/O mount
613 		 * attempt if the error was returned for some other reason.
614 		 */
615 		return (true);
616 	default:
617 		return (false);
618 	}
619 }
620 
621 int
622 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
623 {
624 	struct vfsoptlist *optlist;
625 	struct vfsopt *opt, *tmp_opt;
626 	char *fstype, *fspath, *errmsg;
627 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
628 	bool autoro;
629 
630 	errmsg = fspath = NULL;
631 	errmsg_len = fspathlen = 0;
632 	errmsg_pos = -1;
633 	autoro = default_autoro;
634 
635 	error = vfs_buildopts(fsoptions, &optlist);
636 	if (error)
637 		return (error);
638 
639 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
640 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
641 
642 	/*
643 	 * We need these two options before the others,
644 	 * and they are mandatory for any filesystem.
645 	 * Ensure they are NUL terminated as well.
646 	 */
647 	fstypelen = 0;
648 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
649 	if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
650 		error = EINVAL;
651 		if (errmsg != NULL)
652 			strncpy(errmsg, "Invalid fstype", errmsg_len);
653 		goto bail;
654 	}
655 	fspathlen = 0;
656 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
657 	if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
658 		error = EINVAL;
659 		if (errmsg != NULL)
660 			strncpy(errmsg, "Invalid fspath", errmsg_len);
661 		goto bail;
662 	}
663 
664 	/*
665 	 * We need to see if we have the "update" option
666 	 * before we call vfs_domount(), since vfs_domount() has special
667 	 * logic based on MNT_UPDATE.  This is very important
668 	 * when we want to update the root filesystem.
669 	 */
670 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
671 		int do_freeopt = 0;
672 
673 		if (strcmp(opt->name, "update") == 0) {
674 			fsflags |= MNT_UPDATE;
675 			do_freeopt = 1;
676 		}
677 		else if (strcmp(opt->name, "async") == 0)
678 			fsflags |= MNT_ASYNC;
679 		else if (strcmp(opt->name, "force") == 0) {
680 			fsflags |= MNT_FORCE;
681 			do_freeopt = 1;
682 		}
683 		else if (strcmp(opt->name, "reload") == 0) {
684 			fsflags |= MNT_RELOAD;
685 			do_freeopt = 1;
686 		}
687 		else if (strcmp(opt->name, "multilabel") == 0)
688 			fsflags |= MNT_MULTILABEL;
689 		else if (strcmp(opt->name, "noasync") == 0)
690 			fsflags &= ~MNT_ASYNC;
691 		else if (strcmp(opt->name, "noatime") == 0)
692 			fsflags |= MNT_NOATIME;
693 		else if (strcmp(opt->name, "atime") == 0) {
694 			free(opt->name, M_MOUNT);
695 			opt->name = strdup("nonoatime", M_MOUNT);
696 		}
697 		else if (strcmp(opt->name, "noclusterr") == 0)
698 			fsflags |= MNT_NOCLUSTERR;
699 		else if (strcmp(opt->name, "clusterr") == 0) {
700 			free(opt->name, M_MOUNT);
701 			opt->name = strdup("nonoclusterr", M_MOUNT);
702 		}
703 		else if (strcmp(opt->name, "noclusterw") == 0)
704 			fsflags |= MNT_NOCLUSTERW;
705 		else if (strcmp(opt->name, "clusterw") == 0) {
706 			free(opt->name, M_MOUNT);
707 			opt->name = strdup("nonoclusterw", M_MOUNT);
708 		}
709 		else if (strcmp(opt->name, "noexec") == 0)
710 			fsflags |= MNT_NOEXEC;
711 		else if (strcmp(opt->name, "exec") == 0) {
712 			free(opt->name, M_MOUNT);
713 			opt->name = strdup("nonoexec", M_MOUNT);
714 		}
715 		else if (strcmp(opt->name, "nosuid") == 0)
716 			fsflags |= MNT_NOSUID;
717 		else if (strcmp(opt->name, "suid") == 0) {
718 			free(opt->name, M_MOUNT);
719 			opt->name = strdup("nonosuid", M_MOUNT);
720 		}
721 		else if (strcmp(opt->name, "nosymfollow") == 0)
722 			fsflags |= MNT_NOSYMFOLLOW;
723 		else if (strcmp(opt->name, "symfollow") == 0) {
724 			free(opt->name, M_MOUNT);
725 			opt->name = strdup("nonosymfollow", M_MOUNT);
726 		}
727 		else if (strcmp(opt->name, "noro") == 0) {
728 			fsflags &= ~MNT_RDONLY;
729 			autoro = false;
730 		}
731 		else if (strcmp(opt->name, "rw") == 0) {
732 			fsflags &= ~MNT_RDONLY;
733 			autoro = false;
734 		}
735 		else if (strcmp(opt->name, "ro") == 0) {
736 			fsflags |= MNT_RDONLY;
737 			autoro = false;
738 		}
739 		else if (strcmp(opt->name, "rdonly") == 0) {
740 			free(opt->name, M_MOUNT);
741 			opt->name = strdup("ro", M_MOUNT);
742 			fsflags |= MNT_RDONLY;
743 			autoro = false;
744 		}
745 		else if (strcmp(opt->name, "autoro") == 0) {
746 			do_freeopt = 1;
747 			autoro = true;
748 		}
749 		else if (strcmp(opt->name, "suiddir") == 0)
750 			fsflags |= MNT_SUIDDIR;
751 		else if (strcmp(opt->name, "sync") == 0)
752 			fsflags |= MNT_SYNCHRONOUS;
753 		else if (strcmp(opt->name, "union") == 0)
754 			fsflags |= MNT_UNION;
755 		else if (strcmp(opt->name, "automounted") == 0) {
756 			fsflags |= MNT_AUTOMOUNTED;
757 			do_freeopt = 1;
758 		} else if (strcmp(opt->name, "nocover") == 0) {
759 			fsflags |= MNT_NOCOVER;
760 			do_freeopt = 1;
761 		} else if (strcmp(opt->name, "cover") == 0) {
762 			fsflags &= ~MNT_NOCOVER;
763 			do_freeopt = 1;
764 		} else if (strcmp(opt->name, "emptydir") == 0) {
765 			fsflags |= MNT_EMPTYDIR;
766 			do_freeopt = 1;
767 		} else if (strcmp(opt->name, "noemptydir") == 0) {
768 			fsflags &= ~MNT_EMPTYDIR;
769 			do_freeopt = 1;
770 		}
771 		if (do_freeopt)
772 			vfs_freeopt(optlist, opt);
773 	}
774 
775 	/*
776 	 * Be ultra-paranoid about making sure the type and fspath
777 	 * variables will fit in our mp buffers, including the
778 	 * terminating NUL.
779 	 */
780 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
781 		error = ENAMETOOLONG;
782 		goto bail;
783 	}
784 
785 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
786 
787 	/*
788 	 * See if we can mount in the read-only mode if the error code suggests
789 	 * that it could be possible and the mount options allow for that.
790 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
791 	 * overridden by "autoro".
792 	 */
793 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
794 		printf("%s: R/W mount failed, possibly R/O media,"
795 		    " trying R/O mount\n", __func__);
796 		fsflags |= MNT_RDONLY;
797 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
798 	}
799 bail:
800 	/* copyout the errmsg */
801 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
802 	    && errmsg_len > 0 && errmsg != NULL) {
803 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
804 			bcopy(errmsg,
805 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
806 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
807 		} else {
808 			copyout(errmsg,
809 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
810 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
811 		}
812 	}
813 
814 	if (optlist != NULL)
815 		vfs_freeopts(optlist);
816 	return (error);
817 }
818 
819 /*
820  * Old mount API.
821  */
822 #ifndef _SYS_SYSPROTO_H_
823 struct mount_args {
824 	char	*type;
825 	char	*path;
826 	int	flags;
827 	caddr_t	data;
828 };
829 #endif
830 /* ARGSUSED */
831 int
832 sys_mount(struct thread *td, struct mount_args *uap)
833 {
834 	char *fstype;
835 	struct vfsconf *vfsp = NULL;
836 	struct mntarg *ma = NULL;
837 	uint64_t flags;
838 	int error;
839 
840 	/*
841 	 * Mount flags are now 64-bits. On 32-bit architectures only
842 	 * 32-bits are passed in, but from here on everything handles
843 	 * 64-bit flags correctly.
844 	 */
845 	flags = uap->flags;
846 
847 	AUDIT_ARG_FFLAGS(flags);
848 
849 	/*
850 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
851 	 * userspace to set this flag, but we must filter it out if we want
852 	 * MNT_UPDATE on the root file system to work.
853 	 * MNT_ROOTFS should only be set by the kernel when mounting its
854 	 * root file system.
855 	 */
856 	flags &= ~MNT_ROOTFS;
857 
858 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
859 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
860 	if (error) {
861 		free(fstype, M_TEMP);
862 		return (error);
863 	}
864 
865 	AUDIT_ARG_TEXT(fstype);
866 	vfsp = vfs_byname_kld(fstype, td, &error);
867 	free(fstype, M_TEMP);
868 	if (vfsp == NULL)
869 		return (ENOENT);
870 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
871 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
872 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
873 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
874 		return (EOPNOTSUPP);
875 
876 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
877 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
878 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
879 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
880 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
881 
882 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
883 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
884 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
885 }
886 
887 /*
888  * vfs_domount_first(): first file system mount (not update)
889  */
890 static int
891 vfs_domount_first(
892 	struct thread *td,		/* Calling thread. */
893 	struct vfsconf *vfsp,		/* File system type. */
894 	char *fspath,			/* Mount path. */
895 	struct vnode *vp,		/* Vnode to be covered. */
896 	uint64_t fsflags,		/* Flags common to all filesystems. */
897 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
898 	)
899 {
900 	struct vattr va;
901 	struct mount *mp;
902 	struct vnode *newdp;
903 	int error, error1;
904 
905 	ASSERT_VOP_ELOCKED(vp, __func__);
906 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
907 
908 	if ((fsflags & MNT_EMPTYDIR) != 0) {
909 		error = vfs_emptydir(vp);
910 		if (error != 0) {
911 			vput(vp);
912 			return (error);
913 		}
914 	}
915 
916 	/*
917 	 * If the jail of the calling thread lacks permission for this type of
918 	 * file system, deny immediately.
919 	 */
920 	if (jailed(td->td_ucred) && !prison_allow(td->td_ucred,
921 	    vfsp->vfc_prison_flag)) {
922 		vput(vp);
923 		return (EPERM);
924 	}
925 
926 	/*
927 	 * If the user is not root, ensure that they own the directory
928 	 * onto which we are attempting to mount.
929 	 */
930 	error = VOP_GETATTR(vp, &va, td->td_ucred);
931 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
932 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
933 	if (error == 0)
934 		error = vinvalbuf(vp, V_SAVE, 0, 0);
935 	if (error == 0 && vp->v_type != VDIR)
936 		error = ENOTDIR;
937 	if (error == 0) {
938 		VI_LOCK(vp);
939 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
940 			vp->v_iflag |= VI_MOUNT;
941 		else
942 			error = EBUSY;
943 		VI_UNLOCK(vp);
944 	}
945 	if (error != 0) {
946 		vput(vp);
947 		return (error);
948 	}
949 	VOP_UNLOCK(vp, 0);
950 
951 	/* Allocate and initialize the filesystem. */
952 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
953 	/* XXXMAC: pass to vfs_mount_alloc? */
954 	mp->mnt_optnew = *optlist;
955 	/* Set the mount level flags. */
956 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
957 
958 	/*
959 	 * Mount the filesystem.
960 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
961 	 * get.  No freeing of cn_pnbuf.
962 	 */
963 	error1 = 0;
964 	if ((error = VFS_MOUNT(mp)) != 0 ||
965 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
966 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
967 		if (error1 != 0) {
968 			error = error1;
969 			if ((error1 = VFS_UNMOUNT(mp, 0)) != 0)
970 				printf("VFS_UNMOUNT returned %d\n", error1);
971 		}
972 		vfs_unbusy(mp);
973 		mp->mnt_vnodecovered = NULL;
974 		vfs_mount_destroy(mp);
975 		VI_LOCK(vp);
976 		vp->v_iflag &= ~VI_MOUNT;
977 		VI_UNLOCK(vp);
978 		vrele(vp);
979 		return (error);
980 	}
981 	VOP_UNLOCK(newdp, 0);
982 
983 	if (mp->mnt_opt != NULL)
984 		vfs_freeopts(mp->mnt_opt);
985 	mp->mnt_opt = mp->mnt_optnew;
986 	*optlist = NULL;
987 
988 	/*
989 	 * Prevent external consumers of mount options from reading mnt_optnew.
990 	 */
991 	mp->mnt_optnew = NULL;
992 
993 	MNT_ILOCK(mp);
994 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
995 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
996 		mp->mnt_kern_flag |= MNTK_ASYNC;
997 	else
998 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
999 	MNT_IUNLOCK(mp);
1000 
1001 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1002 	cache_purge(vp);
1003 	VI_LOCK(vp);
1004 	vp->v_iflag &= ~VI_MOUNT;
1005 	VI_UNLOCK(vp);
1006 	vp->v_mountedhere = mp;
1007 	/* Place the new filesystem at the end of the mount list. */
1008 	mtx_lock(&mountlist_mtx);
1009 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1010 	mtx_unlock(&mountlist_mtx);
1011 	vfs_event_signal(NULL, VQ_MOUNT, 0);
1012 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1013 	VOP_UNLOCK(vp, 0);
1014 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1015 	VOP_UNLOCK(newdp, 0);
1016 	mountcheckdirs(vp, newdp);
1017 	vrele(newdp);
1018 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1019 		vfs_allocate_syncvnode(mp);
1020 	vfs_op_exit(mp);
1021 	vfs_unbusy(mp);
1022 	return (0);
1023 }
1024 
1025 /*
1026  * vfs_domount_update(): update of mounted file system
1027  */
1028 static int
1029 vfs_domount_update(
1030 	struct thread *td,		/* Calling thread. */
1031 	struct vnode *vp,		/* Mount point vnode. */
1032 	uint64_t fsflags,		/* Flags common to all filesystems. */
1033 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1034 	)
1035 {
1036 	struct export_args export;
1037 	void *bufp;
1038 	struct mount *mp;
1039 	int error, export_error, len;
1040 	uint64_t flag;
1041 
1042 	ASSERT_VOP_ELOCKED(vp, __func__);
1043 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1044 	mp = vp->v_mount;
1045 
1046 	if ((vp->v_vflag & VV_ROOT) == 0) {
1047 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1048 		    == 0)
1049 			error = EXDEV;
1050 		else
1051 			error = EINVAL;
1052 		vput(vp);
1053 		return (error);
1054 	}
1055 
1056 	/*
1057 	 * We only allow the filesystem to be reloaded if it
1058 	 * is currently mounted read-only.
1059 	 */
1060 	flag = mp->mnt_flag;
1061 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1062 		vput(vp);
1063 		return (EOPNOTSUPP);	/* Needs translation */
1064 	}
1065 	/*
1066 	 * Only privileged root, or (if MNT_USER is set) the user that
1067 	 * did the original mount is permitted to update it.
1068 	 */
1069 	error = vfs_suser(mp, td);
1070 	if (error != 0) {
1071 		vput(vp);
1072 		return (error);
1073 	}
1074 	if (vfs_busy(mp, MBF_NOWAIT)) {
1075 		vput(vp);
1076 		return (EBUSY);
1077 	}
1078 	VI_LOCK(vp);
1079 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1080 		VI_UNLOCK(vp);
1081 		vfs_unbusy(mp);
1082 		vput(vp);
1083 		return (EBUSY);
1084 	}
1085 	vp->v_iflag |= VI_MOUNT;
1086 	VI_UNLOCK(vp);
1087 	VOP_UNLOCK(vp, 0);
1088 
1089 	vfs_op_enter(mp);
1090 
1091 	MNT_ILOCK(mp);
1092 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1093 		MNT_IUNLOCK(mp);
1094 		error = EBUSY;
1095 		goto end;
1096 	}
1097 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1098 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1099 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1100 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1101 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1102 	MNT_IUNLOCK(mp);
1103 	mp->mnt_optnew = *optlist;
1104 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1105 
1106 	/*
1107 	 * Mount the filesystem.
1108 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1109 	 * get.  No freeing of cn_pnbuf.
1110 	 */
1111 	error = VFS_MOUNT(mp);
1112 
1113 	export_error = 0;
1114 	/* Process the export option. */
1115 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1116 	    &len) == 0) {
1117 		/* Assume that there is only 1 ABI for each length. */
1118 		switch (len) {
1119 		case (sizeof(struct oexport_args)):
1120 			bzero(&export, sizeof(export));
1121 			/* FALLTHROUGH */
1122 		case (sizeof(export)):
1123 			bcopy(bufp, &export, len);
1124 			export_error = vfs_export(mp, &export);
1125 			break;
1126 		default:
1127 			export_error = EINVAL;
1128 			break;
1129 		}
1130 	}
1131 
1132 	MNT_ILOCK(mp);
1133 	if (error == 0) {
1134 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1135 		    MNT_SNAPSHOT);
1136 	} else {
1137 		/*
1138 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1139 		 * because it is not part of MNT_UPDATEMASK, but it could have
1140 		 * changed in the meantime if quotactl(2) was called.
1141 		 * All in all we want current value of MNT_QUOTA, not the old
1142 		 * one.
1143 		 */
1144 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1145 	}
1146 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1147 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1148 		mp->mnt_kern_flag |= MNTK_ASYNC;
1149 	else
1150 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1151 	MNT_IUNLOCK(mp);
1152 
1153 	if (error != 0)
1154 		goto end;
1155 
1156 	if (mp->mnt_opt != NULL)
1157 		vfs_freeopts(mp->mnt_opt);
1158 	mp->mnt_opt = mp->mnt_optnew;
1159 	*optlist = NULL;
1160 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1161 	/*
1162 	 * Prevent external consumers of mount options from reading
1163 	 * mnt_optnew.
1164 	 */
1165 	mp->mnt_optnew = NULL;
1166 
1167 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1168 		vfs_allocate_syncvnode(mp);
1169 	else
1170 		vfs_deallocate_syncvnode(mp);
1171 end:
1172 	vfs_op_exit(mp);
1173 	vfs_unbusy(mp);
1174 	VI_LOCK(vp);
1175 	vp->v_iflag &= ~VI_MOUNT;
1176 	VI_UNLOCK(vp);
1177 	vrele(vp);
1178 	return (error != 0 ? error : export_error);
1179 }
1180 
1181 /*
1182  * vfs_domount(): actually attempt a filesystem mount.
1183  */
1184 static int
1185 vfs_domount(
1186 	struct thread *td,		/* Calling thread. */
1187 	const char *fstype,		/* Filesystem type. */
1188 	char *fspath,			/* Mount path. */
1189 	uint64_t fsflags,		/* Flags common to all filesystems. */
1190 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1191 	)
1192 {
1193 	struct vfsconf *vfsp;
1194 	struct nameidata nd;
1195 	struct vnode *vp;
1196 	char *pathbuf;
1197 	int error;
1198 
1199 	/*
1200 	 * Be ultra-paranoid about making sure the type and fspath
1201 	 * variables will fit in our mp buffers, including the
1202 	 * terminating NUL.
1203 	 */
1204 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1205 		return (ENAMETOOLONG);
1206 
1207 	if (jailed(td->td_ucred) || usermount == 0) {
1208 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1209 			return (error);
1210 	}
1211 
1212 	/*
1213 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1214 	 */
1215 	if (fsflags & MNT_EXPORTED) {
1216 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1217 		if (error)
1218 			return (error);
1219 	}
1220 	if (fsflags & MNT_SUIDDIR) {
1221 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1222 		if (error)
1223 			return (error);
1224 	}
1225 	/*
1226 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1227 	 */
1228 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1229 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1230 			fsflags |= MNT_NOSUID | MNT_USER;
1231 	}
1232 
1233 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1234 	vfsp = NULL;
1235 	if ((fsflags & MNT_UPDATE) == 0) {
1236 		/* Don't try to load KLDs if we're mounting the root. */
1237 		if (fsflags & MNT_ROOTFS)
1238 			vfsp = vfs_byname(fstype);
1239 		else
1240 			vfsp = vfs_byname_kld(fstype, td, &error);
1241 		if (vfsp == NULL)
1242 			return (ENODEV);
1243 	}
1244 
1245 	/*
1246 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1247 	 */
1248 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1249 	    UIO_SYSSPACE, fspath, td);
1250 	error = namei(&nd);
1251 	if (error != 0)
1252 		return (error);
1253 	NDFREE(&nd, NDF_ONLY_PNBUF);
1254 	vp = nd.ni_vp;
1255 	if ((fsflags & MNT_UPDATE) == 0) {
1256 		if ((vp->v_vflag & VV_ROOT) != 0 &&
1257 		    (fsflags & MNT_NOCOVER) != 0) {
1258 			vput(vp);
1259 			return (EBUSY);
1260 		}
1261 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1262 		strcpy(pathbuf, fspath);
1263 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1264 		/* debug.disablefullpath == 1 results in ENODEV */
1265 		if (error == 0 || error == ENODEV) {
1266 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1267 			    fsflags, optlist);
1268 		}
1269 		free(pathbuf, M_TEMP);
1270 	} else
1271 		error = vfs_domount_update(td, vp, fsflags, optlist);
1272 
1273 	return (error);
1274 }
1275 
1276 /*
1277  * Unmount a filesystem.
1278  *
1279  * Note: unmount takes a path to the vnode mounted on as argument, not
1280  * special file (as before).
1281  */
1282 #ifndef _SYS_SYSPROTO_H_
1283 struct unmount_args {
1284 	char	*path;
1285 	int	flags;
1286 };
1287 #endif
1288 /* ARGSUSED */
1289 int
1290 sys_unmount(struct thread *td, struct unmount_args *uap)
1291 {
1292 	struct nameidata nd;
1293 	struct mount *mp;
1294 	char *pathbuf;
1295 	int error, id0, id1;
1296 
1297 	AUDIT_ARG_VALUE(uap->flags);
1298 	if (jailed(td->td_ucred) || usermount == 0) {
1299 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1300 		if (error)
1301 			return (error);
1302 	}
1303 
1304 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1305 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1306 	if (error) {
1307 		free(pathbuf, M_TEMP);
1308 		return (error);
1309 	}
1310 	if (uap->flags & MNT_BYFSID) {
1311 		AUDIT_ARG_TEXT(pathbuf);
1312 		/* Decode the filesystem ID. */
1313 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1314 			free(pathbuf, M_TEMP);
1315 			return (EINVAL);
1316 		}
1317 
1318 		mtx_lock(&mountlist_mtx);
1319 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1320 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1321 			    mp->mnt_stat.f_fsid.val[1] == id1) {
1322 				vfs_ref(mp);
1323 				break;
1324 			}
1325 		}
1326 		mtx_unlock(&mountlist_mtx);
1327 	} else {
1328 		/*
1329 		 * Try to find global path for path argument.
1330 		 */
1331 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1332 		    UIO_SYSSPACE, pathbuf, td);
1333 		if (namei(&nd) == 0) {
1334 			NDFREE(&nd, NDF_ONLY_PNBUF);
1335 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1336 			    MNAMELEN);
1337 			if (error == 0 || error == ENODEV)
1338 				vput(nd.ni_vp);
1339 		}
1340 		mtx_lock(&mountlist_mtx);
1341 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1342 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1343 				vfs_ref(mp);
1344 				break;
1345 			}
1346 		}
1347 		mtx_unlock(&mountlist_mtx);
1348 	}
1349 	free(pathbuf, M_TEMP);
1350 	if (mp == NULL) {
1351 		/*
1352 		 * Previously we returned ENOENT for a nonexistent path and
1353 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1354 		 * now, so in the !MNT_BYFSID case return the more likely
1355 		 * EINVAL for compatibility.
1356 		 */
1357 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1358 	}
1359 
1360 	/*
1361 	 * Don't allow unmounting the root filesystem.
1362 	 */
1363 	if (mp->mnt_flag & MNT_ROOTFS) {
1364 		vfs_rel(mp);
1365 		return (EINVAL);
1366 	}
1367 	error = dounmount(mp, uap->flags, td);
1368 	return (error);
1369 }
1370 
1371 /*
1372  * Return error if any of the vnodes, ignoring the root vnode
1373  * and the syncer vnode, have non-zero usecount.
1374  *
1375  * This function is purely advisory - it can return false positives
1376  * and negatives.
1377  */
1378 static int
1379 vfs_check_usecounts(struct mount *mp)
1380 {
1381 	struct vnode *vp, *mvp;
1382 
1383 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1384 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1385 		    vp->v_usecount != 0) {
1386 			VI_UNLOCK(vp);
1387 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1388 			return (EBUSY);
1389 		}
1390 		VI_UNLOCK(vp);
1391 	}
1392 
1393 	return (0);
1394 }
1395 
1396 static void
1397 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1398 {
1399 
1400 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1401 	mp->mnt_kern_flag &= ~mntkflags;
1402 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1403 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1404 		wakeup(mp);
1405 	}
1406 	vfs_op_exit_locked(mp);
1407 	MNT_IUNLOCK(mp);
1408 	if (coveredvp != NULL) {
1409 		VOP_UNLOCK(coveredvp, 0);
1410 		vdrop(coveredvp);
1411 	}
1412 	vn_finished_write(mp);
1413 }
1414 
1415 /*
1416  * There are various reference counters associated with the mount point.
1417  * Normally it is permitted to modify them without taking the mnt ilock,
1418  * but this behavior can be temporarily disabled if stable value is needed
1419  * or callers are expected to block (e.g. to not allow new users during
1420  * forced unmount).
1421  */
1422 void
1423 vfs_op_enter(struct mount *mp)
1424 {
1425 	int cpu;
1426 
1427 	MNT_ILOCK(mp);
1428 	mp->mnt_vfs_ops++;
1429 	if (mp->mnt_vfs_ops > 1) {
1430 		MNT_IUNLOCK(mp);
1431 		return;
1432 	}
1433 	/*
1434 	 * Paired with a fence in vfs_op_thread_enter(). See the comment
1435 	 * above it for details.
1436 	 */
1437 	atomic_thread_fence_seq_cst();
1438 	vfs_op_barrier_wait(mp);
1439 	/*
1440 	 * Paired with a fence in vfs_op_thread_exit().
1441 	 */
1442 	atomic_thread_fence_acq();
1443 	CPU_FOREACH(cpu) {
1444 		mp->mnt_ref +=
1445 		    zpcpu_replace_cpu(mp->mnt_ref_pcpu, 0, cpu);
1446 		mp->mnt_lockref +=
1447 		    zpcpu_replace_cpu(mp->mnt_lockref_pcpu, 0, cpu);
1448 		mp->mnt_writeopcount +=
1449 		    zpcpu_replace_cpu(mp->mnt_writeopcount_pcpu, 0, cpu);
1450 	}
1451 	MNT_IUNLOCK(mp);
1452 	vfs_assert_mount_counters(mp);
1453 }
1454 
1455 void
1456 vfs_op_exit_locked(struct mount *mp)
1457 {
1458 
1459 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1460 
1461 	if (mp->mnt_vfs_ops <= 0)
1462 		panic("%s: invalid vfs_ops count %d for mp %p\n",
1463 		    __func__, mp->mnt_vfs_ops, mp);
1464 	mp->mnt_vfs_ops--;
1465 }
1466 
1467 void
1468 vfs_op_exit(struct mount *mp)
1469 {
1470 
1471 	MNT_ILOCK(mp);
1472 	vfs_op_exit_locked(mp);
1473 	MNT_IUNLOCK(mp);
1474 }
1475 
1476 /*
1477  * It is assumed the caller already posted at least an acquire barrier.
1478  */
1479 void
1480 vfs_op_barrier_wait(struct mount *mp)
1481 {
1482 	int *in_op;
1483 	int cpu;
1484 
1485 	CPU_FOREACH(cpu) {
1486 		in_op = zpcpu_get_cpu(mp->mnt_thread_in_ops_pcpu, cpu);
1487 		while (atomic_load_int(in_op))
1488 			cpu_spinwait();
1489 	}
1490 }
1491 
1492 #ifdef DIAGNOSTIC
1493 void
1494 vfs_assert_mount_counters(struct mount *mp)
1495 {
1496 	int cpu;
1497 
1498 	if (mp->mnt_vfs_ops == 0)
1499 		return;
1500 
1501 	CPU_FOREACH(cpu) {
1502 		if (*(int *)zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu) != 0 ||
1503 		    *(int *)zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu) != 0 ||
1504 		    *(int *)zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu) != 0)
1505 			vfs_dump_mount_counters(mp);
1506 	}
1507 }
1508 
1509 void
1510 vfs_dump_mount_counters(struct mount *mp)
1511 {
1512 	int cpu, *count;
1513 	int ref, lockref, writeopcount;
1514 
1515 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1516 
1517 	printf("        ref : ");
1518 	ref = mp->mnt_ref;
1519 	CPU_FOREACH(cpu) {
1520 		count = zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu);
1521 		printf("%d ", *count);
1522 		ref += *count;
1523 	}
1524 	printf("\n");
1525 	printf("    lockref : ");
1526 	lockref = mp->mnt_lockref;
1527 	CPU_FOREACH(cpu) {
1528 		count = zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu);
1529 		printf("%d ", *count);
1530 		lockref += *count;
1531 	}
1532 	printf("\n");
1533 	printf("writeopcount: ");
1534 	writeopcount = mp->mnt_writeopcount;
1535 	CPU_FOREACH(cpu) {
1536 		count = zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu);
1537 		printf("%d ", *count);
1538 		writeopcount += *count;
1539 	}
1540 	printf("\n");
1541 
1542 	printf("counter       struct total\n");
1543 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1544 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1545 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1546 
1547 	panic("invalid counts on struct mount");
1548 }
1549 #endif
1550 
1551 int
1552 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1553 {
1554 	int *base, *pcpu;
1555 	int cpu, sum;
1556 
1557 	switch (which) {
1558 	case MNT_COUNT_REF:
1559 		base = &mp->mnt_ref;
1560 		pcpu = mp->mnt_ref_pcpu;
1561 		break;
1562 	case MNT_COUNT_LOCKREF:
1563 		base = &mp->mnt_lockref;
1564 		pcpu = mp->mnt_lockref_pcpu;
1565 		break;
1566 	case MNT_COUNT_WRITEOPCOUNT:
1567 		base = &mp->mnt_writeopcount;
1568 		pcpu = mp->mnt_writeopcount_pcpu;
1569 		break;
1570 	}
1571 
1572 	sum = *base;
1573 	CPU_FOREACH(cpu) {
1574 		sum += *(int *)zpcpu_get_cpu(pcpu, cpu);
1575 	}
1576 	return (sum);
1577 }
1578 
1579 /*
1580  * Do the actual filesystem unmount.
1581  */
1582 int
1583 dounmount(struct mount *mp, int flags, struct thread *td)
1584 {
1585 	struct vnode *coveredvp;
1586 	int error;
1587 	uint64_t async_flag;
1588 	int mnt_gen_r;
1589 
1590 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1591 		mnt_gen_r = mp->mnt_gen;
1592 		VI_LOCK(coveredvp);
1593 		vholdl(coveredvp);
1594 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1595 		/*
1596 		 * Check for mp being unmounted while waiting for the
1597 		 * covered vnode lock.
1598 		 */
1599 		if (coveredvp->v_mountedhere != mp ||
1600 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1601 			VOP_UNLOCK(coveredvp, 0);
1602 			vdrop(coveredvp);
1603 			vfs_rel(mp);
1604 			return (EBUSY);
1605 		}
1606 	}
1607 
1608 	/*
1609 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1610 	 * original mount is permitted to unmount this filesystem.
1611 	 */
1612 	error = vfs_suser(mp, td);
1613 	if (error != 0) {
1614 		if (coveredvp != NULL) {
1615 			VOP_UNLOCK(coveredvp, 0);
1616 			vdrop(coveredvp);
1617 		}
1618 		vfs_rel(mp);
1619 		return (error);
1620 	}
1621 
1622 	vfs_op_enter(mp);
1623 
1624 	vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1625 	MNT_ILOCK(mp);
1626 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1627 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
1628 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
1629 		dounmount_cleanup(mp, coveredvp, 0);
1630 		return (EBUSY);
1631 	}
1632 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1633 	if (flags & MNT_NONBUSY) {
1634 		MNT_IUNLOCK(mp);
1635 		error = vfs_check_usecounts(mp);
1636 		MNT_ILOCK(mp);
1637 		if (error != 0) {
1638 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1639 			return (error);
1640 		}
1641 	}
1642 	/* Allow filesystems to detect that a forced unmount is in progress. */
1643 	if (flags & MNT_FORCE) {
1644 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1645 		MNT_IUNLOCK(mp);
1646 		/*
1647 		 * Must be done after setting MNTK_UNMOUNTF and before
1648 		 * waiting for mnt_lockref to become 0.
1649 		 */
1650 		VFS_PURGE(mp);
1651 		MNT_ILOCK(mp);
1652 	}
1653 	error = 0;
1654 	if (mp->mnt_lockref) {
1655 		mp->mnt_kern_flag |= MNTK_DRAINING;
1656 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1657 		    "mount drain", 0);
1658 	}
1659 	MNT_IUNLOCK(mp);
1660 	KASSERT(mp->mnt_lockref == 0,
1661 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1662 	    __func__, __FILE__, __LINE__));
1663 	KASSERT(error == 0,
1664 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1665 	    __func__, __FILE__, __LINE__));
1666 
1667 	if (mp->mnt_flag & MNT_EXPUBLIC)
1668 		vfs_setpublicfs(NULL, NULL, NULL);
1669 
1670 	/*
1671 	 * From now, we can claim that the use reference on the
1672 	 * coveredvp is ours, and the ref can be released only by
1673 	 * successfull unmount by us, or left for later unmount
1674 	 * attempt.  The previously acquired hold reference is no
1675 	 * longer needed to protect the vnode from reuse.
1676 	 */
1677 	if (coveredvp != NULL)
1678 		vdrop(coveredvp);
1679 
1680 	vfs_msync(mp, MNT_WAIT);
1681 	MNT_ILOCK(mp);
1682 	async_flag = mp->mnt_flag & MNT_ASYNC;
1683 	mp->mnt_flag &= ~MNT_ASYNC;
1684 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1685 	MNT_IUNLOCK(mp);
1686 	cache_purgevfs(mp, false); /* remove cache entries for this file sys */
1687 	vfs_deallocate_syncvnode(mp);
1688 	if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 ||
1689 	    (error = VFS_SYNC(mp, MNT_WAIT)) == 0)
1690 		error = VFS_UNMOUNT(mp, flags);
1691 	vn_finished_write(mp);
1692 	/*
1693 	 * If we failed to flush the dirty blocks for this mount point,
1694 	 * undo all the cdir/rdir and rootvnode changes we made above.
1695 	 * Unless we failed to do so because the device is reporting that
1696 	 * it doesn't exist anymore.
1697 	 */
1698 	if (error && error != ENXIO) {
1699 		MNT_ILOCK(mp);
1700 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1701 			MNT_IUNLOCK(mp);
1702 			vfs_allocate_syncvnode(mp);
1703 			MNT_ILOCK(mp);
1704 		}
1705 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1706 		mp->mnt_flag |= async_flag;
1707 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1708 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1709 			mp->mnt_kern_flag |= MNTK_ASYNC;
1710 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1711 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1712 			wakeup(mp);
1713 		}
1714 		vfs_op_exit_locked(mp);
1715 		MNT_IUNLOCK(mp);
1716 		if (coveredvp)
1717 			VOP_UNLOCK(coveredvp, 0);
1718 		return (error);
1719 	}
1720 	mtx_lock(&mountlist_mtx);
1721 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1722 	mtx_unlock(&mountlist_mtx);
1723 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1724 	if (coveredvp != NULL) {
1725 		coveredvp->v_mountedhere = NULL;
1726 		VOP_UNLOCK(coveredvp, 0);
1727 	}
1728 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1729 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
1730 		vrele(rootvnode);
1731 		rootvnode = NULL;
1732 	}
1733 	if (mp == rootdevmp)
1734 		rootdevmp = NULL;
1735 	vfs_mount_destroy(mp);
1736 	return (0);
1737 }
1738 
1739 /*
1740  * Report errors during filesystem mounting.
1741  */
1742 void
1743 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1744 {
1745 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1746 	va_list ap;
1747 	int error, len;
1748 	char *errmsg;
1749 
1750 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1751 	if (error || errmsg == NULL || len <= 0)
1752 		return;
1753 
1754 	va_start(ap, fmt);
1755 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1756 	va_end(ap);
1757 }
1758 
1759 void
1760 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1761 {
1762 	va_list ap;
1763 	int error, len;
1764 	char *errmsg;
1765 
1766 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1767 	if (error || errmsg == NULL || len <= 0)
1768 		return;
1769 
1770 	va_start(ap, fmt);
1771 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1772 	va_end(ap);
1773 }
1774 
1775 /*
1776  * ---------------------------------------------------------------------
1777  * Functions for querying mount options/arguments from filesystems.
1778  */
1779 
1780 /*
1781  * Check that no unknown options are given
1782  */
1783 int
1784 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1785 {
1786 	struct vfsopt *opt;
1787 	char errmsg[255];
1788 	const char **t, *p, *q;
1789 	int ret = 0;
1790 
1791 	TAILQ_FOREACH(opt, opts, link) {
1792 		p = opt->name;
1793 		q = NULL;
1794 		if (p[0] == 'n' && p[1] == 'o')
1795 			q = p + 2;
1796 		for(t = global_opts; *t != NULL; t++) {
1797 			if (strcmp(*t, p) == 0)
1798 				break;
1799 			if (q != NULL) {
1800 				if (strcmp(*t, q) == 0)
1801 					break;
1802 			}
1803 		}
1804 		if (*t != NULL)
1805 			continue;
1806 		for(t = legal; *t != NULL; t++) {
1807 			if (strcmp(*t, p) == 0)
1808 				break;
1809 			if (q != NULL) {
1810 				if (strcmp(*t, q) == 0)
1811 					break;
1812 			}
1813 		}
1814 		if (*t != NULL)
1815 			continue;
1816 		snprintf(errmsg, sizeof(errmsg),
1817 		    "mount option <%s> is unknown", p);
1818 		ret = EINVAL;
1819 	}
1820 	if (ret != 0) {
1821 		TAILQ_FOREACH(opt, opts, link) {
1822 			if (strcmp(opt->name, "errmsg") == 0) {
1823 				strncpy((char *)opt->value, errmsg, opt->len);
1824 				break;
1825 			}
1826 		}
1827 		if (opt == NULL)
1828 			printf("%s\n", errmsg);
1829 	}
1830 	return (ret);
1831 }
1832 
1833 /*
1834  * Get a mount option by its name.
1835  *
1836  * Return 0 if the option was found, ENOENT otherwise.
1837  * If len is non-NULL it will be filled with the length
1838  * of the option. If buf is non-NULL, it will be filled
1839  * with the address of the option.
1840  */
1841 int
1842 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
1843 {
1844 	struct vfsopt *opt;
1845 
1846 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1847 
1848 	TAILQ_FOREACH(opt, opts, link) {
1849 		if (strcmp(name, opt->name) == 0) {
1850 			opt->seen = 1;
1851 			if (len != NULL)
1852 				*len = opt->len;
1853 			if (buf != NULL)
1854 				*buf = opt->value;
1855 			return (0);
1856 		}
1857 	}
1858 	return (ENOENT);
1859 }
1860 
1861 int
1862 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1863 {
1864 	struct vfsopt *opt;
1865 
1866 	if (opts == NULL)
1867 		return (-1);
1868 
1869 	TAILQ_FOREACH(opt, opts, link) {
1870 		if (strcmp(name, opt->name) == 0) {
1871 			opt->seen = 1;
1872 			return (opt->pos);
1873 		}
1874 	}
1875 	return (-1);
1876 }
1877 
1878 int
1879 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1880 {
1881 	char *opt_value, *vtp;
1882 	quad_t iv;
1883 	int error, opt_len;
1884 
1885 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1886 	if (error != 0)
1887 		return (error);
1888 	if (opt_len == 0 || opt_value == NULL)
1889 		return (EINVAL);
1890 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1891 		return (EINVAL);
1892 	iv = strtoq(opt_value, &vtp, 0);
1893 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1894 		return (EINVAL);
1895 	if (iv < 0)
1896 		return (EINVAL);
1897 	switch (vtp[0]) {
1898 	case 't': case 'T':
1899 		iv *= 1024;
1900 		/* FALLTHROUGH */
1901 	case 'g': case 'G':
1902 		iv *= 1024;
1903 		/* FALLTHROUGH */
1904 	case 'm': case 'M':
1905 		iv *= 1024;
1906 		/* FALLTHROUGH */
1907 	case 'k': case 'K':
1908 		iv *= 1024;
1909 	case '\0':
1910 		break;
1911 	default:
1912 		return (EINVAL);
1913 	}
1914 	*value = iv;
1915 
1916 	return (0);
1917 }
1918 
1919 char *
1920 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1921 {
1922 	struct vfsopt *opt;
1923 
1924 	*error = 0;
1925 	TAILQ_FOREACH(opt, opts, link) {
1926 		if (strcmp(name, opt->name) != 0)
1927 			continue;
1928 		opt->seen = 1;
1929 		if (opt->len == 0 ||
1930 		    ((char *)opt->value)[opt->len - 1] != '\0') {
1931 			*error = EINVAL;
1932 			return (NULL);
1933 		}
1934 		return (opt->value);
1935 	}
1936 	*error = ENOENT;
1937 	return (NULL);
1938 }
1939 
1940 int
1941 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1942 	uint64_t val)
1943 {
1944 	struct vfsopt *opt;
1945 
1946 	TAILQ_FOREACH(opt, opts, link) {
1947 		if (strcmp(name, opt->name) == 0) {
1948 			opt->seen = 1;
1949 			if (w != NULL)
1950 				*w |= val;
1951 			return (1);
1952 		}
1953 	}
1954 	if (w != NULL)
1955 		*w &= ~val;
1956 	return (0);
1957 }
1958 
1959 int
1960 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1961 {
1962 	va_list ap;
1963 	struct vfsopt *opt;
1964 	int ret;
1965 
1966 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1967 
1968 	TAILQ_FOREACH(opt, opts, link) {
1969 		if (strcmp(name, opt->name) != 0)
1970 			continue;
1971 		opt->seen = 1;
1972 		if (opt->len == 0 || opt->value == NULL)
1973 			return (0);
1974 		if (((char *)opt->value)[opt->len - 1] != '\0')
1975 			return (0);
1976 		va_start(ap, fmt);
1977 		ret = vsscanf(opt->value, fmt, ap);
1978 		va_end(ap);
1979 		return (ret);
1980 	}
1981 	return (0);
1982 }
1983 
1984 int
1985 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
1986 {
1987 	struct vfsopt *opt;
1988 
1989 	TAILQ_FOREACH(opt, opts, link) {
1990 		if (strcmp(name, opt->name) != 0)
1991 			continue;
1992 		opt->seen = 1;
1993 		if (opt->value == NULL)
1994 			opt->len = len;
1995 		else {
1996 			if (opt->len != len)
1997 				return (EINVAL);
1998 			bcopy(value, opt->value, len);
1999 		}
2000 		return (0);
2001 	}
2002 	return (ENOENT);
2003 }
2004 
2005 int
2006 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2007 {
2008 	struct vfsopt *opt;
2009 
2010 	TAILQ_FOREACH(opt, opts, link) {
2011 		if (strcmp(name, opt->name) != 0)
2012 			continue;
2013 		opt->seen = 1;
2014 		if (opt->value == NULL)
2015 			opt->len = len;
2016 		else {
2017 			if (opt->len < len)
2018 				return (EINVAL);
2019 			opt->len = len;
2020 			bcopy(value, opt->value, len);
2021 		}
2022 		return (0);
2023 	}
2024 	return (ENOENT);
2025 }
2026 
2027 int
2028 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2029 {
2030 	struct vfsopt *opt;
2031 
2032 	TAILQ_FOREACH(opt, opts, link) {
2033 		if (strcmp(name, opt->name) != 0)
2034 			continue;
2035 		opt->seen = 1;
2036 		if (opt->value == NULL)
2037 			opt->len = strlen(value) + 1;
2038 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2039 			return (EINVAL);
2040 		return (0);
2041 	}
2042 	return (ENOENT);
2043 }
2044 
2045 /*
2046  * Find and copy a mount option.
2047  *
2048  * The size of the buffer has to be specified
2049  * in len, if it is not the same length as the
2050  * mount option, EINVAL is returned.
2051  * Returns ENOENT if the option is not found.
2052  */
2053 int
2054 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2055 {
2056 	struct vfsopt *opt;
2057 
2058 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2059 
2060 	TAILQ_FOREACH(opt, opts, link) {
2061 		if (strcmp(name, opt->name) == 0) {
2062 			opt->seen = 1;
2063 			if (len != opt->len)
2064 				return (EINVAL);
2065 			bcopy(opt->value, dest, opt->len);
2066 			return (0);
2067 		}
2068 	}
2069 	return (ENOENT);
2070 }
2071 
2072 int
2073 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2074 {
2075 
2076 	/*
2077 	 * Filesystems only fill in part of the structure for updates, we
2078 	 * have to read the entirety first to get all content.
2079 	 */
2080 	memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2081 
2082 	/*
2083 	 * Set these in case the underlying filesystem fails to do so.
2084 	 */
2085 	sbp->f_version = STATFS_VERSION;
2086 	sbp->f_namemax = NAME_MAX;
2087 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2088 
2089 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2090 }
2091 
2092 void
2093 vfs_mountedfrom(struct mount *mp, const char *from)
2094 {
2095 
2096 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2097 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2098 	    sizeof mp->mnt_stat.f_mntfromname);
2099 }
2100 
2101 /*
2102  * ---------------------------------------------------------------------
2103  * This is the api for building mount args and mounting filesystems from
2104  * inside the kernel.
2105  *
2106  * The API works by accumulation of individual args.  First error is
2107  * latched.
2108  *
2109  * XXX: should be documented in new manpage kernel_mount(9)
2110  */
2111 
2112 /* A memory allocation which must be freed when we are done */
2113 struct mntaarg {
2114 	SLIST_ENTRY(mntaarg)	next;
2115 };
2116 
2117 /* The header for the mount arguments */
2118 struct mntarg {
2119 	struct iovec *v;
2120 	int len;
2121 	int error;
2122 	SLIST_HEAD(, mntaarg)	list;
2123 };
2124 
2125 /*
2126  * Add a boolean argument.
2127  *
2128  * flag is the boolean value.
2129  * name must start with "no".
2130  */
2131 struct mntarg *
2132 mount_argb(struct mntarg *ma, int flag, const char *name)
2133 {
2134 
2135 	KASSERT(name[0] == 'n' && name[1] == 'o',
2136 	    ("mount_argb(...,%s): name must start with 'no'", name));
2137 
2138 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2139 }
2140 
2141 /*
2142  * Add an argument printf style
2143  */
2144 struct mntarg *
2145 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2146 {
2147 	va_list ap;
2148 	struct mntaarg *maa;
2149 	struct sbuf *sb;
2150 	int len;
2151 
2152 	if (ma == NULL) {
2153 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2154 		SLIST_INIT(&ma->list);
2155 	}
2156 	if (ma->error)
2157 		return (ma);
2158 
2159 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2160 	    M_MOUNT, M_WAITOK);
2161 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2162 	ma->v[ma->len].iov_len = strlen(name) + 1;
2163 	ma->len++;
2164 
2165 	sb = sbuf_new_auto();
2166 	va_start(ap, fmt);
2167 	sbuf_vprintf(sb, fmt, ap);
2168 	va_end(ap);
2169 	sbuf_finish(sb);
2170 	len = sbuf_len(sb) + 1;
2171 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2172 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2173 	bcopy(sbuf_data(sb), maa + 1, len);
2174 	sbuf_delete(sb);
2175 
2176 	ma->v[ma->len].iov_base = maa + 1;
2177 	ma->v[ma->len].iov_len = len;
2178 	ma->len++;
2179 
2180 	return (ma);
2181 }
2182 
2183 /*
2184  * Add an argument which is a userland string.
2185  */
2186 struct mntarg *
2187 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2188 {
2189 	struct mntaarg *maa;
2190 	char *tbuf;
2191 
2192 	if (val == NULL)
2193 		return (ma);
2194 	if (ma == NULL) {
2195 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2196 		SLIST_INIT(&ma->list);
2197 	}
2198 	if (ma->error)
2199 		return (ma);
2200 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2201 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2202 	tbuf = (void *)(maa + 1);
2203 	ma->error = copyinstr(val, tbuf, len, NULL);
2204 	return (mount_arg(ma, name, tbuf, -1));
2205 }
2206 
2207 /*
2208  * Plain argument.
2209  *
2210  * If length is -1, treat value as a C string.
2211  */
2212 struct mntarg *
2213 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2214 {
2215 
2216 	if (ma == NULL) {
2217 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2218 		SLIST_INIT(&ma->list);
2219 	}
2220 	if (ma->error)
2221 		return (ma);
2222 
2223 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2224 	    M_MOUNT, M_WAITOK);
2225 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2226 	ma->v[ma->len].iov_len = strlen(name) + 1;
2227 	ma->len++;
2228 
2229 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2230 	if (len < 0)
2231 		ma->v[ma->len].iov_len = strlen(val) + 1;
2232 	else
2233 		ma->v[ma->len].iov_len = len;
2234 	ma->len++;
2235 	return (ma);
2236 }
2237 
2238 /*
2239  * Free a mntarg structure
2240  */
2241 static void
2242 free_mntarg(struct mntarg *ma)
2243 {
2244 	struct mntaarg *maa;
2245 
2246 	while (!SLIST_EMPTY(&ma->list)) {
2247 		maa = SLIST_FIRST(&ma->list);
2248 		SLIST_REMOVE_HEAD(&ma->list, next);
2249 		free(maa, M_MOUNT);
2250 	}
2251 	free(ma->v, M_MOUNT);
2252 	free(ma, M_MOUNT);
2253 }
2254 
2255 /*
2256  * Mount a filesystem
2257  */
2258 int
2259 kernel_mount(struct mntarg *ma, uint64_t flags)
2260 {
2261 	struct uio auio;
2262 	int error;
2263 
2264 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2265 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2266 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2267 
2268 	auio.uio_iov = ma->v;
2269 	auio.uio_iovcnt = ma->len;
2270 	auio.uio_segflg = UIO_SYSSPACE;
2271 
2272 	error = ma->error;
2273 	if (!error)
2274 		error = vfs_donmount(curthread, flags, &auio);
2275 	free_mntarg(ma);
2276 	return (error);
2277 }
2278 
2279 /*
2280  * A printflike function to mount a filesystem.
2281  */
2282 int
2283 kernel_vmount(int flags, ...)
2284 {
2285 	struct mntarg *ma = NULL;
2286 	va_list ap;
2287 	const char *cp;
2288 	const void *vp;
2289 	int error;
2290 
2291 	va_start(ap, flags);
2292 	for (;;) {
2293 		cp = va_arg(ap, const char *);
2294 		if (cp == NULL)
2295 			break;
2296 		vp = va_arg(ap, const void *);
2297 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2298 	}
2299 	va_end(ap);
2300 
2301 	error = kernel_mount(ma, flags);
2302 	return (error);
2303 }
2304 
2305 void
2306 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
2307 {
2308 
2309 	bcopy(oexp, exp, sizeof(*oexp));
2310 	exp->ex_numsecflavors = 0;
2311 }
2312