xref: /freebsd/sys/kern/vfs_mount.c (revision db612abe8df3355d1eb23bb3b50fdd97bc21e979)
1 /*-
2  * Copyright (c) 1999-2004 Poul-Henning Kamp
3  * Copyright (c) 1999 Michael Smith
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/conf.h>
42 #include <sys/clock.h>
43 #include <sys/fcntl.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/libkern.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mutex.h>
50 #include <sys/namei.h>
51 #include <sys/priv.h>
52 #include <sys/proc.h>
53 #include <sys/filedesc.h>
54 #include <sys/reboot.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysproto.h>
57 #include <sys/sx.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/systm.h>
61 #include <sys/vnode.h>
62 #include <vm/uma.h>
63 
64 #include <geom/geom.h>
65 
66 #include <machine/stdarg.h>
67 
68 #include <security/audit/audit.h>
69 #include <security/mac/mac_framework.h>
70 
71 #include "opt_rootdevname.h"
72 #include "opt_ddb.h"
73 #include "opt_mac.h"
74 
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #endif
78 
79 #define	ROOTNAME		"root_device"
80 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
81 
82 static int	vfs_domount(struct thread *td, const char *fstype,
83 		    char *fspath, int fsflags, void *fsdata);
84 static int	vfs_mountroot_ask(void);
85 static int	vfs_mountroot_try(const char *mountfrom);
86 static int	vfs_donmount(struct thread *td, int fsflags,
87 		    struct uio *fsoptions);
88 static void	free_mntarg(struct mntarg *ma);
89 static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
90 
91 static int	usermount = 0;
92 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
93     "Unprivileged users may mount and unmount file systems");
94 
95 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
96 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
97 static uma_zone_t mount_zone;
98 
99 /* List of mounted filesystems. */
100 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
101 
102 /* For any iteration/modification of mountlist */
103 struct mtx mountlist_mtx;
104 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
105 
106 TAILQ_HEAD(vfsoptlist, vfsopt);
107 struct vfsopt {
108 	TAILQ_ENTRY(vfsopt) link;
109 	char	*name;
110 	void	*value;
111 	int	len;
112 };
113 
114 /*
115  * The vnode of the system's root (/ in the filesystem, without chroot
116  * active.)
117  */
118 struct vnode	*rootvnode;
119 
120 /*
121  * The root filesystem is detailed in the kernel environment variable
122  * vfs.root.mountfrom, which is expected to be in the general format
123  *
124  * <vfsname>:[<path>]
125  * vfsname   := the name of a VFS known to the kernel and capable
126  *              of being mounted as root
127  * path      := disk device name or other data used by the filesystem
128  *              to locate its physical store
129  */
130 
131 /*
132  * Global opts, taken by all filesystems
133  */
134 static const char *global_opts[] = {
135 	"errmsg",
136 	"fstype",
137 	"fspath",
138 	"ro",
139 	"rw",
140 	"nosuid",
141 	"noexec",
142 	"update",
143 	NULL
144 };
145 
146 /*
147  * The root specifiers we will try if RB_CDROM is specified.
148  */
149 static char *cdrom_rootdevnames[] = {
150 	"cd9660:cd0",
151 	"cd9660:acd0",
152 	NULL
153 };
154 
155 /* legacy find-root code */
156 char		*rootdevnames[2] = {NULL, NULL};
157 #ifndef ROOTDEVNAME
158 #  define ROOTDEVNAME NULL
159 #endif
160 static const char	*ctrootdevname = ROOTDEVNAME;
161 
162 /*
163  * ---------------------------------------------------------------------
164  * Functions for building and sanitizing the mount options
165  */
166 
167 /* Remove one mount option. */
168 static void
169 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
170 {
171 
172 	TAILQ_REMOVE(opts, opt, link);
173 	free(opt->name, M_MOUNT);
174 	if (opt->value != NULL)
175 		free(opt->value, M_MOUNT);
176 #ifdef INVARIANTS
177 	else if (opt->len != 0)
178 		panic("%s: mount option with NULL value but length != 0",
179 		    __func__);
180 #endif
181 	free(opt, M_MOUNT);
182 }
183 
184 /* Release all resources related to the mount options. */
185 void
186 vfs_freeopts(struct vfsoptlist *opts)
187 {
188 	struct vfsopt *opt;
189 
190 	while (!TAILQ_EMPTY(opts)) {
191 		opt = TAILQ_FIRST(opts);
192 		vfs_freeopt(opts, opt);
193 	}
194 	free(opts, M_MOUNT);
195 }
196 
197 void
198 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
199 {
200 	struct vfsopt *opt, *temp;
201 
202 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
203 		if (strcmp(opt->name, name) == 0)
204 			vfs_freeopt(opts, opt);
205 	}
206 }
207 
208 /*
209  * Check if options are equal (with or without the "no" prefix).
210  */
211 static int
212 vfs_equalopts(const char *opt1, const char *opt2)
213 {
214 
215 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
216 	if (strcmp(opt1, opt2) == 0)
217 		return (1);
218 	/* "noopt" vs. "opt" */
219 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
220 		return (1);
221 	/* "opt" vs. "noopt" */
222 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
223 		return (1);
224 	return (0);
225 }
226 
227 /*
228  * If a mount option is specified several times,
229  * (with or without the "no" prefix) only keep
230  * the last occurence of it.
231  */
232 static void
233 vfs_sanitizeopts(struct vfsoptlist *opts)
234 {
235 	struct vfsopt *opt, *opt2, *tmp;
236 
237 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
238 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
239 		while (opt2 != NULL) {
240 			if (vfs_equalopts(opt->name, opt2->name)) {
241 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
242 				vfs_freeopt(opts, opt2);
243 				opt2 = tmp;
244 			} else {
245 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
246 			}
247 		}
248 	}
249 }
250 
251 /*
252  * Build a linked list of mount options from a struct uio.
253  */
254 static int
255 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
256 {
257 	struct vfsoptlist *opts;
258 	struct vfsopt *opt;
259 	size_t memused;
260 	unsigned int i, iovcnt;
261 	int error, namelen, optlen;
262 
263 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
264 	TAILQ_INIT(opts);
265 	memused = 0;
266 	iovcnt = auio->uio_iovcnt;
267 	for (i = 0; i < iovcnt; i += 2) {
268 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
269 		namelen = auio->uio_iov[i].iov_len;
270 		optlen = auio->uio_iov[i + 1].iov_len;
271 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
272 		opt->value = NULL;
273 		opt->len = 0;
274 
275 		/*
276 		 * Do this early, so jumps to "bad" will free the current
277 		 * option.
278 		 */
279 		TAILQ_INSERT_TAIL(opts, opt, link);
280 		memused += sizeof(struct vfsopt) + optlen + namelen;
281 
282 		/*
283 		 * Avoid consuming too much memory, and attempts to overflow
284 		 * memused.
285 		 */
286 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
287 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
288 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
289 			error = EINVAL;
290 			goto bad;
291 		}
292 
293 		if (auio->uio_segflg == UIO_SYSSPACE) {
294 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
295 		} else {
296 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
297 			    namelen);
298 			if (error)
299 				goto bad;
300 		}
301 		/* Ensure names are null-terminated strings. */
302 		if (opt->name[namelen - 1] != '\0') {
303 			error = EINVAL;
304 			goto bad;
305 		}
306 		if (optlen != 0) {
307 			opt->len = optlen;
308 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
309 			if (auio->uio_segflg == UIO_SYSSPACE) {
310 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
311 				    optlen);
312 			} else {
313 				error = copyin(auio->uio_iov[i + 1].iov_base,
314 				    opt->value, optlen);
315 				if (error)
316 					goto bad;
317 			}
318 		}
319 	}
320 	vfs_sanitizeopts(opts);
321 	*options = opts;
322 	return (0);
323 bad:
324 	vfs_freeopts(opts);
325 	return (error);
326 }
327 
328 /*
329  * Merge the old mount options with the new ones passed
330  * in the MNT_UPDATE case.
331  *
332  * XXX This function will keep a "nofoo" option in the
333  *     new options if there is no matching "foo" option
334  *     to be cancelled in the old options.  This is a bug
335  *     if the option's canonical name is "foo".  E.g., "noro"
336  *     shouldn't end up in the mount point's active options,
337  *     but it can.
338  */
339 static void
340 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
341 {
342 	struct vfsopt *opt, *opt2, *new;
343 
344 	TAILQ_FOREACH(opt, opts, link) {
345 		/*
346 		 * Check that this option hasn't been redefined
347 		 * nor cancelled with a "no" mount option.
348 		 */
349 		opt2 = TAILQ_FIRST(toopts);
350 		while (opt2 != NULL) {
351 			if (strcmp(opt2->name, opt->name) == 0)
352 				goto next;
353 			if (strncmp(opt2->name, "no", 2) == 0 &&
354 			    strcmp(opt2->name + 2, opt->name) == 0) {
355 				vfs_freeopt(toopts, opt2);
356 				goto next;
357 			}
358 			opt2 = TAILQ_NEXT(opt2, link);
359 		}
360 		/* We want this option, duplicate it. */
361 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
362 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
363 		strcpy(new->name, opt->name);
364 		if (opt->len != 0) {
365 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
366 			bcopy(opt->value, new->value, opt->len);
367 		} else {
368 			new->value = NULL;
369 		}
370 		new->len = opt->len;
371 		TAILQ_INSERT_TAIL(toopts, new, link);
372 next:
373 		continue;
374 	}
375 }
376 
377 /*
378  * Mount a filesystem.
379  */
380 int
381 nmount(td, uap)
382 	struct thread *td;
383 	struct nmount_args /* {
384 		struct iovec *iovp;
385 		unsigned int iovcnt;
386 		int flags;
387 	} */ *uap;
388 {
389 	struct uio *auio;
390 	struct iovec *iov;
391 	unsigned int i;
392 	int error;
393 	u_int iovcnt;
394 
395 	AUDIT_ARG(fflags, uap->flags);
396 
397 	/*
398 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
399 	 * userspace to set this flag, but we must filter it out if we want
400 	 * MNT_UPDATE on the root file system to work.
401 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
402 	 */
403 	uap->flags &= ~MNT_ROOTFS;
404 
405 	iovcnt = uap->iovcnt;
406 	/*
407 	 * Check that we have an even number of iovec's
408 	 * and that we have at least two options.
409 	 */
410 	if ((iovcnt & 1) || (iovcnt < 4))
411 		return (EINVAL);
412 
413 	error = copyinuio(uap->iovp, iovcnt, &auio);
414 	if (error)
415 		return (error);
416 	iov = auio->uio_iov;
417 	for (i = 0; i < iovcnt; i++) {
418 		if (iov->iov_len > MMAXOPTIONLEN) {
419 			free(auio, M_IOV);
420 			return (EINVAL);
421 		}
422 		iov++;
423 	}
424 	error = vfs_donmount(td, uap->flags, auio);
425 
426 	free(auio, M_IOV);
427 	return (error);
428 }
429 
430 /*
431  * ---------------------------------------------------------------------
432  * Various utility functions
433  */
434 
435 void
436 vfs_ref(struct mount *mp)
437 {
438 
439 	MNT_ILOCK(mp);
440 	MNT_REF(mp);
441 	MNT_IUNLOCK(mp);
442 }
443 
444 void
445 vfs_rel(struct mount *mp)
446 {
447 
448 	MNT_ILOCK(mp);
449 	MNT_REL(mp);
450 	MNT_IUNLOCK(mp);
451 }
452 
453 static int
454 mount_init(void *mem, int size, int flags)
455 {
456 	struct mount *mp;
457 
458 	mp = (struct mount *)mem;
459 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
460 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
461 	return (0);
462 }
463 
464 static void
465 mount_fini(void *mem, int size)
466 {
467 	struct mount *mp;
468 
469 	mp = (struct mount *)mem;
470 	lockdestroy(&mp->mnt_lock);
471 	mtx_destroy(&mp->mnt_mtx);
472 }
473 
474 /*
475  * Allocate and initialize the mount point struct.
476  */
477 struct mount *
478 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
479     const char *fspath, struct thread *td)
480 {
481 	struct mount *mp;
482 
483 	mp = uma_zalloc(mount_zone, M_WAITOK);
484 	bzero(&mp->mnt_startzero,
485 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
486 	TAILQ_INIT(&mp->mnt_nvnodelist);
487 	mp->mnt_nvnodelistsize = 0;
488 	mp->mnt_ref = 0;
489 	(void) vfs_busy(mp, LK_NOWAIT, 0, td);
490 	mp->mnt_op = vfsp->vfc_vfsops;
491 	mp->mnt_vfc = vfsp;
492 	vfsp->vfc_refcount++;	/* XXX Unlocked */
493 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
494 	mp->mnt_gen++;
495 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
496 	mp->mnt_vnodecovered = vp;
497 	mp->mnt_cred = crdup(td->td_ucred);
498 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
499 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
500 	mp->mnt_iosize_max = DFLTPHYS;
501 #ifdef MAC
502 	mac_mount_init(mp);
503 	mac_mount_create(td->td_ucred, mp);
504 #endif
505 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
506 	return (mp);
507 }
508 
509 /*
510  * Destroy the mount struct previously allocated by vfs_mount_alloc().
511  */
512 void
513 vfs_mount_destroy(struct mount *mp)
514 {
515 	int i;
516 
517 	MNT_ILOCK(mp);
518 	for (i = 0; mp->mnt_ref && i < 3; i++)
519 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
520 	/*
521 	 * This will always cause a 3 second delay in rebooting due to
522 	 * refs on the root mountpoint that never go away.  Most of these
523 	 * are held by init which never exits.
524 	 */
525 	if (i == 3 && (!rebooting || bootverbose))
526 		printf("Mount point %s had %d dangling refs\n",
527 		    mp->mnt_stat.f_mntonname, mp->mnt_ref);
528 	if (mp->mnt_holdcnt != 0) {
529 		printf("Waiting for mount point to be unheld\n");
530 		while (mp->mnt_holdcnt != 0) {
531 			mp->mnt_holdcntwaiters++;
532 			msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
533 			       PZERO, "mntdestroy", 0);
534 			mp->mnt_holdcntwaiters--;
535 		}
536 		printf("mount point unheld\n");
537 	}
538 	if (mp->mnt_writeopcount > 0) {
539 		printf("Waiting for mount point write ops\n");
540 		while (mp->mnt_writeopcount > 0) {
541 			mp->mnt_kern_flag |= MNTK_SUSPEND;
542 			msleep(&mp->mnt_writeopcount,
543 			       MNT_MTX(mp),
544 			       PZERO, "mntdestroy2", 0);
545 		}
546 		printf("mount point write ops completed\n");
547 	}
548 	if (mp->mnt_secondary_writes > 0) {
549 		printf("Waiting for mount point secondary write ops\n");
550 		while (mp->mnt_secondary_writes > 0) {
551 			mp->mnt_kern_flag |= MNTK_SUSPEND;
552 			msleep(&mp->mnt_secondary_writes,
553 			       MNT_MTX(mp),
554 			       PZERO, "mntdestroy3", 0);
555 		}
556 		printf("mount point secondary write ops completed\n");
557 	}
558 	MNT_IUNLOCK(mp);
559 	mp->mnt_vfc->vfc_refcount--;
560 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
561 		struct vnode *vp;
562 
563 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
564 			vprint("", vp);
565 		panic("unmount: dangling vnode");
566 	}
567 	MNT_ILOCK(mp);
568 	if (mp->mnt_kern_flag & MNTK_MWAIT)
569 		wakeup(mp);
570 	if (mp->mnt_writeopcount != 0)
571 		panic("vfs_mount_destroy: nonzero writeopcount");
572 	if (mp->mnt_secondary_writes != 0)
573 		panic("vfs_mount_destroy: nonzero secondary_writes");
574 	if (mp->mnt_nvnodelistsize != 0)
575 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
576 	mp->mnt_writeopcount = -1000;
577 	mp->mnt_nvnodelistsize = -1000;
578 	mp->mnt_secondary_writes = -1000;
579 	MNT_IUNLOCK(mp);
580 #ifdef MAC
581 	mac_mount_destroy(mp);
582 #endif
583 	if (mp->mnt_opt != NULL)
584 		vfs_freeopts(mp->mnt_opt);
585 	crfree(mp->mnt_cred);
586 	uma_zfree(mount_zone, mp);
587 }
588 
589 static int
590 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
591 {
592 	struct vfsoptlist *optlist;
593 	struct vfsopt *opt, *noro_opt;
594 	char *fstype, *fspath, *errmsg;
595 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
596 	int has_rw, has_noro;
597 
598 	errmsg = NULL;
599 	errmsg_len = 0;
600 	errmsg_pos = -1;
601 	has_rw = 0;
602 	has_noro = 0;
603 
604 	error = vfs_buildopts(fsoptions, &optlist);
605 	if (error)
606 		return (error);
607 
608 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
609 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
610 
611 	/*
612 	 * We need these two options before the others,
613 	 * and they are mandatory for any filesystem.
614 	 * Ensure they are NUL terminated as well.
615 	 */
616 	fstypelen = 0;
617 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
618 	if (error || fstype[fstypelen - 1] != '\0') {
619 		error = EINVAL;
620 		if (errmsg != NULL)
621 			strncpy(errmsg, "Invalid fstype", errmsg_len);
622 		goto bail;
623 	}
624 	fspathlen = 0;
625 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
626 	if (error || fspath[fspathlen - 1] != '\0') {
627 		error = EINVAL;
628 		if (errmsg != NULL)
629 			strncpy(errmsg, "Invalid fspath", errmsg_len);
630 		goto bail;
631 	}
632 
633 	/*
634 	 * We need to see if we have the "update" option
635 	 * before we call vfs_domount(), since vfs_domount() has special
636 	 * logic based on MNT_UPDATE.  This is very important
637 	 * when we want to update the root filesystem.
638 	 */
639 	TAILQ_FOREACH(opt, optlist, link) {
640 		if (strcmp(opt->name, "update") == 0)
641 			fsflags |= MNT_UPDATE;
642 		else if (strcmp(opt->name, "async") == 0)
643 			fsflags |= MNT_ASYNC;
644 		else if (strcmp(opt->name, "force") == 0)
645 			fsflags |= MNT_FORCE;
646 		else if (strcmp(opt->name, "multilabel") == 0)
647 			fsflags |= MNT_MULTILABEL;
648 		else if (strcmp(opt->name, "noasync") == 0)
649 			fsflags &= ~MNT_ASYNC;
650 		else if (strcmp(opt->name, "noatime") == 0)
651 			fsflags |= MNT_NOATIME;
652 		else if (strcmp(opt->name, "atime") == 0) {
653 			free(opt->name, M_MOUNT);
654 			opt->name = strdup("nonoatime", M_MOUNT);
655 		}
656 		else if (strcmp(opt->name, "noclusterr") == 0)
657 			fsflags |= MNT_NOCLUSTERR;
658 		else if (strcmp(opt->name, "clusterr") == 0) {
659 			free(opt->name, M_MOUNT);
660 			opt->name = strdup("nonoclusterr", M_MOUNT);
661 		}
662 		else if (strcmp(opt->name, "noclusterw") == 0)
663 			fsflags |= MNT_NOCLUSTERW;
664 		else if (strcmp(opt->name, "clusterw") == 0) {
665 			free(opt->name, M_MOUNT);
666 			opt->name = strdup("nonoclusterw", M_MOUNT);
667 		}
668 		else if (strcmp(opt->name, "noexec") == 0)
669 			fsflags |= MNT_NOEXEC;
670 		else if (strcmp(opt->name, "exec") == 0) {
671 			free(opt->name, M_MOUNT);
672 			opt->name = strdup("nonoexec", M_MOUNT);
673 		}
674 		else if (strcmp(opt->name, "nosuid") == 0)
675 			fsflags |= MNT_NOSUID;
676 		else if (strcmp(opt->name, "suid") == 0) {
677 			free(opt->name, M_MOUNT);
678 			opt->name = strdup("nonosuid", M_MOUNT);
679 		}
680 		else if (strcmp(opt->name, "nosymfollow") == 0)
681 			fsflags |= MNT_NOSYMFOLLOW;
682 		else if (strcmp(opt->name, "symfollow") == 0) {
683 			free(opt->name, M_MOUNT);
684 			opt->name = strdup("nonosymfollow", M_MOUNT);
685 		}
686 		else if (strcmp(opt->name, "noro") == 0) {
687 			fsflags &= ~MNT_RDONLY;
688 			has_noro = 1;
689 		}
690 		else if (strcmp(opt->name, "rw") == 0) {
691 			fsflags &= ~MNT_RDONLY;
692 			has_rw = 1;
693 		}
694 		else if (strcmp(opt->name, "ro") == 0)
695 			fsflags |= MNT_RDONLY;
696 		else if (strcmp(opt->name, "rdonly") == 0) {
697 			free(opt->name, M_MOUNT);
698 			opt->name = strdup("ro", M_MOUNT);
699 			fsflags |= MNT_RDONLY;
700 		}
701 		else if (strcmp(opt->name, "snapshot") == 0)
702 			fsflags |= MNT_SNAPSHOT;
703 		else if (strcmp(opt->name, "suiddir") == 0)
704 			fsflags |= MNT_SUIDDIR;
705 		else if (strcmp(opt->name, "sync") == 0)
706 			fsflags |= MNT_SYNCHRONOUS;
707 		else if (strcmp(opt->name, "union") == 0)
708 			fsflags |= MNT_UNION;
709 	}
710 
711 	/*
712 	 * If "rw" was specified as a mount option, and we
713 	 * are trying to update a mount-point from "ro" to "rw",
714 	 * we need a mount option "noro", since in vfs_mergeopts(),
715 	 * "noro" will cancel "ro", but "rw" will not do anything.
716 	 */
717 	if (has_rw && !has_noro) {
718 		noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
719 		noro_opt->name = strdup("noro", M_MOUNT);
720 		noro_opt->value = NULL;
721 		noro_opt->len = 0;
722 		TAILQ_INSERT_TAIL(optlist, noro_opt, link);
723 	}
724 
725 	/*
726 	 * Be ultra-paranoid about making sure the type and fspath
727 	 * variables will fit in our mp buffers, including the
728 	 * terminating NUL.
729 	 */
730 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
731 		error = ENAMETOOLONG;
732 		goto bail;
733 	}
734 
735 	mtx_lock(&Giant);
736 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
737 	mtx_unlock(&Giant);
738 bail:
739 	/* copyout the errmsg */
740 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
741 	    && errmsg_len > 0 && errmsg != NULL) {
742 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
743 			bcopy(errmsg,
744 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
745 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
746 		} else {
747 			copyout(errmsg,
748 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
749 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
750 		}
751 	}
752 
753 	if (error != 0)
754 		vfs_freeopts(optlist);
755 	return (error);
756 }
757 
758 /*
759  * Old mount API.
760  */
761 #ifndef _SYS_SYSPROTO_H_
762 struct mount_args {
763 	char	*type;
764 	char	*path;
765 	int	flags;
766 	caddr_t	data;
767 };
768 #endif
769 /* ARGSUSED */
770 int
771 mount(td, uap)
772 	struct thread *td;
773 	struct mount_args /* {
774 		char *type;
775 		char *path;
776 		int flags;
777 		caddr_t data;
778 	} */ *uap;
779 {
780 	char *fstype;
781 	struct vfsconf *vfsp = NULL;
782 	struct mntarg *ma = NULL;
783 	int error;
784 
785 	AUDIT_ARG(fflags, uap->flags);
786 
787 	/*
788 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
789 	 * userspace to set this flag, but we must filter it out if we want
790 	 * MNT_UPDATE on the root file system to work.
791 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
792 	 */
793 	uap->flags &= ~MNT_ROOTFS;
794 
795 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
796 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
797 	if (error) {
798 		free(fstype, M_TEMP);
799 		return (error);
800 	}
801 
802 	AUDIT_ARG(text, fstype);
803 	mtx_lock(&Giant);
804 	vfsp = vfs_byname_kld(fstype, td, &error);
805 	free(fstype, M_TEMP);
806 	if (vfsp == NULL) {
807 		mtx_unlock(&Giant);
808 		return (ENOENT);
809 	}
810 	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
811 		mtx_unlock(&Giant);
812 		return (EOPNOTSUPP);
813 	}
814 
815 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
816 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
817 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
818 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
819 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
820 
821 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
822 	mtx_unlock(&Giant);
823 	return (error);
824 }
825 
826 
827 /*
828  * vfs_domount(): actually attempt a filesystem mount.
829  */
830 static int
831 vfs_domount(
832 	struct thread *td,	/* Calling thread. */
833 	const char *fstype,	/* Filesystem type. */
834 	char *fspath,		/* Mount path. */
835 	int fsflags,		/* Flags common to all filesystems. */
836 	void *fsdata		/* Options local to the filesystem. */
837 	)
838 {
839 	struct vnode *vp;
840 	struct mount *mp;
841 	struct vfsconf *vfsp;
842 	struct export_args export;
843 	int error, flag = 0;
844 	struct vattr va;
845 	struct nameidata nd;
846 
847 	mtx_assert(&Giant, MA_OWNED);
848 	/*
849 	 * Be ultra-paranoid about making sure the type and fspath
850 	 * variables will fit in our mp buffers, including the
851 	 * terminating NUL.
852 	 */
853 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
854 		return (ENAMETOOLONG);
855 
856 	if (jailed(td->td_ucred) || usermount == 0) {
857 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
858 			return (error);
859 	}
860 
861 	/*
862 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
863 	 */
864 	if (fsflags & MNT_EXPORTED) {
865 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
866 		if (error)
867 			return (error);
868 	}
869 	if (fsflags & MNT_SUIDDIR) {
870 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
871 		if (error)
872 			return (error);
873 	}
874 	/*
875 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
876 	 */
877 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
878 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
879 			fsflags |= MNT_NOSUID | MNT_USER;
880 	}
881 
882 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
883 	vfsp = NULL;
884 	if ((fsflags & MNT_UPDATE) == 0) {
885 		/* Don't try to load KLDs if we're mounting the root. */
886 		if (fsflags & MNT_ROOTFS)
887 			vfsp = vfs_byname(fstype);
888 		else
889 			vfsp = vfs_byname_kld(fstype, td, &error);
890 		if (vfsp == NULL)
891 			return (ENODEV);
892 		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
893 			return (EPERM);
894 	}
895 	/*
896 	 * Get vnode to be covered
897 	 */
898 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
899 	    fspath, td);
900 	if ((error = namei(&nd)) != 0)
901 		return (error);
902 	NDFREE(&nd, NDF_ONLY_PNBUF);
903 	vp = nd.ni_vp;
904 	if (fsflags & MNT_UPDATE) {
905 		if ((vp->v_vflag & VV_ROOT) == 0) {
906 			vput(vp);
907 			return (EINVAL);
908 		}
909 		mp = vp->v_mount;
910 		MNT_ILOCK(mp);
911 		flag = mp->mnt_flag;
912 		/*
913 		 * We only allow the filesystem to be reloaded if it
914 		 * is currently mounted read-only.
915 		 */
916 		if ((fsflags & MNT_RELOAD) &&
917 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
918 			MNT_IUNLOCK(mp);
919 			vput(vp);
920 			return (EOPNOTSUPP);	/* Needs translation */
921 		}
922 		MNT_IUNLOCK(mp);
923 		/*
924 		 * Only privileged root, or (if MNT_USER is set) the user that
925 		 * did the original mount is permitted to update it.
926 		 */
927 		error = vfs_suser(mp, td);
928 		if (error) {
929 			vput(vp);
930 			return (error);
931 		}
932 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
933 			vput(vp);
934 			return (EBUSY);
935 		}
936 		VI_LOCK(vp);
937 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
938 		    vp->v_mountedhere != NULL) {
939 			VI_UNLOCK(vp);
940 			vfs_unbusy(mp, td);
941 			vput(vp);
942 			return (EBUSY);
943 		}
944 		vp->v_iflag |= VI_MOUNT;
945 		VI_UNLOCK(vp);
946 		MNT_ILOCK(mp);
947 		mp->mnt_flag |= fsflags &
948 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
949 		MNT_IUNLOCK(mp);
950 		VOP_UNLOCK(vp, 0);
951 		mp->mnt_optnew = fsdata;
952 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
953 	} else {
954 		/*
955 		 * If the user is not root, ensure that they own the directory
956 		 * onto which we are attempting to mount.
957 		 */
958 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
959 		if (error) {
960 			vput(vp);
961 			return (error);
962 		}
963 		if (va.va_uid != td->td_ucred->cr_uid) {
964 			error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
965 			    0);
966 			if (error) {
967 				vput(vp);
968 				return (error);
969 			}
970 		}
971 		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
972 		if (error != 0) {
973 			vput(vp);
974 			return (error);
975 		}
976 		if (vp->v_type != VDIR) {
977 			vput(vp);
978 			return (ENOTDIR);
979 		}
980 		VI_LOCK(vp);
981 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
982 		    vp->v_mountedhere != NULL) {
983 			VI_UNLOCK(vp);
984 			vput(vp);
985 			return (EBUSY);
986 		}
987 		vp->v_iflag |= VI_MOUNT;
988 		VI_UNLOCK(vp);
989 
990 		/*
991 		 * Allocate and initialize the filesystem.
992 		 */
993 		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
994 		VOP_UNLOCK(vp, 0);
995 
996 		/* XXXMAC: pass to vfs_mount_alloc? */
997 		mp->mnt_optnew = fsdata;
998 	}
999 
1000 	/*
1001 	 * Set the mount level flags.
1002 	 */
1003 	MNT_ILOCK(mp);
1004 	mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
1005 		(fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
1006 			    MNT_RDONLY));
1007 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1008 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1009 	MNT_IUNLOCK(mp);
1010 	/*
1011 	 * Mount the filesystem.
1012 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1013 	 * get.  No freeing of cn_pnbuf.
1014 	 */
1015         error = VFS_MOUNT(mp, td);
1016 
1017 	/*
1018 	 * Process the export option only if we are
1019 	 * updating mount options.
1020 	 */
1021 	if (!error && (fsflags & MNT_UPDATE)) {
1022 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
1023 		    sizeof(export)) == 0)
1024 			error = vfs_export(mp, &export);
1025 	}
1026 
1027 	if (!error) {
1028 		if (mp->mnt_opt != NULL)
1029 			vfs_freeopts(mp->mnt_opt);
1030 		mp->mnt_opt = mp->mnt_optnew;
1031 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
1032 	}
1033 	/*
1034 	 * Prevent external consumers of mount options from reading
1035 	 * mnt_optnew.
1036 	*/
1037 	mp->mnt_optnew = NULL;
1038 	if (mp->mnt_flag & MNT_UPDATE) {
1039 		MNT_ILOCK(mp);
1040 		if (error)
1041 			mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1042 				(flag & ~MNT_QUOTA);
1043 		else
1044 			mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD |
1045 					  MNT_FORCE | MNT_SNAPSHOT);
1046 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1047 			mp->mnt_kern_flag |= MNTK_ASYNC;
1048 		else
1049 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1050 		MNT_IUNLOCK(mp);
1051 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1052 			if (mp->mnt_syncer == NULL)
1053 				error = vfs_allocate_syncvnode(mp);
1054 		} else {
1055 			if (mp->mnt_syncer != NULL)
1056 				vrele(mp->mnt_syncer);
1057 			mp->mnt_syncer = NULL;
1058 		}
1059 		vfs_unbusy(mp, td);
1060 		VI_LOCK(vp);
1061 		vp->v_iflag &= ~VI_MOUNT;
1062 		VI_UNLOCK(vp);
1063 		vrele(vp);
1064 		return (error);
1065 	}
1066 	MNT_ILOCK(mp);
1067 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1068 		mp->mnt_kern_flag |= MNTK_ASYNC;
1069 	else
1070 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1071 	MNT_IUNLOCK(mp);
1072 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1073 	/*
1074 	 * Put the new filesystem on the mount list after root.
1075 	 */
1076 	cache_purge(vp);
1077 	if (!error) {
1078 		struct vnode *newdp;
1079 
1080 		VI_LOCK(vp);
1081 		vp->v_iflag &= ~VI_MOUNT;
1082 		VI_UNLOCK(vp);
1083 		vp->v_mountedhere = mp;
1084 		mtx_lock(&mountlist_mtx);
1085 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1086 		mtx_unlock(&mountlist_mtx);
1087 		vfs_event_signal(NULL, VQ_MOUNT, 0);
1088 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1089 			panic("mount: lost mount");
1090 		mountcheckdirs(vp, newdp);
1091 		vput(newdp);
1092 		VOP_UNLOCK(vp, 0);
1093 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1094 			error = vfs_allocate_syncvnode(mp);
1095 		vfs_unbusy(mp, td);
1096 		if (error)
1097 			vrele(vp);
1098 	} else {
1099 		VI_LOCK(vp);
1100 		vp->v_iflag &= ~VI_MOUNT;
1101 		VI_UNLOCK(vp);
1102 		vfs_unbusy(mp, td);
1103 		vfs_mount_destroy(mp);
1104 		vput(vp);
1105 	}
1106 	return (error);
1107 }
1108 
1109 /*
1110  * Unmount a filesystem.
1111  *
1112  * Note: unmount takes a path to the vnode mounted on as argument, not
1113  * special file (as before).
1114  */
1115 #ifndef _SYS_SYSPROTO_H_
1116 struct unmount_args {
1117 	char	*path;
1118 	int	flags;
1119 };
1120 #endif
1121 /* ARGSUSED */
1122 int
1123 unmount(td, uap)
1124 	struct thread *td;
1125 	register struct unmount_args /* {
1126 		char *path;
1127 		int flags;
1128 	} */ *uap;
1129 {
1130 	struct mount *mp;
1131 	char *pathbuf;
1132 	int error, id0, id1;
1133 
1134 	if (jailed(td->td_ucred) || usermount == 0) {
1135 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1136 		if (error)
1137 			return (error);
1138 	}
1139 
1140 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1141 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1142 	if (error) {
1143 		free(pathbuf, M_TEMP);
1144 		return (error);
1145 	}
1146 	AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1147 	mtx_lock(&Giant);
1148 	if (uap->flags & MNT_BYFSID) {
1149 		/* Decode the filesystem ID. */
1150 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1151 			mtx_unlock(&Giant);
1152 			free(pathbuf, M_TEMP);
1153 			return (EINVAL);
1154 		}
1155 
1156 		mtx_lock(&mountlist_mtx);
1157 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1158 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1159 			    mp->mnt_stat.f_fsid.val[1] == id1)
1160 				break;
1161 		}
1162 		mtx_unlock(&mountlist_mtx);
1163 	} else {
1164 		mtx_lock(&mountlist_mtx);
1165 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1166 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1167 				break;
1168 		}
1169 		mtx_unlock(&mountlist_mtx);
1170 	}
1171 	free(pathbuf, M_TEMP);
1172 	if (mp == NULL) {
1173 		/*
1174 		 * Previously we returned ENOENT for a nonexistent path and
1175 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1176 		 * now, so in the !MNT_BYFSID case return the more likely
1177 		 * EINVAL for compatibility.
1178 		 */
1179 		mtx_unlock(&Giant);
1180 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1181 	}
1182 
1183 	/*
1184 	 * Don't allow unmounting the root filesystem.
1185 	 */
1186 	if (mp->mnt_flag & MNT_ROOTFS) {
1187 		mtx_unlock(&Giant);
1188 		return (EINVAL);
1189 	}
1190 	error = dounmount(mp, uap->flags, td);
1191 	mtx_unlock(&Giant);
1192 	return (error);
1193 }
1194 
1195 /*
1196  * Do the actual filesystem unmount.
1197  */
1198 int
1199 dounmount(mp, flags, td)
1200 	struct mount *mp;
1201 	int flags;
1202 	struct thread *td;
1203 {
1204 	struct vnode *coveredvp, *fsrootvp;
1205 	int error;
1206 	int async_flag;
1207 	int mnt_gen_r;
1208 
1209 	mtx_assert(&Giant, MA_OWNED);
1210 
1211 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1212 		mnt_gen_r = mp->mnt_gen;
1213 		VI_LOCK(coveredvp);
1214 		vholdl(coveredvp);
1215 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1216 		vdrop(coveredvp);
1217 		/*
1218 		 * Check for mp being unmounted while waiting for the
1219 		 * covered vnode lock.
1220 		 */
1221 		if (coveredvp->v_mountedhere != mp ||
1222 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1223 			VOP_UNLOCK(coveredvp, 0);
1224 			return (EBUSY);
1225 		}
1226 	}
1227 	/*
1228 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1229 	 * original mount is permitted to unmount this filesystem.
1230 	 */
1231 	error = vfs_suser(mp, td);
1232 	if (error) {
1233 		if (coveredvp)
1234 			VOP_UNLOCK(coveredvp, 0);
1235 		return (error);
1236 	}
1237 
1238 	MNT_ILOCK(mp);
1239 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1240 		MNT_IUNLOCK(mp);
1241 		if (coveredvp)
1242 			VOP_UNLOCK(coveredvp, 0);
1243 		return (EBUSY);
1244 	}
1245 	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1246 	/* Allow filesystems to detect that a forced unmount is in progress. */
1247 	if (flags & MNT_FORCE)
1248 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1249 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1250 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp));
1251 	if (error) {
1252 		MNT_ILOCK(mp);
1253 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ |
1254 		    MNTK_UNMOUNTF);
1255 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1256 			wakeup(mp);
1257 		MNT_IUNLOCK(mp);
1258 		if (coveredvp)
1259 			VOP_UNLOCK(coveredvp, 0);
1260 		return (error);
1261 	}
1262 	vn_start_write(NULL, &mp, V_WAIT);
1263 
1264 	if (mp->mnt_flag & MNT_EXPUBLIC)
1265 		vfs_setpublicfs(NULL, NULL, NULL);
1266 
1267 	vfs_msync(mp, MNT_WAIT);
1268 	MNT_ILOCK(mp);
1269 	async_flag = mp->mnt_flag & MNT_ASYNC;
1270 	mp->mnt_flag &= ~MNT_ASYNC;
1271 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1272 	MNT_IUNLOCK(mp);
1273 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1274 	if (mp->mnt_syncer != NULL)
1275 		vrele(mp->mnt_syncer);
1276 	/*
1277 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1278 	 * vnode to the covered vnode.  For non-forced unmounts we want
1279 	 * such references to cause an EBUSY error.
1280 	 */
1281 	if ((flags & MNT_FORCE) &&
1282 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1283 		if (mp->mnt_vnodecovered != NULL)
1284 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1285 		if (fsrootvp == rootvnode) {
1286 			vrele(rootvnode);
1287 			rootvnode = NULL;
1288 		}
1289 		vput(fsrootvp);
1290 	}
1291 	if (((mp->mnt_flag & MNT_RDONLY) ||
1292 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1293 	    (flags & MNT_FORCE)) {
1294 		error = VFS_UNMOUNT(mp, flags, td);
1295 	}
1296 	vn_finished_write(mp);
1297 	/*
1298 	 * If we failed to flush the dirty blocks for this mount point,
1299 	 * undo all the cdir/rdir and rootvnode changes we made above.
1300 	 * Unless we failed to do so because the device is reporting that
1301 	 * it doesn't exist anymore.
1302 	 */
1303 	if (error && error != ENXIO) {
1304 		if ((flags & MNT_FORCE) &&
1305 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1306 			if (mp->mnt_vnodecovered != NULL)
1307 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1308 			if (rootvnode == NULL) {
1309 				rootvnode = fsrootvp;
1310 				vref(rootvnode);
1311 			}
1312 			vput(fsrootvp);
1313 		}
1314 		MNT_ILOCK(mp);
1315 		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1316 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) {
1317 			MNT_IUNLOCK(mp);
1318 			(void) vfs_allocate_syncvnode(mp);
1319 			MNT_ILOCK(mp);
1320 		}
1321 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1322 		mp->mnt_flag |= async_flag;
1323 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1324 			mp->mnt_kern_flag |= MNTK_ASYNC;
1325 		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1326 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1327 			wakeup(mp);
1328 		MNT_IUNLOCK(mp);
1329 		if (coveredvp)
1330 			VOP_UNLOCK(coveredvp, 0);
1331 		return (error);
1332 	}
1333 	mtx_lock(&mountlist_mtx);
1334 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1335 	mtx_unlock(&mountlist_mtx);
1336 	if (coveredvp != NULL) {
1337 		coveredvp->v_mountedhere = NULL;
1338 		vput(coveredvp);
1339 	}
1340 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1341 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1342 	vfs_mount_destroy(mp);
1343 	return (0);
1344 }
1345 
1346 /*
1347  * ---------------------------------------------------------------------
1348  * Mounting of root filesystem
1349  *
1350  */
1351 
1352 struct root_hold_token {
1353 	const char			*who;
1354 	LIST_ENTRY(root_hold_token)	list;
1355 };
1356 
1357 static LIST_HEAD(, root_hold_token)	root_holds =
1358     LIST_HEAD_INITIALIZER(&root_holds);
1359 
1360 static int root_mount_complete;
1361 
1362 /*
1363  * Hold root mount.
1364  */
1365 struct root_hold_token *
1366 root_mount_hold(const char *identifier)
1367 {
1368 	struct root_hold_token *h;
1369 
1370 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1371 	h->who = identifier;
1372 	mtx_lock(&mountlist_mtx);
1373 	LIST_INSERT_HEAD(&root_holds, h, list);
1374 	mtx_unlock(&mountlist_mtx);
1375 	return (h);
1376 }
1377 
1378 /*
1379  * Release root mount.
1380  */
1381 void
1382 root_mount_rel(struct root_hold_token *h)
1383 {
1384 
1385 	mtx_lock(&mountlist_mtx);
1386 	LIST_REMOVE(h, list);
1387 	wakeup(&root_holds);
1388 	mtx_unlock(&mountlist_mtx);
1389 	free(h, M_DEVBUF);
1390 }
1391 
1392 /*
1393  * Wait for all subsystems to release root mount.
1394  */
1395 static void
1396 root_mount_prepare(void)
1397 {
1398 	struct root_hold_token *h;
1399 
1400 	for (;;) {
1401 		DROP_GIANT();
1402 		g_waitidle();
1403 		PICKUP_GIANT();
1404 		mtx_lock(&mountlist_mtx);
1405 		if (LIST_EMPTY(&root_holds)) {
1406 			mtx_unlock(&mountlist_mtx);
1407 			break;
1408 		}
1409 		printf("Root mount waiting for:");
1410 		LIST_FOREACH(h, &root_holds, list)
1411 			printf(" %s", h->who);
1412 		printf("\n");
1413 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1414 		    hz);
1415 	}
1416 }
1417 
1418 /*
1419  * Root was mounted, share the good news.
1420  */
1421 static void
1422 root_mount_done(void)
1423 {
1424 
1425 	/*
1426 	 * Use a mutex to prevent the wakeup being missed and waiting for
1427 	 * an extra 1 second sleep.
1428 	 */
1429 	mtx_lock(&mountlist_mtx);
1430 	root_mount_complete = 1;
1431 	wakeup(&root_mount_complete);
1432 	mtx_unlock(&mountlist_mtx);
1433 }
1434 
1435 /*
1436  * Return true if root is already mounted.
1437  */
1438 int
1439 root_mounted(void)
1440 {
1441 
1442 	/* No mutex is acquired here because int stores are atomic. */
1443 	return (root_mount_complete);
1444 }
1445 
1446 /*
1447  * Wait until root is mounted.
1448  */
1449 void
1450 root_mount_wait(void)
1451 {
1452 
1453 	/*
1454 	 * Panic on an obvious deadlock - the function can't be called from
1455 	 * a thread which is doing the whole SYSINIT stuff.
1456 	 */
1457 	KASSERT(curthread->td_proc->p_pid != 0,
1458 	    ("root_mount_wait: cannot be called from the swapper thread"));
1459 	mtx_lock(&mountlist_mtx);
1460 	while (!root_mount_complete) {
1461 		msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
1462 		    hz);
1463 	}
1464 	mtx_unlock(&mountlist_mtx);
1465 }
1466 
1467 static void
1468 set_rootvnode(struct thread *td)
1469 {
1470 	struct proc *p;
1471 
1472 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1473 		panic("Cannot find root vnode");
1474 
1475 	p = td->td_proc;
1476 	FILEDESC_SLOCK(p->p_fd);
1477 
1478 	if (p->p_fd->fd_cdir != NULL)
1479 		vrele(p->p_fd->fd_cdir);
1480 	p->p_fd->fd_cdir = rootvnode;
1481 	VREF(rootvnode);
1482 
1483 	if (p->p_fd->fd_rdir != NULL)
1484 		vrele(p->p_fd->fd_rdir);
1485 	p->p_fd->fd_rdir = rootvnode;
1486 	VREF(rootvnode);
1487 
1488 	FILEDESC_SUNLOCK(p->p_fd);
1489 
1490 	VOP_UNLOCK(rootvnode, 0);
1491 
1492 	EVENTHANDLER_INVOKE(mountroot);
1493 }
1494 
1495 /*
1496  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1497  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1498  */
1499 
1500 static void
1501 devfs_first(void)
1502 {
1503 	struct thread *td = curthread;
1504 	struct vfsoptlist *opts;
1505 	struct vfsconf *vfsp;
1506 	struct mount *mp = NULL;
1507 	int error;
1508 
1509 	vfsp = vfs_byname("devfs");
1510 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1511 	if (vfsp == NULL)
1512 		return;
1513 
1514 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1515 
1516 	error = VFS_MOUNT(mp, td);
1517 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1518 	if (error)
1519 		return;
1520 
1521 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1522 	TAILQ_INIT(opts);
1523 	mp->mnt_opt = opts;
1524 
1525 	mtx_lock(&mountlist_mtx);
1526 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1527 	mtx_unlock(&mountlist_mtx);
1528 
1529 	set_rootvnode(td);
1530 
1531 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1532 	if (error)
1533 		printf("kern_symlink /dev -> / returns %d\n", error);
1534 }
1535 
1536 /*
1537  * Surgically move our devfs to be mounted on /dev.
1538  */
1539 
1540 static void
1541 devfs_fixup(struct thread *td)
1542 {
1543 	struct nameidata nd;
1544 	int error;
1545 	struct vnode *vp, *dvp;
1546 	struct mount *mp;
1547 
1548 	/* Remove our devfs mount from the mountlist and purge the cache */
1549 	mtx_lock(&mountlist_mtx);
1550 	mp = TAILQ_FIRST(&mountlist);
1551 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1552 	mtx_unlock(&mountlist_mtx);
1553 	cache_purgevfs(mp);
1554 
1555 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1556 	VI_LOCK(dvp);
1557 	dvp->v_iflag &= ~VI_MOUNT;
1558 	VI_UNLOCK(dvp);
1559 	dvp->v_mountedhere = NULL;
1560 
1561 	/* Set up the real rootvnode, and purge the cache */
1562 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1563 	set_rootvnode(td);
1564 	cache_purgevfs(rootvnode->v_mount);
1565 
1566 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1567 	error = namei(&nd);
1568 	if (error) {
1569 		printf("Lookup of /dev for devfs, error: %d\n", error);
1570 		return;
1571 	}
1572 	NDFREE(&nd, NDF_ONLY_PNBUF);
1573 	vp = nd.ni_vp;
1574 	if (vp->v_type != VDIR) {
1575 		vput(vp);
1576 	}
1577 	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1578 	if (error) {
1579 		vput(vp);
1580 	}
1581 	cache_purge(vp);
1582 	mp->mnt_vnodecovered = vp;
1583 	vp->v_mountedhere = mp;
1584 	mtx_lock(&mountlist_mtx);
1585 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1586 	mtx_unlock(&mountlist_mtx);
1587 	VOP_UNLOCK(vp, 0);
1588 	vput(dvp);
1589 	vfs_unbusy(mp, td);
1590 
1591 	/* Unlink the no longer needed /dev/dev -> / symlink */
1592 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1593 }
1594 
1595 /*
1596  * Report errors during filesystem mounting.
1597  */
1598 void
1599 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1600 {
1601 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1602 	va_list ap;
1603 	int error, len;
1604 	char *errmsg;
1605 
1606 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1607 	if (error || errmsg == NULL || len <= 0)
1608 		return;
1609 
1610 	va_start(ap, fmt);
1611 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1612 	va_end(ap);
1613 }
1614 
1615 /*
1616  * Find and mount the root filesystem
1617  */
1618 void
1619 vfs_mountroot(void)
1620 {
1621 	char *cp;
1622 	int error, i, asked = 0;
1623 
1624 	root_mount_prepare();
1625 
1626 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1627 	    NULL, NULL, mount_init, mount_fini,
1628 	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1629 	devfs_first();
1630 
1631 	/*
1632 	 * We are booted with instructions to prompt for the root filesystem.
1633 	 */
1634 	if (boothowto & RB_ASKNAME) {
1635 		if (!vfs_mountroot_ask())
1636 			goto mounted;
1637 		asked = 1;
1638 	}
1639 
1640 	/*
1641 	 * The root filesystem information is compiled in, and we are
1642 	 * booted with instructions to use it.
1643 	 */
1644 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1645 		if (!vfs_mountroot_try(ctrootdevname))
1646 			goto mounted;
1647 		ctrootdevname = NULL;
1648 	}
1649 
1650 	/*
1651 	 * We've been given the generic "use CDROM as root" flag.  This is
1652 	 * necessary because one media may be used in many different
1653 	 * devices, so we need to search for them.
1654 	 */
1655 	if (boothowto & RB_CDROM) {
1656 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1657 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1658 				goto mounted;
1659 		}
1660 	}
1661 
1662 	/*
1663 	 * Try to use the value read by the loader from /etc/fstab, or
1664 	 * supplied via some other means.  This is the preferred
1665 	 * mechanism.
1666 	 */
1667 	cp = getenv("vfs.root.mountfrom");
1668 	if (cp != NULL) {
1669 		error = vfs_mountroot_try(cp);
1670 		freeenv(cp);
1671 		if (!error)
1672 			goto mounted;
1673 	}
1674 
1675 	/*
1676 	 * Try values that may have been computed by code during boot
1677 	 */
1678 	if (!vfs_mountroot_try(rootdevnames[0]))
1679 		goto mounted;
1680 	if (!vfs_mountroot_try(rootdevnames[1]))
1681 		goto mounted;
1682 
1683 	/*
1684 	 * If we (still) have a compiled-in default, try it.
1685 	 */
1686 	if (ctrootdevname != NULL)
1687 		if (!vfs_mountroot_try(ctrootdevname))
1688 			goto mounted;
1689 	/*
1690 	 * Everything so far has failed, prompt on the console if we haven't
1691 	 * already tried that.
1692 	 */
1693 	if (!asked)
1694 		if (!vfs_mountroot_ask())
1695 			goto mounted;
1696 
1697 	panic("Root mount failed, startup aborted.");
1698 
1699 mounted:
1700 	root_mount_done();
1701 }
1702 
1703 /*
1704  * Mount (mountfrom) as the root filesystem.
1705  */
1706 static int
1707 vfs_mountroot_try(const char *mountfrom)
1708 {
1709 	struct mount	*mp;
1710 	char		*vfsname, *path;
1711 	time_t		timebase;
1712 	int		error;
1713 	char		patt[32];
1714 
1715 	vfsname = NULL;
1716 	path    = NULL;
1717 	mp      = NULL;
1718 	error   = EINVAL;
1719 
1720 	if (mountfrom == NULL)
1721 		return (error);		/* don't complain */
1722 	printf("Trying to mount root from %s\n", mountfrom);
1723 
1724 	/* parse vfs name and path */
1725 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1726 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1727 	vfsname[0] = path[0] = 0;
1728 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1729 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1730 		goto out;
1731 
1732 	if (path[0] == '\0')
1733 		strcpy(path, ROOTNAME);
1734 
1735 	error = kernel_vmount(
1736 	    MNT_RDONLY | MNT_ROOTFS,
1737 	    "fstype", vfsname,
1738 	    "fspath", "/",
1739 	    "from", path,
1740 	    NULL);
1741 	if (error == 0) {
1742 		/*
1743 		 * We mount devfs prior to mounting the / FS, so the first
1744 		 * entry will typically be devfs.
1745 		 */
1746 		mp = TAILQ_FIRST(&mountlist);
1747 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1748 
1749 		/*
1750 		 * Iterate over all currently mounted file systems and use
1751 		 * the time stamp found to check and/or initialize the RTC.
1752 		 * Typically devfs has no time stamp and the only other FS
1753 		 * is the actual / FS.
1754 		 * Call inittodr() only once and pass it the largest of the
1755 		 * timestamps we encounter.
1756 		 */
1757 		timebase = 0;
1758 		do {
1759 			if (mp->mnt_time > timebase)
1760 				timebase = mp->mnt_time;
1761 			mp = TAILQ_NEXT(mp, mnt_list);
1762 		} while (mp != NULL);
1763 		inittodr(timebase);
1764 
1765 		devfs_fixup(curthread);
1766 	}
1767 out:
1768 	free(path, M_MOUNT);
1769 	free(vfsname, M_MOUNT);
1770 	return (error);
1771 }
1772 
1773 /*
1774  * ---------------------------------------------------------------------
1775  * Interactive root filesystem selection code.
1776  */
1777 
1778 static int
1779 vfs_mountroot_ask(void)
1780 {
1781 	char name[128];
1782 
1783 	for(;;) {
1784 		printf("\nManual root filesystem specification:\n");
1785 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1786 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1787 		printf("                       eg. ufs:da0s1a\n");
1788 #else
1789 		printf("                       eg. ufs:/dev/da0a\n");
1790 #endif
1791 		printf("  ?                  List valid disk boot devices\n");
1792 		printf("  <empty line>       Abort manual input\n");
1793 		printf("\nmountroot> ");
1794 		gets(name, sizeof(name), 1);
1795 		if (name[0] == '\0')
1796 			return (1);
1797 		if (name[0] == '?') {
1798 			printf("\nList of GEOM managed disk devices:\n  ");
1799 			g_dev_print();
1800 			continue;
1801 		}
1802 		if (!vfs_mountroot_try(name))
1803 			return (0);
1804 	}
1805 }
1806 
1807 /*
1808  * ---------------------------------------------------------------------
1809  * Functions for querying mount options/arguments from filesystems.
1810  */
1811 
1812 /*
1813  * Check that no unknown options are given
1814  */
1815 int
1816 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1817 {
1818 	struct vfsopt *opt;
1819 	char errmsg[255];
1820 	const char **t, *p, *q;
1821 	int ret = 0;
1822 
1823 	TAILQ_FOREACH(opt, opts, link) {
1824 		p = opt->name;
1825 		q = NULL;
1826 		if (p[0] == 'n' && p[1] == 'o')
1827 			q = p + 2;
1828 		for(t = global_opts; *t != NULL; t++) {
1829 			if (strcmp(*t, p) == 0)
1830 				break;
1831 			if (q != NULL) {
1832 				if (strcmp(*t, q) == 0)
1833 					break;
1834 			}
1835 		}
1836 		if (*t != NULL)
1837 			continue;
1838 		for(t = legal; *t != NULL; t++) {
1839 			if (strcmp(*t, p) == 0)
1840 				break;
1841 			if (q != NULL) {
1842 				if (strcmp(*t, q) == 0)
1843 					break;
1844 			}
1845 		}
1846 		if (*t != NULL)
1847 			continue;
1848 		sprintf(errmsg, "mount option <%s> is unknown", p);
1849 		printf("%s\n", errmsg);
1850 		ret = EINVAL;
1851 	}
1852 	if (ret != 0) {
1853 		TAILQ_FOREACH(opt, opts, link) {
1854 			if (strcmp(opt->name, "errmsg") == 0) {
1855 				strncpy((char *)opt->value, errmsg, opt->len);
1856 			}
1857 		}
1858 	}
1859 	return (ret);
1860 }
1861 
1862 /*
1863  * Get a mount option by its name.
1864  *
1865  * Return 0 if the option was found, ENOENT otherwise.
1866  * If len is non-NULL it will be filled with the length
1867  * of the option. If buf is non-NULL, it will be filled
1868  * with the address of the option.
1869  */
1870 int
1871 vfs_getopt(opts, name, buf, len)
1872 	struct vfsoptlist *opts;
1873 	const char *name;
1874 	void **buf;
1875 	int *len;
1876 {
1877 	struct vfsopt *opt;
1878 
1879 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1880 
1881 	TAILQ_FOREACH(opt, opts, link) {
1882 		if (strcmp(name, opt->name) == 0) {
1883 			if (len != NULL)
1884 				*len = opt->len;
1885 			if (buf != NULL)
1886 				*buf = opt->value;
1887 			return (0);
1888 		}
1889 	}
1890 	return (ENOENT);
1891 }
1892 
1893 static int
1894 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1895 {
1896 	struct vfsopt *opt;
1897 	int i;
1898 
1899 	if (opts == NULL)
1900 		return (-1);
1901 
1902 	i = 0;
1903 	TAILQ_FOREACH(opt, opts, link) {
1904 		if (strcmp(name, opt->name) == 0)
1905 			return (i);
1906 		++i;
1907 	}
1908 	return (-1);
1909 }
1910 
1911 char *
1912 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1913 {
1914 	struct vfsopt *opt;
1915 
1916 	*error = 0;
1917 	TAILQ_FOREACH(opt, opts, link) {
1918 		if (strcmp(name, opt->name) != 0)
1919 			continue;
1920 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1921 			*error = EINVAL;
1922 			return (NULL);
1923 		}
1924 		return (opt->value);
1925 	}
1926 	*error = ENOENT;
1927 	return (NULL);
1928 }
1929 
1930 int
1931 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1932 {
1933 	struct vfsopt *opt;
1934 
1935 	TAILQ_FOREACH(opt, opts, link) {
1936 		if (strcmp(name, opt->name) == 0) {
1937 			if (w != NULL)
1938 				*w |= val;
1939 			return (1);
1940 		}
1941 	}
1942 	if (w != NULL)
1943 		*w &= ~val;
1944 	return (0);
1945 }
1946 
1947 int
1948 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1949 {
1950 	va_list ap;
1951 	struct vfsopt *opt;
1952 	int ret;
1953 
1954 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1955 
1956 	TAILQ_FOREACH(opt, opts, link) {
1957 		if (strcmp(name, opt->name) != 0)
1958 			continue;
1959 		if (opt->len == 0 || opt->value == NULL)
1960 			return (0);
1961 		if (((char *)opt->value)[opt->len - 1] != '\0')
1962 			return (0);
1963 		va_start(ap, fmt);
1964 		ret = vsscanf(opt->value, fmt, ap);
1965 		va_end(ap);
1966 		return (ret);
1967 	}
1968 	return (0);
1969 }
1970 
1971 /*
1972  * Find and copy a mount option.
1973  *
1974  * The size of the buffer has to be specified
1975  * in len, if it is not the same length as the
1976  * mount option, EINVAL is returned.
1977  * Returns ENOENT if the option is not found.
1978  */
1979 int
1980 vfs_copyopt(opts, name, dest, len)
1981 	struct vfsoptlist *opts;
1982 	const char *name;
1983 	void *dest;
1984 	int len;
1985 {
1986 	struct vfsopt *opt;
1987 
1988 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1989 
1990 	TAILQ_FOREACH(opt, opts, link) {
1991 		if (strcmp(name, opt->name) == 0) {
1992 			if (len != opt->len)
1993 				return (EINVAL);
1994 			bcopy(opt->value, dest, opt->len);
1995 			return (0);
1996 		}
1997 	}
1998 	return (ENOENT);
1999 }
2000 
2001 /*
2002  * This is a helper function for filesystems to traverse their
2003  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
2004  */
2005 
2006 struct vnode *
2007 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
2008 {
2009 	struct vnode *vp;
2010 
2011 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2012 
2013 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2014 	if ((*mvp)->v_yield++ == 500) {
2015 		MNT_IUNLOCK(mp);
2016 		(*mvp)->v_yield = 0;
2017 		uio_yield();
2018 		MNT_ILOCK(mp);
2019 	}
2020 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
2021 	while (vp != NULL && vp->v_type == VMARKER)
2022 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2023 
2024 	/* Check if we are done */
2025 	if (vp == NULL) {
2026 		__mnt_vnode_markerfree(mvp, mp);
2027 		return (NULL);
2028 	}
2029 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2030 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2031 	return (vp);
2032 }
2033 
2034 struct vnode *
2035 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
2036 {
2037 	struct vnode *vp;
2038 
2039 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2040 
2041 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2042 	while (vp != NULL && vp->v_type == VMARKER)
2043 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2044 
2045 	/* Check if we are done */
2046 	if (vp == NULL) {
2047 		*mvp = NULL;
2048 		return (NULL);
2049 	}
2050 	mp->mnt_holdcnt++;
2051 	MNT_IUNLOCK(mp);
2052 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
2053 				       M_VNODE_MARKER,
2054 				       M_WAITOK | M_ZERO);
2055 	MNT_ILOCK(mp);
2056 	(*mvp)->v_type = VMARKER;
2057 
2058 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2059 	while (vp != NULL && vp->v_type == VMARKER)
2060 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2061 
2062 	/* Check if we are done */
2063 	if (vp == NULL) {
2064 		MNT_IUNLOCK(mp);
2065 		free(*mvp, M_VNODE_MARKER);
2066 		MNT_ILOCK(mp);
2067 		*mvp = NULL;
2068 		mp->mnt_holdcnt--;
2069 		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2070 			wakeup(&mp->mnt_holdcnt);
2071 		return (NULL);
2072 	}
2073 	mp->mnt_markercnt++;
2074 	(*mvp)->v_mount = mp;
2075 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2076 	return (vp);
2077 }
2078 
2079 
2080 void
2081 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
2082 {
2083 
2084 	if (*mvp == NULL)
2085 		return;
2086 
2087 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2088 
2089 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2090 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2091 	MNT_IUNLOCK(mp);
2092 	free(*mvp, M_VNODE_MARKER);
2093 	MNT_ILOCK(mp);
2094 	*mvp = NULL;
2095 
2096 	mp->mnt_markercnt--;
2097 	mp->mnt_holdcnt--;
2098 	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2099 		wakeup(&mp->mnt_holdcnt);
2100 }
2101 
2102 
2103 int
2104 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
2105 {
2106 	int error;
2107 
2108 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
2109 	if (sbp != &mp->mnt_stat)
2110 		*sbp = mp->mnt_stat;
2111 	return (error);
2112 }
2113 
2114 void
2115 vfs_mountedfrom(struct mount *mp, const char *from)
2116 {
2117 
2118 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2119 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2120 	    sizeof mp->mnt_stat.f_mntfromname);
2121 }
2122 
2123 /*
2124  * ---------------------------------------------------------------------
2125  * This is the api for building mount args and mounting filesystems from
2126  * inside the kernel.
2127  *
2128  * The API works by accumulation of individual args.  First error is
2129  * latched.
2130  *
2131  * XXX: should be documented in new manpage kernel_mount(9)
2132  */
2133 
2134 /* A memory allocation which must be freed when we are done */
2135 struct mntaarg {
2136 	SLIST_ENTRY(mntaarg)	next;
2137 };
2138 
2139 /* The header for the mount arguments */
2140 struct mntarg {
2141 	struct iovec *v;
2142 	int len;
2143 	int error;
2144 	SLIST_HEAD(, mntaarg)	list;
2145 };
2146 
2147 /*
2148  * Add a boolean argument.
2149  *
2150  * flag is the boolean value.
2151  * name must start with "no".
2152  */
2153 struct mntarg *
2154 mount_argb(struct mntarg *ma, int flag, const char *name)
2155 {
2156 
2157 	KASSERT(name[0] == 'n' && name[1] == 'o',
2158 	    ("mount_argb(...,%s): name must start with 'no'", name));
2159 
2160 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2161 }
2162 
2163 /*
2164  * Add an argument printf style
2165  */
2166 struct mntarg *
2167 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2168 {
2169 	va_list ap;
2170 	struct mntaarg *maa;
2171 	struct sbuf *sb;
2172 	int len;
2173 
2174 	if (ma == NULL) {
2175 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2176 		SLIST_INIT(&ma->list);
2177 	}
2178 	if (ma->error)
2179 		return (ma);
2180 
2181 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2182 	    M_MOUNT, M_WAITOK);
2183 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2184 	ma->v[ma->len].iov_len = strlen(name) + 1;
2185 	ma->len++;
2186 
2187 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
2188 	va_start(ap, fmt);
2189 	sbuf_vprintf(sb, fmt, ap);
2190 	va_end(ap);
2191 	sbuf_finish(sb);
2192 	len = sbuf_len(sb) + 1;
2193 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2194 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2195 	bcopy(sbuf_data(sb), maa + 1, len);
2196 	sbuf_delete(sb);
2197 
2198 	ma->v[ma->len].iov_base = maa + 1;
2199 	ma->v[ma->len].iov_len = len;
2200 	ma->len++;
2201 
2202 	return (ma);
2203 }
2204 
2205 /*
2206  * Add an argument which is a userland string.
2207  */
2208 struct mntarg *
2209 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2210 {
2211 	struct mntaarg *maa;
2212 	char *tbuf;
2213 
2214 	if (val == NULL)
2215 		return (ma);
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 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2223 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2224 	tbuf = (void *)(maa + 1);
2225 	ma->error = copyinstr(val, tbuf, len, NULL);
2226 	return (mount_arg(ma, name, tbuf, -1));
2227 }
2228 
2229 /*
2230  * Plain argument.
2231  *
2232  * If length is -1, treat value as a C string.
2233  */
2234 struct mntarg *
2235 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2236 {
2237 
2238 	if (ma == NULL) {
2239 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2240 		SLIST_INIT(&ma->list);
2241 	}
2242 	if (ma->error)
2243 		return (ma);
2244 
2245 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2246 	    M_MOUNT, M_WAITOK);
2247 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2248 	ma->v[ma->len].iov_len = strlen(name) + 1;
2249 	ma->len++;
2250 
2251 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2252 	if (len < 0)
2253 		ma->v[ma->len].iov_len = strlen(val) + 1;
2254 	else
2255 		ma->v[ma->len].iov_len = len;
2256 	ma->len++;
2257 	return (ma);
2258 }
2259 
2260 /*
2261  * Free a mntarg structure
2262  */
2263 static void
2264 free_mntarg(struct mntarg *ma)
2265 {
2266 	struct mntaarg *maa;
2267 
2268 	while (!SLIST_EMPTY(&ma->list)) {
2269 		maa = SLIST_FIRST(&ma->list);
2270 		SLIST_REMOVE_HEAD(&ma->list, next);
2271 		free(maa, M_MOUNT);
2272 	}
2273 	free(ma->v, M_MOUNT);
2274 	free(ma, M_MOUNT);
2275 }
2276 
2277 /*
2278  * Mount a filesystem
2279  */
2280 int
2281 kernel_mount(struct mntarg *ma, int flags)
2282 {
2283 	struct uio auio;
2284 	int error;
2285 
2286 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2287 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2288 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2289 
2290 	auio.uio_iov = ma->v;
2291 	auio.uio_iovcnt = ma->len;
2292 	auio.uio_segflg = UIO_SYSSPACE;
2293 
2294 	error = ma->error;
2295 	if (!error)
2296 		error = vfs_donmount(curthread, flags, &auio);
2297 	free_mntarg(ma);
2298 	return (error);
2299 }
2300 
2301 /*
2302  * A printflike function to mount a filesystem.
2303  */
2304 int
2305 kernel_vmount(int flags, ...)
2306 {
2307 	struct mntarg *ma = NULL;
2308 	va_list ap;
2309 	const char *cp;
2310 	const void *vp;
2311 	int error;
2312 
2313 	va_start(ap, flags);
2314 	for (;;) {
2315 		cp = va_arg(ap, const char *);
2316 		if (cp == NULL)
2317 			break;
2318 		vp = va_arg(ap, const void *);
2319 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2320 	}
2321 	va_end(ap);
2322 
2323 	error = kernel_mount(ma, flags);
2324 	return (error);
2325 }
2326