xref: /freebsd/sys/kern/vfs_mount.c (revision 2b743a9e9ddc6736208dc8ca1ce06ce64ad20a19)
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/jail.h>
44 #include <sys/kernel.h>
45 #include <sys/libkern.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/filedesc.h>
53 #include <sys/reboot.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysproto.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/systm.h>
60 #include <sys/vnode.h>
61 #include <vm/uma.h>
62 
63 #include <geom/geom.h>
64 
65 #include <machine/stdarg.h>
66 
67 #include <security/audit/audit.h>
68 #include <security/mac/mac_framework.h>
69 
70 #include "opt_rootdevname.h"
71 #include "opt_ddb.h"
72 #include "opt_mac.h"
73 
74 #ifdef DDB
75 #include <ddb/ddb.h>
76 #endif
77 
78 #define	ROOTNAME		"root_device"
79 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
80 
81 static int	vfs_domount(struct thread *td, const char *fstype,
82 		    char *fspath, int fsflags, void *fsdata);
83 static struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
84 		    const char *fspath, struct thread *td);
85 static int	vfs_mountroot_ask(void);
86 static int	vfs_mountroot_try(const char *mountfrom);
87 static int	vfs_donmount(struct thread *td, int fsflags,
88 		    struct uio *fsoptions);
89 static void	free_mntarg(struct mntarg *ma);
90 static void	vfs_mount_destroy(struct mount *);
91 static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
92 
93 static int	usermount = 0;
94 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
95     "Unprivileged users may mount and unmount file systems");
96 
97 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
98 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
99 static uma_zone_t mount_zone;
100 
101 /* List of mounted filesystems. */
102 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
103 
104 /* For any iteration/modification of mountlist */
105 struct mtx mountlist_mtx;
106 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
107 
108 TAILQ_HEAD(vfsoptlist, vfsopt);
109 struct vfsopt {
110 	TAILQ_ENTRY(vfsopt) link;
111 	char	*name;
112 	void	*value;
113 	int	len;
114 };
115 
116 /*
117  * The vnode of the system's root (/ in the filesystem, without chroot
118  * active.)
119  */
120 struct vnode	*rootvnode;
121 
122 /*
123  * The root filesystem is detailed in the kernel environment variable
124  * vfs.root.mountfrom, which is expected to be in the general format
125  *
126  * <vfsname>:[<path>]
127  * vfsname   := the name of a VFS known to the kernel and capable
128  *              of being mounted as root
129  * path      := disk device name or other data used by the filesystem
130  *              to locate its physical store
131  */
132 
133 /*
134  * Global opts, taken by all filesystems
135  */
136 static const char *global_opts[] = {
137 	"errmsg",
138 	"fstype",
139 	"fspath",
140 	"rdonly",
141 	"ro",
142 	"rw",
143 	"suid",
144 	"exec",
145 	"update",
146 	NULL
147 };
148 
149 /*
150  * The root specifiers we will try if RB_CDROM is specified.
151  */
152 static char *cdrom_rootdevnames[] = {
153 	"cd9660:cd0",
154 	"cd9660:acd0",
155 	NULL
156 };
157 
158 /* legacy find-root code */
159 char		*rootdevnames[2] = {NULL, NULL};
160 #ifndef ROOTDEVNAME
161 #  define ROOTDEVNAME NULL
162 #endif
163 static const char	*ctrootdevname = ROOTDEVNAME;
164 
165 /*
166  * ---------------------------------------------------------------------
167  * Functions for building and sanitizing the mount options
168  */
169 
170 /* Remove one mount option. */
171 static void
172 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
173 {
174 
175 	TAILQ_REMOVE(opts, opt, link);
176 	free(opt->name, M_MOUNT);
177 	if (opt->value != NULL)
178 		free(opt->value, M_MOUNT);
179 #ifdef INVARIANTS
180 	else if (opt->len != 0)
181 		panic("%s: mount option with NULL value but length != 0",
182 		    __func__);
183 #endif
184 	free(opt, M_MOUNT);
185 }
186 
187 /* Release all resources related to the mount options. */
188 static void
189 vfs_freeopts(struct vfsoptlist *opts)
190 {
191 	struct vfsopt *opt;
192 
193 	while (!TAILQ_EMPTY(opts)) {
194 		opt = TAILQ_FIRST(opts);
195 		vfs_freeopt(opts, opt);
196 	}
197 	free(opts, M_MOUNT);
198 }
199 
200 void
201 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
202 {
203 	struct vfsopt *opt, *temp;
204 
205 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
206 		if (strcmp(opt->name, name) == 0)
207 			vfs_freeopt(opts, opt);
208 	}
209 }
210 
211 /*
212  * Check if options are equal (with or without the "no" prefix).
213  */
214 static int
215 vfs_equalopts(const char *opt1, const char *opt2)
216 {
217 
218 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
219 	if (strcmp(opt1, opt2) == 0)
220 		return (1);
221 	/* "noopt" vs. "opt" */
222 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
223 		return (1);
224 	/* "opt" vs. "noopt" */
225 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
226 		return (1);
227 	return (0);
228 }
229 
230 /*
231  * If a mount option is specified several times,
232  * (with or without the "no" prefix) only keep
233  * the last occurence of it.
234  */
235 static void
236 vfs_sanitizeopts(struct vfsoptlist *opts)
237 {
238 	struct vfsopt *opt, *opt2, *tmp;
239 
240 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
241 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
242 		while (opt2 != NULL) {
243 			if (vfs_equalopts(opt->name, opt2->name)) {
244 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
245 				vfs_freeopt(opts, opt2);
246 				opt2 = tmp;
247 			} else {
248 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
249 			}
250 		}
251 	}
252 }
253 
254 /*
255  * Build a linked list of mount options from a struct uio.
256  */
257 static int
258 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
259 {
260 	struct vfsoptlist *opts;
261 	struct vfsopt *opt;
262 	size_t memused;
263 	unsigned int i, iovcnt;
264 	int error, namelen, optlen;
265 
266 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
267 	TAILQ_INIT(opts);
268 	memused = 0;
269 	iovcnt = auio->uio_iovcnt;
270 	for (i = 0; i < iovcnt; i += 2) {
271 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
272 		namelen = auio->uio_iov[i].iov_len;
273 		optlen = auio->uio_iov[i + 1].iov_len;
274 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
275 		opt->value = NULL;
276 		opt->len = 0;
277 
278 		/*
279 		 * Do this early, so jumps to "bad" will free the current
280 		 * option.
281 		 */
282 		TAILQ_INSERT_TAIL(opts, opt, link);
283 		memused += sizeof(struct vfsopt) + optlen + namelen;
284 
285 		/*
286 		 * Avoid consuming too much memory, and attempts to overflow
287 		 * memused.
288 		 */
289 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
290 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
291 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
292 			error = EINVAL;
293 			goto bad;
294 		}
295 
296 		if (auio->uio_segflg == UIO_SYSSPACE) {
297 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
298 		} else {
299 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
300 			    namelen);
301 			if (error)
302 				goto bad;
303 		}
304 		/* Ensure names are null-terminated strings. */
305 		if (opt->name[namelen - 1] != '\0') {
306 			error = EINVAL;
307 			goto bad;
308 		}
309 		if (optlen != 0) {
310 			opt->len = optlen;
311 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
312 			if (auio->uio_segflg == UIO_SYSSPACE) {
313 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
314 				    optlen);
315 			} else {
316 				error = copyin(auio->uio_iov[i + 1].iov_base,
317 				    opt->value, optlen);
318 				if (error)
319 					goto bad;
320 			}
321 		}
322 	}
323 	vfs_sanitizeopts(opts);
324 	*options = opts;
325 	return (0);
326 bad:
327 	vfs_freeopts(opts);
328 	return (error);
329 }
330 
331 /*
332  * Merge the old mount options with the new ones passed
333  * in the MNT_UPDATE case.
334  */
335 static void
336 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
337 {
338 	struct vfsopt *opt, *opt2, *new;
339 
340 	TAILQ_FOREACH(opt, opts, link) {
341 		/*
342 		 * Check that this option hasn't been redefined
343 		 * nor cancelled with a "no" mount option.
344 		 */
345 		opt2 = TAILQ_FIRST(toopts);
346 		while (opt2 != NULL) {
347 			if (strcmp(opt2->name, opt->name) == 0)
348 				goto next;
349 			if (strncmp(opt2->name, "no", 2) == 0 &&
350 			    strcmp(opt2->name + 2, opt->name) == 0) {
351 				vfs_freeopt(toopts, opt2);
352 				goto next;
353 			}
354 			opt2 = TAILQ_NEXT(opt2, link);
355 		}
356 		/* We want this option, duplicate it. */
357 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
358 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
359 		strcpy(new->name, opt->name);
360 		if (opt->len != 0) {
361 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
362 			bcopy(opt->value, new->value, opt->len);
363 		} else {
364 			new->value = NULL;
365 		}
366 		new->len = opt->len;
367 		TAILQ_INSERT_TAIL(toopts, new, link);
368 next:
369 		continue;
370 	}
371 }
372 
373 /*
374  * ---------------------------------------------------------------------
375  * Mount a filesystem
376  */
377 int
378 nmount(td, uap)
379 	struct thread *td;
380 	struct nmount_args /* {
381 		struct iovec *iovp;
382 		unsigned int iovcnt;
383 		int flags;
384 	} */ *uap;
385 {
386 	struct uio *auio;
387 	struct iovec *iov;
388 	unsigned int i;
389 	int error;
390 	u_int iovcnt;
391 
392 	AUDIT_ARG(fflags, uap->flags);
393 
394 	/* Kick out MNT_ROOTFS early as it is legal internally */
395 	if (uap->flags & MNT_ROOTFS)
396 		return (EINVAL);
397 
398 	iovcnt = uap->iovcnt;
399 	/*
400 	 * Check that we have an even number of iovec's
401 	 * and that we have at least two options.
402 	 */
403 	if ((iovcnt & 1) || (iovcnt < 4))
404 		return (EINVAL);
405 
406 	error = copyinuio(uap->iovp, iovcnt, &auio);
407 	if (error)
408 		return (error);
409 	iov = auio->uio_iov;
410 	for (i = 0; i < iovcnt; i++) {
411 		if (iov->iov_len > MMAXOPTIONLEN) {
412 			free(auio, M_IOV);
413 			return (EINVAL);
414 		}
415 		iov++;
416 	}
417 	error = vfs_donmount(td, uap->flags, auio);
418 
419 	free(auio, M_IOV);
420 	return (error);
421 }
422 
423 /*
424  * ---------------------------------------------------------------------
425  * Various utility functions
426  */
427 
428 void
429 vfs_ref(struct mount *mp)
430 {
431 
432 	MNT_ILOCK(mp);
433 	MNT_REF(mp);
434 	MNT_IUNLOCK(mp);
435 }
436 
437 void
438 vfs_rel(struct mount *mp)
439 {
440 
441 	MNT_ILOCK(mp);
442 	MNT_REL(mp);
443 	MNT_IUNLOCK(mp);
444 }
445 
446 static int
447 mount_init(void *mem, int size, int flags)
448 {
449 	struct mount *mp;
450 
451 	mp = (struct mount *)mem;
452 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
453 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
454 	return (0);
455 }
456 
457 static void
458 mount_fini(void *mem, int size)
459 {
460 	struct mount *mp;
461 
462 	mp = (struct mount *)mem;
463 	lockdestroy(&mp->mnt_lock);
464 	mtx_destroy(&mp->mnt_mtx);
465 }
466 
467 /*
468  * Allocate and initialize the mount point struct.
469  */
470 static struct mount *
471 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
472     const char *fspath, struct thread *td)
473 {
474 	struct mount *mp;
475 
476 	mp = uma_zalloc(mount_zone, M_WAITOK);
477 	bzero(&mp->mnt_startzero,
478 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
479 	TAILQ_INIT(&mp->mnt_nvnodelist);
480 	mp->mnt_nvnodelistsize = 0;
481 	mp->mnt_ref = 0;
482 	(void) vfs_busy(mp, LK_NOWAIT, 0, td);
483 	mp->mnt_op = vfsp->vfc_vfsops;
484 	mp->mnt_vfc = vfsp;
485 	vfsp->vfc_refcount++;	/* XXX Unlocked */
486 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
487 	MNT_ILOCK(mp);
488 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
489 	MNT_IUNLOCK(mp);
490 	mp->mnt_gen++;
491 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
492 	mp->mnt_vnodecovered = vp;
493 	mp->mnt_cred = crdup(td->td_ucred);
494 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
495 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
496 	mp->mnt_iosize_max = DFLTPHYS;
497 #ifdef MAC
498 	mac_init_mount(mp);
499 	mac_create_mount(td->td_ucred, mp);
500 #endif
501 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
502 	return (mp);
503 }
504 
505 /*
506  * Destroy the mount struct previously allocated by vfs_mount_alloc().
507  */
508 static void
509 vfs_mount_destroy(struct mount *mp)
510 {
511 	int i;
512 
513 	MNT_ILOCK(mp);
514 	for (i = 0; mp->mnt_ref && i < 3; i++)
515 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
516 	/*
517 	 * This will always cause a 3 second delay in rebooting due to
518 	 * refs on the root mountpoint that never go away.  Most of these
519 	 * are held by init which never exits.
520 	 */
521 	if (i == 3 && (!rebooting || bootverbose))
522 		printf("Mount point %s had %d dangling refs\n",
523 		    mp->mnt_stat.f_mntonname, mp->mnt_ref);
524 	if (mp->mnt_holdcnt != 0) {
525 		printf("Waiting for mount point to be unheld\n");
526 		while (mp->mnt_holdcnt != 0) {
527 			mp->mnt_holdcntwaiters++;
528 			msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
529 			       PZERO, "mntdestroy", 0);
530 			mp->mnt_holdcntwaiters--;
531 		}
532 		printf("mount point unheld\n");
533 	}
534 	if (mp->mnt_writeopcount > 0) {
535 		printf("Waiting for mount point write ops\n");
536 		while (mp->mnt_writeopcount > 0) {
537 			mp->mnt_kern_flag |= MNTK_SUSPEND;
538 			msleep(&mp->mnt_writeopcount,
539 			       MNT_MTX(mp),
540 			       PZERO, "mntdestroy2", 0);
541 		}
542 		printf("mount point write ops completed\n");
543 	}
544 	if (mp->mnt_secondary_writes > 0) {
545 		printf("Waiting for mount point secondary write ops\n");
546 		while (mp->mnt_secondary_writes > 0) {
547 			mp->mnt_kern_flag |= MNTK_SUSPEND;
548 			msleep(&mp->mnt_secondary_writes,
549 			       MNT_MTX(mp),
550 			       PZERO, "mntdestroy3", 0);
551 		}
552 		printf("mount point secondary write ops completed\n");
553 	}
554 	MNT_IUNLOCK(mp);
555 	mp->mnt_vfc->vfc_refcount--;
556 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
557 		struct vnode *vp;
558 
559 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
560 			vprint("", vp);
561 		panic("unmount: dangling vnode");
562 	}
563 	MNT_ILOCK(mp);
564 	if (mp->mnt_kern_flag & MNTK_MWAIT)
565 		wakeup(mp);
566 	if (mp->mnt_writeopcount != 0)
567 		panic("vfs_mount_destroy: nonzero writeopcount");
568 	if (mp->mnt_secondary_writes != 0)
569 		panic("vfs_mount_destroy: nonzero secondary_writes");
570 	if (mp->mnt_nvnodelistsize != 0)
571 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
572 	mp->mnt_writeopcount = -1000;
573 	mp->mnt_nvnodelistsize = -1000;
574 	mp->mnt_secondary_writes = -1000;
575 	MNT_IUNLOCK(mp);
576 #ifdef MAC
577 	mac_destroy_mount(mp);
578 #endif
579 	if (mp->mnt_opt != NULL)
580 		vfs_freeopts(mp->mnt_opt);
581 	crfree(mp->mnt_cred);
582 	uma_zfree(mount_zone, mp);
583 }
584 
585 static int
586 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
587 {
588 	struct vfsoptlist *optlist;
589 	struct vfsopt *opt, *noro_opt;
590 	char *fstype, *fspath, *errmsg;
591 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
592 	int has_rw, has_noro;
593 
594 	errmsg = NULL;
595 	errmsg_len = 0;
596 	errmsg_pos = -1;
597 	has_rw = 0;
598 	has_noro = 0;
599 
600 	error = vfs_buildopts(fsoptions, &optlist);
601 	if (error)
602 		return (error);
603 
604 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
605 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
606 
607 	/*
608 	 * We need these two options before the others,
609 	 * and they are mandatory for any filesystem.
610 	 * Ensure they are NUL terminated as well.
611 	 */
612 	fstypelen = 0;
613 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
614 	if (error || fstype[fstypelen - 1] != '\0') {
615 		error = EINVAL;
616 		if (errmsg != NULL)
617 			strncpy(errmsg, "Invalid fstype", errmsg_len);
618 		goto bail;
619 	}
620 	fspathlen = 0;
621 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
622 	if (error || fspath[fspathlen - 1] != '\0') {
623 		error = EINVAL;
624 		if (errmsg != NULL)
625 			strncpy(errmsg, "Invalid fspath", errmsg_len);
626 		goto bail;
627 	}
628 
629 	/*
630 	 * We need to see if we have the "update" option
631 	 * before we call vfs_domount(), since vfs_domount() has special
632 	 * logic based on MNT_UPDATE.  This is very important
633 	 * when we want to update the root filesystem.
634 	 */
635 	TAILQ_FOREACH(opt, optlist, link) {
636 		if (strcmp(opt->name, "update") == 0)
637 			fsflags |= MNT_UPDATE;
638 		else if (strcmp(opt->name, "async") == 0)
639 			fsflags |= MNT_ASYNC;
640 		else if (strcmp(opt->name, "force") == 0)
641 			fsflags |= MNT_FORCE;
642 		else if (strcmp(opt->name, "multilabel") == 0)
643 			fsflags |= MNT_MULTILABEL;
644 		else if (strcmp(opt->name, "noasync") == 0)
645 			fsflags &= ~MNT_ASYNC;
646 		else if (strcmp(opt->name, "noatime") == 0)
647 			fsflags |= MNT_NOATIME;
648 		else if (strcmp(opt->name, "noclusterr") == 0)
649 			fsflags |= MNT_NOCLUSTERR;
650 		else if (strcmp(opt->name, "noclusterw") == 0)
651 			fsflags |= MNT_NOCLUSTERW;
652 		else if (strcmp(opt->name, "noexec") == 0)
653 			fsflags |= MNT_NOEXEC;
654 		else if (strcmp(opt->name, "nosuid") == 0)
655 			fsflags |= MNT_NOSUID;
656 		else if (strcmp(opt->name, "nosymfollow") == 0)
657 			fsflags |= MNT_NOSYMFOLLOW;
658 		else if (strcmp(opt->name, "noro") == 0) {
659 			fsflags &= ~MNT_RDONLY;
660 			has_noro = 1;
661 		}
662 		else if (strcmp(opt->name, "rw") == 0) {
663 			fsflags &= ~MNT_RDONLY;
664 			has_rw = 1;
665 		}
666 		else if (strcmp(opt->name, "ro") == 0 ||
667 		    strcmp(opt->name, "rdonly") == 0)
668 			fsflags |= MNT_RDONLY;
669 		else if (strcmp(opt->name, "snapshot") == 0)
670 			fsflags |= MNT_SNAPSHOT;
671 		else if (strcmp(opt->name, "suiddir") == 0)
672 			fsflags |= MNT_SUIDDIR;
673 		else if (strcmp(opt->name, "sync") == 0)
674 			fsflags |= MNT_SYNCHRONOUS;
675 		else if (strcmp(opt->name, "union") == 0)
676 			fsflags |= MNT_UNION;
677 	}
678 
679 	/*
680 	 * If "rw" was specified as a mount option, and we
681 	 * are trying to update a mount-point from "ro" to "rw",
682 	 * we need a mount option "noro", since in vfs_mergeopts(),
683 	 * "noro" will cancel "ro", but "rw" will not do anything.
684 	 */
685 	if (has_rw && !has_noro) {
686 		noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
687 		noro_opt->name = strdup("noro", M_MOUNT);
688 		noro_opt->value = NULL;
689 		noro_opt->len = 0;
690 		TAILQ_INSERT_TAIL(optlist, noro_opt, link);
691 	}
692 
693 	/*
694 	 * Be ultra-paranoid about making sure the type and fspath
695 	 * variables will fit in our mp buffers, including the
696 	 * terminating NUL.
697 	 */
698 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
699 		error = ENAMETOOLONG;
700 		goto bail;
701 	}
702 
703 	mtx_lock(&Giant);
704 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
705 	mtx_unlock(&Giant);
706 bail:
707 	/* copyout the errmsg */
708 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
709 	    && errmsg_len > 0 && errmsg != NULL) {
710 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
711 			bcopy(errmsg,
712 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
713 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
714 		} else {
715 			copyout(errmsg,
716 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
717 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
718 		}
719 	}
720 
721 	if (error != 0)
722 		vfs_freeopts(optlist);
723 	return (error);
724 }
725 
726 /*
727  * ---------------------------------------------------------------------
728  * Old mount API.
729  */
730 #ifndef _SYS_SYSPROTO_H_
731 struct mount_args {
732 	char	*type;
733 	char	*path;
734 	int	flags;
735 	caddr_t	data;
736 };
737 #endif
738 /* ARGSUSED */
739 int
740 mount(td, uap)
741 	struct thread *td;
742 	struct mount_args /* {
743 		char *type;
744 		char *path;
745 		int flags;
746 		caddr_t data;
747 	} */ *uap;
748 {
749 	char *fstype;
750 	struct vfsconf *vfsp = NULL;
751 	struct mntarg *ma = NULL;
752 	int error;
753 
754 	AUDIT_ARG(fflags, uap->flags);
755 
756 	/* Kick out MNT_ROOTFS early as it is legal internally */
757 	uap->flags &= ~MNT_ROOTFS;
758 
759 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
760 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
761 	if (error) {
762 		free(fstype, M_TEMP);
763 		return (error);
764 	}
765 
766 	AUDIT_ARG(text, fstype);
767 	mtx_lock(&Giant);
768 	vfsp = vfs_byname_kld(fstype, td, &error);
769 	free(fstype, M_TEMP);
770 	if (vfsp == NULL) {
771 		mtx_unlock(&Giant);
772 		return (ENOENT);
773 	}
774 	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
775 		mtx_unlock(&Giant);
776 		return (EOPNOTSUPP);
777 	}
778 
779 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
780 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
781 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
782 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
783 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
784 
785 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
786 	mtx_unlock(&Giant);
787 	return (error);
788 }
789 
790 
791 /*
792  * vfs_domount(): actually attempt a filesystem mount.
793  */
794 static int
795 vfs_domount(
796 	struct thread *td,	/* Calling thread. */
797 	const char *fstype,	/* Filesystem type. */
798 	char *fspath,		/* Mount path. */
799 	int fsflags,		/* Flags common to all filesystems. */
800 	void *fsdata		/* Options local to the filesystem. */
801 	)
802 {
803 	struct vnode *vp;
804 	struct mount *mp;
805 	struct vfsconf *vfsp;
806 	struct export_args export;
807 	int error, flag = 0;
808 	struct vattr va;
809 	struct nameidata nd;
810 
811 	mtx_assert(&Giant, MA_OWNED);
812 	/*
813 	 * Be ultra-paranoid about making sure the type and fspath
814 	 * variables will fit in our mp buffers, including the
815 	 * terminating NUL.
816 	 */
817 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
818 		return (ENAMETOOLONG);
819 
820 	if (jailed(td->td_ucred))
821 		return (EPERM);
822 	if (usermount == 0) {
823 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
824 			return (error);
825 	}
826 
827 	/*
828 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
829 	 */
830 	if (fsflags & MNT_EXPORTED) {
831 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
832 		if (error)
833 			return (error);
834 	}
835 	if (fsflags & MNT_SUIDDIR) {
836 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
837 		if (error)
838 			return (error);
839 
840 	}
841 	/*
842 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
843 	 */
844 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
845 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
846 			fsflags |= MNT_NOSUID | MNT_USER;
847 	}
848 
849 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
850 	vfsp = NULL;
851 	if ((fsflags & MNT_UPDATE) == 0) {
852 		/* Don't try to load KLDs if we're mounting the root. */
853 		if (fsflags & MNT_ROOTFS)
854 			vfsp = vfs_byname(fstype);
855 		else
856 			vfsp = vfs_byname_kld(fstype, td, &error);
857 		if (vfsp == NULL)
858 			return (ENODEV);
859 	}
860 	/*
861 	 * Get vnode to be covered
862 	 */
863 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
864 	    fspath, td);
865 	if ((error = namei(&nd)) != 0)
866 		return (error);
867 	NDFREE(&nd, NDF_ONLY_PNBUF);
868 	vp = nd.ni_vp;
869 	if (fsflags & MNT_UPDATE) {
870 		if ((vp->v_vflag & VV_ROOT) == 0) {
871 			vput(vp);
872 			return (EINVAL);
873 		}
874 		mp = vp->v_mount;
875 		MNT_ILOCK(mp);
876 		flag = mp->mnt_flag;
877 		/*
878 		 * We only allow the filesystem to be reloaded if it
879 		 * is currently mounted read-only.
880 		 */
881 		if ((fsflags & MNT_RELOAD) &&
882 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
883 			MNT_IUNLOCK(mp);
884 			vput(vp);
885 			return (EOPNOTSUPP);	/* Needs translation */
886 		}
887 		MNT_IUNLOCK(mp);
888 		/*
889 		 * Only privileged root, or (if MNT_USER is set) the user that
890 		 * did the original mount is permitted to update it.
891 		 */
892 		error = vfs_suser(mp, td);
893 		if (error) {
894 			vput(vp);
895 			return (error);
896 		}
897 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
898 			vput(vp);
899 			return (EBUSY);
900 		}
901 		VI_LOCK(vp);
902 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
903 		    vp->v_mountedhere != NULL) {
904 			VI_UNLOCK(vp);
905 			vfs_unbusy(mp, td);
906 			vput(vp);
907 			return (EBUSY);
908 		}
909 		vp->v_iflag |= VI_MOUNT;
910 		VI_UNLOCK(vp);
911 		MNT_ILOCK(mp);
912 		mp->mnt_flag |= fsflags &
913 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
914 		MNT_IUNLOCK(mp);
915 		VOP_UNLOCK(vp, 0, td);
916 		mp->mnt_optnew = fsdata;
917 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
918 	} else {
919 		/*
920 		 * If the user is not root, ensure that they own the directory
921 		 * onto which we are attempting to mount.
922 		 */
923 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
924 		if (error) {
925 			vput(vp);
926 			return (error);
927 		}
928 		if (va.va_uid != td->td_ucred->cr_uid) {
929 			error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
930 			    SUSER_ALLOWJAIL);
931 			if (error) {
932 				vput(vp);
933 				return (error);
934 			}
935 		}
936 		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
937 		if (error != 0) {
938 			vput(vp);
939 			return (error);
940 		}
941 		if (vp->v_type != VDIR) {
942 			vput(vp);
943 			return (ENOTDIR);
944 		}
945 		VI_LOCK(vp);
946 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
947 		    vp->v_mountedhere != NULL) {
948 			VI_UNLOCK(vp);
949 			vput(vp);
950 			return (EBUSY);
951 		}
952 		vp->v_iflag |= VI_MOUNT;
953 		VI_UNLOCK(vp);
954 
955 		/*
956 		 * Allocate and initialize the filesystem.
957 		 */
958 		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
959 		VOP_UNLOCK(vp, 0, td);
960 
961 		/* XXXMAC: pass to vfs_mount_alloc? */
962 		mp->mnt_optnew = fsdata;
963 	}
964 
965 	/*
966 	 * Set the mount level flags.
967 	 */
968 	MNT_ILOCK(mp);
969 	mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
970 		(fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
971 			    MNT_RDONLY));
972 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
973 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
974 	MNT_IUNLOCK(mp);
975 	/*
976 	 * Mount the filesystem.
977 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
978 	 * get.  No freeing of cn_pnbuf.
979 	 */
980         error = VFS_MOUNT(mp, td);
981 
982 	/*
983 	 * Process the export option only if we are
984 	 * updating mount options.
985 	 */
986 	if (!error && (fsflags & MNT_UPDATE)) {
987 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
988 		    sizeof(export)) == 0)
989 			error = vfs_export(mp, &export);
990 	}
991 
992 	if (!error) {
993 		if (mp->mnt_opt != NULL)
994 			vfs_freeopts(mp->mnt_opt);
995 		mp->mnt_opt = mp->mnt_optnew;
996 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
997 	}
998 	/*
999 	 * Prevent external consumers of mount options from reading
1000 	 * mnt_optnew.
1001 	*/
1002 	mp->mnt_optnew = NULL;
1003 	if (mp->mnt_flag & MNT_UPDATE) {
1004 		MNT_ILOCK(mp);
1005 		if (error)
1006 			mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1007 				(flag & ~MNT_QUOTA);
1008 		else
1009 			mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD |
1010 					  MNT_FORCE | MNT_SNAPSHOT);
1011 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1012 			mp->mnt_kern_flag |= MNTK_ASYNC;
1013 		else
1014 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1015 		MNT_IUNLOCK(mp);
1016 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1017 			if (mp->mnt_syncer == NULL)
1018 				error = vfs_allocate_syncvnode(mp);
1019 		} else {
1020 			if (mp->mnt_syncer != NULL)
1021 				vrele(mp->mnt_syncer);
1022 			mp->mnt_syncer = NULL;
1023 		}
1024 		vfs_unbusy(mp, td);
1025 		VI_LOCK(vp);
1026 		vp->v_iflag &= ~VI_MOUNT;
1027 		VI_UNLOCK(vp);
1028 		vrele(vp);
1029 		return (error);
1030 	}
1031 	MNT_ILOCK(mp);
1032 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1033 		mp->mnt_kern_flag |= MNTK_ASYNC;
1034 	else
1035 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1036 	MNT_IUNLOCK(mp);
1037 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1038 	/*
1039 	 * Put the new filesystem on the mount list after root.
1040 	 */
1041 	cache_purge(vp);
1042 	if (!error) {
1043 		struct vnode *newdp;
1044 
1045 		VI_LOCK(vp);
1046 		vp->v_iflag &= ~VI_MOUNT;
1047 		VI_UNLOCK(vp);
1048 		vp->v_mountedhere = mp;
1049 		mtx_lock(&mountlist_mtx);
1050 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1051 		mtx_unlock(&mountlist_mtx);
1052 		vfs_event_signal(NULL, VQ_MOUNT, 0);
1053 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1054 			panic("mount: lost mount");
1055 		mountcheckdirs(vp, newdp);
1056 		vput(newdp);
1057 		VOP_UNLOCK(vp, 0, td);
1058 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1059 			error = vfs_allocate_syncvnode(mp);
1060 		vfs_unbusy(mp, td);
1061 		if (error)
1062 			vrele(vp);
1063 	} else {
1064 		VI_LOCK(vp);
1065 		vp->v_iflag &= ~VI_MOUNT;
1066 		VI_UNLOCK(vp);
1067 		vfs_unbusy(mp, td);
1068 		vfs_mount_destroy(mp);
1069 		vput(vp);
1070 	}
1071 	return (error);
1072 }
1073 
1074 /*
1075  * ---------------------------------------------------------------------
1076  * Unmount a filesystem.
1077  *
1078  * Note: unmount takes a path to the vnode mounted on as argument,
1079  * not special file (as before).
1080  */
1081 #ifndef _SYS_SYSPROTO_H_
1082 struct unmount_args {
1083 	char	*path;
1084 	int	flags;
1085 };
1086 #endif
1087 /* ARGSUSED */
1088 int
1089 unmount(td, uap)
1090 	struct thread *td;
1091 	register struct unmount_args /* {
1092 		char *path;
1093 		int flags;
1094 	} */ *uap;
1095 {
1096 	struct mount *mp;
1097 	char *pathbuf;
1098 	int error, id0, id1;
1099 
1100 	if (jailed(td->td_ucred))
1101 		return (EPERM);
1102 	if (usermount == 0) {
1103 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1104 		if (error)
1105 			return (error);
1106 	}
1107 
1108 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1109 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1110 	if (error) {
1111 		free(pathbuf, M_TEMP);
1112 		return (error);
1113 	}
1114 	AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1115 	mtx_lock(&Giant);
1116 	if (uap->flags & MNT_BYFSID) {
1117 		/* Decode the filesystem ID. */
1118 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1119 			mtx_unlock(&Giant);
1120 			free(pathbuf, M_TEMP);
1121 			return (EINVAL);
1122 		}
1123 
1124 		mtx_lock(&mountlist_mtx);
1125 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1126 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1127 			    mp->mnt_stat.f_fsid.val[1] == id1)
1128 				break;
1129 		}
1130 		mtx_unlock(&mountlist_mtx);
1131 	} else {
1132 		mtx_lock(&mountlist_mtx);
1133 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1134 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1135 				break;
1136 		}
1137 		mtx_unlock(&mountlist_mtx);
1138 	}
1139 	free(pathbuf, M_TEMP);
1140 	if (mp == NULL) {
1141 		/*
1142 		 * Previously we returned ENOENT for a nonexistent path and
1143 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1144 		 * now, so in the !MNT_BYFSID case return the more likely
1145 		 * EINVAL for compatibility.
1146 		 */
1147 		mtx_unlock(&Giant);
1148 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1149 	}
1150 
1151 	/*
1152 	 * Don't allow unmounting the root filesystem.
1153 	 */
1154 	if (mp->mnt_flag & MNT_ROOTFS) {
1155 		mtx_unlock(&Giant);
1156 		return (EINVAL);
1157 	}
1158 	error = dounmount(mp, uap->flags, td);
1159 	mtx_unlock(&Giant);
1160 	return (error);
1161 }
1162 
1163 /*
1164  * Do the actual filesystem unmount.
1165  */
1166 int
1167 dounmount(mp, flags, td)
1168 	struct mount *mp;
1169 	int flags;
1170 	struct thread *td;
1171 {
1172 	struct vnode *coveredvp, *fsrootvp;
1173 	int error;
1174 	int async_flag;
1175 	int mnt_gen_r;
1176 
1177 	mtx_assert(&Giant, MA_OWNED);
1178 
1179 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1180 		mnt_gen_r = mp->mnt_gen;
1181 		VI_LOCK(coveredvp);
1182 		vholdl(coveredvp);
1183 		error = vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK, td);
1184 		vdrop(coveredvp);
1185 		/*
1186 		 * Check for mp being unmounted while waiting for the
1187 		 * covered vnode lock.
1188 		 */
1189 		if (error)
1190 			return (error);
1191 		if (coveredvp->v_mountedhere != mp ||
1192 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1193 			VOP_UNLOCK(coveredvp, 0, td);
1194 			return (EBUSY);
1195 		}
1196 	}
1197 	/*
1198 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1199 	 * original mount is permitted to unmount this filesystem.
1200 	 */
1201 	error = vfs_suser(mp, td);
1202 	if (error) {
1203 		if (coveredvp)
1204 			VOP_UNLOCK(coveredvp, 0, td);
1205 		return (error);
1206 	}
1207 
1208 	MNT_ILOCK(mp);
1209 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1210 		MNT_IUNLOCK(mp);
1211 		if (coveredvp)
1212 			VOP_UNLOCK(coveredvp, 0, td);
1213 		return (EBUSY);
1214 	}
1215 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1216 	/* Allow filesystems to detect that a forced unmount is in progress. */
1217 	if (flags & MNT_FORCE)
1218 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1219 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1220 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1221 	if (error) {
1222 		MNT_ILOCK(mp);
1223 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1224 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1225 			wakeup(mp);
1226 		MNT_IUNLOCK(mp);
1227 		if (coveredvp)
1228 			VOP_UNLOCK(coveredvp, 0, td);
1229 		return (error);
1230 	}
1231 	vn_start_write(NULL, &mp, V_WAIT);
1232 
1233 	if (mp->mnt_flag & MNT_EXPUBLIC)
1234 		vfs_setpublicfs(NULL, NULL, NULL);
1235 
1236 	vfs_msync(mp, MNT_WAIT);
1237 	MNT_ILOCK(mp);
1238 	async_flag = mp->mnt_flag & MNT_ASYNC;
1239 	mp->mnt_flag &= ~MNT_ASYNC;
1240 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1241 	MNT_IUNLOCK(mp);
1242 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1243 	if (mp->mnt_syncer != NULL)
1244 		vrele(mp->mnt_syncer);
1245 	/*
1246 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1247 	 * vnode to the covered vnode.  For non-forced unmounts we want
1248 	 * such references to cause an EBUSY error.
1249 	 */
1250 	if ((flags & MNT_FORCE) &&
1251 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1252 		if (mp->mnt_vnodecovered != NULL)
1253 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1254 		if (fsrootvp == rootvnode) {
1255 			vrele(rootvnode);
1256 			rootvnode = NULL;
1257 		}
1258 		vput(fsrootvp);
1259 	}
1260 	if (((mp->mnt_flag & MNT_RDONLY) ||
1261 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1262 	    (flags & MNT_FORCE)) {
1263 		error = VFS_UNMOUNT(mp, flags, td);
1264 	}
1265 	vn_finished_write(mp);
1266 	if (error) {
1267 		/* Undo cdir/rdir and rootvnode changes made above. */
1268 		if ((flags & MNT_FORCE) &&
1269 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1270 			if (mp->mnt_vnodecovered != NULL)
1271 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1272 			if (rootvnode == NULL) {
1273 				rootvnode = fsrootvp;
1274 				vref(rootvnode);
1275 			}
1276 			vput(fsrootvp);
1277 		}
1278 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1279 			(void) vfs_allocate_syncvnode(mp);
1280 		MNT_ILOCK(mp);
1281 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1282 		mp->mnt_flag |= async_flag;
1283 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1284 			mp->mnt_kern_flag |= MNTK_ASYNC;
1285 		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1286 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1287 			wakeup(mp);
1288 		MNT_IUNLOCK(mp);
1289 		if (coveredvp)
1290 			VOP_UNLOCK(coveredvp, 0, td);
1291 		return (error);
1292 	}
1293 	mtx_lock(&mountlist_mtx);
1294 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1295 	mtx_unlock(&mountlist_mtx);
1296 	if (coveredvp != NULL) {
1297 		coveredvp->v_mountedhere = NULL;
1298 		vput(coveredvp);
1299 	}
1300 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1301 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1302 	vfs_mount_destroy(mp);
1303 	return (0);
1304 }
1305 
1306 /*
1307  * ---------------------------------------------------------------------
1308  * Mounting of root filesystem
1309  *
1310  */
1311 
1312 struct root_hold_token {
1313 	const char 			*who;
1314 	LIST_ENTRY(root_hold_token)	list;
1315 };
1316 
1317 static LIST_HEAD(, root_hold_token)	root_holds =
1318     LIST_HEAD_INITIALIZER(&root_holds);
1319 
1320 struct root_hold_token *
1321 root_mount_hold(const char *identifier)
1322 {
1323 	struct root_hold_token *h;
1324 
1325 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1326 	h->who = identifier;
1327 	mtx_lock(&mountlist_mtx);
1328 	LIST_INSERT_HEAD(&root_holds, h, list);
1329 	mtx_unlock(&mountlist_mtx);
1330 	return (h);
1331 }
1332 
1333 void
1334 root_mount_rel(struct root_hold_token *h)
1335 {
1336 
1337 	mtx_lock(&mountlist_mtx);
1338 	LIST_REMOVE(h, list);
1339 	wakeup(&root_holds);
1340 	mtx_unlock(&mountlist_mtx);
1341 	free(h, M_DEVBUF);
1342 }
1343 
1344 static void
1345 root_mount_wait(void)
1346 {
1347 	struct root_hold_token *h;
1348 
1349 	for (;;) {
1350 		DROP_GIANT();
1351 		g_waitidle();
1352 		PICKUP_GIANT();
1353 		mtx_lock(&mountlist_mtx);
1354 		if (LIST_EMPTY(&root_holds)) {
1355 			mtx_unlock(&mountlist_mtx);
1356 			break;
1357 		}
1358 		printf("Root mount waiting for:");
1359 		LIST_FOREACH(h, &root_holds, list)
1360 			printf(" %s", h->who);
1361 		printf("\n");
1362 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1363 		    hz);
1364 	}
1365 }
1366 
1367 static void
1368 set_rootvnode(struct thread *td)
1369 {
1370 	struct proc *p;
1371 
1372 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1373 		panic("Cannot find root vnode");
1374 
1375 	p = td->td_proc;
1376 	FILEDESC_LOCK(p->p_fd);
1377 
1378 	if (p->p_fd->fd_cdir != NULL)
1379 		vrele(p->p_fd->fd_cdir);
1380 	p->p_fd->fd_cdir = rootvnode;
1381 	VREF(rootvnode);
1382 
1383 	if (p->p_fd->fd_rdir != NULL)
1384 		vrele(p->p_fd->fd_rdir);
1385 	p->p_fd->fd_rdir = rootvnode;
1386 	VREF(rootvnode);
1387 
1388 	FILEDESC_UNLOCK(p->p_fd);
1389 
1390 	VOP_UNLOCK(rootvnode, 0, td);
1391 }
1392 
1393 /*
1394  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1395  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1396  */
1397 
1398 static void
1399 devfs_first(void)
1400 {
1401 	struct thread *td = curthread;
1402 	struct vfsoptlist *opts;
1403 	struct vfsconf *vfsp;
1404 	struct mount *mp = NULL;
1405 	int error;
1406 
1407 	vfsp = vfs_byname("devfs");
1408 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1409 	if (vfsp == NULL)
1410 		return;
1411 
1412 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1413 
1414 	error = VFS_MOUNT(mp, td);
1415 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1416 	if (error)
1417 		return;
1418 
1419 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1420 	TAILQ_INIT(opts);
1421 	mp->mnt_opt = opts;
1422 
1423 	mtx_lock(&mountlist_mtx);
1424 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1425 	mtx_unlock(&mountlist_mtx);
1426 
1427 	set_rootvnode(td);
1428 
1429 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1430 	if (error)
1431 		printf("kern_symlink /dev -> / returns %d\n", error);
1432 }
1433 
1434 /*
1435  * Surgically move our devfs to be mounted on /dev.
1436  */
1437 
1438 static void
1439 devfs_fixup(struct thread *td)
1440 {
1441 	struct nameidata nd;
1442 	int error;
1443 	struct vnode *vp, *dvp;
1444 	struct mount *mp;
1445 
1446 	/* Remove our devfs mount from the mountlist and purge the cache */
1447 	mtx_lock(&mountlist_mtx);
1448 	mp = TAILQ_FIRST(&mountlist);
1449 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1450 	mtx_unlock(&mountlist_mtx);
1451 	cache_purgevfs(mp);
1452 
1453 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1454 	VI_LOCK(dvp);
1455 	dvp->v_iflag &= ~VI_MOUNT;
1456 	dvp->v_mountedhere = NULL;
1457 	VI_UNLOCK(dvp);
1458 
1459 	/* Set up the real rootvnode, and purge the cache */
1460 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1461 	set_rootvnode(td);
1462 	cache_purgevfs(rootvnode->v_mount);
1463 
1464 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1465 	error = namei(&nd);
1466 	if (error) {
1467 		printf("Lookup of /dev for devfs, error: %d\n", error);
1468 		return;
1469 	}
1470 	NDFREE(&nd, NDF_ONLY_PNBUF);
1471 	vp = nd.ni_vp;
1472 	if (vp->v_type != VDIR) {
1473 		vput(vp);
1474 	}
1475 	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1476 	if (error) {
1477 		vput(vp);
1478 	}
1479 	cache_purge(vp);
1480 	mp->mnt_vnodecovered = vp;
1481 	vp->v_mountedhere = mp;
1482 	mtx_lock(&mountlist_mtx);
1483 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1484 	mtx_unlock(&mountlist_mtx);
1485 	VOP_UNLOCK(vp, 0, td);
1486 	vput(dvp);
1487 	vfs_unbusy(mp, td);
1488 
1489 	/* Unlink the no longer needed /dev/dev -> / symlink */
1490 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1491 }
1492 
1493 /*
1494  * Report errors during filesystem mounting.
1495  */
1496 void
1497 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1498 {
1499 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1500 	va_list ap;
1501 	int error, len;
1502 	char *errmsg;
1503 
1504 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1505 	if (error || errmsg == NULL || len <= 0)
1506 		return;
1507 
1508 	va_start(ap, fmt);
1509 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1510 	va_end(ap);
1511 }
1512 
1513 /*
1514  * Find and mount the root filesystem
1515  */
1516 void
1517 vfs_mountroot(void)
1518 {
1519 	char *cp;
1520 	int error, i, asked = 0;
1521 
1522 	root_mount_wait();
1523 
1524 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1525 	    NULL, NULL, mount_init, mount_fini,
1526 	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1527 	devfs_first();
1528 
1529 	/*
1530 	 * We are booted with instructions to prompt for the root filesystem.
1531 	 */
1532 	if (boothowto & RB_ASKNAME) {
1533 		if (!vfs_mountroot_ask())
1534 			return;
1535 		asked = 1;
1536 	}
1537 
1538 	/*
1539 	 * The root filesystem information is compiled in, and we are
1540 	 * booted with instructions to use it.
1541 	 */
1542 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1543 		if (!vfs_mountroot_try(ctrootdevname))
1544 			return;
1545 		ctrootdevname = NULL;
1546 	}
1547 
1548 	/*
1549 	 * We've been given the generic "use CDROM as root" flag.  This is
1550 	 * necessary because one media may be used in many different
1551 	 * devices, so we need to search for them.
1552 	 */
1553 	if (boothowto & RB_CDROM) {
1554 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1555 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1556 				return;
1557 		}
1558 	}
1559 
1560 	/*
1561 	 * Try to use the value read by the loader from /etc/fstab, or
1562 	 * supplied via some other means.  This is the preferred
1563 	 * mechanism.
1564 	 */
1565 	cp = getenv("vfs.root.mountfrom");
1566 	if (cp != NULL) {
1567 		error = vfs_mountroot_try(cp);
1568 		freeenv(cp);
1569 		if (!error)
1570 			return;
1571 	}
1572 
1573 	/*
1574 	 * Try values that may have been computed by code during boot
1575 	 */
1576 	if (!vfs_mountroot_try(rootdevnames[0]))
1577 		return;
1578 	if (!vfs_mountroot_try(rootdevnames[1]))
1579 		return;
1580 
1581 	/*
1582 	 * If we (still) have a compiled-in default, try it.
1583 	 */
1584 	if (ctrootdevname != NULL)
1585 		if (!vfs_mountroot_try(ctrootdevname))
1586 			return;
1587 	/*
1588 	 * Everything so far has failed, prompt on the console if we haven't
1589 	 * already tried that.
1590 	 */
1591 	if (!asked)
1592 		if (!vfs_mountroot_ask())
1593 			return;
1594 
1595 	panic("Root mount failed, startup aborted.");
1596 }
1597 
1598 /*
1599  * Mount (mountfrom) as the root filesystem.
1600  */
1601 static int
1602 vfs_mountroot_try(const char *mountfrom)
1603 {
1604 	struct mount	*mp;
1605 	char		*vfsname, *path;
1606 	time_t		timebase;
1607 	int		error;
1608 	char		patt[32];
1609 
1610 	vfsname = NULL;
1611 	path    = NULL;
1612 	mp      = NULL;
1613 	error   = EINVAL;
1614 
1615 	if (mountfrom == NULL)
1616 		return (error);		/* don't complain */
1617 	printf("Trying to mount root from %s\n", mountfrom);
1618 
1619 	/* parse vfs name and path */
1620 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1621 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1622 	vfsname[0] = path[0] = 0;
1623 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1624 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1625 		goto out;
1626 
1627 	if (path[0] == '\0')
1628 		strcpy(path, ROOTNAME);
1629 
1630 	error = kernel_vmount(
1631 	    MNT_RDONLY | MNT_ROOTFS,
1632 	    "fstype", vfsname,
1633 	    "fspath", "/",
1634 	    "from", path,
1635 	    NULL);
1636 	if (error == 0) {
1637 		/*
1638 		 * We mount devfs prior to mounting the / FS, so the first
1639 		 * entry will typically be devfs.
1640 		 */
1641 		mp = TAILQ_FIRST(&mountlist);
1642 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1643 
1644 		/*
1645 		 * Iterate over all currently mounted file systems and use
1646 		 * the time stamp found to check and/or initialize the RTC.
1647 		 * Typically devfs has no time stamp and the only other FS
1648 		 * is the actual / FS.
1649 		 * Call inittodr() only once and pass it the largest of the
1650 		 * timestamps we encounter.
1651 		 */
1652 		timebase = 0;
1653 		do {
1654 			if (mp->mnt_time > timebase)
1655 				timebase = mp->mnt_time;
1656 			mp = TAILQ_NEXT(mp, mnt_list);
1657 		} while (mp != NULL);
1658 		inittodr(timebase);
1659 
1660 		devfs_fixup(curthread);
1661 	}
1662 out:
1663 	free(path, M_MOUNT);
1664 	free(vfsname, M_MOUNT);
1665 	return (error);
1666 }
1667 
1668 /*
1669  * ---------------------------------------------------------------------
1670  * Interactive root filesystem selection code.
1671  */
1672 
1673 static int
1674 vfs_mountroot_ask(void)
1675 {
1676 	char name[128];
1677 
1678 	for(;;) {
1679 		printf("\nManual root filesystem specification:\n");
1680 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1681 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1682 		printf("                       eg. ufs:da0s1a\n");
1683 #else
1684 		printf("                       eg. ufs:/dev/da0a\n");
1685 #endif
1686 		printf("  ?                  List valid disk boot devices\n");
1687 		printf("  <empty line>       Abort manual input\n");
1688 		printf("\nmountroot> ");
1689 		gets(name, sizeof(name), 1);
1690 		if (name[0] == '\0')
1691 			return (1);
1692 		if (name[0] == '?') {
1693 			printf("\nList of GEOM managed disk devices:\n  ");
1694 			g_dev_print();
1695 			continue;
1696 		}
1697 		if (!vfs_mountroot_try(name))
1698 			return (0);
1699 	}
1700 }
1701 
1702 /*
1703  * ---------------------------------------------------------------------
1704  * Functions for querying mount options/arguments from filesystems.
1705  */
1706 
1707 /*
1708  * Check that no unknown options are given
1709  */
1710 int
1711 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1712 {
1713 	struct vfsopt *opt;
1714 	const char **t, *p;
1715 
1716 
1717 	TAILQ_FOREACH(opt, opts, link) {
1718 		p = opt->name;
1719 		if (p[0] == 'n' && p[1] == 'o')
1720 			p += 2;
1721 		for(t = global_opts; *t != NULL; t++)
1722 			if (!strcmp(*t, p))
1723 				break;
1724 		if (*t != NULL)
1725 			continue;
1726 		for(t = legal; *t != NULL; t++)
1727 			if (!strcmp(*t, p))
1728 				break;
1729 		if (*t != NULL)
1730 			continue;
1731 		printf("mount option <%s> is unknown\n", p);
1732 		return (EINVAL);
1733 	}
1734 	return (0);
1735 }
1736 
1737 /*
1738  * Get a mount option by its name.
1739  *
1740  * Return 0 if the option was found, ENOENT otherwise.
1741  * If len is non-NULL it will be filled with the length
1742  * of the option. If buf is non-NULL, it will be filled
1743  * with the address of the option.
1744  */
1745 int
1746 vfs_getopt(opts, name, buf, len)
1747 	struct vfsoptlist *opts;
1748 	const char *name;
1749 	void **buf;
1750 	int *len;
1751 {
1752 	struct vfsopt *opt;
1753 
1754 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1755 
1756 	TAILQ_FOREACH(opt, opts, link) {
1757 		if (strcmp(name, opt->name) == 0) {
1758 			if (len != NULL)
1759 				*len = opt->len;
1760 			if (buf != NULL)
1761 				*buf = opt->value;
1762 			return (0);
1763 		}
1764 	}
1765 	return (ENOENT);
1766 }
1767 
1768 static int
1769 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1770 {
1771 	struct vfsopt *opt;
1772 	int i;
1773 
1774 	if (opts == NULL)
1775 		return (-1);
1776 
1777 	i = 0;
1778 	TAILQ_FOREACH(opt, opts, link) {
1779 		if (strcmp(name, opt->name) == 0)
1780 			return (i);
1781 		++i;
1782 	}
1783 	return (-1);
1784 }
1785 
1786 char *
1787 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1788 {
1789 	struct vfsopt *opt;
1790 
1791 	*error = 0;
1792 	TAILQ_FOREACH(opt, opts, link) {
1793 		if (strcmp(name, opt->name) != 0)
1794 			continue;
1795 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1796 			*error = EINVAL;
1797 			return (NULL);
1798 		}
1799 		return (opt->value);
1800 	}
1801 	*error = ENOENT;
1802 	return (NULL);
1803 }
1804 
1805 int
1806 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1807 {
1808 	struct vfsopt *opt;
1809 
1810 	TAILQ_FOREACH(opt, opts, link) {
1811 		if (strcmp(name, opt->name) == 0) {
1812 			if (w != NULL)
1813 				*w |= val;
1814 			return (1);
1815 		}
1816 	}
1817 	if (w != NULL)
1818 		*w &= ~val;
1819 	return (0);
1820 }
1821 
1822 int
1823 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1824 {
1825 	va_list ap;
1826 	struct vfsopt *opt;
1827 	int ret;
1828 
1829 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1830 
1831 	TAILQ_FOREACH(opt, opts, link) {
1832 		if (strcmp(name, opt->name) != 0)
1833 			continue;
1834 		if (((char *)opt->value)[opt->len - 1] != '\0')
1835 			return (0);
1836 		va_start(ap, fmt);
1837 		ret = vsscanf(opt->value, fmt, ap);
1838 		va_end(ap);
1839 		return (ret);
1840 	}
1841 	return (0);
1842 }
1843 
1844 /*
1845  * Find and copy a mount option.
1846  *
1847  * The size of the buffer has to be specified
1848  * in len, if it is not the same length as the
1849  * mount option, EINVAL is returned.
1850  * Returns ENOENT if the option is not found.
1851  */
1852 int
1853 vfs_copyopt(opts, name, dest, len)
1854 	struct vfsoptlist *opts;
1855 	const char *name;
1856 	void *dest;
1857 	int len;
1858 {
1859 	struct vfsopt *opt;
1860 
1861 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1862 
1863 	TAILQ_FOREACH(opt, opts, link) {
1864 		if (strcmp(name, opt->name) == 0) {
1865 			if (len != opt->len)
1866 				return (EINVAL);
1867 			bcopy(opt->value, dest, opt->len);
1868 			return (0);
1869 		}
1870 	}
1871 	return (ENOENT);
1872 }
1873 
1874 /*
1875  * This is a helper function for filesystems to traverse their
1876  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1877  */
1878 
1879 struct vnode *
1880 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1881 {
1882 	struct vnode *vp;
1883 
1884 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1885 
1886 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1887 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1888 	while (vp != NULL && vp->v_type == VMARKER)
1889 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1890 
1891 	/* Check if we are done */
1892 	if (vp == NULL) {
1893 		__mnt_vnode_markerfree(mvp, mp);
1894 		return (NULL);
1895 	}
1896 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1897 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1898 	return (vp);
1899 }
1900 
1901 struct vnode *
1902 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
1903 {
1904 	struct vnode *vp;
1905 
1906 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1907 
1908 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1909 	while (vp != NULL && vp->v_type == VMARKER)
1910 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1911 
1912 	/* Check if we are done */
1913 	if (vp == NULL) {
1914 		*mvp = NULL;
1915 		return (NULL);
1916 	}
1917 	mp->mnt_holdcnt++;
1918 	MNT_IUNLOCK(mp);
1919 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
1920 				       M_VNODE_MARKER,
1921 				       M_WAITOK | M_ZERO);
1922 	MNT_ILOCK(mp);
1923 	(*mvp)->v_type = VMARKER;
1924 
1925 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1926 	while (vp != NULL && vp->v_type == VMARKER)
1927 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1928 
1929 	/* Check if we are done */
1930 	if (vp == NULL) {
1931 		MNT_IUNLOCK(mp);
1932 		free(*mvp, M_VNODE_MARKER);
1933 		MNT_ILOCK(mp);
1934 		*mvp = NULL;
1935 		mp->mnt_holdcnt--;
1936 		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1937 			wakeup(&mp->mnt_holdcnt);
1938 		return (NULL);
1939 	}
1940 	mp->mnt_markercnt++;
1941 	(*mvp)->v_mount = mp;
1942 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1943 	return (vp);
1944 }
1945 
1946 
1947 void
1948 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
1949 {
1950 
1951 	if (*mvp == NULL)
1952 		return;
1953 
1954 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1955 
1956 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1957 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1958 	MNT_IUNLOCK(mp);
1959 	free(*mvp, M_VNODE_MARKER);
1960 	MNT_ILOCK(mp);
1961 	*mvp = NULL;
1962 
1963 	mp->mnt_markercnt--;
1964 	mp->mnt_holdcnt--;
1965 	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1966 		wakeup(&mp->mnt_holdcnt);
1967 }
1968 
1969 
1970 int
1971 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
1972 {
1973 	int error;
1974 
1975 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
1976 	if (sbp != &mp->mnt_stat)
1977 		*sbp = mp->mnt_stat;
1978 	return (error);
1979 }
1980 
1981 void
1982 vfs_mountedfrom(struct mount *mp, const char *from)
1983 {
1984 
1985 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1986 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1987 	    sizeof mp->mnt_stat.f_mntfromname);
1988 }
1989 
1990 /*
1991  * ---------------------------------------------------------------------
1992  * This is the api for building mount args and mounting filesystems from
1993  * inside the kernel.
1994  *
1995  * The API works by accumulation of individual args.  First error is
1996  * latched.
1997  *
1998  * XXX: should be documented in new manpage kernel_mount(9)
1999  */
2000 
2001 /* A memory allocation which must be freed when we are done */
2002 struct mntaarg {
2003 	SLIST_ENTRY(mntaarg)	next;
2004 };
2005 
2006 /* The header for the mount arguments */
2007 struct mntarg {
2008 	struct iovec *v;
2009 	int len;
2010 	int error;
2011 	SLIST_HEAD(, mntaarg)	list;
2012 };
2013 
2014 /*
2015  * Add a boolean argument.
2016  *
2017  * flag is the boolean value.
2018  * name must start with "no".
2019  */
2020 struct mntarg *
2021 mount_argb(struct mntarg *ma, int flag, const char *name)
2022 {
2023 
2024 	KASSERT(name[0] == 'n' && name[1] == 'o',
2025 	    ("mount_argb(...,%s): name must start with 'no'", name));
2026 
2027 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2028 }
2029 
2030 /*
2031  * Add an argument printf style
2032  */
2033 struct mntarg *
2034 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2035 {
2036 	va_list ap;
2037 	struct mntaarg *maa;
2038 	struct sbuf *sb;
2039 	int len;
2040 
2041 	if (ma == NULL) {
2042 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2043 		SLIST_INIT(&ma->list);
2044 	}
2045 	if (ma->error)
2046 		return (ma);
2047 
2048 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2049 	    M_MOUNT, M_WAITOK);
2050 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2051 	ma->v[ma->len].iov_len = strlen(name) + 1;
2052 	ma->len++;
2053 
2054 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
2055 	va_start(ap, fmt);
2056 	sbuf_vprintf(sb, fmt, ap);
2057 	va_end(ap);
2058 	sbuf_finish(sb);
2059 	len = sbuf_len(sb) + 1;
2060 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2061 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2062 	bcopy(sbuf_data(sb), maa + 1, len);
2063 	sbuf_delete(sb);
2064 
2065 	ma->v[ma->len].iov_base = maa + 1;
2066 	ma->v[ma->len].iov_len = len;
2067 	ma->len++;
2068 
2069 	return (ma);
2070 }
2071 
2072 /*
2073  * Add an argument which is a userland string.
2074  */
2075 struct mntarg *
2076 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2077 {
2078 	struct mntaarg *maa;
2079 	char *tbuf;
2080 
2081 	if (val == NULL)
2082 		return (ma);
2083 	if (ma == NULL) {
2084 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2085 		SLIST_INIT(&ma->list);
2086 	}
2087 	if (ma->error)
2088 		return (ma);
2089 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2090 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2091 	tbuf = (void *)(maa + 1);
2092 	ma->error = copyinstr(val, tbuf, len, NULL);
2093 	return (mount_arg(ma, name, tbuf, -1));
2094 }
2095 
2096 /*
2097  * Plain argument.
2098  *
2099  * If length is -1, use printf.
2100  */
2101 struct mntarg *
2102 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2103 {
2104 
2105 	if (ma == NULL) {
2106 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2107 		SLIST_INIT(&ma->list);
2108 	}
2109 	if (ma->error)
2110 		return (ma);
2111 
2112 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2113 	    M_MOUNT, M_WAITOK);
2114 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2115 	ma->v[ma->len].iov_len = strlen(name) + 1;
2116 	ma->len++;
2117 
2118 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2119 	if (len < 0)
2120 		ma->v[ma->len].iov_len = strlen(val) + 1;
2121 	else
2122 		ma->v[ma->len].iov_len = len;
2123 	ma->len++;
2124 	return (ma);
2125 }
2126 
2127 /*
2128  * Free a mntarg structure
2129  */
2130 static void
2131 free_mntarg(struct mntarg *ma)
2132 {
2133 	struct mntaarg *maa;
2134 
2135 	while (!SLIST_EMPTY(&ma->list)) {
2136 		maa = SLIST_FIRST(&ma->list);
2137 		SLIST_REMOVE_HEAD(&ma->list, next);
2138 		free(maa, M_MOUNT);
2139 	}
2140 	free(ma->v, M_MOUNT);
2141 	free(ma, M_MOUNT);
2142 }
2143 
2144 /*
2145  * Mount a filesystem
2146  */
2147 int
2148 kernel_mount(struct mntarg *ma, int flags)
2149 {
2150 	struct uio auio;
2151 	int error;
2152 
2153 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2154 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2155 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2156 
2157 	auio.uio_iov = ma->v;
2158 	auio.uio_iovcnt = ma->len;
2159 	auio.uio_segflg = UIO_SYSSPACE;
2160 
2161 	error = ma->error;
2162 	if (!error)
2163 		error = vfs_donmount(curthread, flags, &auio);
2164 	free_mntarg(ma);
2165 	return (error);
2166 }
2167 
2168 /*
2169  * A printflike function to mount a filesystem.
2170  */
2171 int
2172 kernel_vmount(int flags, ...)
2173 {
2174 	struct mntarg *ma = NULL;
2175 	va_list ap;
2176 	const char *cp;
2177 	const void *vp;
2178 	int error;
2179 
2180 	va_start(ap, flags);
2181 	for (;;) {
2182 		cp = va_arg(ap, const char *);
2183 		if (cp == NULL)
2184 			break;
2185 		vp = va_arg(ap, const void *);
2186 		ma = mount_arg(ma, cp, vp, -1);
2187 	}
2188 	va_end(ap);
2189 
2190 	error = kernel_mount(ma, flags);
2191 	return (error);
2192 }
2193