xref: /freebsd/sys/kern/vfs_mount.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
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/jail.h>
43 #include <sys/kernel.h>
44 #include <sys/libkern.h>
45 #include <sys/mac.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/proc.h>
51 #include <sys/filedesc.h>
52 #include <sys/reboot.h>
53 #include <sys/syscallsubr.h>
54 #include <sys/sysproto.h>
55 #include <sys/sx.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysent.h>
58 #include <sys/systm.h>
59 #include <sys/vnode.h>
60 
61 #include <geom/geom.h>
62 
63 #include <machine/stdarg.h>
64 
65 #include "opt_rootdevname.h"
66 #include "opt_ddb.h"
67 #include "opt_mac.h"
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif
72 
73 #define	ROOTNAME		"root_device"
74 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
75 
76 static int	vfs_domount(struct thread *td, const char *fstype,
77 		    char *fspath, int fsflags, void *fsdata);
78 static struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
79 		    const char *fspath, struct thread *td);
80 static int	vfs_mountroot_ask(void);
81 static int	vfs_mountroot_try(const char *mountfrom);
82 static int	vfs_donmount(struct thread *td, int fsflags,
83 		    struct uio *fsoptions);
84 static void	free_mntarg(struct mntarg *ma);
85 static void	vfs_mount_destroy(struct mount *, struct thread *);
86 static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
87 
88 static int	usermount = 0;
89 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
90     "Unprivileged users may mount and unmount file systems");
91 
92 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
93 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
94 
95 /* List of mounted filesystems. */
96 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
97 
98 /* For any iteration/modification of mountlist */
99 struct mtx mountlist_mtx;
100 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
101 
102 TAILQ_HEAD(vfsoptlist, vfsopt);
103 struct vfsopt {
104 	TAILQ_ENTRY(vfsopt) link;
105 	char	*name;
106 	void	*value;
107 	int	len;
108 };
109 
110 /*
111  * The vnode of the system's root (/ in the filesystem, without chroot
112  * active.)
113  */
114 struct vnode	*rootvnode;
115 
116 /*
117  * The root filesystem is detailed in the kernel environment variable
118  * vfs.root.mountfrom, which is expected to be in the general format
119  *
120  * <vfsname>:[<path>]
121  * vfsname   := the name of a VFS known to the kernel and capable
122  *              of being mounted as root
123  * path      := disk device name or other data used by the filesystem
124  *              to locate its physical store
125  */
126 
127 /*
128  * Global opts, taken by all filesystems
129  */
130 static const char *global_opts[] = {
131 	"errmsg",
132 	"fstype",
133 	"fspath",
134 	"rdonly",
135 	"ro",
136 	"rw",
137 	"suid",
138 	"exec",
139 	NULL
140 };
141 
142 /*
143  * The root specifiers we will try if RB_CDROM is specified.
144  */
145 static char *cdrom_rootdevnames[] = {
146 	"cd9660:cd0",
147 	"cd9660:acd0",
148 	NULL
149 };
150 
151 /* legacy find-root code */
152 char		*rootdevnames[2] = {NULL, NULL};
153 #ifndef ROOTDEVNAME
154 #  define ROOTDEVNAME NULL
155 #endif
156 static const char	*ctrootdevname = ROOTDEVNAME;
157 
158 /*
159  * ---------------------------------------------------------------------
160  * Functions for building and sanitizing the mount options
161  */
162 
163 /* Remove one mount option. */
164 static void
165 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
166 {
167 
168 	TAILQ_REMOVE(opts, opt, link);
169 	free(opt->name, M_MOUNT);
170 	if (opt->value != NULL)
171 		free(opt->value, M_MOUNT);
172 #ifdef INVARIANTS
173 	else if (opt->len != 0)
174 		panic("%s: mount option with NULL value but length != 0",
175 		    __func__);
176 #endif
177 	free(opt, M_MOUNT);
178 }
179 
180 /* Release all resources related to the mount options. */
181 static void
182 vfs_freeopts(struct vfsoptlist *opts)
183 {
184 	struct vfsopt *opt;
185 
186 	while (!TAILQ_EMPTY(opts)) {
187 		opt = TAILQ_FIRST(opts);
188 		vfs_freeopt(opts, opt);
189 	}
190 	free(opts, M_MOUNT);
191 }
192 
193 /*
194  * Check if options are equal (with or without the "no" prefix).
195  */
196 static int
197 vfs_equalopts(const char *opt1, const char *opt2)
198 {
199 
200 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
201 	if (strcmp(opt1, opt2) == 0)
202 		return (1);
203 	/* "noopt" vs. "opt" */
204 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
205 		return (1);
206 	/* "opt" vs. "noopt" */
207 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
208 		return (1);
209 	return (0);
210 }
211 
212 /*
213  * If a mount option is specified several times,
214  * (with or without the "no" prefix) only keep
215  * the last occurence of it.
216  */
217 static void
218 vfs_sanitizeopts(struct vfsoptlist *opts)
219 {
220 	struct vfsopt *opt, *opt2, *tmp;
221 
222 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
223 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
224 		while (opt2 != NULL) {
225 			if (vfs_equalopts(opt->name, opt2->name)) {
226 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
227 				vfs_freeopt(opts, opt2);
228 				opt2 = tmp;
229 			} else {
230 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
231 			}
232 		}
233 	}
234 }
235 
236 /*
237  * Build a linked list of mount options from a struct uio.
238  */
239 static int
240 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
241 {
242 	struct vfsoptlist *opts;
243 	struct vfsopt *opt;
244 	size_t memused;
245 	unsigned int i, iovcnt;
246 	int error, namelen, optlen;
247 
248 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
249 	TAILQ_INIT(opts);
250 	memused = 0;
251 	iovcnt = auio->uio_iovcnt;
252 	for (i = 0; i < iovcnt; i += 2) {
253 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
254 		namelen = auio->uio_iov[i].iov_len;
255 		optlen = auio->uio_iov[i + 1].iov_len;
256 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
257 		opt->value = NULL;
258 		opt->len = 0;
259 
260 		/*
261 		 * Do this early, so jumps to "bad" will free the current
262 		 * option.
263 		 */
264 		TAILQ_INSERT_TAIL(opts, opt, link);
265 		memused += sizeof(struct vfsopt) + optlen + namelen;
266 
267 		/*
268 		 * Avoid consuming too much memory, and attempts to overflow
269 		 * memused.
270 		 */
271 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
272 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
273 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
274 			error = EINVAL;
275 			goto bad;
276 		}
277 
278 		if (auio->uio_segflg == UIO_SYSSPACE) {
279 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
280 		} else {
281 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
282 			    namelen);
283 			if (error)
284 				goto bad;
285 		}
286 		/* Ensure names are null-terminated strings. */
287 		if (opt->name[namelen - 1] != '\0') {
288 			error = EINVAL;
289 			goto bad;
290 		}
291 		if (optlen != 0) {
292 			opt->len = optlen;
293 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
294 			if (auio->uio_segflg == UIO_SYSSPACE) {
295 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
296 				    optlen);
297 			} else {
298 				error = copyin(auio->uio_iov[i + 1].iov_base,
299 				    opt->value, optlen);
300 				if (error)
301 					goto bad;
302 			}
303 		}
304 	}
305 	vfs_sanitizeopts(opts);
306 	*options = opts;
307 	return (0);
308 bad:
309 	vfs_freeopts(opts);
310 	return (error);
311 }
312 
313 /*
314  * Merge the old mount options with the new ones passed
315  * in the MNT_UPDATE case.
316  */
317 static void
318 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
319 {
320 	struct vfsopt *opt, *opt2, *new;
321 
322 	TAILQ_FOREACH(opt, opts, link) {
323 		/*
324 		 * Check that this option hasn't been redefined
325 		 * nor cancelled with a "no" mount option.
326 		 */
327 		opt2 = TAILQ_FIRST(toopts);
328 		while (opt2 != NULL) {
329 			if (strcmp(opt2->name, opt->name) == 0)
330 				goto next;
331 			if (strncmp(opt2->name, "no", 2) == 0 &&
332 			    strcmp(opt2->name + 2, opt->name) == 0) {
333 				vfs_freeopt(toopts, opt2);
334 				goto next;
335 			}
336 			opt2 = TAILQ_NEXT(opt2, link);
337 		}
338 		/* We want this option, duplicate it. */
339 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
340 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
341 		strcpy(new->name, opt->name);
342 		if (opt->len != 0) {
343 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
344 			bcopy(opt->value, new->value, opt->len);
345 		} else {
346 			new->value = NULL;
347 		}
348 		new->len = opt->len;
349 		TAILQ_INSERT_TAIL(toopts, new, link);
350 next:
351 		continue;
352 	}
353 }
354 
355 /*
356  * ---------------------------------------------------------------------
357  * Mount a filesystem
358  */
359 int
360 nmount(td, uap)
361 	struct thread *td;
362 	struct nmount_args /* {
363 		struct iovec *iovp;
364 		unsigned int iovcnt;
365 		int flags;
366 	} */ *uap;
367 {
368 	struct uio *auio;
369 	struct iovec *iov;
370 	unsigned int i;
371 	int error;
372 	u_int iovcnt;
373 
374 	/* Kick out MNT_ROOTFS early as it is legal internally */
375 	if (uap->flags & MNT_ROOTFS)
376 		return (EINVAL);
377 
378 	iovcnt = uap->iovcnt;
379 	/*
380 	 * Check that we have an even number of iovec's
381 	 * and that we have at least two options.
382 	 */
383 	if ((iovcnt & 1) || (iovcnt < 4))
384 		return (EINVAL);
385 
386 	error = copyinuio(uap->iovp, iovcnt, &auio);
387 	if (error)
388 		return (error);
389 	iov = auio->uio_iov;
390 	for (i = 0; i < iovcnt; i++) {
391 		if (iov->iov_len > MMAXOPTIONLEN) {
392 			free(auio, M_IOV);
393 			return (EINVAL);
394 		}
395 		iov++;
396 	}
397 	error = vfs_donmount(td, uap->flags, auio);
398 
399 	free(auio, M_IOV);
400 	return (error);
401 }
402 
403 /*
404  * ---------------------------------------------------------------------
405  * Various utility functions
406  */
407 
408 void
409 vfs_ref(struct mount *mp)
410 {
411 
412 	MNT_ILOCK(mp);
413 	MNT_REF(mp);
414 	MNT_IUNLOCK(mp);
415 }
416 
417 void
418 vfs_rel(struct mount *mp)
419 {
420 
421 	MNT_ILOCK(mp);
422 	MNT_REL(mp);
423 	MNT_IUNLOCK(mp);
424 }
425 
426 /*
427  * Allocate and initialize the mount point struct.
428  */
429 static struct mount *
430 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
431     const char *fspath, struct thread *td)
432 {
433 	struct mount *mp;
434 
435 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
436 	TAILQ_INIT(&mp->mnt_nvnodelist);
437 	mp->mnt_nvnodelistsize = 0;
438 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
439 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
440 	(void) vfs_busy(mp, LK_NOWAIT, 0, td);
441 	mp->mnt_ref = 0;
442 	mp->mnt_op = vfsp->vfc_vfsops;
443 	mp->mnt_vfc = vfsp;
444 	vfsp->vfc_refcount++;	/* XXX Unlocked */
445 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
446 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
447 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
448 	mp->mnt_vnodecovered = vp;
449 	mp->mnt_cred = crdup(td->td_ucred);
450 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
451 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
452 	mp->mnt_iosize_max = DFLTPHYS;
453 #ifdef MAC
454 	mac_init_mount(mp);
455 	mac_create_mount(td->td_ucred, mp);
456 #endif
457 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
458 	return (mp);
459 }
460 
461 /*
462  * Destroy the mount struct previously allocated by vfs_mount_alloc().
463  */
464 static void
465 vfs_mount_destroy(struct mount *mp, struct thread *td)
466 {
467 	int i;
468 
469 	vfs_unbusy(mp, td);
470 	MNT_ILOCK(mp);
471 	for (i = 0; mp->mnt_ref && i < 3; i++)
472 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
473 	/*
474 	 * This will always cause a 3 second delay in rebooting due to
475 	 * refs on the root mountpoint that never go away.  Most of these
476 	 * are held by init which never exits.
477 	 */
478 	if (i == 3 && (!rebooting || bootverbose))
479 		printf("Mount point %s had %d dangling refs\n",
480 		    mp->mnt_stat.f_mntonname, mp->mnt_ref);
481 	if (mp->mnt_holdcnt != 0) {
482 		printf("Waiting for mount point to be unheld\n");
483 		while (mp->mnt_holdcnt != 0) {
484 			mp->mnt_holdcntwaiters++;
485 			msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
486 			       PZERO, "mntdestroy", 0);
487 			mp->mnt_holdcntwaiters--;
488 		}
489 		printf("mount point unheld\n");
490 	}
491 	if (mp->mnt_writeopcount > 0) {
492 		printf("Waiting for mount point write ops\n");
493 		while (mp->mnt_writeopcount > 0) {
494 			mp->mnt_kern_flag |= MNTK_SUSPEND;
495 			msleep(&mp->mnt_writeopcount,
496 			       MNT_MTX(mp),
497 			       PZERO, "mntdestroy2", 0);
498 		}
499 		printf("mount point write ops completed\n");
500 	}
501 	if (mp->mnt_secondary_writes > 0) {
502 		printf("Waiting for mount point secondary write ops\n");
503 		while (mp->mnt_secondary_writes > 0) {
504 			mp->mnt_kern_flag |= MNTK_SUSPEND;
505 			msleep(&mp->mnt_secondary_writes,
506 			       MNT_MTX(mp),
507 			       PZERO, "mntdestroy3", 0);
508 		}
509 		printf("mount point secondary write ops completed\n");
510 	}
511 	MNT_IUNLOCK(mp);
512 	mp->mnt_vfc->vfc_refcount--;
513 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
514 		panic("unmount: dangling vnode");
515 	lockdestroy(&mp->mnt_lock);
516 	MNT_ILOCK(mp);
517 	if (mp->mnt_kern_flag & MNTK_MWAIT)
518 		wakeup(mp);
519 	if (mp->mnt_writeopcount != 0)
520 		panic("vfs_mount_destroy: nonzero writeopcount");
521 	if (mp->mnt_secondary_writes != 0)
522 		panic("vfs_mount_destroy: nonzero secondary_writes");
523 	if (mp->mnt_nvnodelistsize != 0)
524 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
525 	mp->mnt_writeopcount = -1000;
526 	mp->mnt_nvnodelistsize = -1000;
527 	mp->mnt_secondary_writes = -1000;
528 	MNT_IUNLOCK(mp);
529 	mtx_destroy(&mp->mnt_mtx);
530 #ifdef MAC
531 	mac_destroy_mount(mp);
532 #endif
533 	if (mp->mnt_opt != NULL)
534 		vfs_freeopts(mp->mnt_opt);
535 	crfree(mp->mnt_cred);
536 	free(mp, M_MOUNT);
537 }
538 
539 static int
540 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
541 {
542 	struct vfsoptlist *optlist;
543 	struct vfsopt *opt;
544 	char *fstype, *fspath, *errmsg;
545 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
546 
547 	errmsg_len = 0;
548 	errmsg_pos = -1;
549 
550 	error = vfs_buildopts(fsoptions, &optlist);
551 	if (error)
552 		return (error);
553 
554 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
555 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
556 	else
557 		errmsg_len = 0;
558 
559 	/*
560 	 * We need these two options before the others,
561 	 * and they are mandatory for any filesystem.
562 	 * Ensure they are NUL terminated as well.
563 	 */
564 	fstypelen = 0;
565 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
566 	if (error || fstype[fstypelen - 1] != '\0') {
567 		error = EINVAL;
568 		if (errmsg != NULL)
569 			strncpy(errmsg, "Invalid fstype", errmsg_len);
570 		goto bail;
571 	}
572 	fspathlen = 0;
573 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
574 	if (error || fspath[fspathlen - 1] != '\0') {
575 		error = EINVAL;
576 		if (errmsg != NULL)
577 			strncpy(errmsg, "Invalid fspath", errmsg_len);
578 		goto bail;
579 	}
580 
581 	/*
582 	 * We need to see if we have the "update" option
583 	 * before we call vfs_domount(), since vfs_domount() has special
584 	 * logic based on MNT_UPDATE.  This is very important
585 	 * when we want to update the root filesystem.
586 	 */
587 	TAILQ_FOREACH(opt, optlist, link) {
588 		if (strcmp(opt->name, "update") == 0)
589 			fsflags |= MNT_UPDATE;
590 		else if (strcmp(opt->name, "async") == 0)
591 			fsflags |= MNT_ASYNC;
592 		else if (strcmp(opt->name, "force") == 0)
593 			fsflags |= MNT_FORCE;
594 		else if (strcmp(opt->name, "multilabel") == 0)
595 			fsflags |= MNT_MULTILABEL;
596 		else if (strcmp(opt->name, "noasync") == 0)
597 			fsflags &= ~MNT_ASYNC;
598 		else if (strcmp(opt->name, "noatime") == 0)
599 			fsflags |= MNT_NOATIME;
600 		else if (strcmp(opt->name, "noclusterr") == 0)
601 			fsflags |= MNT_NOCLUSTERR;
602 		else if (strcmp(opt->name, "noclusterw") == 0)
603 			fsflags |= MNT_NOCLUSTERW;
604 		else if (strcmp(opt->name, "noexec") == 0)
605 			fsflags |= MNT_NOEXEC;
606 		else if (strcmp(opt->name, "nosuid") == 0)
607 			fsflags |= MNT_NOSUID;
608 		else if (strcmp(opt->name, "nosymfollow") == 0)
609 			fsflags |= MNT_NOSYMFOLLOW;
610 		else if (strcmp(opt->name, "noro") == 0 ||
611 		    strcmp(opt->name, "rw") == 0)
612 			fsflags &= ~MNT_RDONLY;
613 		else if (strcmp(opt->name, "ro") == 0 ||
614 		    strcmp(opt->name, "rdonly") == 0)
615 			fsflags |= MNT_RDONLY;
616 		else if (strcmp(opt->name, "snapshot") == 0)
617 			fsflags |= MNT_SNAPSHOT;
618 		else if (strcmp(opt->name, "suiddir") == 0)
619 			fsflags |= MNT_SUIDDIR;
620 		else if (strcmp(opt->name, "sync") == 0)
621 			fsflags |= MNT_SYNCHRONOUS;
622 		else if (strcmp(opt->name, "union") == 0)
623 			fsflags |= MNT_UNION;
624 	}
625 
626 	/*
627 	 * Be ultra-paranoid about making sure the type and fspath
628 	 * variables will fit in our mp buffers, including the
629 	 * terminating NUL.
630 	 */
631 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
632 		error = ENAMETOOLONG;
633 		goto bail;
634 	}
635 
636 	mtx_lock(&Giant);
637 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
638 	mtx_unlock(&Giant);
639 bail:
640 	/* copyout the errmsg */
641 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
642 	    && errmsg_len > 0 && errmsg != NULL) {
643 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
644 			strncpy(fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
645 			    errmsg,
646 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
647 		} else {
648 			copystr(errmsg,
649 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
650 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len,
651 			    NULL);
652 		}
653 	}
654 
655 	if (error != 0)
656 		vfs_freeopts(optlist);
657 	return (error);
658 }
659 
660 /*
661  * ---------------------------------------------------------------------
662  * Old mount API.
663  */
664 #ifndef _SYS_SYSPROTO_H_
665 struct mount_args {
666 	char	*type;
667 	char	*path;
668 	int	flags;
669 	caddr_t	data;
670 };
671 #endif
672 /* ARGSUSED */
673 int
674 mount(td, uap)
675 	struct thread *td;
676 	struct mount_args /* {
677 		char *type;
678 		char *path;
679 		int flags;
680 		caddr_t data;
681 	} */ *uap;
682 {
683 	char *fstype;
684 	struct vfsconf *vfsp = NULL;
685 	struct mntarg *ma = NULL;
686 	int error;
687 
688 	/* Kick out MNT_ROOTFS early as it is legal internally */
689 	uap->flags &= ~MNT_ROOTFS;
690 
691 	if (uap->data == NULL)
692 		return (EINVAL);
693 
694 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
695 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
696 	if (!error) {
697 		mtx_lock(&Giant);	/* XXX ? */
698 		vfsp = vfs_byname_kld(fstype, td, &error);
699 		mtx_unlock(&Giant);
700 	}
701 	free(fstype, M_TEMP);
702 	if (error)
703 		return (error);
704 	if (vfsp == NULL)
705 		return (ENOENT);
706 	if (vfsp->vfc_vfsops->vfs_cmount == NULL)
707 		return (EOPNOTSUPP);
708 
709 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
710 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
711 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
712 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
713 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
714 
715 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
716 	return (error);
717 }
718 
719 
720 /*
721  * vfs_domount(): actually attempt a filesystem mount.
722  */
723 static int
724 vfs_domount(
725 	struct thread *td,	/* Flags common to all filesystems. */
726 	const char *fstype,	/* Filesystem type. */
727 	char *fspath,		/* Mount path. */
728 	int fsflags,		/* Flags common to all filesystems. */
729 	void *fsdata		/* Options local to the filesystem. */
730 	)
731 {
732 	struct vnode *vp;
733 	struct mount *mp;
734 	struct vfsconf *vfsp;
735 	int error, flag = 0, kern_flag = 0;
736 	struct vattr va;
737 	struct nameidata nd;
738 
739 	mtx_assert(&Giant, MA_OWNED);
740 	/*
741 	 * Be ultra-paranoid about making sure the type and fspath
742 	 * variables will fit in our mp buffers, including the
743 	 * terminating NUL.
744 	 */
745 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
746 		return (ENAMETOOLONG);
747 
748 	if (jailed(td->td_ucred))
749 		return (EPERM);
750 	if (usermount == 0) {
751 		if ((error = suser(td)) != 0)
752 			return (error);
753 	}
754 
755 	/*
756 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
757 	 */
758 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
759 		if ((error = suser(td)) != 0)
760 			return (error);
761 	}
762 	/*
763 	 * Silently enforce MNT_NOSUID and MNT_USER for
764 	 * unprivileged users.
765 	 */
766 	if (suser(td) != 0)
767 		fsflags |= MNT_NOSUID | MNT_USER;
768 
769 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
770 	vfsp = NULL;
771 	if ((fsflags & MNT_UPDATE) == 0) {
772 		/* Don't try to load KLDs if we're mounting the root. */
773 		if (fsflags & MNT_ROOTFS)
774 			vfsp = vfs_byname(fstype);
775 		else
776 			vfsp = vfs_byname_kld(fstype, td, &error);
777 		if (vfsp == NULL)
778 			return (ENODEV);
779 	}
780 	/*
781 	 * Get vnode to be covered
782 	 */
783 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
784 	if ((error = namei(&nd)) != 0)
785 		return (error);
786 	NDFREE(&nd, NDF_ONLY_PNBUF);
787 	vp = nd.ni_vp;
788 	if (fsflags & MNT_UPDATE) {
789 		if ((vp->v_vflag & VV_ROOT) == 0) {
790 			vput(vp);
791 			return (EINVAL);
792 		}
793 		mp = vp->v_mount;
794 		flag = mp->mnt_flag;
795 		kern_flag = mp->mnt_kern_flag;
796 		/*
797 		 * We only allow the filesystem to be reloaded if it
798 		 * is currently mounted read-only.
799 		 */
800 		if ((fsflags & MNT_RELOAD) &&
801 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
802 			vput(vp);
803 			return (EOPNOTSUPP);	/* Needs translation */
804 		}
805 		/*
806 		 * Only privileged root, or (if MNT_USER is set) the user that
807 		 * did the original mount is permitted to update it.
808 		 */
809 		error = vfs_suser(mp, td);
810 		if (error) {
811 			vput(vp);
812 			return (error);
813 		}
814 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
815 			vput(vp);
816 			return (EBUSY);
817 		}
818 		VI_LOCK(vp);
819 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
820 		    vp->v_mountedhere != NULL) {
821 			VI_UNLOCK(vp);
822 			vfs_unbusy(mp, td);
823 			vput(vp);
824 			return (EBUSY);
825 		}
826 		vp->v_iflag |= VI_MOUNT;
827 		VI_UNLOCK(vp);
828 		mp->mnt_flag |= fsflags &
829 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
830 		VOP_UNLOCK(vp, 0, td);
831 		mp->mnt_optnew = fsdata;
832 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
833 	} else {
834 		/*
835 		 * If the user is not root, ensure that they own the directory
836 		 * onto which we are attempting to mount.
837 		 */
838 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
839 		if (error) {
840 			vput(vp);
841 			return (error);
842 		}
843 		if (va.va_uid != td->td_ucred->cr_uid) {
844 			if ((error = suser(td)) != 0) {
845 				vput(vp);
846 				return (error);
847 			}
848 		}
849 		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
850 		if (error != 0) {
851 			vput(vp);
852 			return (error);
853 		}
854 		if (vp->v_type != VDIR) {
855 			vput(vp);
856 			return (ENOTDIR);
857 		}
858 		VI_LOCK(vp);
859 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
860 		    vp->v_mountedhere != NULL) {
861 			VI_UNLOCK(vp);
862 			vput(vp);
863 			return (EBUSY);
864 		}
865 		vp->v_iflag |= VI_MOUNT;
866 		VI_UNLOCK(vp);
867 
868 		/*
869 		 * Allocate and initialize the filesystem.
870 		 */
871 		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
872 		VOP_UNLOCK(vp, 0, td);
873 
874 		/* XXXMAC: pass to vfs_mount_alloc? */
875 		mp->mnt_optnew = fsdata;
876 	}
877 
878 	/*
879 	 * Set the mount level flags.
880 	 */
881 	if (fsflags & MNT_RDONLY)
882 		mp->mnt_flag |= MNT_RDONLY;
883 	mp->mnt_flag &=~ MNT_UPDATEMASK;
884 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
885 	/*
886 	 * Mount the filesystem.
887 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
888 	 * get.  No freeing of cn_pnbuf.
889 	 */
890         error = VFS_MOUNT(mp, td);
891 	if (!error) {
892 		if (mp->mnt_opt != NULL)
893 			vfs_freeopts(mp->mnt_opt);
894 		mp->mnt_opt = mp->mnt_optnew;
895 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
896 	}
897 	/*
898 	 * Prevent external consumers of mount options from reading
899 	 * mnt_optnew.
900 	*/
901 	mp->mnt_optnew = NULL;
902 	if (mp->mnt_flag & MNT_UPDATE) {
903 		mp->mnt_flag &=
904 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
905 		if (error) {
906 			mp->mnt_flag = flag;
907 			mp->mnt_kern_flag = kern_flag;
908 		}
909 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
910 			if (mp->mnt_syncer == NULL)
911 				error = vfs_allocate_syncvnode(mp);
912 		} else {
913 			if (mp->mnt_syncer != NULL)
914 				vrele(mp->mnt_syncer);
915 			mp->mnt_syncer = NULL;
916 		}
917 		vfs_unbusy(mp, td);
918 		VI_LOCK(vp);
919 		vp->v_iflag &= ~VI_MOUNT;
920 		VI_UNLOCK(vp);
921 		vrele(vp);
922 		return (error);
923 	}
924 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
925 	/*
926 	 * Put the new filesystem on the mount list after root.
927 	 */
928 	cache_purge(vp);
929 	if (!error) {
930 		struct vnode *newdp;
931 
932 		VI_LOCK(vp);
933 		vp->v_iflag &= ~VI_MOUNT;
934 		VI_UNLOCK(vp);
935 		vp->v_mountedhere = mp;
936 		mtx_lock(&mountlist_mtx);
937 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
938 		mtx_unlock(&mountlist_mtx);
939 		vfs_event_signal(NULL, VQ_MOUNT, 0);
940 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
941 			panic("mount: lost mount");
942 		mountcheckdirs(vp, newdp);
943 		vput(newdp);
944 		VOP_UNLOCK(vp, 0, td);
945 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
946 			error = vfs_allocate_syncvnode(mp);
947 		vfs_unbusy(mp, td);
948 		if (error)
949 			vrele(vp);
950 	} else {
951 		VI_LOCK(vp);
952 		vp->v_iflag &= ~VI_MOUNT;
953 		VI_UNLOCK(vp);
954 		vfs_mount_destroy(mp, td);
955 		vput(vp);
956 	}
957 	return (error);
958 }
959 
960 /*
961  * ---------------------------------------------------------------------
962  * Unmount a filesystem.
963  *
964  * Note: unmount takes a path to the vnode mounted on as argument,
965  * not special file (as before).
966  */
967 #ifndef _SYS_SYSPROTO_H_
968 struct unmount_args {
969 	char	*path;
970 	int	flags;
971 };
972 #endif
973 /* ARGSUSED */
974 int
975 unmount(td, uap)
976 	struct thread *td;
977 	register struct unmount_args /* {
978 		char *path;
979 		int flags;
980 	} */ *uap;
981 {
982 	struct mount *mp;
983 	char *pathbuf;
984 	int error, id0, id1;
985 
986 	if (jailed(td->td_ucred))
987 		return (EPERM);
988 	if (usermount == 0) {
989 		if ((error = suser(td)) != 0)
990 			return (error);
991 	}
992 
993 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
994 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
995 	if (error) {
996 		free(pathbuf, M_TEMP);
997 		return (error);
998 	}
999 	if (uap->flags & MNT_BYFSID) {
1000 		/* Decode the filesystem ID. */
1001 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1002 			free(pathbuf, M_TEMP);
1003 			return (EINVAL);
1004 		}
1005 
1006 		mtx_lock(&mountlist_mtx);
1007 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1008 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1009 			    mp->mnt_stat.f_fsid.val[1] == id1)
1010 				break;
1011 		}
1012 		mtx_unlock(&mountlist_mtx);
1013 	} else {
1014 		mtx_lock(&mountlist_mtx);
1015 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1016 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1017 				break;
1018 		}
1019 		mtx_unlock(&mountlist_mtx);
1020 	}
1021 	free(pathbuf, M_TEMP);
1022 	if (mp == NULL) {
1023 		/*
1024 		 * Previously we returned ENOENT for a nonexistent path and
1025 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1026 		 * now, so in the !MNT_BYFSID case return the more likely
1027 		 * EINVAL for compatibility.
1028 		 */
1029 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1030 	}
1031 
1032 	/*
1033 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1034 	 * original mount is permitted to unmount this filesystem.
1035 	 */
1036 	error = vfs_suser(mp, td);
1037 	if (error)
1038 		return (error);
1039 
1040 	/*
1041 	 * Don't allow unmounting the root filesystem.
1042 	 */
1043 	if (mp->mnt_flag & MNT_ROOTFS)
1044 		return (EINVAL);
1045 	mtx_lock(&Giant);
1046 	error = dounmount(mp, uap->flags, td);
1047 	mtx_unlock(&Giant);
1048 	return (error);
1049 }
1050 
1051 /*
1052  * Do the actual filesystem unmount.
1053  */
1054 int
1055 dounmount(mp, flags, td)
1056 	struct mount *mp;
1057 	int flags;
1058 	struct thread *td;
1059 {
1060 	struct vnode *coveredvp, *fsrootvp;
1061 	int error;
1062 	int async_flag;
1063 
1064 	mtx_assert(&Giant, MA_OWNED);
1065 
1066 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1067 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY, td);
1068 	MNT_ILOCK(mp);
1069 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1070 		MNT_IUNLOCK(mp);
1071 		if (coveredvp)
1072 			VOP_UNLOCK(coveredvp, 0, td);
1073 		return (EBUSY);
1074 	}
1075 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1076 	/* Allow filesystems to detect that a forced unmount is in progress. */
1077 	if (flags & MNT_FORCE)
1078 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1079 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1080 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1081 	if (error) {
1082 		MNT_ILOCK(mp);
1083 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1084 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1085 			wakeup(mp);
1086 		MNT_IUNLOCK(mp);
1087 		if (coveredvp)
1088 			VOP_UNLOCK(coveredvp, 0, td);
1089 		return (error);
1090 	}
1091 	vn_start_write(NULL, &mp, V_WAIT);
1092 
1093 	if (mp->mnt_flag & MNT_EXPUBLIC)
1094 		vfs_setpublicfs(NULL, NULL, NULL);
1095 
1096 	vfs_msync(mp, MNT_WAIT);
1097 	async_flag = mp->mnt_flag & MNT_ASYNC;
1098 	mp->mnt_flag &= ~MNT_ASYNC;
1099 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1100 	if (mp->mnt_syncer != NULL)
1101 		vrele(mp->mnt_syncer);
1102 	/*
1103 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1104 	 * vnode to the covered vnode.  For non-forced unmounts we want
1105 	 * such references to cause an EBUSY error.
1106 	 */
1107 	if ((flags & MNT_FORCE) &&
1108 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1109 		if (mp->mnt_vnodecovered != NULL)
1110 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1111 		if (fsrootvp == rootvnode) {
1112 			vrele(rootvnode);
1113 			rootvnode = NULL;
1114 		}
1115 		vput(fsrootvp);
1116 	}
1117 	if (((mp->mnt_flag & MNT_RDONLY) ||
1118 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1119 	    (flags & MNT_FORCE)) {
1120 		error = VFS_UNMOUNT(mp, flags, td);
1121 	}
1122 	vn_finished_write(mp);
1123 	if (error) {
1124 		/* Undo cdir/rdir and rootvnode changes made above. */
1125 		if ((flags & MNT_FORCE) &&
1126 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1127 			if (mp->mnt_vnodecovered != NULL)
1128 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1129 			if (rootvnode == NULL) {
1130 				rootvnode = fsrootvp;
1131 				vref(rootvnode);
1132 			}
1133 			vput(fsrootvp);
1134 		}
1135 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1136 			(void) vfs_allocate_syncvnode(mp);
1137 		MNT_ILOCK(mp);
1138 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1139 		mp->mnt_flag |= async_flag;
1140 		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1141 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1142 			wakeup(mp);
1143 		MNT_IUNLOCK(mp);
1144 		if (coveredvp)
1145 			VOP_UNLOCK(coveredvp, 0, td);
1146 		return (error);
1147 	}
1148 	mtx_lock(&mountlist_mtx);
1149 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1150 	mtx_unlock(&mountlist_mtx);
1151 	if (coveredvp != NULL) {
1152 		coveredvp->v_mountedhere = NULL;
1153 		vput(coveredvp);
1154 	}
1155 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1156 	vfs_mount_destroy(mp, td);
1157 	return (0);
1158 }
1159 
1160 /*
1161  * ---------------------------------------------------------------------
1162  * Mounting of root filesystem
1163  *
1164  */
1165 
1166 struct root_hold_token {
1167 	const char 			*who;
1168 	LIST_ENTRY(root_hold_token)	list;
1169 };
1170 
1171 static LIST_HEAD(, root_hold_token)	root_holds =
1172     LIST_HEAD_INITIALIZER(&root_holds);
1173 
1174 struct root_hold_token *
1175 root_mount_hold(const char *identifier)
1176 {
1177 	struct root_hold_token *h;
1178 
1179 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1180 	h->who = identifier;
1181 	mtx_lock(&mountlist_mtx);
1182 	LIST_INSERT_HEAD(&root_holds, h, list);
1183 	mtx_unlock(&mountlist_mtx);
1184 	return (h);
1185 }
1186 
1187 void
1188 root_mount_rel(struct root_hold_token *h)
1189 {
1190 
1191 	mtx_lock(&mountlist_mtx);
1192 	LIST_REMOVE(h, list);
1193 	wakeup(&root_holds);
1194 	mtx_unlock(&mountlist_mtx);
1195 	free(h, M_DEVBUF);
1196 }
1197 
1198 static void
1199 root_mount_wait(void)
1200 {
1201 	struct root_hold_token *h;
1202 
1203 	for (;;) {
1204 		DROP_GIANT();
1205 		g_waitidle();
1206 		PICKUP_GIANT();
1207 		mtx_lock(&mountlist_mtx);
1208 		if (LIST_EMPTY(&root_holds)) {
1209 			mtx_unlock(&mountlist_mtx);
1210 			break;
1211 		}
1212 		printf("Root mount waiting for:");
1213 		LIST_FOREACH(h, &root_holds, list)
1214 			printf(" %s", h->who);
1215 		printf("\n");
1216 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1217 		    hz);
1218 	}
1219 }
1220 
1221 static void
1222 set_rootvnode(struct thread *td)
1223 {
1224 	struct proc *p;
1225 
1226 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1227 		panic("Cannot find root vnode");
1228 
1229 	p = td->td_proc;
1230 	FILEDESC_LOCK(p->p_fd);
1231 
1232 	if (p->p_fd->fd_cdir != NULL)
1233 		vrele(p->p_fd->fd_cdir);
1234 	p->p_fd->fd_cdir = rootvnode;
1235 	VREF(rootvnode);
1236 
1237 	if (p->p_fd->fd_rdir != NULL)
1238 		vrele(p->p_fd->fd_rdir);
1239 	p->p_fd->fd_rdir = rootvnode;
1240 	VREF(rootvnode);
1241 
1242 	FILEDESC_UNLOCK(p->p_fd);
1243 
1244 	VOP_UNLOCK(rootvnode, 0, td);
1245 }
1246 
1247 /*
1248  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1249  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1250  */
1251 
1252 static void
1253 devfs_first(void)
1254 {
1255 	struct thread *td = curthread;
1256 	struct vfsoptlist *opts;
1257 	struct vfsconf *vfsp;
1258 	struct mount *mp = NULL;
1259 	int error;
1260 
1261 	vfsp = vfs_byname("devfs");
1262 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1263 	if (vfsp == NULL)
1264 		return;
1265 
1266 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1267 
1268 	error = VFS_MOUNT(mp, td);
1269 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1270 	if (error)
1271 		return;
1272 
1273 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1274 	TAILQ_INIT(opts);
1275 	mp->mnt_opt = opts;
1276 
1277 	mtx_lock(&mountlist_mtx);
1278 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1279 	mtx_unlock(&mountlist_mtx);
1280 
1281 	set_rootvnode(td);
1282 
1283 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1284 	if (error)
1285 		printf("kern_symlink /dev -> / returns %d\n", error);
1286 }
1287 
1288 /*
1289  * Surgically move our devfs to be mounted on /dev.
1290  */
1291 
1292 static void
1293 devfs_fixup(struct thread *td)
1294 {
1295 	struct nameidata nd;
1296 	int error;
1297 	struct vnode *vp, *dvp;
1298 	struct mount *mp;
1299 
1300 	/* Remove our devfs mount from the mountlist and purge the cache */
1301 	mtx_lock(&mountlist_mtx);
1302 	mp = TAILQ_FIRST(&mountlist);
1303 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1304 	mtx_unlock(&mountlist_mtx);
1305 	cache_purgevfs(mp);
1306 
1307 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1308 	VI_LOCK(dvp);
1309 	dvp->v_iflag &= ~VI_MOUNT;
1310 	dvp->v_mountedhere = NULL;
1311 	VI_UNLOCK(dvp);
1312 
1313 	/* Set up the real rootvnode, and purge the cache */
1314 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1315 	set_rootvnode(td);
1316 	cache_purgevfs(rootvnode->v_mount);
1317 
1318 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1319 	error = namei(&nd);
1320 	if (error) {
1321 		printf("Lookup of /dev for devfs, error: %d\n", error);
1322 		return;
1323 	}
1324 	NDFREE(&nd, NDF_ONLY_PNBUF);
1325 	vp = nd.ni_vp;
1326 	if (vp->v_type != VDIR) {
1327 		vput(vp);
1328 	}
1329 	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1330 	if (error) {
1331 		vput(vp);
1332 	}
1333 	cache_purge(vp);
1334 	mp->mnt_vnodecovered = vp;
1335 	vp->v_mountedhere = mp;
1336 	mtx_lock(&mountlist_mtx);
1337 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1338 	mtx_unlock(&mountlist_mtx);
1339 	VOP_UNLOCK(vp, 0, td);
1340 	vput(dvp);
1341 	vfs_unbusy(mp, td);
1342 
1343 	/* Unlink the no longer needed /dev/dev -> / symlink */
1344 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1345 }
1346 
1347 /*
1348  * Report errors during filesystem mounting.
1349  */
1350 void
1351 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1352 {
1353 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1354 	va_list ap;
1355 	int error, len;
1356 	char *errmsg;
1357 
1358 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1359 	if (error || errmsg == NULL || len <= 0)
1360 		return;
1361 
1362 	va_start(ap, fmt);
1363 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1364 	va_end(ap);
1365 }
1366 
1367 /*
1368  * Find and mount the root filesystem
1369  */
1370 void
1371 vfs_mountroot(void)
1372 {
1373 	char *cp;
1374 	int error, i, asked = 0;
1375 
1376 	root_mount_wait();
1377 
1378 	devfs_first();
1379 
1380 	/*
1381 	 * We are booted with instructions to prompt for the root filesystem.
1382 	 */
1383 	if (boothowto & RB_ASKNAME) {
1384 		if (!vfs_mountroot_ask())
1385 			return;
1386 		asked = 1;
1387 	}
1388 
1389 	/*
1390 	 * The root filesystem information is compiled in, and we are
1391 	 * booted with instructions to use it.
1392 	 */
1393 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1394 		if (!vfs_mountroot_try(ctrootdevname))
1395 			return;
1396 		ctrootdevname = NULL;
1397 	}
1398 
1399 	/*
1400 	 * We've been given the generic "use CDROM as root" flag.  This is
1401 	 * necessary because one media may be used in many different
1402 	 * devices, so we need to search for them.
1403 	 */
1404 	if (boothowto & RB_CDROM) {
1405 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1406 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1407 				return;
1408 		}
1409 	}
1410 
1411 	/*
1412 	 * Try to use the value read by the loader from /etc/fstab, or
1413 	 * supplied via some other means.  This is the preferred
1414 	 * mechanism.
1415 	 */
1416 	cp = getenv("vfs.root.mountfrom");
1417 	if (cp != NULL) {
1418 		error = vfs_mountroot_try(cp);
1419 		freeenv(cp);
1420 		if (!error)
1421 			return;
1422 	}
1423 
1424 	/*
1425 	 * Try values that may have been computed by code during boot
1426 	 */
1427 	if (!vfs_mountroot_try(rootdevnames[0]))
1428 		return;
1429 	if (!vfs_mountroot_try(rootdevnames[1]))
1430 		return;
1431 
1432 	/*
1433 	 * If we (still) have a compiled-in default, try it.
1434 	 */
1435 	if (ctrootdevname != NULL)
1436 		if (!vfs_mountroot_try(ctrootdevname))
1437 			return;
1438 	/*
1439 	 * Everything so far has failed, prompt on the console if we haven't
1440 	 * already tried that.
1441 	 */
1442 	if (!asked)
1443 		if (!vfs_mountroot_ask())
1444 			return;
1445 
1446 	panic("Root mount failed, startup aborted.");
1447 }
1448 
1449 /*
1450  * Mount (mountfrom) as the root filesystem.
1451  */
1452 static int
1453 vfs_mountroot_try(const char *mountfrom)
1454 {
1455 	struct mount	*mp;
1456 	char		*vfsname, *path;
1457 	time_t		timebase;
1458 	int		error;
1459 	char		patt[32];
1460 
1461 	vfsname = NULL;
1462 	path    = NULL;
1463 	mp      = NULL;
1464 	error   = EINVAL;
1465 
1466 	if (mountfrom == NULL)
1467 		return (error);		/* don't complain */
1468 	printf("Trying to mount root from %s\n", mountfrom);
1469 
1470 	/* parse vfs name and path */
1471 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1472 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1473 	vfsname[0] = path[0] = 0;
1474 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1475 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1476 		goto out;
1477 
1478 	if (path[0] == '\0')
1479 		strcpy(path, ROOTNAME);
1480 
1481 	error = kernel_vmount(
1482 	    MNT_RDONLY | MNT_ROOTFS,
1483 	    "fstype", vfsname,
1484 	    "fspath", "/",
1485 	    "from", path,
1486 	    NULL);
1487 	if (error == 0) {
1488 		/*
1489 		 * We mount devfs prior to mounting the / FS, so the first
1490 		 * entry will typically be devfs.
1491 		 */
1492 		mp = TAILQ_FIRST(&mountlist);
1493 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1494 
1495 		/*
1496 		 * Iterate over all currently mounted file systems and use
1497 		 * the time stamp found to check and/or initialize the RTC.
1498 		 * Typically devfs has no time stamp and the only other FS
1499 		 * is the actual / FS.
1500 		 * Call inittodr() only once and pass it the largest of the
1501 		 * timestamps we encounter.
1502 		 */
1503 		timebase = 0;
1504 		do {
1505 			if (mp->mnt_time > timebase)
1506 				timebase = mp->mnt_time;
1507 			mp = TAILQ_NEXT(mp, mnt_list);
1508 		} while (mp != NULL);
1509 		inittodr(timebase);
1510 
1511 		devfs_fixup(curthread);
1512 	}
1513 out:
1514 	free(path, M_MOUNT);
1515 	free(vfsname, M_MOUNT);
1516 	return (error);
1517 }
1518 
1519 /*
1520  * ---------------------------------------------------------------------
1521  * Interactive root filesystem selection code.
1522  */
1523 
1524 static int
1525 vfs_mountroot_ask(void)
1526 {
1527 	char name[128];
1528 
1529 	for(;;) {
1530 		printf("\nManual root filesystem specification:\n");
1531 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1532 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1533 		printf("                       eg. ufs:da0s1a\n");
1534 #else
1535 		printf("                       eg. ufs:/dev/da0a\n");
1536 #endif
1537 		printf("  ?                  List valid disk boot devices\n");
1538 		printf("  <empty line>       Abort manual input\n");
1539 		printf("\nmountroot> ");
1540 		gets(name, sizeof(name), 1);
1541 		if (name[0] == '\0')
1542 			return (1);
1543 		if (name[0] == '?') {
1544 			printf("\nList of GEOM managed disk devices:\n  ");
1545 			g_dev_print();
1546 			continue;
1547 		}
1548 		if (!vfs_mountroot_try(name))
1549 			return (0);
1550 	}
1551 }
1552 
1553 /*
1554  * ---------------------------------------------------------------------
1555  * Functions for querying mount options/arguments from filesystems.
1556  */
1557 
1558 /*
1559  * Check that no unknown options are given
1560  */
1561 int
1562 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1563 {
1564 	struct vfsopt *opt;
1565 	const char **t, *p;
1566 
1567 
1568 	TAILQ_FOREACH(opt, opts, link) {
1569 		p = opt->name;
1570 		if (p[0] == 'n' && p[1] == 'o')
1571 			p += 2;
1572 		for(t = global_opts; *t != NULL; t++)
1573 			if (!strcmp(*t, p))
1574 				break;
1575 		if (*t != NULL)
1576 			continue;
1577 		for(t = legal; *t != NULL; t++)
1578 			if (!strcmp(*t, p))
1579 				break;
1580 		if (*t != NULL)
1581 			continue;
1582 		printf("mount option <%s> is unknown\n", p);
1583 		return (EINVAL);
1584 	}
1585 	return (0);
1586 }
1587 
1588 /*
1589  * Get a mount option by its name.
1590  *
1591  * Return 0 if the option was found, ENOENT otherwise.
1592  * If len is non-NULL it will be filled with the length
1593  * of the option. If buf is non-NULL, it will be filled
1594  * with the address of the option.
1595  */
1596 int
1597 vfs_getopt(opts, name, buf, len)
1598 	struct vfsoptlist *opts;
1599 	const char *name;
1600 	void **buf;
1601 	int *len;
1602 {
1603 	struct vfsopt *opt;
1604 
1605 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1606 
1607 	TAILQ_FOREACH(opt, opts, link) {
1608 		if (strcmp(name, opt->name) == 0) {
1609 			if (len != NULL)
1610 				*len = opt->len;
1611 			if (buf != NULL)
1612 				*buf = opt->value;
1613 			return (0);
1614 		}
1615 	}
1616 	return (ENOENT);
1617 }
1618 
1619 static int
1620 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1621 {
1622 	struct vfsopt *opt;
1623 	int i;
1624 
1625 	if (opts == NULL)
1626 		return (-1);
1627 
1628 	i = 0;
1629 	TAILQ_FOREACH(opt, opts, link) {
1630 		if (strcmp(name, opt->name) == 0)
1631 			return (i);
1632 		++i;
1633 	}
1634 	return (-1);
1635 }
1636 
1637 char *
1638 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1639 {
1640 	struct vfsopt *opt;
1641 
1642 	*error = 0;
1643 	TAILQ_FOREACH(opt, opts, link) {
1644 		if (strcmp(name, opt->name) != 0)
1645 			continue;
1646 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1647 			*error = EINVAL;
1648 			return (NULL);
1649 		}
1650 		return (opt->value);
1651 	}
1652 	return (NULL);
1653 }
1654 
1655 int
1656 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1657 {
1658 	struct vfsopt *opt;
1659 
1660 	TAILQ_FOREACH(opt, opts, link) {
1661 		if (strcmp(name, opt->name) == 0) {
1662 			if (w != NULL)
1663 				*w |= val;
1664 			return (1);
1665 		}
1666 	}
1667 	if (w != NULL)
1668 		*w &= ~val;
1669 	return (0);
1670 }
1671 
1672 int
1673 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1674 {
1675 	va_list ap;
1676 	struct vfsopt *opt;
1677 	int ret;
1678 
1679 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1680 
1681 	TAILQ_FOREACH(opt, opts, link) {
1682 		if (strcmp(name, opt->name) != 0)
1683 			continue;
1684 		if (((char *)opt->value)[opt->len - 1] != '\0')
1685 			return (0);
1686 		va_start(ap, fmt);
1687 		ret = vsscanf(opt->value, fmt, ap);
1688 		va_end(ap);
1689 		return (ret);
1690 	}
1691 	return (0);
1692 }
1693 
1694 /*
1695  * Find and copy a mount option.
1696  *
1697  * The size of the buffer has to be specified
1698  * in len, if it is not the same length as the
1699  * mount option, EINVAL is returned.
1700  * Returns ENOENT if the option is not found.
1701  */
1702 int
1703 vfs_copyopt(opts, name, dest, len)
1704 	struct vfsoptlist *opts;
1705 	const char *name;
1706 	void *dest;
1707 	int len;
1708 {
1709 	struct vfsopt *opt;
1710 
1711 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1712 
1713 	TAILQ_FOREACH(opt, opts, link) {
1714 		if (strcmp(name, opt->name) == 0) {
1715 			if (len != opt->len)
1716 				return (EINVAL);
1717 			bcopy(opt->value, dest, opt->len);
1718 			return (0);
1719 		}
1720 	}
1721 	return (ENOENT);
1722 }
1723 
1724 /*
1725  * This is a helper function for filesystems to traverse their
1726  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1727  */
1728 
1729 struct vnode *
1730 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1731 {
1732 	struct vnode *vp;
1733 
1734 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1735 
1736 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1737 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1738 	while (vp != NULL && vp->v_type == VMARKER)
1739 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1740 
1741 	/* Check if we are done */
1742 	if (vp == NULL) {
1743 		__mnt_vnode_markerfree(mvp, mp);
1744 		return (NULL);
1745 	}
1746 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1747 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1748 	return (vp);
1749 }
1750 
1751 struct vnode *
1752 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
1753 {
1754 	struct vnode *vp;
1755 
1756 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1757 
1758 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1759 	while (vp != NULL && vp->v_type == VMARKER)
1760 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1761 
1762 	/* Check if we are done */
1763 	if (vp == NULL) {
1764 		*mvp = NULL;
1765 		return (NULL);
1766 	}
1767 	mp->mnt_holdcnt++;
1768 	MNT_IUNLOCK(mp);
1769 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
1770 				       M_VNODE_MARKER,
1771 				       M_WAITOK | M_ZERO);
1772 	MNT_ILOCK(mp);
1773 	(*mvp)->v_type = VMARKER;
1774 
1775 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1776 	while (vp != NULL && vp->v_type == VMARKER)
1777 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1778 
1779 	/* Check if we are done */
1780 	if (vp == NULL) {
1781 		MNT_IUNLOCK(mp);
1782 		free(*mvp, M_VNODE_MARKER);
1783 		MNT_ILOCK(mp);
1784 		*mvp = NULL;
1785 		mp->mnt_holdcnt--;
1786 		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1787 			wakeup(&mp->mnt_holdcnt);
1788 		return (NULL);
1789 	}
1790 	mp->mnt_markercnt++;
1791 	(*mvp)->v_mount = mp;
1792 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1793 	return (vp);
1794 }
1795 
1796 
1797 void
1798 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
1799 {
1800 
1801 	if (*mvp == NULL)
1802 		return;
1803 
1804 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1805 
1806 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1807 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1808 	MNT_IUNLOCK(mp);
1809 	free(*mvp, M_VNODE_MARKER);
1810 	MNT_ILOCK(mp);
1811 	*mvp = NULL;
1812 
1813 	mp->mnt_markercnt--;
1814 	mp->mnt_holdcnt--;
1815 	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1816 		wakeup(&mp->mnt_holdcnt);
1817 }
1818 
1819 
1820 int
1821 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
1822 {
1823 	int error;
1824 
1825 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
1826 	if (sbp != &mp->mnt_stat)
1827 		*sbp = mp->mnt_stat;
1828 	return (error);
1829 }
1830 
1831 void
1832 vfs_mountedfrom(struct mount *mp, const char *from)
1833 {
1834 
1835 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1836 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1837 	    sizeof mp->mnt_stat.f_mntfromname);
1838 }
1839 
1840 /*
1841  * ---------------------------------------------------------------------
1842  * This is the api for building mount args and mounting filesystems from
1843  * inside the kernel.
1844  *
1845  * The API works by accumulation of individual args.  First error is
1846  * latched.
1847  *
1848  * XXX: should be documented in new manpage kernel_mount(9)
1849  */
1850 
1851 /* A memory allocation which must be freed when we are done */
1852 struct mntaarg {
1853 	SLIST_ENTRY(mntaarg)	next;
1854 };
1855 
1856 /* The header for the mount arguments */
1857 struct mntarg {
1858 	struct iovec *v;
1859 	int len;
1860 	int error;
1861 	SLIST_HEAD(, mntaarg)	list;
1862 };
1863 
1864 /*
1865  * Add a boolean argument.
1866  *
1867  * flag is the boolean value.
1868  * name must start with "no".
1869  */
1870 struct mntarg *
1871 mount_argb(struct mntarg *ma, int flag, const char *name)
1872 {
1873 
1874 	KASSERT(name[0] == 'n' && name[1] == 'o',
1875 	    ("mount_argb(...,%s): name must start with 'no'", name));
1876 
1877 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1878 }
1879 
1880 /*
1881  * Add an argument printf style
1882  */
1883 struct mntarg *
1884 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1885 {
1886 	va_list ap;
1887 	struct mntaarg *maa;
1888 	struct sbuf *sb;
1889 	int len;
1890 
1891 	if (ma == NULL) {
1892 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1893 		SLIST_INIT(&ma->list);
1894 	}
1895 	if (ma->error)
1896 		return (ma);
1897 
1898 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1899 	    M_MOUNT, M_WAITOK);
1900 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1901 	ma->v[ma->len].iov_len = strlen(name) + 1;
1902 	ma->len++;
1903 
1904 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
1905 	va_start(ap, fmt);
1906 	sbuf_vprintf(sb, fmt, ap);
1907 	va_end(ap);
1908 	sbuf_finish(sb);
1909 	len = sbuf_len(sb) + 1;
1910 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1911 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1912 	bcopy(sbuf_data(sb), maa + 1, len);
1913 	sbuf_delete(sb);
1914 
1915 	ma->v[ma->len].iov_base = maa + 1;
1916 	ma->v[ma->len].iov_len = len;
1917 	ma->len++;
1918 
1919 	return (ma);
1920 }
1921 
1922 /*
1923  * Add an argument which is a userland string.
1924  */
1925 struct mntarg *
1926 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1927 {
1928 	struct mntaarg *maa;
1929 	char *tbuf;
1930 
1931 	if (val == NULL)
1932 		return (ma);
1933 	if (ma == NULL) {
1934 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1935 		SLIST_INIT(&ma->list);
1936 	}
1937 	if (ma->error)
1938 		return (ma);
1939 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1940 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1941 	tbuf = (void *)(maa + 1);
1942 	ma->error = copyinstr(val, tbuf, len, NULL);
1943 	return (mount_arg(ma, name, tbuf, -1));
1944 }
1945 
1946 /*
1947  * Plain argument.
1948  *
1949  * If length is -1, use printf.
1950  */
1951 struct mntarg *
1952 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
1953 {
1954 
1955 	if (ma == NULL) {
1956 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1957 		SLIST_INIT(&ma->list);
1958 	}
1959 	if (ma->error)
1960 		return (ma);
1961 
1962 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1963 	    M_MOUNT, M_WAITOK);
1964 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1965 	ma->v[ma->len].iov_len = strlen(name) + 1;
1966 	ma->len++;
1967 
1968 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
1969 	if (len < 0)
1970 		ma->v[ma->len].iov_len = strlen(val) + 1;
1971 	else
1972 		ma->v[ma->len].iov_len = len;
1973 	ma->len++;
1974 	return (ma);
1975 }
1976 
1977 /*
1978  * Free a mntarg structure
1979  */
1980 static void
1981 free_mntarg(struct mntarg *ma)
1982 {
1983 	struct mntaarg *maa;
1984 
1985 	while (!SLIST_EMPTY(&ma->list)) {
1986 		maa = SLIST_FIRST(&ma->list);
1987 		SLIST_REMOVE_HEAD(&ma->list, next);
1988 		free(maa, M_MOUNT);
1989 	}
1990 	free(ma->v, M_MOUNT);
1991 	free(ma, M_MOUNT);
1992 }
1993 
1994 /*
1995  * Mount a filesystem
1996  */
1997 int
1998 kernel_mount(struct mntarg *ma, int flags)
1999 {
2000 	struct uio auio;
2001 	int error;
2002 
2003 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2004 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2005 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2006 
2007 	auio.uio_iov = ma->v;
2008 	auio.uio_iovcnt = ma->len;
2009 	auio.uio_segflg = UIO_SYSSPACE;
2010 
2011 	error = ma->error;
2012 	if (!error)
2013 		error = vfs_donmount(curthread, flags, &auio);
2014 	free_mntarg(ma);
2015 	return (error);
2016 }
2017 
2018 /*
2019  * A printflike function to mount a filesystem.
2020  */
2021 int
2022 kernel_vmount(int flags, ...)
2023 {
2024 	struct mntarg *ma = NULL;
2025 	va_list ap;
2026 	const char *cp;
2027 	const void *vp;
2028 	int error;
2029 
2030 	va_start(ap, flags);
2031 	for (;;) {
2032 		cp = va_arg(ap, const char *);
2033 		if (cp == NULL)
2034 			break;
2035 		vp = va_arg(ap, const void *);
2036 		ma = mount_arg(ma, cp, vp, -1);
2037 	}
2038 	va_end(ap);
2039 
2040 	error = kernel_mount(ma, flags);
2041 	return (error);
2042 }
2043