xref: /freebsd/sys/kern/vfs_mount.c (revision d056fa046c6a91b90cd98165face0e42a33a5173)
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 #include <vm/uma.h>
61 
62 #include <geom/geom.h>
63 
64 #include <machine/stdarg.h>
65 
66 #include <security/audit/audit.h>
67 
68 #include "opt_rootdevname.h"
69 #include "opt_ddb.h"
70 #include "opt_mac.h"
71 
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75 
76 #define	ROOTNAME		"root_device"
77 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
78 
79 static int	vfs_domount(struct thread *td, const char *fstype,
80 		    char *fspath, int fsflags, void *fsdata);
81 static struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
82 		    const char *fspath, struct thread *td);
83 static int	vfs_mountroot_ask(void);
84 static int	vfs_mountroot_try(const char *mountfrom);
85 static int	vfs_donmount(struct thread *td, int fsflags,
86 		    struct uio *fsoptions);
87 static void	free_mntarg(struct mntarg *ma);
88 static void	vfs_mount_destroy(struct mount *);
89 static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
90 
91 static int	usermount = 0;
92 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
93     "Unprivileged users may mount and unmount file systems");
94 
95 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
96 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
97 static uma_zone_t mount_zone;
98 
99 /* List of mounted filesystems. */
100 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
101 
102 /* For any iteration/modification of mountlist */
103 struct mtx mountlist_mtx;
104 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
105 
106 TAILQ_HEAD(vfsoptlist, vfsopt);
107 struct vfsopt {
108 	TAILQ_ENTRY(vfsopt) link;
109 	char	*name;
110 	void	*value;
111 	int	len;
112 };
113 
114 /*
115  * The vnode of the system's root (/ in the filesystem, without chroot
116  * active.)
117  */
118 struct vnode	*rootvnode;
119 
120 /*
121  * The root filesystem is detailed in the kernel environment variable
122  * vfs.root.mountfrom, which is expected to be in the general format
123  *
124  * <vfsname>:[<path>]
125  * vfsname   := the name of a VFS known to the kernel and capable
126  *              of being mounted as root
127  * path      := disk device name or other data used by the filesystem
128  *              to locate its physical store
129  */
130 
131 /*
132  * Global opts, taken by all filesystems
133  */
134 static const char *global_opts[] = {
135 	"errmsg",
136 	"fstype",
137 	"fspath",
138 	"rdonly",
139 	"ro",
140 	"rw",
141 	"suid",
142 	"exec",
143 	"update",
144 	NULL
145 };
146 
147 /*
148  * The root specifiers we will try if RB_CDROM is specified.
149  */
150 static char *cdrom_rootdevnames[] = {
151 	"cd9660:cd0",
152 	"cd9660:acd0",
153 	NULL
154 };
155 
156 /* legacy find-root code */
157 char		*rootdevnames[2] = {NULL, NULL};
158 #ifndef ROOTDEVNAME
159 #  define ROOTDEVNAME NULL
160 #endif
161 static const char	*ctrootdevname = ROOTDEVNAME;
162 
163 /*
164  * ---------------------------------------------------------------------
165  * Functions for building and sanitizing the mount options
166  */
167 
168 /* Remove one mount option. */
169 static void
170 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
171 {
172 
173 	TAILQ_REMOVE(opts, opt, link);
174 	free(opt->name, M_MOUNT);
175 	if (opt->value != NULL)
176 		free(opt->value, M_MOUNT);
177 #ifdef INVARIANTS
178 	else if (opt->len != 0)
179 		panic("%s: mount option with NULL value but length != 0",
180 		    __func__);
181 #endif
182 	free(opt, M_MOUNT);
183 }
184 
185 /* Release all resources related to the mount options. */
186 static void
187 vfs_freeopts(struct vfsoptlist *opts)
188 {
189 	struct vfsopt *opt;
190 
191 	while (!TAILQ_EMPTY(opts)) {
192 		opt = TAILQ_FIRST(opts);
193 		vfs_freeopt(opts, opt);
194 	}
195 	free(opts, M_MOUNT);
196 }
197 
198 /*
199  * Check if options are equal (with or without the "no" prefix).
200  */
201 static int
202 vfs_equalopts(const char *opt1, const char *opt2)
203 {
204 
205 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
206 	if (strcmp(opt1, opt2) == 0)
207 		return (1);
208 	/* "noopt" vs. "opt" */
209 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
210 		return (1);
211 	/* "opt" vs. "noopt" */
212 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
213 		return (1);
214 	return (0);
215 }
216 
217 /*
218  * If a mount option is specified several times,
219  * (with or without the "no" prefix) only keep
220  * the last occurence of it.
221  */
222 static void
223 vfs_sanitizeopts(struct vfsoptlist *opts)
224 {
225 	struct vfsopt *opt, *opt2, *tmp;
226 
227 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
228 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
229 		while (opt2 != NULL) {
230 			if (vfs_equalopts(opt->name, opt2->name)) {
231 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
232 				vfs_freeopt(opts, opt2);
233 				opt2 = tmp;
234 			} else {
235 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
236 			}
237 		}
238 	}
239 }
240 
241 /*
242  * Build a linked list of mount options from a struct uio.
243  */
244 static int
245 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
246 {
247 	struct vfsoptlist *opts;
248 	struct vfsopt *opt;
249 	size_t memused;
250 	unsigned int i, iovcnt;
251 	int error, namelen, optlen;
252 
253 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
254 	TAILQ_INIT(opts);
255 	memused = 0;
256 	iovcnt = auio->uio_iovcnt;
257 	for (i = 0; i < iovcnt; i += 2) {
258 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
259 		namelen = auio->uio_iov[i].iov_len;
260 		optlen = auio->uio_iov[i + 1].iov_len;
261 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
262 		opt->value = NULL;
263 		opt->len = 0;
264 
265 		/*
266 		 * Do this early, so jumps to "bad" will free the current
267 		 * option.
268 		 */
269 		TAILQ_INSERT_TAIL(opts, opt, link);
270 		memused += sizeof(struct vfsopt) + optlen + namelen;
271 
272 		/*
273 		 * Avoid consuming too much memory, and attempts to overflow
274 		 * memused.
275 		 */
276 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
277 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
278 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
279 			error = EINVAL;
280 			goto bad;
281 		}
282 
283 		if (auio->uio_segflg == UIO_SYSSPACE) {
284 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
285 		} else {
286 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
287 			    namelen);
288 			if (error)
289 				goto bad;
290 		}
291 		/* Ensure names are null-terminated strings. */
292 		if (opt->name[namelen - 1] != '\0') {
293 			error = EINVAL;
294 			goto bad;
295 		}
296 		if (optlen != 0) {
297 			opt->len = optlen;
298 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
299 			if (auio->uio_segflg == UIO_SYSSPACE) {
300 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
301 				    optlen);
302 			} else {
303 				error = copyin(auio->uio_iov[i + 1].iov_base,
304 				    opt->value, optlen);
305 				if (error)
306 					goto bad;
307 			}
308 		}
309 	}
310 	vfs_sanitizeopts(opts);
311 	*options = opts;
312 	return (0);
313 bad:
314 	vfs_freeopts(opts);
315 	return (error);
316 }
317 
318 /*
319  * Merge the old mount options with the new ones passed
320  * in the MNT_UPDATE case.
321  */
322 static void
323 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
324 {
325 	struct vfsopt *opt, *opt2, *new;
326 
327 	TAILQ_FOREACH(opt, opts, link) {
328 		/*
329 		 * Check that this option hasn't been redefined
330 		 * nor cancelled with a "no" mount option.
331 		 */
332 		opt2 = TAILQ_FIRST(toopts);
333 		while (opt2 != NULL) {
334 			if (strcmp(opt2->name, opt->name) == 0)
335 				goto next;
336 			if (strncmp(opt2->name, "no", 2) == 0 &&
337 			    strcmp(opt2->name + 2, opt->name) == 0) {
338 				vfs_freeopt(toopts, opt2);
339 				goto next;
340 			}
341 			opt2 = TAILQ_NEXT(opt2, link);
342 		}
343 		/* We want this option, duplicate it. */
344 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
345 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
346 		strcpy(new->name, opt->name);
347 		if (opt->len != 0) {
348 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
349 			bcopy(opt->value, new->value, opt->len);
350 		} else {
351 			new->value = NULL;
352 		}
353 		new->len = opt->len;
354 		TAILQ_INSERT_TAIL(toopts, new, link);
355 next:
356 		continue;
357 	}
358 }
359 
360 /*
361  * ---------------------------------------------------------------------
362  * Mount a filesystem
363  */
364 int
365 nmount(td, uap)
366 	struct thread *td;
367 	struct nmount_args /* {
368 		struct iovec *iovp;
369 		unsigned int iovcnt;
370 		int flags;
371 	} */ *uap;
372 {
373 	struct uio *auio;
374 	struct iovec *iov;
375 	unsigned int i;
376 	int error;
377 	u_int iovcnt;
378 
379 	AUDIT_ARG(fflags, uap->flags);
380 
381 	/* Kick out MNT_ROOTFS early as it is legal internally */
382 	if (uap->flags & MNT_ROOTFS)
383 		return (EINVAL);
384 
385 	iovcnt = uap->iovcnt;
386 	/*
387 	 * Check that we have an even number of iovec's
388 	 * and that we have at least two options.
389 	 */
390 	if ((iovcnt & 1) || (iovcnt < 4))
391 		return (EINVAL);
392 
393 	error = copyinuio(uap->iovp, iovcnt, &auio);
394 	if (error)
395 		return (error);
396 	iov = auio->uio_iov;
397 	for (i = 0; i < iovcnt; i++) {
398 		if (iov->iov_len > MMAXOPTIONLEN) {
399 			free(auio, M_IOV);
400 			return (EINVAL);
401 		}
402 		iov++;
403 	}
404 	error = vfs_donmount(td, uap->flags, auio);
405 
406 	free(auio, M_IOV);
407 	return (error);
408 }
409 
410 /*
411  * ---------------------------------------------------------------------
412  * Various utility functions
413  */
414 
415 void
416 vfs_ref(struct mount *mp)
417 {
418 
419 	MNT_ILOCK(mp);
420 	MNT_REF(mp);
421 	MNT_IUNLOCK(mp);
422 }
423 
424 void
425 vfs_rel(struct mount *mp)
426 {
427 
428 	MNT_ILOCK(mp);
429 	MNT_REL(mp);
430 	MNT_IUNLOCK(mp);
431 }
432 
433 static int
434 mount_init(void *mem, int size, int flags)
435 {
436 	struct mount *mp;
437 
438 	mp = (struct mount *)mem;
439 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
440 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
441 	return (0);
442 }
443 
444 static void
445 mount_fini(void *mem, int size)
446 {
447 	struct mount *mp;
448 
449 	mp = (struct mount *)mem;
450 	lockdestroy(&mp->mnt_lock);
451 	mtx_destroy(&mp->mnt_mtx);
452 }
453 
454 /*
455  * Allocate and initialize the mount point struct.
456  */
457 static struct mount *
458 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
459     const char *fspath, struct thread *td)
460 {
461 	struct mount *mp;
462 
463 	mp = uma_zalloc(mount_zone, M_WAITOK);
464 	bzero(&mp->mnt_startzero,
465 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
466 	TAILQ_INIT(&mp->mnt_nvnodelist);
467 	mp->mnt_nvnodelistsize = 0;
468 	mp->mnt_ref = 0;
469 	(void) vfs_busy(mp, LK_NOWAIT, 0, td);
470 	mp->mnt_op = vfsp->vfc_vfsops;
471 	mp->mnt_vfc = vfsp;
472 	vfsp->vfc_refcount++;	/* XXX Unlocked */
473 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
474 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
475 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
476 	mp->mnt_vnodecovered = vp;
477 	mp->mnt_cred = crdup(td->td_ucred);
478 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
479 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
480 	mp->mnt_iosize_max = DFLTPHYS;
481 #ifdef MAC
482 	mac_init_mount(mp);
483 	mac_create_mount(td->td_ucred, mp);
484 #endif
485 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
486 	return (mp);
487 }
488 
489 /*
490  * Destroy the mount struct previously allocated by vfs_mount_alloc().
491  */
492 static void
493 vfs_mount_destroy(struct mount *mp)
494 {
495 	int i;
496 
497 	MNT_ILOCK(mp);
498 	for (i = 0; mp->mnt_ref && i < 3; i++)
499 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
500 	/*
501 	 * This will always cause a 3 second delay in rebooting due to
502 	 * refs on the root mountpoint that never go away.  Most of these
503 	 * are held by init which never exits.
504 	 */
505 	if (i == 3 && (!rebooting || bootverbose))
506 		printf("Mount point %s had %d dangling refs\n",
507 		    mp->mnt_stat.f_mntonname, mp->mnt_ref);
508 	if (mp->mnt_holdcnt != 0) {
509 		printf("Waiting for mount point to be unheld\n");
510 		while (mp->mnt_holdcnt != 0) {
511 			mp->mnt_holdcntwaiters++;
512 			msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
513 			       PZERO, "mntdestroy", 0);
514 			mp->mnt_holdcntwaiters--;
515 		}
516 		printf("mount point unheld\n");
517 	}
518 	if (mp->mnt_writeopcount > 0) {
519 		printf("Waiting for mount point write ops\n");
520 		while (mp->mnt_writeopcount > 0) {
521 			mp->mnt_kern_flag |= MNTK_SUSPEND;
522 			msleep(&mp->mnt_writeopcount,
523 			       MNT_MTX(mp),
524 			       PZERO, "mntdestroy2", 0);
525 		}
526 		printf("mount point write ops completed\n");
527 	}
528 	if (mp->mnt_secondary_writes > 0) {
529 		printf("Waiting for mount point secondary write ops\n");
530 		while (mp->mnt_secondary_writes > 0) {
531 			mp->mnt_kern_flag |= MNTK_SUSPEND;
532 			msleep(&mp->mnt_secondary_writes,
533 			       MNT_MTX(mp),
534 			       PZERO, "mntdestroy3", 0);
535 		}
536 		printf("mount point secondary write ops completed\n");
537 	}
538 	MNT_IUNLOCK(mp);
539 	mp->mnt_vfc->vfc_refcount--;
540 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
541 		struct vnode *vp;
542 
543 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
544 			vprint("", vp);
545 		panic("unmount: dangling vnode");
546 	}
547 	MNT_ILOCK(mp);
548 	if (mp->mnt_kern_flag & MNTK_MWAIT)
549 		wakeup(mp);
550 	if (mp->mnt_writeopcount != 0)
551 		panic("vfs_mount_destroy: nonzero writeopcount");
552 	if (mp->mnt_secondary_writes != 0)
553 		panic("vfs_mount_destroy: nonzero secondary_writes");
554 	if (mp->mnt_nvnodelistsize != 0)
555 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
556 	mp->mnt_writeopcount = -1000;
557 	mp->mnt_nvnodelistsize = -1000;
558 	mp->mnt_secondary_writes = -1000;
559 	MNT_IUNLOCK(mp);
560 #ifdef MAC
561 	mac_destroy_mount(mp);
562 #endif
563 	if (mp->mnt_opt != NULL)
564 		vfs_freeopts(mp->mnt_opt);
565 	crfree(mp->mnt_cred);
566 	uma_zfree(mount_zone, mp);
567 }
568 
569 static int
570 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
571 {
572 	struct vfsoptlist *optlist;
573 	struct vfsopt *opt, *noro_opt;
574 	char *fstype, *fspath, *errmsg;
575 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
576 	int has_rw, has_noro;
577 
578 	errmsg_len = 0;
579 	errmsg_pos = -1;
580 	has_rw = 0;
581 	has_noro = 0;
582 
583 	error = vfs_buildopts(fsoptions, &optlist);
584 	if (error)
585 		return (error);
586 
587 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
588 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
589 	else
590 		errmsg_len = 0;
591 
592 	/*
593 	 * We need these two options before the others,
594 	 * and they are mandatory for any filesystem.
595 	 * Ensure they are NUL terminated as well.
596 	 */
597 	fstypelen = 0;
598 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
599 	if (error || fstype[fstypelen - 1] != '\0') {
600 		error = EINVAL;
601 		if (errmsg != NULL)
602 			strncpy(errmsg, "Invalid fstype", errmsg_len);
603 		goto bail;
604 	}
605 	fspathlen = 0;
606 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
607 	if (error || fspath[fspathlen - 1] != '\0') {
608 		error = EINVAL;
609 		if (errmsg != NULL)
610 			strncpy(errmsg, "Invalid fspath", errmsg_len);
611 		goto bail;
612 	}
613 
614 	/*
615 	 * We need to see if we have the "update" option
616 	 * before we call vfs_domount(), since vfs_domount() has special
617 	 * logic based on MNT_UPDATE.  This is very important
618 	 * when we want to update the root filesystem.
619 	 */
620 	TAILQ_FOREACH(opt, optlist, link) {
621 		if (strcmp(opt->name, "update") == 0)
622 			fsflags |= MNT_UPDATE;
623 		else if (strcmp(opt->name, "async") == 0)
624 			fsflags |= MNT_ASYNC;
625 		else if (strcmp(opt->name, "force") == 0)
626 			fsflags |= MNT_FORCE;
627 		else if (strcmp(opt->name, "multilabel") == 0)
628 			fsflags |= MNT_MULTILABEL;
629 		else if (strcmp(opt->name, "noasync") == 0)
630 			fsflags &= ~MNT_ASYNC;
631 		else if (strcmp(opt->name, "noatime") == 0)
632 			fsflags |= MNT_NOATIME;
633 		else if (strcmp(opt->name, "noclusterr") == 0)
634 			fsflags |= MNT_NOCLUSTERR;
635 		else if (strcmp(opt->name, "noclusterw") == 0)
636 			fsflags |= MNT_NOCLUSTERW;
637 		else if (strcmp(opt->name, "noexec") == 0)
638 			fsflags |= MNT_NOEXEC;
639 		else if (strcmp(opt->name, "nosuid") == 0)
640 			fsflags |= MNT_NOSUID;
641 		else if (strcmp(opt->name, "nosymfollow") == 0)
642 			fsflags |= MNT_NOSYMFOLLOW;
643 		else if (strcmp(opt->name, "noro") == 0) {
644 			fsflags &= ~MNT_RDONLY;
645 			has_noro = 1;
646 		}
647 		else if (strcmp(opt->name, "rw") == 0) {
648 			fsflags &= ~MNT_RDONLY;
649 			has_rw = 1;
650 		}
651 		else if (strcmp(opt->name, "ro") == 0 ||
652 		    strcmp(opt->name, "rdonly") == 0)
653 			fsflags |= MNT_RDONLY;
654 		else if (strcmp(opt->name, "snapshot") == 0)
655 			fsflags |= MNT_SNAPSHOT;
656 		else if (strcmp(opt->name, "suiddir") == 0)
657 			fsflags |= MNT_SUIDDIR;
658 		else if (strcmp(opt->name, "sync") == 0)
659 			fsflags |= MNT_SYNCHRONOUS;
660 		else if (strcmp(opt->name, "union") == 0)
661 			fsflags |= MNT_UNION;
662 	}
663 
664 	/*
665 	 * If "rw" was specified as a mount option, and we
666 	 * are trying to update a mount-point from "ro" to "rw",
667 	 * we need a mount option "noro", since in vfs_mergeopts(),
668 	 * "noro" will cancel "ro", but "rw" will not do anything.
669 	 */
670 	if (has_rw && !has_noro) {
671 		noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
672 		noro_opt->name = strdup("noro", M_MOUNT);
673 		noro_opt->value = NULL;
674 		noro_opt->len = 0;
675 		TAILQ_INSERT_TAIL(optlist, noro_opt, link);
676 	}
677 
678 	/*
679 	 * Be ultra-paranoid about making sure the type and fspath
680 	 * variables will fit in our mp buffers, including the
681 	 * terminating NUL.
682 	 */
683 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
684 		error = ENAMETOOLONG;
685 		goto bail;
686 	}
687 
688 	mtx_lock(&Giant);
689 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
690 	mtx_unlock(&Giant);
691 bail:
692 	/* copyout the errmsg */
693 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
694 	    && errmsg_len > 0 && errmsg != NULL) {
695 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
696 			strncpy(fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
697 			    errmsg,
698 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
699 		} else {
700 			copystr(errmsg,
701 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
702 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len,
703 			    NULL);
704 		}
705 	}
706 
707 	if (error != 0)
708 		vfs_freeopts(optlist);
709 	return (error);
710 }
711 
712 /*
713  * ---------------------------------------------------------------------
714  * Old mount API.
715  */
716 #ifndef _SYS_SYSPROTO_H_
717 struct mount_args {
718 	char	*type;
719 	char	*path;
720 	int	flags;
721 	caddr_t	data;
722 };
723 #endif
724 /* ARGSUSED */
725 int
726 mount(td, uap)
727 	struct thread *td;
728 	struct mount_args /* {
729 		char *type;
730 		char *path;
731 		int flags;
732 		caddr_t data;
733 	} */ *uap;
734 {
735 	char *fstype;
736 	struct vfsconf *vfsp = NULL;
737 	struct mntarg *ma = NULL;
738 	int error;
739 
740 	AUDIT_ARG(fflags, uap->flags);
741 
742 	/* Kick out MNT_ROOTFS early as it is legal internally */
743 	uap->flags &= ~MNT_ROOTFS;
744 
745 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
746 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
747 	if (error) {
748 		free(fstype, M_TEMP);
749 		return (error);
750 	}
751 
752 	AUDIT_ARG(text, fstype);
753 	mtx_lock(&Giant);
754 	vfsp = vfs_byname_kld(fstype, td, &error);
755 	free(fstype, M_TEMP);
756 	if (vfsp == NULL) {
757 		mtx_unlock(&Giant);
758 		return (ENOENT);
759 	}
760 	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
761 		mtx_unlock(&Giant);
762 		return (EOPNOTSUPP);
763 	}
764 
765 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
766 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
767 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
768 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
769 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
770 
771 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
772 	mtx_unlock(&Giant);
773 	return (error);
774 }
775 
776 
777 /*
778  * vfs_domount(): actually attempt a filesystem mount.
779  */
780 static int
781 vfs_domount(
782 	struct thread *td,	/* Flags common to all filesystems. */
783 	const char *fstype,	/* Filesystem type. */
784 	char *fspath,		/* Mount path. */
785 	int fsflags,		/* Flags common to all filesystems. */
786 	void *fsdata		/* Options local to the filesystem. */
787 	)
788 {
789 	struct vnode *vp;
790 	struct mount *mp;
791 	struct vfsconf *vfsp;
792 	struct export_args export;
793 	int error, flag = 0, kern_flag = 0;
794 	struct vattr va;
795 	struct nameidata nd;
796 
797 	mtx_assert(&Giant, MA_OWNED);
798 	/*
799 	 * Be ultra-paranoid about making sure the type and fspath
800 	 * variables will fit in our mp buffers, including the
801 	 * terminating NUL.
802 	 */
803 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
804 		return (ENAMETOOLONG);
805 
806 	if (jailed(td->td_ucred))
807 		return (EPERM);
808 	if (usermount == 0) {
809 		if ((error = suser(td)) != 0)
810 			return (error);
811 	}
812 
813 	/*
814 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
815 	 */
816 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
817 		if ((error = suser(td)) != 0)
818 			return (error);
819 	}
820 	/*
821 	 * Silently enforce MNT_NOSUID and MNT_USER for
822 	 * unprivileged users.
823 	 */
824 	if (suser(td) != 0)
825 		fsflags |= MNT_NOSUID | MNT_USER;
826 
827 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
828 	vfsp = NULL;
829 	if ((fsflags & MNT_UPDATE) == 0) {
830 		/* Don't try to load KLDs if we're mounting the root. */
831 		if (fsflags & MNT_ROOTFS)
832 			vfsp = vfs_byname(fstype);
833 		else
834 			vfsp = vfs_byname_kld(fstype, td, &error);
835 		if (vfsp == NULL)
836 			return (ENODEV);
837 	}
838 	/*
839 	 * Get vnode to be covered
840 	 */
841 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
842 	    fspath, td);
843 	if ((error = namei(&nd)) != 0)
844 		return (error);
845 	NDFREE(&nd, NDF_ONLY_PNBUF);
846 	vp = nd.ni_vp;
847 	if (fsflags & MNT_UPDATE) {
848 		if ((vp->v_vflag & VV_ROOT) == 0) {
849 			vput(vp);
850 			return (EINVAL);
851 		}
852 		mp = vp->v_mount;
853 		flag = mp->mnt_flag;
854 		kern_flag = mp->mnt_kern_flag;
855 		/*
856 		 * We only allow the filesystem to be reloaded if it
857 		 * is currently mounted read-only.
858 		 */
859 		if ((fsflags & MNT_RELOAD) &&
860 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
861 			vput(vp);
862 			return (EOPNOTSUPP);	/* Needs translation */
863 		}
864 		/*
865 		 * Only privileged root, or (if MNT_USER is set) the user that
866 		 * did the original mount is permitted to update it.
867 		 */
868 		error = vfs_suser(mp, td);
869 		if (error) {
870 			vput(vp);
871 			return (error);
872 		}
873 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
874 			vput(vp);
875 			return (EBUSY);
876 		}
877 		VI_LOCK(vp);
878 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
879 		    vp->v_mountedhere != NULL) {
880 			VI_UNLOCK(vp);
881 			vfs_unbusy(mp, td);
882 			vput(vp);
883 			return (EBUSY);
884 		}
885 		vp->v_iflag |= VI_MOUNT;
886 		VI_UNLOCK(vp);
887 		mp->mnt_flag |= fsflags &
888 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
889 		VOP_UNLOCK(vp, 0, td);
890 		mp->mnt_optnew = fsdata;
891 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
892 	} else {
893 		/*
894 		 * If the user is not root, ensure that they own the directory
895 		 * onto which we are attempting to mount.
896 		 */
897 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
898 		if (error) {
899 			vput(vp);
900 			return (error);
901 		}
902 		if (va.va_uid != td->td_ucred->cr_uid) {
903 			if ((error = suser(td)) != 0) {
904 				vput(vp);
905 				return (error);
906 			}
907 		}
908 		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
909 		if (error != 0) {
910 			vput(vp);
911 			return (error);
912 		}
913 		if (vp->v_type != VDIR) {
914 			vput(vp);
915 			return (ENOTDIR);
916 		}
917 		VI_LOCK(vp);
918 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
919 		    vp->v_mountedhere != NULL) {
920 			VI_UNLOCK(vp);
921 			vput(vp);
922 			return (EBUSY);
923 		}
924 		vp->v_iflag |= VI_MOUNT;
925 		VI_UNLOCK(vp);
926 
927 		/*
928 		 * Allocate and initialize the filesystem.
929 		 */
930 		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
931 		VOP_UNLOCK(vp, 0, td);
932 
933 		/* XXXMAC: pass to vfs_mount_alloc? */
934 		mp->mnt_optnew = fsdata;
935 	}
936 
937 	/*
938 	 * Set the mount level flags.
939 	 */
940 	if (fsflags & MNT_RDONLY)
941 		mp->mnt_flag |= MNT_RDONLY;
942 	mp->mnt_flag &=~ MNT_UPDATEMASK;
943 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
944 	/*
945 	 * Mount the filesystem.
946 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
947 	 * get.  No freeing of cn_pnbuf.
948 	 */
949         error = VFS_MOUNT(mp, td);
950 
951 	/*
952 	 * Process the export option only if we are
953 	 * updating mount options.
954 	 */
955 	if (!error && (fsflags & MNT_UPDATE)) {
956 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
957 		    sizeof(export)) == 0)
958 			error = vfs_export(mp, &export);
959 	}
960 
961 	if (!error) {
962 		if (mp->mnt_opt != NULL)
963 			vfs_freeopts(mp->mnt_opt);
964 		mp->mnt_opt = mp->mnt_optnew;
965 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
966 	}
967 	/*
968 	 * Prevent external consumers of mount options from reading
969 	 * mnt_optnew.
970 	*/
971 	mp->mnt_optnew = NULL;
972 	if (mp->mnt_flag & MNT_UPDATE) {
973 		mp->mnt_flag &=
974 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
975 		if (error) {
976 			mp->mnt_flag = flag;
977 			mp->mnt_kern_flag = kern_flag;
978 		}
979 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
980 			if (mp->mnt_syncer == NULL)
981 				error = vfs_allocate_syncvnode(mp);
982 		} else {
983 			if (mp->mnt_syncer != NULL)
984 				vrele(mp->mnt_syncer);
985 			mp->mnt_syncer = NULL;
986 		}
987 		vfs_unbusy(mp, td);
988 		VI_LOCK(vp);
989 		vp->v_iflag &= ~VI_MOUNT;
990 		VI_UNLOCK(vp);
991 		vrele(vp);
992 		return (error);
993 	}
994 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
995 	/*
996 	 * Put the new filesystem on the mount list after root.
997 	 */
998 	cache_purge(vp);
999 	if (!error) {
1000 		struct vnode *newdp;
1001 
1002 		VI_LOCK(vp);
1003 		vp->v_iflag &= ~VI_MOUNT;
1004 		VI_UNLOCK(vp);
1005 		vp->v_mountedhere = mp;
1006 		mtx_lock(&mountlist_mtx);
1007 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1008 		mtx_unlock(&mountlist_mtx);
1009 		vfs_event_signal(NULL, VQ_MOUNT, 0);
1010 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1011 			panic("mount: lost mount");
1012 		mountcheckdirs(vp, newdp);
1013 		vput(newdp);
1014 		VOP_UNLOCK(vp, 0, td);
1015 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1016 			error = vfs_allocate_syncvnode(mp);
1017 		vfs_unbusy(mp, td);
1018 		if (error)
1019 			vrele(vp);
1020 	} else {
1021 		VI_LOCK(vp);
1022 		vp->v_iflag &= ~VI_MOUNT;
1023 		VI_UNLOCK(vp);
1024 		vfs_unbusy(mp, td);
1025 		vfs_mount_destroy(mp);
1026 		vput(vp);
1027 	}
1028 	return (error);
1029 }
1030 
1031 /*
1032  * ---------------------------------------------------------------------
1033  * Unmount a filesystem.
1034  *
1035  * Note: unmount takes a path to the vnode mounted on as argument,
1036  * not special file (as before).
1037  */
1038 #ifndef _SYS_SYSPROTO_H_
1039 struct unmount_args {
1040 	char	*path;
1041 	int	flags;
1042 };
1043 #endif
1044 /* ARGSUSED */
1045 int
1046 unmount(td, uap)
1047 	struct thread *td;
1048 	register struct unmount_args /* {
1049 		char *path;
1050 		int flags;
1051 	} */ *uap;
1052 {
1053 	struct mount *mp;
1054 	char *pathbuf;
1055 	int error, id0, id1;
1056 
1057 	if (jailed(td->td_ucred))
1058 		return (EPERM);
1059 	if (usermount == 0) {
1060 		if ((error = suser(td)) != 0)
1061 			return (error);
1062 	}
1063 
1064 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1065 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1066 	if (error) {
1067 		free(pathbuf, M_TEMP);
1068 		return (error);
1069 	}
1070 	AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1071 	mtx_lock(&Giant);
1072 	if (uap->flags & MNT_BYFSID) {
1073 		/* Decode the filesystem ID. */
1074 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1075 			mtx_unlock(&Giant);
1076 			free(pathbuf, M_TEMP);
1077 			return (EINVAL);
1078 		}
1079 
1080 		mtx_lock(&mountlist_mtx);
1081 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1082 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1083 			    mp->mnt_stat.f_fsid.val[1] == id1)
1084 				break;
1085 		}
1086 		mtx_unlock(&mountlist_mtx);
1087 	} else {
1088 		mtx_lock(&mountlist_mtx);
1089 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1090 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1091 				break;
1092 		}
1093 		mtx_unlock(&mountlist_mtx);
1094 	}
1095 	free(pathbuf, M_TEMP);
1096 	if (mp == NULL) {
1097 		/*
1098 		 * Previously we returned ENOENT for a nonexistent path and
1099 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1100 		 * now, so in the !MNT_BYFSID case return the more likely
1101 		 * EINVAL for compatibility.
1102 		 */
1103 		mtx_unlock(&Giant);
1104 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1105 	}
1106 
1107 	/*
1108 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1109 	 * original mount is permitted to unmount this filesystem.
1110 	 */
1111 	error = vfs_suser(mp, td);
1112 	if (error) {
1113 		mtx_unlock(&Giant);
1114 		return (error);
1115 	}
1116 
1117 	/*
1118 	 * Don't allow unmounting the root filesystem.
1119 	 */
1120 	if (mp->mnt_flag & MNT_ROOTFS) {
1121 		mtx_unlock(&Giant);
1122 		return (EINVAL);
1123 	}
1124 	error = dounmount(mp, uap->flags, td);
1125 	mtx_unlock(&Giant);
1126 	return (error);
1127 }
1128 
1129 /*
1130  * Do the actual filesystem unmount.
1131  */
1132 int
1133 dounmount(mp, flags, td)
1134 	struct mount *mp;
1135 	int flags;
1136 	struct thread *td;
1137 {
1138 	struct vnode *coveredvp, *fsrootvp;
1139 	int error;
1140 	int async_flag;
1141 
1142 	mtx_assert(&Giant, MA_OWNED);
1143 
1144 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1145 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY, td);
1146 	MNT_ILOCK(mp);
1147 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1148 		MNT_IUNLOCK(mp);
1149 		if (coveredvp)
1150 			VOP_UNLOCK(coveredvp, 0, td);
1151 		return (EBUSY);
1152 	}
1153 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1154 	/* Allow filesystems to detect that a forced unmount is in progress. */
1155 	if (flags & MNT_FORCE)
1156 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1157 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1158 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1159 	if (error) {
1160 		MNT_ILOCK(mp);
1161 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1162 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1163 			wakeup(mp);
1164 		MNT_IUNLOCK(mp);
1165 		if (coveredvp)
1166 			VOP_UNLOCK(coveredvp, 0, td);
1167 		return (error);
1168 	}
1169 	vn_start_write(NULL, &mp, V_WAIT);
1170 
1171 	if (mp->mnt_flag & MNT_EXPUBLIC)
1172 		vfs_setpublicfs(NULL, NULL, NULL);
1173 
1174 	vfs_msync(mp, MNT_WAIT);
1175 	async_flag = mp->mnt_flag & MNT_ASYNC;
1176 	mp->mnt_flag &= ~MNT_ASYNC;
1177 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1178 	if (mp->mnt_syncer != NULL)
1179 		vrele(mp->mnt_syncer);
1180 	/*
1181 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1182 	 * vnode to the covered vnode.  For non-forced unmounts we want
1183 	 * such references to cause an EBUSY error.
1184 	 */
1185 	if ((flags & MNT_FORCE) &&
1186 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1187 		if (mp->mnt_vnodecovered != NULL)
1188 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1189 		if (fsrootvp == rootvnode) {
1190 			vrele(rootvnode);
1191 			rootvnode = NULL;
1192 		}
1193 		vput(fsrootvp);
1194 	}
1195 	if (((mp->mnt_flag & MNT_RDONLY) ||
1196 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1197 	    (flags & MNT_FORCE)) {
1198 		error = VFS_UNMOUNT(mp, flags, td);
1199 	}
1200 	vn_finished_write(mp);
1201 	if (error) {
1202 		/* Undo cdir/rdir and rootvnode changes made above. */
1203 		if ((flags & MNT_FORCE) &&
1204 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1205 			if (mp->mnt_vnodecovered != NULL)
1206 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1207 			if (rootvnode == NULL) {
1208 				rootvnode = fsrootvp;
1209 				vref(rootvnode);
1210 			}
1211 			vput(fsrootvp);
1212 		}
1213 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1214 			(void) vfs_allocate_syncvnode(mp);
1215 		MNT_ILOCK(mp);
1216 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1217 		mp->mnt_flag |= async_flag;
1218 		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1219 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1220 			wakeup(mp);
1221 		MNT_IUNLOCK(mp);
1222 		if (coveredvp)
1223 			VOP_UNLOCK(coveredvp, 0, td);
1224 		return (error);
1225 	}
1226 	mtx_lock(&mountlist_mtx);
1227 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1228 	mtx_unlock(&mountlist_mtx);
1229 	if (coveredvp != NULL) {
1230 		coveredvp->v_mountedhere = NULL;
1231 		vput(coveredvp);
1232 	}
1233 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1234 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1235 	vfs_mount_destroy(mp);
1236 	return (0);
1237 }
1238 
1239 /*
1240  * ---------------------------------------------------------------------
1241  * Mounting of root filesystem
1242  *
1243  */
1244 
1245 struct root_hold_token {
1246 	const char 			*who;
1247 	LIST_ENTRY(root_hold_token)	list;
1248 };
1249 
1250 static LIST_HEAD(, root_hold_token)	root_holds =
1251     LIST_HEAD_INITIALIZER(&root_holds);
1252 
1253 struct root_hold_token *
1254 root_mount_hold(const char *identifier)
1255 {
1256 	struct root_hold_token *h;
1257 
1258 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1259 	h->who = identifier;
1260 	mtx_lock(&mountlist_mtx);
1261 	LIST_INSERT_HEAD(&root_holds, h, list);
1262 	mtx_unlock(&mountlist_mtx);
1263 	return (h);
1264 }
1265 
1266 void
1267 root_mount_rel(struct root_hold_token *h)
1268 {
1269 
1270 	mtx_lock(&mountlist_mtx);
1271 	LIST_REMOVE(h, list);
1272 	wakeup(&root_holds);
1273 	mtx_unlock(&mountlist_mtx);
1274 	free(h, M_DEVBUF);
1275 }
1276 
1277 static void
1278 root_mount_wait(void)
1279 {
1280 	struct root_hold_token *h;
1281 
1282 	for (;;) {
1283 		DROP_GIANT();
1284 		g_waitidle();
1285 		PICKUP_GIANT();
1286 		mtx_lock(&mountlist_mtx);
1287 		if (LIST_EMPTY(&root_holds)) {
1288 			mtx_unlock(&mountlist_mtx);
1289 			break;
1290 		}
1291 		printf("Root mount waiting for:");
1292 		LIST_FOREACH(h, &root_holds, list)
1293 			printf(" %s", h->who);
1294 		printf("\n");
1295 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1296 		    hz);
1297 	}
1298 }
1299 
1300 static void
1301 set_rootvnode(struct thread *td)
1302 {
1303 	struct proc *p;
1304 
1305 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1306 		panic("Cannot find root vnode");
1307 
1308 	p = td->td_proc;
1309 	FILEDESC_LOCK(p->p_fd);
1310 
1311 	if (p->p_fd->fd_cdir != NULL)
1312 		vrele(p->p_fd->fd_cdir);
1313 	p->p_fd->fd_cdir = rootvnode;
1314 	VREF(rootvnode);
1315 
1316 	if (p->p_fd->fd_rdir != NULL)
1317 		vrele(p->p_fd->fd_rdir);
1318 	p->p_fd->fd_rdir = rootvnode;
1319 	VREF(rootvnode);
1320 
1321 	FILEDESC_UNLOCK(p->p_fd);
1322 
1323 	VOP_UNLOCK(rootvnode, 0, td);
1324 }
1325 
1326 /*
1327  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1328  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1329  */
1330 
1331 static void
1332 devfs_first(void)
1333 {
1334 	struct thread *td = curthread;
1335 	struct vfsoptlist *opts;
1336 	struct vfsconf *vfsp;
1337 	struct mount *mp = NULL;
1338 	int error;
1339 
1340 	vfsp = vfs_byname("devfs");
1341 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1342 	if (vfsp == NULL)
1343 		return;
1344 
1345 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1346 
1347 	error = VFS_MOUNT(mp, td);
1348 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1349 	if (error)
1350 		return;
1351 
1352 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1353 	TAILQ_INIT(opts);
1354 	mp->mnt_opt = opts;
1355 
1356 	mtx_lock(&mountlist_mtx);
1357 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1358 	mtx_unlock(&mountlist_mtx);
1359 
1360 	set_rootvnode(td);
1361 
1362 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1363 	if (error)
1364 		printf("kern_symlink /dev -> / returns %d\n", error);
1365 }
1366 
1367 /*
1368  * Surgically move our devfs to be mounted on /dev.
1369  */
1370 
1371 static void
1372 devfs_fixup(struct thread *td)
1373 {
1374 	struct nameidata nd;
1375 	int error;
1376 	struct vnode *vp, *dvp;
1377 	struct mount *mp;
1378 
1379 	/* Remove our devfs mount from the mountlist and purge the cache */
1380 	mtx_lock(&mountlist_mtx);
1381 	mp = TAILQ_FIRST(&mountlist);
1382 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1383 	mtx_unlock(&mountlist_mtx);
1384 	cache_purgevfs(mp);
1385 
1386 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1387 	VI_LOCK(dvp);
1388 	dvp->v_iflag &= ~VI_MOUNT;
1389 	dvp->v_mountedhere = NULL;
1390 	VI_UNLOCK(dvp);
1391 
1392 	/* Set up the real rootvnode, and purge the cache */
1393 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1394 	set_rootvnode(td);
1395 	cache_purgevfs(rootvnode->v_mount);
1396 
1397 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1398 	error = namei(&nd);
1399 	if (error) {
1400 		printf("Lookup of /dev for devfs, error: %d\n", error);
1401 		return;
1402 	}
1403 	NDFREE(&nd, NDF_ONLY_PNBUF);
1404 	vp = nd.ni_vp;
1405 	if (vp->v_type != VDIR) {
1406 		vput(vp);
1407 	}
1408 	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1409 	if (error) {
1410 		vput(vp);
1411 	}
1412 	cache_purge(vp);
1413 	mp->mnt_vnodecovered = vp;
1414 	vp->v_mountedhere = mp;
1415 	mtx_lock(&mountlist_mtx);
1416 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1417 	mtx_unlock(&mountlist_mtx);
1418 	VOP_UNLOCK(vp, 0, td);
1419 	vput(dvp);
1420 	vfs_unbusy(mp, td);
1421 
1422 	/* Unlink the no longer needed /dev/dev -> / symlink */
1423 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1424 }
1425 
1426 /*
1427  * Report errors during filesystem mounting.
1428  */
1429 void
1430 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1431 {
1432 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1433 	va_list ap;
1434 	int error, len;
1435 	char *errmsg;
1436 
1437 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1438 	if (error || errmsg == NULL || len <= 0)
1439 		return;
1440 
1441 	va_start(ap, fmt);
1442 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1443 	va_end(ap);
1444 }
1445 
1446 /*
1447  * Find and mount the root filesystem
1448  */
1449 void
1450 vfs_mountroot(void)
1451 {
1452 	char *cp;
1453 	int error, i, asked = 0;
1454 
1455 	root_mount_wait();
1456 
1457 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1458 	    NULL, NULL, mount_init, mount_fini,
1459 	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1460 	devfs_first();
1461 
1462 	/*
1463 	 * We are booted with instructions to prompt for the root filesystem.
1464 	 */
1465 	if (boothowto & RB_ASKNAME) {
1466 		if (!vfs_mountroot_ask())
1467 			return;
1468 		asked = 1;
1469 	}
1470 
1471 	/*
1472 	 * The root filesystem information is compiled in, and we are
1473 	 * booted with instructions to use it.
1474 	 */
1475 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1476 		if (!vfs_mountroot_try(ctrootdevname))
1477 			return;
1478 		ctrootdevname = NULL;
1479 	}
1480 
1481 	/*
1482 	 * We've been given the generic "use CDROM as root" flag.  This is
1483 	 * necessary because one media may be used in many different
1484 	 * devices, so we need to search for them.
1485 	 */
1486 	if (boothowto & RB_CDROM) {
1487 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1488 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1489 				return;
1490 		}
1491 	}
1492 
1493 	/*
1494 	 * Try to use the value read by the loader from /etc/fstab, or
1495 	 * supplied via some other means.  This is the preferred
1496 	 * mechanism.
1497 	 */
1498 	cp = getenv("vfs.root.mountfrom");
1499 	if (cp != NULL) {
1500 		error = vfs_mountroot_try(cp);
1501 		freeenv(cp);
1502 		if (!error)
1503 			return;
1504 	}
1505 
1506 	/*
1507 	 * Try values that may have been computed by code during boot
1508 	 */
1509 	if (!vfs_mountroot_try(rootdevnames[0]))
1510 		return;
1511 	if (!vfs_mountroot_try(rootdevnames[1]))
1512 		return;
1513 
1514 	/*
1515 	 * If we (still) have a compiled-in default, try it.
1516 	 */
1517 	if (ctrootdevname != NULL)
1518 		if (!vfs_mountroot_try(ctrootdevname))
1519 			return;
1520 	/*
1521 	 * Everything so far has failed, prompt on the console if we haven't
1522 	 * already tried that.
1523 	 */
1524 	if (!asked)
1525 		if (!vfs_mountroot_ask())
1526 			return;
1527 
1528 	panic("Root mount failed, startup aborted.");
1529 }
1530 
1531 /*
1532  * Mount (mountfrom) as the root filesystem.
1533  */
1534 static int
1535 vfs_mountroot_try(const char *mountfrom)
1536 {
1537 	struct mount	*mp;
1538 	char		*vfsname, *path;
1539 	time_t		timebase;
1540 	int		error;
1541 	char		patt[32];
1542 
1543 	vfsname = NULL;
1544 	path    = NULL;
1545 	mp      = NULL;
1546 	error   = EINVAL;
1547 
1548 	if (mountfrom == NULL)
1549 		return (error);		/* don't complain */
1550 	printf("Trying to mount root from %s\n", mountfrom);
1551 
1552 	/* parse vfs name and path */
1553 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1554 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1555 	vfsname[0] = path[0] = 0;
1556 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1557 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1558 		goto out;
1559 
1560 	if (path[0] == '\0')
1561 		strcpy(path, ROOTNAME);
1562 
1563 	error = kernel_vmount(
1564 	    MNT_RDONLY | MNT_ROOTFS,
1565 	    "fstype", vfsname,
1566 	    "fspath", "/",
1567 	    "from", path,
1568 	    NULL);
1569 	if (error == 0) {
1570 		/*
1571 		 * We mount devfs prior to mounting the / FS, so the first
1572 		 * entry will typically be devfs.
1573 		 */
1574 		mp = TAILQ_FIRST(&mountlist);
1575 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1576 
1577 		/*
1578 		 * Iterate over all currently mounted file systems and use
1579 		 * the time stamp found to check and/or initialize the RTC.
1580 		 * Typically devfs has no time stamp and the only other FS
1581 		 * is the actual / FS.
1582 		 * Call inittodr() only once and pass it the largest of the
1583 		 * timestamps we encounter.
1584 		 */
1585 		timebase = 0;
1586 		do {
1587 			if (mp->mnt_time > timebase)
1588 				timebase = mp->mnt_time;
1589 			mp = TAILQ_NEXT(mp, mnt_list);
1590 		} while (mp != NULL);
1591 		inittodr(timebase);
1592 
1593 		devfs_fixup(curthread);
1594 	}
1595 out:
1596 	free(path, M_MOUNT);
1597 	free(vfsname, M_MOUNT);
1598 	return (error);
1599 }
1600 
1601 /*
1602  * ---------------------------------------------------------------------
1603  * Interactive root filesystem selection code.
1604  */
1605 
1606 static int
1607 vfs_mountroot_ask(void)
1608 {
1609 	char name[128];
1610 
1611 	for(;;) {
1612 		printf("\nManual root filesystem specification:\n");
1613 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1614 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1615 		printf("                       eg. ufs:da0s1a\n");
1616 #else
1617 		printf("                       eg. ufs:/dev/da0a\n");
1618 #endif
1619 		printf("  ?                  List valid disk boot devices\n");
1620 		printf("  <empty line>       Abort manual input\n");
1621 		printf("\nmountroot> ");
1622 		gets(name, sizeof(name), 1);
1623 		if (name[0] == '\0')
1624 			return (1);
1625 		if (name[0] == '?') {
1626 			printf("\nList of GEOM managed disk devices:\n  ");
1627 			g_dev_print();
1628 			continue;
1629 		}
1630 		if (!vfs_mountroot_try(name))
1631 			return (0);
1632 	}
1633 }
1634 
1635 /*
1636  * ---------------------------------------------------------------------
1637  * Functions for querying mount options/arguments from filesystems.
1638  */
1639 
1640 /*
1641  * Check that no unknown options are given
1642  */
1643 int
1644 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1645 {
1646 	struct vfsopt *opt;
1647 	const char **t, *p;
1648 
1649 
1650 	TAILQ_FOREACH(opt, opts, link) {
1651 		p = opt->name;
1652 		if (p[0] == 'n' && p[1] == 'o')
1653 			p += 2;
1654 		for(t = global_opts; *t != NULL; t++)
1655 			if (!strcmp(*t, p))
1656 				break;
1657 		if (*t != NULL)
1658 			continue;
1659 		for(t = legal; *t != NULL; t++)
1660 			if (!strcmp(*t, p))
1661 				break;
1662 		if (*t != NULL)
1663 			continue;
1664 		printf("mount option <%s> is unknown\n", p);
1665 		return (EINVAL);
1666 	}
1667 	return (0);
1668 }
1669 
1670 /*
1671  * Get a mount option by its name.
1672  *
1673  * Return 0 if the option was found, ENOENT otherwise.
1674  * If len is non-NULL it will be filled with the length
1675  * of the option. If buf is non-NULL, it will be filled
1676  * with the address of the option.
1677  */
1678 int
1679 vfs_getopt(opts, name, buf, len)
1680 	struct vfsoptlist *opts;
1681 	const char *name;
1682 	void **buf;
1683 	int *len;
1684 {
1685 	struct vfsopt *opt;
1686 
1687 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1688 
1689 	TAILQ_FOREACH(opt, opts, link) {
1690 		if (strcmp(name, opt->name) == 0) {
1691 			if (len != NULL)
1692 				*len = opt->len;
1693 			if (buf != NULL)
1694 				*buf = opt->value;
1695 			return (0);
1696 		}
1697 	}
1698 	return (ENOENT);
1699 }
1700 
1701 static int
1702 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1703 {
1704 	struct vfsopt *opt;
1705 	int i;
1706 
1707 	if (opts == NULL)
1708 		return (-1);
1709 
1710 	i = 0;
1711 	TAILQ_FOREACH(opt, opts, link) {
1712 		if (strcmp(name, opt->name) == 0)
1713 			return (i);
1714 		++i;
1715 	}
1716 	return (-1);
1717 }
1718 
1719 char *
1720 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1721 {
1722 	struct vfsopt *opt;
1723 
1724 	*error = 0;
1725 	TAILQ_FOREACH(opt, opts, link) {
1726 		if (strcmp(name, opt->name) != 0)
1727 			continue;
1728 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1729 			*error = EINVAL;
1730 			return (NULL);
1731 		}
1732 		return (opt->value);
1733 	}
1734 	return (NULL);
1735 }
1736 
1737 int
1738 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1739 {
1740 	struct vfsopt *opt;
1741 
1742 	TAILQ_FOREACH(opt, opts, link) {
1743 		if (strcmp(name, opt->name) == 0) {
1744 			if (w != NULL)
1745 				*w |= val;
1746 			return (1);
1747 		}
1748 	}
1749 	if (w != NULL)
1750 		*w &= ~val;
1751 	return (0);
1752 }
1753 
1754 int
1755 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1756 {
1757 	va_list ap;
1758 	struct vfsopt *opt;
1759 	int ret;
1760 
1761 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1762 
1763 	TAILQ_FOREACH(opt, opts, link) {
1764 		if (strcmp(name, opt->name) != 0)
1765 			continue;
1766 		if (((char *)opt->value)[opt->len - 1] != '\0')
1767 			return (0);
1768 		va_start(ap, fmt);
1769 		ret = vsscanf(opt->value, fmt, ap);
1770 		va_end(ap);
1771 		return (ret);
1772 	}
1773 	return (0);
1774 }
1775 
1776 /*
1777  * Find and copy a mount option.
1778  *
1779  * The size of the buffer has to be specified
1780  * in len, if it is not the same length as the
1781  * mount option, EINVAL is returned.
1782  * Returns ENOENT if the option is not found.
1783  */
1784 int
1785 vfs_copyopt(opts, name, dest, len)
1786 	struct vfsoptlist *opts;
1787 	const char *name;
1788 	void *dest;
1789 	int len;
1790 {
1791 	struct vfsopt *opt;
1792 
1793 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1794 
1795 	TAILQ_FOREACH(opt, opts, link) {
1796 		if (strcmp(name, opt->name) == 0) {
1797 			if (len != opt->len)
1798 				return (EINVAL);
1799 			bcopy(opt->value, dest, opt->len);
1800 			return (0);
1801 		}
1802 	}
1803 	return (ENOENT);
1804 }
1805 
1806 /*
1807  * This is a helper function for filesystems to traverse their
1808  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1809  */
1810 
1811 struct vnode *
1812 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1813 {
1814 	struct vnode *vp;
1815 
1816 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1817 
1818 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1819 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1820 	while (vp != NULL && vp->v_type == VMARKER)
1821 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1822 
1823 	/* Check if we are done */
1824 	if (vp == NULL) {
1825 		__mnt_vnode_markerfree(mvp, mp);
1826 		return (NULL);
1827 	}
1828 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1829 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1830 	return (vp);
1831 }
1832 
1833 struct vnode *
1834 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
1835 {
1836 	struct vnode *vp;
1837 
1838 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1839 
1840 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1841 	while (vp != NULL && vp->v_type == VMARKER)
1842 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1843 
1844 	/* Check if we are done */
1845 	if (vp == NULL) {
1846 		*mvp = NULL;
1847 		return (NULL);
1848 	}
1849 	mp->mnt_holdcnt++;
1850 	MNT_IUNLOCK(mp);
1851 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
1852 				       M_VNODE_MARKER,
1853 				       M_WAITOK | M_ZERO);
1854 	MNT_ILOCK(mp);
1855 	(*mvp)->v_type = VMARKER;
1856 
1857 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1858 	while (vp != NULL && vp->v_type == VMARKER)
1859 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1860 
1861 	/* Check if we are done */
1862 	if (vp == NULL) {
1863 		MNT_IUNLOCK(mp);
1864 		free(*mvp, M_VNODE_MARKER);
1865 		MNT_ILOCK(mp);
1866 		*mvp = NULL;
1867 		mp->mnt_holdcnt--;
1868 		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1869 			wakeup(&mp->mnt_holdcnt);
1870 		return (NULL);
1871 	}
1872 	mp->mnt_markercnt++;
1873 	(*mvp)->v_mount = mp;
1874 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1875 	return (vp);
1876 }
1877 
1878 
1879 void
1880 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
1881 {
1882 
1883 	if (*mvp == NULL)
1884 		return;
1885 
1886 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1887 
1888 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1889 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1890 	MNT_IUNLOCK(mp);
1891 	free(*mvp, M_VNODE_MARKER);
1892 	MNT_ILOCK(mp);
1893 	*mvp = NULL;
1894 
1895 	mp->mnt_markercnt--;
1896 	mp->mnt_holdcnt--;
1897 	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1898 		wakeup(&mp->mnt_holdcnt);
1899 }
1900 
1901 
1902 int
1903 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
1904 {
1905 	int error;
1906 
1907 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
1908 	if (sbp != &mp->mnt_stat)
1909 		*sbp = mp->mnt_stat;
1910 	return (error);
1911 }
1912 
1913 void
1914 vfs_mountedfrom(struct mount *mp, const char *from)
1915 {
1916 
1917 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1918 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1919 	    sizeof mp->mnt_stat.f_mntfromname);
1920 }
1921 
1922 /*
1923  * ---------------------------------------------------------------------
1924  * This is the api for building mount args and mounting filesystems from
1925  * inside the kernel.
1926  *
1927  * The API works by accumulation of individual args.  First error is
1928  * latched.
1929  *
1930  * XXX: should be documented in new manpage kernel_mount(9)
1931  */
1932 
1933 /* A memory allocation which must be freed when we are done */
1934 struct mntaarg {
1935 	SLIST_ENTRY(mntaarg)	next;
1936 };
1937 
1938 /* The header for the mount arguments */
1939 struct mntarg {
1940 	struct iovec *v;
1941 	int len;
1942 	int error;
1943 	SLIST_HEAD(, mntaarg)	list;
1944 };
1945 
1946 /*
1947  * Add a boolean argument.
1948  *
1949  * flag is the boolean value.
1950  * name must start with "no".
1951  */
1952 struct mntarg *
1953 mount_argb(struct mntarg *ma, int flag, const char *name)
1954 {
1955 
1956 	KASSERT(name[0] == 'n' && name[1] == 'o',
1957 	    ("mount_argb(...,%s): name must start with 'no'", name));
1958 
1959 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1960 }
1961 
1962 /*
1963  * Add an argument printf style
1964  */
1965 struct mntarg *
1966 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1967 {
1968 	va_list ap;
1969 	struct mntaarg *maa;
1970 	struct sbuf *sb;
1971 	int len;
1972 
1973 	if (ma == NULL) {
1974 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1975 		SLIST_INIT(&ma->list);
1976 	}
1977 	if (ma->error)
1978 		return (ma);
1979 
1980 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1981 	    M_MOUNT, M_WAITOK);
1982 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1983 	ma->v[ma->len].iov_len = strlen(name) + 1;
1984 	ma->len++;
1985 
1986 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
1987 	va_start(ap, fmt);
1988 	sbuf_vprintf(sb, fmt, ap);
1989 	va_end(ap);
1990 	sbuf_finish(sb);
1991 	len = sbuf_len(sb) + 1;
1992 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1993 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1994 	bcopy(sbuf_data(sb), maa + 1, len);
1995 	sbuf_delete(sb);
1996 
1997 	ma->v[ma->len].iov_base = maa + 1;
1998 	ma->v[ma->len].iov_len = len;
1999 	ma->len++;
2000 
2001 	return (ma);
2002 }
2003 
2004 /*
2005  * Add an argument which is a userland string.
2006  */
2007 struct mntarg *
2008 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2009 {
2010 	struct mntaarg *maa;
2011 	char *tbuf;
2012 
2013 	if (val == NULL)
2014 		return (ma);
2015 	if (ma == NULL) {
2016 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2017 		SLIST_INIT(&ma->list);
2018 	}
2019 	if (ma->error)
2020 		return (ma);
2021 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2022 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2023 	tbuf = (void *)(maa + 1);
2024 	ma->error = copyinstr(val, tbuf, len, NULL);
2025 	return (mount_arg(ma, name, tbuf, -1));
2026 }
2027 
2028 /*
2029  * Plain argument.
2030  *
2031  * If length is -1, use printf.
2032  */
2033 struct mntarg *
2034 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2035 {
2036 
2037 	if (ma == NULL) {
2038 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2039 		SLIST_INIT(&ma->list);
2040 	}
2041 	if (ma->error)
2042 		return (ma);
2043 
2044 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2045 	    M_MOUNT, M_WAITOK);
2046 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2047 	ma->v[ma->len].iov_len = strlen(name) + 1;
2048 	ma->len++;
2049 
2050 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2051 	if (len < 0)
2052 		ma->v[ma->len].iov_len = strlen(val) + 1;
2053 	else
2054 		ma->v[ma->len].iov_len = len;
2055 	ma->len++;
2056 	return (ma);
2057 }
2058 
2059 /*
2060  * Free a mntarg structure
2061  */
2062 static void
2063 free_mntarg(struct mntarg *ma)
2064 {
2065 	struct mntaarg *maa;
2066 
2067 	while (!SLIST_EMPTY(&ma->list)) {
2068 		maa = SLIST_FIRST(&ma->list);
2069 		SLIST_REMOVE_HEAD(&ma->list, next);
2070 		free(maa, M_MOUNT);
2071 	}
2072 	free(ma->v, M_MOUNT);
2073 	free(ma, M_MOUNT);
2074 }
2075 
2076 /*
2077  * Mount a filesystem
2078  */
2079 int
2080 kernel_mount(struct mntarg *ma, int flags)
2081 {
2082 	struct uio auio;
2083 	int error;
2084 
2085 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2086 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2087 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2088 
2089 	auio.uio_iov = ma->v;
2090 	auio.uio_iovcnt = ma->len;
2091 	auio.uio_segflg = UIO_SYSSPACE;
2092 
2093 	error = ma->error;
2094 	if (!error)
2095 		error = vfs_donmount(curthread, flags, &auio);
2096 	free_mntarg(ma);
2097 	return (error);
2098 }
2099 
2100 /*
2101  * A printflike function to mount a filesystem.
2102  */
2103 int
2104 kernel_vmount(int flags, ...)
2105 {
2106 	struct mntarg *ma = NULL;
2107 	va_list ap;
2108 	const char *cp;
2109 	const void *vp;
2110 	int error;
2111 
2112 	va_start(ap, flags);
2113 	for (;;) {
2114 		cp = va_arg(ap, const char *);
2115 		if (cp == NULL)
2116 			break;
2117 		vp = va_arg(ap, const void *);
2118 		ma = mount_arg(ma, cp, vp, -1);
2119 	}
2120 	va_end(ap);
2121 
2122 	error = kernel_mount(ma, flags);
2123 	return (error);
2124 }
2125