xref: /freebsd/sys/kern/vfs_mount.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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/fcntl.h>
43 #include <sys/jail.h>
44 #include <sys/kernel.h>
45 #include <sys/libkern.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/filedesc.h>
53 #include <sys/reboot.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysproto.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/systm.h>
60 #include <sys/vnode.h>
61 #include <vm/uma.h>
62 
63 #include <geom/geom.h>
64 
65 #include <machine/stdarg.h>
66 
67 #include <security/audit/audit.h>
68 #include <security/mac/mac_framework.h>
69 
70 #include "opt_rootdevname.h"
71 
72 #define	ROOTNAME		"root_device"
73 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
74 
75 static void	set_rootvnode(void);
76 static int	vfs_domount(struct thread *td, const char *fstype,
77 		    char *fspath, int fsflags, void *fsdata);
78 static int	vfs_mountroot_ask(void);
79 static int	vfs_mountroot_try(const char *mountfrom, const char *options);
80 static void	free_mntarg(struct mntarg *ma);
81 
82 static int	usermount = 0;
83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
84     "Unprivileged users may mount and unmount file systems");
85 
86 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
87 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
88 static uma_zone_t mount_zone;
89 
90 /* List of mounted filesystems. */
91 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
92 
93 /* For any iteration/modification of mountlist */
94 struct mtx mountlist_mtx;
95 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
96 
97 /*
98  * The vnode of the system's root (/ in the filesystem, without chroot
99  * active.)
100  */
101 struct vnode	*rootvnode;
102 
103 /*
104  * The root filesystem is detailed in the kernel environment variable
105  * vfs.root.mountfrom, which is expected to be in the general format
106  *
107  * <vfsname>:[<path>]
108  * vfsname   := the name of a VFS known to the kernel and capable
109  *              of being mounted as root
110  * path      := disk device name or other data used by the filesystem
111  *              to locate its physical store
112  *
113  * The environment variable vfs.root.mountfrom options is a comma delimited
114  * set of string mount options.  These mount options must be parseable
115  * by nmount() in the kernel.
116  */
117 
118 /*
119  * Global opts, taken by all filesystems
120  */
121 static const char *global_opts[] = {
122 	"errmsg",
123 	"fstype",
124 	"fspath",
125 	"ro",
126 	"rw",
127 	"nosuid",
128 	"noexec",
129 	NULL
130 };
131 
132 /*
133  * The root specifiers we will try if RB_CDROM is specified.
134  */
135 static char *cdrom_rootdevnames[] = {
136 	"cd9660:cd0",
137 	"cd9660:acd0",
138 	NULL
139 };
140 
141 /* legacy find-root code */
142 char		*rootdevnames[2] = {NULL, NULL};
143 #ifndef ROOTDEVNAME
144 #  define ROOTDEVNAME NULL
145 #endif
146 static const char	*ctrootdevname = ROOTDEVNAME;
147 
148 /*
149  * ---------------------------------------------------------------------
150  * Functions for building and sanitizing the mount options
151  */
152 
153 /* Remove one mount option. */
154 static void
155 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
156 {
157 
158 	TAILQ_REMOVE(opts, opt, link);
159 	free(opt->name, M_MOUNT);
160 	if (opt->value != NULL)
161 		free(opt->value, M_MOUNT);
162 	free(opt, M_MOUNT);
163 }
164 
165 /* Release all resources related to the mount options. */
166 void
167 vfs_freeopts(struct vfsoptlist *opts)
168 {
169 	struct vfsopt *opt;
170 
171 	while (!TAILQ_EMPTY(opts)) {
172 		opt = TAILQ_FIRST(opts);
173 		vfs_freeopt(opts, opt);
174 	}
175 	free(opts, M_MOUNT);
176 }
177 
178 void
179 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
180 {
181 	struct vfsopt *opt, *temp;
182 
183 	if (opts == NULL)
184 		return;
185 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
186 		if (strcmp(opt->name, name) == 0)
187 			vfs_freeopt(opts, opt);
188 	}
189 }
190 
191 /*
192  * Check if options are equal (with or without the "no" prefix).
193  */
194 static int
195 vfs_equalopts(const char *opt1, const char *opt2)
196 {
197 	char *p;
198 
199 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
200 	if (strcmp(opt1, opt2) == 0)
201 		return (1);
202 	/* "noopt" vs. "opt" */
203 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
204 		return (1);
205 	/* "opt" vs. "noopt" */
206 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
207 		return (1);
208 	while ((p = strchr(opt1, '.')) != NULL &&
209 	    !strncmp(opt1, opt2, ++p - opt1)) {
210 		opt2 += p - opt1;
211 		opt1 = p;
212 		/* "foo.noopt" vs. "foo.opt" */
213 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
214 			return (1);
215 		/* "foo.opt" vs. "foo.noopt" */
216 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
217 			return (1);
218 	}
219 	return (0);
220 }
221 
222 /*
223  * If a mount option is specified several times,
224  * (with or without the "no" prefix) only keep
225  * the last occurence of it.
226  */
227 static void
228 vfs_sanitizeopts(struct vfsoptlist *opts)
229 {
230 	struct vfsopt *opt, *opt2, *tmp;
231 
232 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
233 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
234 		while (opt2 != NULL) {
235 			if (vfs_equalopts(opt->name, opt2->name)) {
236 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
237 				vfs_freeopt(opts, opt2);
238 				opt2 = tmp;
239 			} else {
240 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
241 			}
242 		}
243 	}
244 }
245 
246 /*
247  * Build a linked list of mount options from a struct uio.
248  */
249 int
250 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
251 {
252 	struct vfsoptlist *opts;
253 	struct vfsopt *opt;
254 	size_t memused, namelen, optlen;
255 	unsigned int i, iovcnt;
256 	int error;
257 
258 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
259 	TAILQ_INIT(opts);
260 	memused = 0;
261 	iovcnt = auio->uio_iovcnt;
262 	for (i = 0; i < iovcnt; i += 2) {
263 		namelen = auio->uio_iov[i].iov_len;
264 		optlen = auio->uio_iov[i + 1].iov_len;
265 		memused += sizeof(struct vfsopt) + optlen + namelen;
266 		/*
267 		 * Avoid consuming too much memory, and attempts to overflow
268 		 * memused.
269 		 */
270 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
271 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
272 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
273 			error = EINVAL;
274 			goto bad;
275 		}
276 
277 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
278 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
279 		opt->value = NULL;
280 		opt->len = 0;
281 		opt->pos = i / 2;
282 		opt->seen = 0;
283 
284 		/*
285 		 * Do this early, so jumps to "bad" will free the current
286 		 * option.
287 		 */
288 		TAILQ_INSERT_TAIL(opts, opt, link);
289 
290 		if (auio->uio_segflg == UIO_SYSSPACE) {
291 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
292 		} else {
293 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
294 			    namelen);
295 			if (error)
296 				goto bad;
297 		}
298 		/* Ensure names are null-terminated strings. */
299 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
300 			error = EINVAL;
301 			goto bad;
302 		}
303 		if (optlen != 0) {
304 			opt->len = optlen;
305 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
306 			if (auio->uio_segflg == UIO_SYSSPACE) {
307 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
308 				    optlen);
309 			} else {
310 				error = copyin(auio->uio_iov[i + 1].iov_base,
311 				    opt->value, optlen);
312 				if (error)
313 					goto bad;
314 			}
315 		}
316 	}
317 	vfs_sanitizeopts(opts);
318 	*options = opts;
319 	return (0);
320 bad:
321 	vfs_freeopts(opts);
322 	return (error);
323 }
324 
325 /*
326  * Merge the old mount options with the new ones passed
327  * in the MNT_UPDATE case.
328  *
329  * XXX This function will keep a "nofoo" option in the
330  *     new options if there is no matching "foo" option
331  *     to be cancelled in the old options.  This is a bug
332  *     if the option's canonical name is "foo".  E.g., "noro"
333  *     shouldn't end up in the mount point's active options,
334  *     but it can.
335  */
336 static void
337 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
338 {
339 	struct vfsopt *opt, *opt2, *new;
340 
341 	TAILQ_FOREACH(opt, opts, link) {
342 		/*
343 		 * Check that this option hasn't been redefined
344 		 * nor cancelled with a "no" mount option.
345 		 */
346 		opt2 = TAILQ_FIRST(toopts);
347 		while (opt2 != NULL) {
348 			if (strcmp(opt2->name, opt->name) == 0)
349 				goto next;
350 			if (strncmp(opt2->name, "no", 2) == 0 &&
351 			    strcmp(opt2->name + 2, opt->name) == 0) {
352 				vfs_freeopt(toopts, opt2);
353 				goto next;
354 			}
355 			opt2 = TAILQ_NEXT(opt2, link);
356 		}
357 		/* We want this option, duplicate it. */
358 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
359 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
360 		strcpy(new->name, opt->name);
361 		if (opt->len != 0) {
362 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
363 			bcopy(opt->value, new->value, opt->len);
364 		} else {
365 			new->value = NULL;
366 		}
367 		new->len = opt->len;
368 		new->seen = opt->seen;
369 		TAILQ_INSERT_TAIL(toopts, new, link);
370 next:
371 		continue;
372 	}
373 }
374 
375 /*
376  * Mount a filesystem.
377  */
378 int
379 nmount(td, uap)
380 	struct thread *td;
381 	struct nmount_args /* {
382 		struct iovec *iovp;
383 		unsigned int iovcnt;
384 		int flags;
385 	} */ *uap;
386 {
387 	struct uio *auio;
388 	int error;
389 	u_int iovcnt;
390 
391 	AUDIT_ARG_FFLAGS(uap->flags);
392 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
393 	    uap->iovp, uap->iovcnt, uap->flags);
394 
395 	/*
396 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
397 	 * userspace to set this flag, but we must filter it out if we want
398 	 * MNT_UPDATE on the root file system to work.
399 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
400 	 */
401 	uap->flags &= ~MNT_ROOTFS;
402 
403 	iovcnt = uap->iovcnt;
404 	/*
405 	 * Check that we have an even number of iovec's
406 	 * and that we have at least two options.
407 	 */
408 	if ((iovcnt & 1) || (iovcnt < 4)) {
409 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
410 		    uap->iovcnt);
411 		return (EINVAL);
412 	}
413 
414 	error = copyinuio(uap->iovp, iovcnt, &auio);
415 	if (error) {
416 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
417 		    __func__, error);
418 		return (error);
419 	}
420 	error = vfs_donmount(td, uap->flags, auio);
421 
422 	free(auio, M_IOV);
423 	return (error);
424 }
425 
426 /*
427  * ---------------------------------------------------------------------
428  * Various utility functions
429  */
430 
431 void
432 vfs_ref(struct mount *mp)
433 {
434 
435 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
436 	MNT_ILOCK(mp);
437 	MNT_REF(mp);
438 	MNT_IUNLOCK(mp);
439 }
440 
441 void
442 vfs_rel(struct mount *mp)
443 {
444 
445 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
446 	MNT_ILOCK(mp);
447 	MNT_REL(mp);
448 	MNT_IUNLOCK(mp);
449 }
450 
451 static int
452 mount_init(void *mem, int size, int flags)
453 {
454 	struct mount *mp;
455 
456 	mp = (struct mount *)mem;
457 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
458 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
459 	return (0);
460 }
461 
462 static void
463 mount_fini(void *mem, int size)
464 {
465 	struct mount *mp;
466 
467 	mp = (struct mount *)mem;
468 	lockdestroy(&mp->mnt_explock);
469 	mtx_destroy(&mp->mnt_mtx);
470 }
471 
472 /*
473  * Allocate and initialize the mount point struct.
474  */
475 struct mount *
476 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
477     struct ucred *cred)
478 {
479 	struct mount *mp;
480 
481 	mp = uma_zalloc(mount_zone, M_WAITOK);
482 	bzero(&mp->mnt_startzero,
483 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
484 	TAILQ_INIT(&mp->mnt_nvnodelist);
485 	mp->mnt_nvnodelistsize = 0;
486 	mp->mnt_ref = 0;
487 	(void) vfs_busy(mp, MBF_NOWAIT);
488 	mp->mnt_op = vfsp->vfc_vfsops;
489 	mp->mnt_vfc = vfsp;
490 	vfsp->vfc_refcount++;	/* XXX Unlocked */
491 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
492 	mp->mnt_gen++;
493 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
494 	mp->mnt_vnodecovered = vp;
495 	mp->mnt_cred = crdup(cred);
496 	mp->mnt_stat.f_owner = cred->cr_uid;
497 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
498 	mp->mnt_iosize_max = DFLTPHYS;
499 #ifdef MAC
500 	mac_mount_init(mp);
501 	mac_mount_create(cred, mp);
502 #endif
503 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
504 	return (mp);
505 }
506 
507 /*
508  * Destroy the mount struct previously allocated by vfs_mount_alloc().
509  */
510 void
511 vfs_mount_destroy(struct mount *mp)
512 {
513 
514 	MNT_ILOCK(mp);
515 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
516 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
517 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
518 		wakeup(mp);
519 	}
520 	while (mp->mnt_ref)
521 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
522 	KASSERT(mp->mnt_ref == 0,
523 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
524 	    __FILE__, __LINE__));
525 	if (mp->mnt_writeopcount != 0)
526 		panic("vfs_mount_destroy: nonzero writeopcount");
527 	if (mp->mnt_secondary_writes != 0)
528 		panic("vfs_mount_destroy: nonzero secondary_writes");
529 	mp->mnt_vfc->vfc_refcount--;
530 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
531 		struct vnode *vp;
532 
533 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
534 			vprint("", vp);
535 		panic("unmount: dangling vnode");
536 	}
537 	if (mp->mnt_nvnodelistsize != 0)
538 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
539 	if (mp->mnt_lockref != 0)
540 		panic("vfs_mount_destroy: nonzero lock refcount");
541 	MNT_IUNLOCK(mp);
542 #ifdef MAC
543 	mac_mount_destroy(mp);
544 #endif
545 	if (mp->mnt_opt != NULL)
546 		vfs_freeopts(mp->mnt_opt);
547 	crfree(mp->mnt_cred);
548 	uma_zfree(mount_zone, mp);
549 }
550 
551 int
552 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
553 {
554 	struct vfsoptlist *optlist;
555 	struct vfsopt *opt, *noro_opt, *tmp_opt;
556 	char *fstype, *fspath, *errmsg;
557 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
558 	int has_rw, has_noro;
559 
560 	errmsg = fspath = NULL;
561 	errmsg_len = has_noro = has_rw = fspathlen = 0;
562 	errmsg_pos = -1;
563 
564 	error = vfs_buildopts(fsoptions, &optlist);
565 	if (error)
566 		return (error);
567 
568 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
569 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
570 
571 	/*
572 	 * We need these two options before the others,
573 	 * and they are mandatory for any filesystem.
574 	 * Ensure they are NUL terminated as well.
575 	 */
576 	fstypelen = 0;
577 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
578 	if (error || fstype[fstypelen - 1] != '\0') {
579 		error = EINVAL;
580 		if (errmsg != NULL)
581 			strncpy(errmsg, "Invalid fstype", errmsg_len);
582 		goto bail;
583 	}
584 	fspathlen = 0;
585 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
586 	if (error || fspath[fspathlen - 1] != '\0') {
587 		error = EINVAL;
588 		if (errmsg != NULL)
589 			strncpy(errmsg, "Invalid fspath", errmsg_len);
590 		goto bail;
591 	}
592 
593 	/*
594 	 * We need to see if we have the "update" option
595 	 * before we call vfs_domount(), since vfs_domount() has special
596 	 * logic based on MNT_UPDATE.  This is very important
597 	 * when we want to update the root filesystem.
598 	 */
599 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
600 		if (strcmp(opt->name, "update") == 0) {
601 			fsflags |= MNT_UPDATE;
602 			vfs_freeopt(optlist, opt);
603 		}
604 		else if (strcmp(opt->name, "async") == 0)
605 			fsflags |= MNT_ASYNC;
606 		else if (strcmp(opt->name, "force") == 0) {
607 			fsflags |= MNT_FORCE;
608 			vfs_freeopt(optlist, opt);
609 		}
610 		else if (strcmp(opt->name, "reload") == 0) {
611 			fsflags |= MNT_RELOAD;
612 			vfs_freeopt(optlist, opt);
613 		}
614 		else if (strcmp(opt->name, "multilabel") == 0)
615 			fsflags |= MNT_MULTILABEL;
616 		else if (strcmp(opt->name, "noasync") == 0)
617 			fsflags &= ~MNT_ASYNC;
618 		else if (strcmp(opt->name, "noatime") == 0)
619 			fsflags |= MNT_NOATIME;
620 		else if (strcmp(opt->name, "atime") == 0) {
621 			free(opt->name, M_MOUNT);
622 			opt->name = strdup("nonoatime", M_MOUNT);
623 		}
624 		else if (strcmp(opt->name, "noclusterr") == 0)
625 			fsflags |= MNT_NOCLUSTERR;
626 		else if (strcmp(opt->name, "clusterr") == 0) {
627 			free(opt->name, M_MOUNT);
628 			opt->name = strdup("nonoclusterr", M_MOUNT);
629 		}
630 		else if (strcmp(opt->name, "noclusterw") == 0)
631 			fsflags |= MNT_NOCLUSTERW;
632 		else if (strcmp(opt->name, "clusterw") == 0) {
633 			free(opt->name, M_MOUNT);
634 			opt->name = strdup("nonoclusterw", M_MOUNT);
635 		}
636 		else if (strcmp(opt->name, "noexec") == 0)
637 			fsflags |= MNT_NOEXEC;
638 		else if (strcmp(opt->name, "exec") == 0) {
639 			free(opt->name, M_MOUNT);
640 			opt->name = strdup("nonoexec", M_MOUNT);
641 		}
642 		else if (strcmp(opt->name, "nosuid") == 0)
643 			fsflags |= MNT_NOSUID;
644 		else if (strcmp(opt->name, "suid") == 0) {
645 			free(opt->name, M_MOUNT);
646 			opt->name = strdup("nonosuid", M_MOUNT);
647 		}
648 		else if (strcmp(opt->name, "nosymfollow") == 0)
649 			fsflags |= MNT_NOSYMFOLLOW;
650 		else if (strcmp(opt->name, "symfollow") == 0) {
651 			free(opt->name, M_MOUNT);
652 			opt->name = strdup("nonosymfollow", M_MOUNT);
653 		}
654 		else if (strcmp(opt->name, "noro") == 0) {
655 			fsflags &= ~MNT_RDONLY;
656 			has_noro = 1;
657 		}
658 		else if (strcmp(opt->name, "rw") == 0) {
659 			fsflags &= ~MNT_RDONLY;
660 			has_rw = 1;
661 		}
662 		else if (strcmp(opt->name, "ro") == 0)
663 			fsflags |= MNT_RDONLY;
664 		else if (strcmp(opt->name, "rdonly") == 0) {
665 			free(opt->name, M_MOUNT);
666 			opt->name = strdup("ro", M_MOUNT);
667 			fsflags |= MNT_RDONLY;
668 		}
669 		else if (strcmp(opt->name, "suiddir") == 0)
670 			fsflags |= MNT_SUIDDIR;
671 		else if (strcmp(opt->name, "sync") == 0)
672 			fsflags |= MNT_SYNCHRONOUS;
673 		else if (strcmp(opt->name, "union") == 0)
674 			fsflags |= MNT_UNION;
675 	}
676 
677 	/*
678 	 * If "rw" was specified as a mount option, and we
679 	 * are trying to update a mount-point from "ro" to "rw",
680 	 * we need a mount option "noro", since in vfs_mergeopts(),
681 	 * "noro" will cancel "ro", but "rw" will not do anything.
682 	 */
683 	if (has_rw && !has_noro) {
684 		noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
685 		noro_opt->name = strdup("noro", M_MOUNT);
686 		noro_opt->value = NULL;
687 		noro_opt->len = 0;
688 		noro_opt->pos = -1;
689 		noro_opt->seen = 1;
690 		TAILQ_INSERT_TAIL(optlist, noro_opt, link);
691 	}
692 
693 	/*
694 	 * Be ultra-paranoid about making sure the type and fspath
695 	 * variables will fit in our mp buffers, including the
696 	 * terminating NUL.
697 	 */
698 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
699 		error = ENAMETOOLONG;
700 		goto bail;
701 	}
702 
703 	mtx_lock(&Giant);
704 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
705 	mtx_unlock(&Giant);
706 bail:
707 	/* copyout the errmsg */
708 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
709 	    && errmsg_len > 0 && errmsg != NULL) {
710 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
711 			bcopy(errmsg,
712 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
713 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
714 		} else {
715 			copyout(errmsg,
716 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
717 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
718 		}
719 	}
720 
721 	if (error != 0)
722 		vfs_freeopts(optlist);
723 	return (error);
724 }
725 
726 /*
727  * Old mount API.
728  */
729 #ifndef _SYS_SYSPROTO_H_
730 struct mount_args {
731 	char	*type;
732 	char	*path;
733 	int	flags;
734 	caddr_t	data;
735 };
736 #endif
737 /* ARGSUSED */
738 int
739 mount(td, uap)
740 	struct thread *td;
741 	struct mount_args /* {
742 		char *type;
743 		char *path;
744 		int flags;
745 		caddr_t data;
746 	} */ *uap;
747 {
748 	char *fstype;
749 	struct vfsconf *vfsp = NULL;
750 	struct mntarg *ma = NULL;
751 	int error;
752 
753 	AUDIT_ARG_FFLAGS(uap->flags);
754 
755 	/*
756 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
757 	 * userspace to set this flag, but we must filter it out if we want
758 	 * MNT_UPDATE on the root file system to work.
759 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
760 	 */
761 	uap->flags &= ~MNT_ROOTFS;
762 
763 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
764 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
765 	if (error) {
766 		free(fstype, M_TEMP);
767 		return (error);
768 	}
769 
770 	AUDIT_ARG_TEXT(fstype);
771 	mtx_lock(&Giant);
772 	vfsp = vfs_byname_kld(fstype, td, &error);
773 	free(fstype, M_TEMP);
774 	if (vfsp == NULL) {
775 		mtx_unlock(&Giant);
776 		return (ENOENT);
777 	}
778 	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
779 		mtx_unlock(&Giant);
780 		return (EOPNOTSUPP);
781 	}
782 
783 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
784 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
785 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
786 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
787 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
788 
789 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags);
790 	mtx_unlock(&Giant);
791 	return (error);
792 }
793 
794 
795 /*
796  * vfs_domount(): actually attempt a filesystem mount.
797  */
798 static int
799 vfs_domount(
800 	struct thread *td,	/* Calling thread. */
801 	const char *fstype,	/* Filesystem type. */
802 	char *fspath,		/* Mount path. */
803 	int fsflags,		/* Flags common to all filesystems. */
804 	void *fsdata		/* Options local to the filesystem. */
805 	)
806 {
807 	struct vnode *vp;
808 	struct mount *mp;
809 	struct vfsconf *vfsp;
810 	struct oexport_args oexport;
811 	struct export_args export;
812 	int error, flag = 0;
813 	struct vattr va;
814 	struct nameidata nd;
815 
816 	mtx_assert(&Giant, MA_OWNED);
817 	/*
818 	 * Be ultra-paranoid about making sure the type and fspath
819 	 * variables will fit in our mp buffers, including the
820 	 * terminating NUL.
821 	 */
822 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
823 		return (ENAMETOOLONG);
824 
825 	if (jailed(td->td_ucred) || usermount == 0) {
826 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
827 			return (error);
828 	}
829 
830 	/*
831 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
832 	 */
833 	if (fsflags & MNT_EXPORTED) {
834 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
835 		if (error)
836 			return (error);
837 	}
838 	if (fsflags & MNT_SUIDDIR) {
839 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
840 		if (error)
841 			return (error);
842 	}
843 	/*
844 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
845 	 */
846 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
847 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
848 			fsflags |= MNT_NOSUID | MNT_USER;
849 	}
850 
851 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
852 	vfsp = NULL;
853 	if ((fsflags & MNT_UPDATE) == 0) {
854 		/* Don't try to load KLDs if we're mounting the root. */
855 		if (fsflags & MNT_ROOTFS)
856 			vfsp = vfs_byname(fstype);
857 		else
858 			vfsp = vfs_byname_kld(fstype, td, &error);
859 		if (vfsp == NULL)
860 			return (ENODEV);
861 		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
862 			return (EPERM);
863 	}
864 	/*
865 	 * Get vnode to be covered
866 	 */
867 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
868 	    fspath, td);
869 	if ((error = namei(&nd)) != 0)
870 		return (error);
871 	NDFREE(&nd, NDF_ONLY_PNBUF);
872 	vp = nd.ni_vp;
873 	if (fsflags & MNT_UPDATE) {
874 		if ((vp->v_vflag & VV_ROOT) == 0) {
875 			vput(vp);
876 			return (EINVAL);
877 		}
878 		mp = vp->v_mount;
879 		MNT_ILOCK(mp);
880 		flag = mp->mnt_flag;
881 		/*
882 		 * We only allow the filesystem to be reloaded if it
883 		 * is currently mounted read-only.
884 		 */
885 		if ((fsflags & MNT_RELOAD) &&
886 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
887 			MNT_IUNLOCK(mp);
888 			vput(vp);
889 			return (EOPNOTSUPP);	/* Needs translation */
890 		}
891 		MNT_IUNLOCK(mp);
892 		/*
893 		 * Only privileged root, or (if MNT_USER is set) the user that
894 		 * did the original mount is permitted to update it.
895 		 */
896 		error = vfs_suser(mp, td);
897 		if (error) {
898 			vput(vp);
899 			return (error);
900 		}
901 		if (vfs_busy(mp, MBF_NOWAIT)) {
902 			vput(vp);
903 			return (EBUSY);
904 		}
905 		VI_LOCK(vp);
906 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
907 		    vp->v_mountedhere != NULL) {
908 			VI_UNLOCK(vp);
909 			vfs_unbusy(mp);
910 			vput(vp);
911 			return (EBUSY);
912 		}
913 		vp->v_iflag |= VI_MOUNT;
914 		VI_UNLOCK(vp);
915 		MNT_ILOCK(mp);
916 		mp->mnt_flag |= fsflags &
917 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
918 		MNT_IUNLOCK(mp);
919 		VOP_UNLOCK(vp, 0);
920 		mp->mnt_optnew = fsdata;
921 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
922 	} else {
923 		/*
924 		 * If the user is not root, ensure that they own the directory
925 		 * onto which we are attempting to mount.
926 		 */
927 		error = VOP_GETATTR(vp, &va, td->td_ucred);
928 		if (error) {
929 			vput(vp);
930 			return (error);
931 		}
932 		if (va.va_uid != td->td_ucred->cr_uid) {
933 			error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
934 			    0);
935 			if (error) {
936 				vput(vp);
937 				return (error);
938 			}
939 		}
940 		error = vinvalbuf(vp, V_SAVE, 0, 0);
941 		if (error != 0) {
942 			vput(vp);
943 			return (error);
944 		}
945 		if (vp->v_type != VDIR) {
946 			vput(vp);
947 			return (ENOTDIR);
948 		}
949 		VI_LOCK(vp);
950 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
951 		    vp->v_mountedhere != NULL) {
952 			VI_UNLOCK(vp);
953 			vput(vp);
954 			return (EBUSY);
955 		}
956 		vp->v_iflag |= VI_MOUNT;
957 		VI_UNLOCK(vp);
958 
959 		/*
960 		 * Allocate and initialize the filesystem.
961 		 */
962 		mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
963 		VOP_UNLOCK(vp, 0);
964 
965 		/* XXXMAC: pass to vfs_mount_alloc? */
966 		mp->mnt_optnew = fsdata;
967 	}
968 
969 	/*
970 	 * Set the mount level flags.
971 	 */
972 	MNT_ILOCK(mp);
973 	mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
974 		(fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
975 			    MNT_RDONLY));
976 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
977 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
978 	MNT_IUNLOCK(mp);
979 	/*
980 	 * Mount the filesystem.
981 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
982 	 * get.  No freeing of cn_pnbuf.
983 	 */
984         error = VFS_MOUNT(mp);
985 
986 	/*
987 	 * Process the export option only if we are
988 	 * updating mount options.
989 	 */
990 	if (!error && (fsflags & MNT_UPDATE)) {
991 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
992 		    sizeof(export)) == 0)
993 			error = vfs_export(mp, &export);
994 		else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport,
995 			sizeof(oexport)) == 0) {
996 			export.ex_flags = oexport.ex_flags;
997 			export.ex_root = oexport.ex_root;
998 			export.ex_anon = oexport.ex_anon;
999 			export.ex_addr = oexport.ex_addr;
1000 			export.ex_addrlen = oexport.ex_addrlen;
1001 			export.ex_mask = oexport.ex_mask;
1002 			export.ex_masklen = oexport.ex_masklen;
1003 			export.ex_indexfile = oexport.ex_indexfile;
1004 			export.ex_numsecflavors = 0;
1005 			error = vfs_export(mp, &export);
1006 		}
1007 	}
1008 
1009 	if (!error) {
1010 		if (mp->mnt_opt != NULL)
1011 			vfs_freeopts(mp->mnt_opt);
1012 		mp->mnt_opt = mp->mnt_optnew;
1013 		(void)VFS_STATFS(mp, &mp->mnt_stat);
1014 	}
1015 	/*
1016 	 * Prevent external consumers of mount options from reading
1017 	 * mnt_optnew.
1018 	*/
1019 	mp->mnt_optnew = NULL;
1020 	if (mp->mnt_flag & MNT_UPDATE) {
1021 		MNT_ILOCK(mp);
1022 		if (error)
1023 			mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1024 				(flag & ~MNT_QUOTA);
1025 		else
1026 			mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD |
1027 					  MNT_FORCE | MNT_SNAPSHOT);
1028 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1029 			mp->mnt_kern_flag |= MNTK_ASYNC;
1030 		else
1031 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1032 		MNT_IUNLOCK(mp);
1033 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1034 			if (mp->mnt_syncer == NULL)
1035 				error = vfs_allocate_syncvnode(mp);
1036 		} else {
1037 			if (mp->mnt_syncer != NULL)
1038 				vrele(mp->mnt_syncer);
1039 			mp->mnt_syncer = NULL;
1040 		}
1041 		vfs_unbusy(mp);
1042 		VI_LOCK(vp);
1043 		vp->v_iflag &= ~VI_MOUNT;
1044 		VI_UNLOCK(vp);
1045 		vrele(vp);
1046 		return (error);
1047 	}
1048 	MNT_ILOCK(mp);
1049 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1050 		mp->mnt_kern_flag |= MNTK_ASYNC;
1051 	else
1052 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1053 	MNT_IUNLOCK(mp);
1054 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1055 	/*
1056 	 * Put the new filesystem on the mount list after root.
1057 	 */
1058 	cache_purge(vp);
1059 	if (!error) {
1060 		struct vnode *newdp;
1061 
1062 		VI_LOCK(vp);
1063 		vp->v_iflag &= ~VI_MOUNT;
1064 		VI_UNLOCK(vp);
1065 		vp->v_mountedhere = mp;
1066 		mtx_lock(&mountlist_mtx);
1067 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1068 		mtx_unlock(&mountlist_mtx);
1069 		vfs_event_signal(NULL, VQ_MOUNT, 0);
1070 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp))
1071 			panic("mount: lost mount");
1072 		VOP_UNLOCK(newdp, 0);
1073 		VOP_UNLOCK(vp, 0);
1074 		mountcheckdirs(vp, newdp);
1075 		vrele(newdp);
1076 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1077 			error = vfs_allocate_syncvnode(mp);
1078 		vfs_unbusy(mp);
1079 		if (error)
1080 			vrele(vp);
1081 	} else {
1082 		VI_LOCK(vp);
1083 		vp->v_iflag &= ~VI_MOUNT;
1084 		VI_UNLOCK(vp);
1085 		vfs_unbusy(mp);
1086 		vfs_mount_destroy(mp);
1087 		vput(vp);
1088 	}
1089 	return (error);
1090 }
1091 
1092 /*
1093  * Unmount a filesystem.
1094  *
1095  * Note: unmount takes a path to the vnode mounted on as argument, not
1096  * special file (as before).
1097  */
1098 #ifndef _SYS_SYSPROTO_H_
1099 struct unmount_args {
1100 	char	*path;
1101 	int	flags;
1102 };
1103 #endif
1104 /* ARGSUSED */
1105 int
1106 unmount(td, uap)
1107 	struct thread *td;
1108 	register struct unmount_args /* {
1109 		char *path;
1110 		int flags;
1111 	} */ *uap;
1112 {
1113 	struct mount *mp;
1114 	char *pathbuf;
1115 	int error, id0, id1;
1116 
1117 	AUDIT_ARG_VALUE(uap->flags);
1118 	if (jailed(td->td_ucred) || usermount == 0) {
1119 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1120 		if (error)
1121 			return (error);
1122 	}
1123 
1124 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1125 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1126 	if (error) {
1127 		free(pathbuf, M_TEMP);
1128 		return (error);
1129 	}
1130 	mtx_lock(&Giant);
1131 	if (uap->flags & MNT_BYFSID) {
1132 		AUDIT_ARG_TEXT(pathbuf);
1133 		/* Decode the filesystem ID. */
1134 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1135 			mtx_unlock(&Giant);
1136 			free(pathbuf, M_TEMP);
1137 			return (EINVAL);
1138 		}
1139 
1140 		mtx_lock(&mountlist_mtx);
1141 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1142 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1143 			    mp->mnt_stat.f_fsid.val[1] == id1)
1144 				break;
1145 		}
1146 		mtx_unlock(&mountlist_mtx);
1147 	} else {
1148 		AUDIT_ARG_UPATH1(td, pathbuf);
1149 		mtx_lock(&mountlist_mtx);
1150 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1151 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1152 				break;
1153 		}
1154 		mtx_unlock(&mountlist_mtx);
1155 	}
1156 	free(pathbuf, M_TEMP);
1157 	if (mp == NULL) {
1158 		/*
1159 		 * Previously we returned ENOENT for a nonexistent path and
1160 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1161 		 * now, so in the !MNT_BYFSID case return the more likely
1162 		 * EINVAL for compatibility.
1163 		 */
1164 		mtx_unlock(&Giant);
1165 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1166 	}
1167 
1168 	/*
1169 	 * Don't allow unmounting the root filesystem.
1170 	 */
1171 	if (mp->mnt_flag & MNT_ROOTFS) {
1172 		mtx_unlock(&Giant);
1173 		return (EINVAL);
1174 	}
1175 	error = dounmount(mp, uap->flags, td);
1176 	mtx_unlock(&Giant);
1177 	return (error);
1178 }
1179 
1180 /*
1181  * Do the actual filesystem unmount.
1182  */
1183 int
1184 dounmount(mp, flags, td)
1185 	struct mount *mp;
1186 	int flags;
1187 	struct thread *td;
1188 {
1189 	struct vnode *coveredvp, *fsrootvp;
1190 	int error;
1191 	int async_flag;
1192 	int mnt_gen_r;
1193 
1194 	mtx_assert(&Giant, MA_OWNED);
1195 
1196 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1197 		mnt_gen_r = mp->mnt_gen;
1198 		VI_LOCK(coveredvp);
1199 		vholdl(coveredvp);
1200 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1201 		vdrop(coveredvp);
1202 		/*
1203 		 * Check for mp being unmounted while waiting for the
1204 		 * covered vnode lock.
1205 		 */
1206 		if (coveredvp->v_mountedhere != mp ||
1207 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1208 			VOP_UNLOCK(coveredvp, 0);
1209 			return (EBUSY);
1210 		}
1211 	}
1212 	/*
1213 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1214 	 * original mount is permitted to unmount this filesystem.
1215 	 */
1216 	error = vfs_suser(mp, td);
1217 	if (error) {
1218 		if (coveredvp)
1219 			VOP_UNLOCK(coveredvp, 0);
1220 		return (error);
1221 	}
1222 
1223 	MNT_ILOCK(mp);
1224 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1225 		MNT_IUNLOCK(mp);
1226 		if (coveredvp)
1227 			VOP_UNLOCK(coveredvp, 0);
1228 		return (EBUSY);
1229 	}
1230 	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1231 	/* Allow filesystems to detect that a forced unmount is in progress. */
1232 	if (flags & MNT_FORCE)
1233 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1234 	error = 0;
1235 	if (mp->mnt_lockref) {
1236 		if ((flags & MNT_FORCE) == 0) {
1237 			mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ |
1238 			    MNTK_UNMOUNTF);
1239 			if (mp->mnt_kern_flag & MNTK_MWAIT) {
1240 				mp->mnt_kern_flag &= ~MNTK_MWAIT;
1241 				wakeup(mp);
1242 			}
1243 			MNT_IUNLOCK(mp);
1244 			if (coveredvp)
1245 				VOP_UNLOCK(coveredvp, 0);
1246 			return (EBUSY);
1247 		}
1248 		mp->mnt_kern_flag |= MNTK_DRAINING;
1249 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1250 		    "mount drain", 0);
1251 	}
1252 	MNT_IUNLOCK(mp);
1253 	KASSERT(mp->mnt_lockref == 0,
1254 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1255 	    __func__, __FILE__, __LINE__));
1256 	KASSERT(error == 0,
1257 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1258 	    __func__, __FILE__, __LINE__));
1259 	vn_start_write(NULL, &mp, V_WAIT);
1260 
1261 	if (mp->mnt_flag & MNT_EXPUBLIC)
1262 		vfs_setpublicfs(NULL, NULL, NULL);
1263 
1264 	vfs_msync(mp, MNT_WAIT);
1265 	MNT_ILOCK(mp);
1266 	async_flag = mp->mnt_flag & MNT_ASYNC;
1267 	mp->mnt_flag &= ~MNT_ASYNC;
1268 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1269 	MNT_IUNLOCK(mp);
1270 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1271 	if (mp->mnt_syncer != NULL)
1272 		vrele(mp->mnt_syncer);
1273 	/*
1274 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1275 	 * vnode to the covered vnode.  For non-forced unmounts we want
1276 	 * such references to cause an EBUSY error.
1277 	 */
1278 	if ((flags & MNT_FORCE) &&
1279 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1280 		if (mp->mnt_vnodecovered != NULL)
1281 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1282 		if (fsrootvp == rootvnode) {
1283 			vrele(rootvnode);
1284 			rootvnode = NULL;
1285 		}
1286 		vput(fsrootvp);
1287 	}
1288 	if (((mp->mnt_flag & MNT_RDONLY) ||
1289 	     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0)
1290 		error = VFS_UNMOUNT(mp, flags);
1291 	vn_finished_write(mp);
1292 	/*
1293 	 * If we failed to flush the dirty blocks for this mount point,
1294 	 * undo all the cdir/rdir and rootvnode changes we made above.
1295 	 * Unless we failed to do so because the device is reporting that
1296 	 * it doesn't exist anymore.
1297 	 */
1298 	if (error && error != ENXIO) {
1299 		if ((flags & MNT_FORCE) &&
1300 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1301 			if (mp->mnt_vnodecovered != NULL)
1302 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1303 			if (rootvnode == NULL) {
1304 				rootvnode = fsrootvp;
1305 				vref(rootvnode);
1306 			}
1307 			vput(fsrootvp);
1308 		}
1309 		MNT_ILOCK(mp);
1310 		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1311 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) {
1312 			MNT_IUNLOCK(mp);
1313 			(void) vfs_allocate_syncvnode(mp);
1314 			MNT_ILOCK(mp);
1315 		}
1316 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1317 		mp->mnt_flag |= async_flag;
1318 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1319 			mp->mnt_kern_flag |= MNTK_ASYNC;
1320 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1321 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1322 			wakeup(mp);
1323 		}
1324 		MNT_IUNLOCK(mp);
1325 		if (coveredvp)
1326 			VOP_UNLOCK(coveredvp, 0);
1327 		return (error);
1328 	}
1329 	mtx_lock(&mountlist_mtx);
1330 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1331 	mtx_unlock(&mountlist_mtx);
1332 	if (coveredvp != NULL) {
1333 		coveredvp->v_mountedhere = NULL;
1334 		vput(coveredvp);
1335 	}
1336 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1337 	vfs_mount_destroy(mp);
1338 	return (0);
1339 }
1340 
1341 /*
1342  * ---------------------------------------------------------------------
1343  * Mounting of root filesystem
1344  *
1345  */
1346 
1347 struct root_hold_token {
1348 	const char			*who;
1349 	LIST_ENTRY(root_hold_token)	list;
1350 };
1351 
1352 static LIST_HEAD(, root_hold_token)	root_holds =
1353     LIST_HEAD_INITIALIZER(&root_holds);
1354 
1355 static int root_mount_complete;
1356 
1357 /*
1358  * Hold root mount.
1359  */
1360 struct root_hold_token *
1361 root_mount_hold(const char *identifier)
1362 {
1363 	struct root_hold_token *h;
1364 
1365 	if (root_mounted())
1366 		return (NULL);
1367 
1368 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1369 	h->who = identifier;
1370 	mtx_lock(&mountlist_mtx);
1371 	LIST_INSERT_HEAD(&root_holds, h, list);
1372 	mtx_unlock(&mountlist_mtx);
1373 	return (h);
1374 }
1375 
1376 /*
1377  * Release root mount.
1378  */
1379 void
1380 root_mount_rel(struct root_hold_token *h)
1381 {
1382 
1383 	if (h == NULL)
1384 		return;
1385 	mtx_lock(&mountlist_mtx);
1386 	LIST_REMOVE(h, list);
1387 	wakeup(&root_holds);
1388 	mtx_unlock(&mountlist_mtx);
1389 	free(h, M_DEVBUF);
1390 }
1391 
1392 /*
1393  * Wait for all subsystems to release root mount.
1394  */
1395 static void
1396 root_mount_prepare(void)
1397 {
1398 	struct root_hold_token *h;
1399 	struct timeval lastfail;
1400 	int curfail = 0;
1401 
1402 	for (;;) {
1403 		DROP_GIANT();
1404 		g_waitidle();
1405 		PICKUP_GIANT();
1406 		mtx_lock(&mountlist_mtx);
1407 		if (LIST_EMPTY(&root_holds)) {
1408 			mtx_unlock(&mountlist_mtx);
1409 			break;
1410 		}
1411 		if (ppsratecheck(&lastfail, &curfail, 1)) {
1412 			printf("Root mount waiting for:");
1413 			LIST_FOREACH(h, &root_holds, list)
1414 				printf(" %s", h->who);
1415 			printf("\n");
1416 		}
1417 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1418 		    hz);
1419 	}
1420 }
1421 
1422 /*
1423  * Root was mounted, share the good news.
1424  */
1425 static void
1426 root_mount_done(void)
1427 {
1428 
1429 	/* Keep prison0's root in sync with the global rootvnode. */
1430 	mtx_lock(&prison0.pr_mtx);
1431 	prison0.pr_root = rootvnode;
1432 	vref(prison0.pr_root);
1433 	mtx_unlock(&prison0.pr_mtx);
1434 	/*
1435 	 * Use a mutex to prevent the wakeup being missed and waiting for
1436 	 * an extra 1 second sleep.
1437 	 */
1438 	mtx_lock(&mountlist_mtx);
1439 	root_mount_complete = 1;
1440 	wakeup(&root_mount_complete);
1441 	mtx_unlock(&mountlist_mtx);
1442 }
1443 
1444 /*
1445  * Return true if root is already mounted.
1446  */
1447 int
1448 root_mounted(void)
1449 {
1450 
1451 	/* No mutex is acquired here because int stores are atomic. */
1452 	return (root_mount_complete);
1453 }
1454 
1455 /*
1456  * Wait until root is mounted.
1457  */
1458 void
1459 root_mount_wait(void)
1460 {
1461 
1462 	/*
1463 	 * Panic on an obvious deadlock - the function can't be called from
1464 	 * a thread which is doing the whole SYSINIT stuff.
1465 	 */
1466 	KASSERT(curthread->td_proc->p_pid != 0,
1467 	    ("root_mount_wait: cannot be called from the swapper thread"));
1468 	mtx_lock(&mountlist_mtx);
1469 	while (!root_mount_complete) {
1470 		msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
1471 		    hz);
1472 	}
1473 	mtx_unlock(&mountlist_mtx);
1474 }
1475 
1476 static void
1477 set_rootvnode()
1478 {
1479 	struct proc *p;
1480 
1481 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
1482 		panic("Cannot find root vnode");
1483 
1484 	VOP_UNLOCK(rootvnode, 0);
1485 
1486 	p = curthread->td_proc;
1487 	FILEDESC_XLOCK(p->p_fd);
1488 
1489 	if (p->p_fd->fd_cdir != NULL)
1490 		vrele(p->p_fd->fd_cdir);
1491 	p->p_fd->fd_cdir = rootvnode;
1492 	VREF(rootvnode);
1493 
1494 	if (p->p_fd->fd_rdir != NULL)
1495 		vrele(p->p_fd->fd_rdir);
1496 	p->p_fd->fd_rdir = rootvnode;
1497 	VREF(rootvnode);
1498 
1499 	FILEDESC_XUNLOCK(p->p_fd);
1500 
1501 	EVENTHANDLER_INVOKE(mountroot);
1502 }
1503 
1504 /*
1505  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1506  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1507  */
1508 
1509 static void
1510 devfs_first(void)
1511 {
1512 	struct thread *td = curthread;
1513 	struct vfsoptlist *opts;
1514 	struct vfsconf *vfsp;
1515 	struct mount *mp = NULL;
1516 	int error;
1517 
1518 	vfsp = vfs_byname("devfs");
1519 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1520 	if (vfsp == NULL)
1521 		return;
1522 
1523 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
1524 
1525 	error = VFS_MOUNT(mp);
1526 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1527 	if (error)
1528 		return;
1529 
1530 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1531 	TAILQ_INIT(opts);
1532 	mp->mnt_opt = opts;
1533 
1534 	mtx_lock(&mountlist_mtx);
1535 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1536 	mtx_unlock(&mountlist_mtx);
1537 
1538 	set_rootvnode();
1539 
1540 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1541 	if (error)
1542 		printf("kern_symlink /dev -> / returns %d\n", error);
1543 }
1544 
1545 /*
1546  * Surgically move our devfs to be mounted on /dev.
1547  */
1548 
1549 static void
1550 devfs_fixup(struct thread *td)
1551 {
1552 	struct nameidata nd;
1553 	int error;
1554 	struct vnode *vp, *dvp;
1555 	struct mount *mp;
1556 
1557 	/* Remove our devfs mount from the mountlist and purge the cache */
1558 	mtx_lock(&mountlist_mtx);
1559 	mp = TAILQ_FIRST(&mountlist);
1560 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1561 	mtx_unlock(&mountlist_mtx);
1562 	cache_purgevfs(mp);
1563 
1564 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp);
1565 	VI_LOCK(dvp);
1566 	dvp->v_iflag &= ~VI_MOUNT;
1567 	VI_UNLOCK(dvp);
1568 	dvp->v_mountedhere = NULL;
1569 
1570 	/* Set up the real rootvnode, and purge the cache */
1571 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1572 	set_rootvnode();
1573 	cache_purgevfs(rootvnode->v_mount);
1574 
1575 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1576 	error = namei(&nd);
1577 	if (error) {
1578 		printf("Lookup of /dev for devfs, error: %d\n", error);
1579 		return;
1580 	}
1581 	NDFREE(&nd, NDF_ONLY_PNBUF);
1582 	vp = nd.ni_vp;
1583 	if (vp->v_type != VDIR) {
1584 		vput(vp);
1585 	}
1586 	error = vinvalbuf(vp, V_SAVE, 0, 0);
1587 	if (error) {
1588 		vput(vp);
1589 	}
1590 	cache_purge(vp);
1591 	mp->mnt_vnodecovered = vp;
1592 	vp->v_mountedhere = mp;
1593 	mtx_lock(&mountlist_mtx);
1594 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1595 	mtx_unlock(&mountlist_mtx);
1596 	VOP_UNLOCK(vp, 0);
1597 	vput(dvp);
1598 	vfs_unbusy(mp);
1599 
1600 	/* Unlink the no longer needed /dev/dev -> / symlink */
1601 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1602 }
1603 
1604 /*
1605  * Report errors during filesystem mounting.
1606  */
1607 void
1608 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1609 {
1610 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1611 	va_list ap;
1612 	int error, len;
1613 	char *errmsg;
1614 
1615 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1616 	if (error || errmsg == NULL || len <= 0)
1617 		return;
1618 
1619 	va_start(ap, fmt);
1620 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1621 	va_end(ap);
1622 }
1623 
1624 void
1625 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1626 {
1627 	va_list ap;
1628 	int error, len;
1629 	char *errmsg;
1630 
1631 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1632 	if (error || errmsg == NULL || len <= 0)
1633 		return;
1634 
1635 	va_start(ap, fmt);
1636 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1637 	va_end(ap);
1638 }
1639 
1640 /*
1641  * Find and mount the root filesystem
1642  */
1643 void
1644 vfs_mountroot(void)
1645 {
1646 	char *cp, *options;
1647 	int error, i, asked = 0;
1648 
1649 	options = NULL;
1650 
1651 	root_mount_prepare();
1652 
1653 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1654 	    NULL, NULL, mount_init, mount_fini,
1655 	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1656 	devfs_first();
1657 
1658 	/*
1659 	 * We are booted with instructions to prompt for the root filesystem.
1660 	 */
1661 	if (boothowto & RB_ASKNAME) {
1662 		if (!vfs_mountroot_ask())
1663 			goto mounted;
1664 		asked = 1;
1665 	}
1666 
1667 	options = getenv("vfs.root.mountfrom.options");
1668 
1669 	/*
1670 	 * The root filesystem information is compiled in, and we are
1671 	 * booted with instructions to use it.
1672 	 */
1673 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1674 		if (!vfs_mountroot_try(ctrootdevname, options))
1675 			goto mounted;
1676 		ctrootdevname = NULL;
1677 	}
1678 
1679 	/*
1680 	 * We've been given the generic "use CDROM as root" flag.  This is
1681 	 * necessary because one media may be used in many different
1682 	 * devices, so we need to search for them.
1683 	 */
1684 	if (boothowto & RB_CDROM) {
1685 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1686 			if (!vfs_mountroot_try(cdrom_rootdevnames[i], options))
1687 				goto mounted;
1688 		}
1689 	}
1690 
1691 	/*
1692 	 * Try to use the value read by the loader from /etc/fstab, or
1693 	 * supplied via some other means.  This is the preferred
1694 	 * mechanism.
1695 	 */
1696 	cp = getenv("vfs.root.mountfrom");
1697 	if (cp != NULL) {
1698 		error = vfs_mountroot_try(cp, options);
1699 		freeenv(cp);
1700 		if (!error)
1701 			goto mounted;
1702 	}
1703 
1704 	/*
1705 	 * Try values that may have been computed by code during boot
1706 	 */
1707 	if (!vfs_mountroot_try(rootdevnames[0], options))
1708 		goto mounted;
1709 	if (!vfs_mountroot_try(rootdevnames[1], options))
1710 		goto mounted;
1711 
1712 	/*
1713 	 * If we (still) have a compiled-in default, try it.
1714 	 */
1715 	if (ctrootdevname != NULL)
1716 		if (!vfs_mountroot_try(ctrootdevname, options))
1717 			goto mounted;
1718 	/*
1719 	 * Everything so far has failed, prompt on the console if we haven't
1720 	 * already tried that.
1721 	 */
1722 	if (!asked)
1723 		if (!vfs_mountroot_ask())
1724 			goto mounted;
1725 
1726 	panic("Root mount failed, startup aborted.");
1727 
1728 mounted:
1729 	root_mount_done();
1730 	freeenv(options);
1731 }
1732 
1733 static struct mntarg *
1734 parse_mountroot_options(struct mntarg *ma, const char *options)
1735 {
1736 	char *p;
1737 	char *name, *name_arg;
1738 	char *val, *val_arg;
1739 	char *opts;
1740 
1741 	if (options == NULL || options[0] == '\0')
1742 		return (ma);
1743 
1744 	p = opts = strdup(options, M_MOUNT);
1745 	if (opts == NULL) {
1746 		return (ma);
1747 	}
1748 
1749 	while((name = strsep(&p, ",")) != NULL) {
1750 		if (name[0] == '\0')
1751 			break;
1752 
1753 		val = strchr(name, '=');
1754 		if (val != NULL) {
1755 			*val = '\0';
1756 			++val;
1757 		}
1758 		if( strcmp(name, "rw") == 0 ||
1759 		    strcmp(name, "noro") == 0) {
1760 			/*
1761 			 * The first time we mount the root file system,
1762 			 * we need to mount 'ro', so We need to ignore
1763 			 * 'rw' and 'noro' mount options.
1764 			 */
1765 			continue;
1766 		}
1767 		name_arg = strdup(name, M_MOUNT);
1768 		val_arg = NULL;
1769 		if (val != NULL)
1770 			val_arg = strdup(val, M_MOUNT);
1771 
1772 		ma = mount_arg(ma, name_arg, val_arg,
1773 		    (val_arg != NULL ? -1 : 0));
1774 	}
1775 	free(opts, M_MOUNT);
1776 	return (ma);
1777 }
1778 
1779 /*
1780  * Mount (mountfrom) as the root filesystem.
1781  */
1782 static int
1783 vfs_mountroot_try(const char *mountfrom, const char *options)
1784 {
1785 	struct mount	*mp;
1786 	struct mntarg	*ma;
1787 	char		*vfsname, *path;
1788 	time_t		timebase;
1789 	int		error;
1790 	char		patt[32];
1791 	char		errmsg[255];
1792 
1793 	vfsname = NULL;
1794 	path    = NULL;
1795 	mp      = NULL;
1796 	ma	= NULL;
1797 	error   = EINVAL;
1798 	bzero(errmsg, sizeof(errmsg));
1799 
1800 	if (mountfrom == NULL)
1801 		return (error);		/* don't complain */
1802 	printf("Trying to mount root from %s\n", mountfrom);
1803 
1804 	/* parse vfs name and path */
1805 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1806 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1807 	vfsname[0] = path[0] = 0;
1808 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1809 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1810 		goto out;
1811 
1812 	if (path[0] == '\0')
1813 		strcpy(path, ROOTNAME);
1814 
1815 	ma = mount_arg(ma, "fstype", vfsname, -1);
1816 	ma = mount_arg(ma, "fspath", "/", -1);
1817 	ma = mount_arg(ma, "from", path, -1);
1818 	ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg));
1819 	ma = mount_arg(ma, "ro", NULL, 0);
1820 	ma = parse_mountroot_options(ma, options);
1821 	error = kernel_mount(ma, MNT_ROOTFS);
1822 
1823 	if (error == 0) {
1824 		/*
1825 		 * We mount devfs prior to mounting the / FS, so the first
1826 		 * entry will typically be devfs.
1827 		 */
1828 		mp = TAILQ_FIRST(&mountlist);
1829 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1830 
1831 		/*
1832 		 * Iterate over all currently mounted file systems and use
1833 		 * the time stamp found to check and/or initialize the RTC.
1834 		 * Typically devfs has no time stamp and the only other FS
1835 		 * is the actual / FS.
1836 		 * Call inittodr() only once and pass it the largest of the
1837 		 * timestamps we encounter.
1838 		 */
1839 		timebase = 0;
1840 		do {
1841 			if (mp->mnt_time > timebase)
1842 				timebase = mp->mnt_time;
1843 			mp = TAILQ_NEXT(mp, mnt_list);
1844 		} while (mp != NULL);
1845 		inittodr(timebase);
1846 
1847 		devfs_fixup(curthread);
1848 	}
1849 
1850 	if (error != 0 ) {
1851 		printf("ROOT MOUNT ERROR: %s\n", errmsg);
1852 		printf("If you have invalid mount options, reboot, and ");
1853 		printf("first try the following from\n");
1854 		printf("the loader prompt:\n\n");
1855 		printf("     set vfs.root.mountfrom.options=rw\n\n");
1856 		printf("and then remove invalid mount options from ");
1857 		printf("/etc/fstab.\n\n");
1858 	}
1859 out:
1860 	free(path, M_MOUNT);
1861 	free(vfsname, M_MOUNT);
1862 	return (error);
1863 }
1864 
1865 /*
1866  * ---------------------------------------------------------------------
1867  * Interactive root filesystem selection code.
1868  */
1869 
1870 static int
1871 vfs_mountroot_ask(void)
1872 {
1873 	char name[128];
1874 	char *mountfrom;
1875 	char *options;
1876 
1877 	for(;;) {
1878 		printf("Loader variables:\n");
1879 		printf("vfs.root.mountfrom=");
1880 		mountfrom = getenv("vfs.root.mountfrom");
1881 		if (mountfrom != NULL) {
1882 			printf("%s", mountfrom);
1883 		}
1884 		printf("\n");
1885 		printf("vfs.root.mountfrom.options=");
1886 		options = getenv("vfs.root.mountfrom.options");
1887 		if (options != NULL) {
1888 			printf("%s", options);
1889 		}
1890 		printf("\n");
1891 		freeenv(mountfrom);
1892 		freeenv(options);
1893 		printf("\nManual root filesystem specification:\n");
1894 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1895 		printf("                       eg. ufs:/dev/da0s1a\n");
1896 		printf("                       eg. cd9660:/dev/acd0\n");
1897 		printf("                       This is equivalent to: ");
1898 		printf("mount -t cd9660 /dev/acd0 /\n");
1899 		printf("\n");
1900 		printf("  ?                  List valid disk boot devices\n");
1901 		printf("  <empty line>       Abort manual input\n");
1902 		printf("\nmountroot> ");
1903 		gets(name, sizeof(name), 1);
1904 		if (name[0] == '\0')
1905 			return (1);
1906 		if (name[0] == '?') {
1907 			printf("\nList of GEOM managed disk devices:\n  ");
1908 			g_dev_print();
1909 			continue;
1910 		}
1911 		if (!vfs_mountroot_try(name, NULL))
1912 			return (0);
1913 	}
1914 }
1915 
1916 /*
1917  * ---------------------------------------------------------------------
1918  * Functions for querying mount options/arguments from filesystems.
1919  */
1920 
1921 /*
1922  * Check that no unknown options are given
1923  */
1924 int
1925 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1926 {
1927 	struct vfsopt *opt;
1928 	char errmsg[255];
1929 	const char **t, *p, *q;
1930 	int ret = 0;
1931 
1932 	TAILQ_FOREACH(opt, opts, link) {
1933 		p = opt->name;
1934 		q = NULL;
1935 		if (p[0] == 'n' && p[1] == 'o')
1936 			q = p + 2;
1937 		for(t = global_opts; *t != NULL; t++) {
1938 			if (strcmp(*t, p) == 0)
1939 				break;
1940 			if (q != NULL) {
1941 				if (strcmp(*t, q) == 0)
1942 					break;
1943 			}
1944 		}
1945 		if (*t != NULL)
1946 			continue;
1947 		for(t = legal; *t != NULL; t++) {
1948 			if (strcmp(*t, p) == 0)
1949 				break;
1950 			if (q != NULL) {
1951 				if (strcmp(*t, q) == 0)
1952 					break;
1953 			}
1954 		}
1955 		if (*t != NULL)
1956 			continue;
1957 		snprintf(errmsg, sizeof(errmsg),
1958 		    "mount option <%s> is unknown", p);
1959 		printf("%s\n", errmsg);
1960 		ret = EINVAL;
1961 	}
1962 	if (ret != 0) {
1963 		TAILQ_FOREACH(opt, opts, link) {
1964 			if (strcmp(opt->name, "errmsg") == 0) {
1965 				strncpy((char *)opt->value, errmsg, opt->len);
1966 			}
1967 		}
1968 	}
1969 	return (ret);
1970 }
1971 
1972 /*
1973  * Get a mount option by its name.
1974  *
1975  * Return 0 if the option was found, ENOENT otherwise.
1976  * If len is non-NULL it will be filled with the length
1977  * of the option. If buf is non-NULL, it will be filled
1978  * with the address of the option.
1979  */
1980 int
1981 vfs_getopt(opts, name, buf, len)
1982 	struct vfsoptlist *opts;
1983 	const char *name;
1984 	void **buf;
1985 	int *len;
1986 {
1987 	struct vfsopt *opt;
1988 
1989 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1990 
1991 	TAILQ_FOREACH(opt, opts, link) {
1992 		if (strcmp(name, opt->name) == 0) {
1993 			opt->seen = 1;
1994 			if (len != NULL)
1995 				*len = opt->len;
1996 			if (buf != NULL)
1997 				*buf = opt->value;
1998 			return (0);
1999 		}
2000 	}
2001 	return (ENOENT);
2002 }
2003 
2004 int
2005 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2006 {
2007 	struct vfsopt *opt;
2008 
2009 	if (opts == NULL)
2010 		return (-1);
2011 
2012 	TAILQ_FOREACH(opt, opts, link) {
2013 		if (strcmp(name, opt->name) == 0) {
2014 			opt->seen = 1;
2015 			return (opt->pos);
2016 		}
2017 	}
2018 	return (-1);
2019 }
2020 
2021 char *
2022 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2023 {
2024 	struct vfsopt *opt;
2025 
2026 	*error = 0;
2027 	TAILQ_FOREACH(opt, opts, link) {
2028 		if (strcmp(name, opt->name) != 0)
2029 			continue;
2030 		opt->seen = 1;
2031 		if (opt->len == 0 ||
2032 		    ((char *)opt->value)[opt->len - 1] != '\0') {
2033 			*error = EINVAL;
2034 			return (NULL);
2035 		}
2036 		return (opt->value);
2037 	}
2038 	*error = ENOENT;
2039 	return (NULL);
2040 }
2041 
2042 int
2043 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
2044 {
2045 	struct vfsopt *opt;
2046 
2047 	TAILQ_FOREACH(opt, opts, link) {
2048 		if (strcmp(name, opt->name) == 0) {
2049 			opt->seen = 1;
2050 			if (w != NULL)
2051 				*w |= val;
2052 			return (1);
2053 		}
2054 	}
2055 	if (w != NULL)
2056 		*w &= ~val;
2057 	return (0);
2058 }
2059 
2060 int
2061 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2062 {
2063 	va_list ap;
2064 	struct vfsopt *opt;
2065 	int ret;
2066 
2067 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2068 
2069 	TAILQ_FOREACH(opt, opts, link) {
2070 		if (strcmp(name, opt->name) != 0)
2071 			continue;
2072 		opt->seen = 1;
2073 		if (opt->len == 0 || opt->value == NULL)
2074 			return (0);
2075 		if (((char *)opt->value)[opt->len - 1] != '\0')
2076 			return (0);
2077 		va_start(ap, fmt);
2078 		ret = vsscanf(opt->value, fmt, ap);
2079 		va_end(ap);
2080 		return (ret);
2081 	}
2082 	return (0);
2083 }
2084 
2085 int
2086 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2087 {
2088 	struct vfsopt *opt;
2089 
2090 	TAILQ_FOREACH(opt, opts, link) {
2091 		if (strcmp(name, opt->name) != 0)
2092 			continue;
2093 		opt->seen = 1;
2094 		if (opt->value == NULL)
2095 			opt->len = len;
2096 		else {
2097 			if (opt->len != len)
2098 				return (EINVAL);
2099 			bcopy(value, opt->value, len);
2100 		}
2101 		return (0);
2102 	}
2103 	return (ENOENT);
2104 }
2105 
2106 int
2107 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2108 {
2109 	struct vfsopt *opt;
2110 
2111 	TAILQ_FOREACH(opt, opts, link) {
2112 		if (strcmp(name, opt->name) != 0)
2113 			continue;
2114 		opt->seen = 1;
2115 		if (opt->value == NULL)
2116 			opt->len = len;
2117 		else {
2118 			if (opt->len < len)
2119 				return (EINVAL);
2120 			opt->len = len;
2121 			bcopy(value, opt->value, len);
2122 		}
2123 		return (0);
2124 	}
2125 	return (ENOENT);
2126 }
2127 
2128 int
2129 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2130 {
2131 	struct vfsopt *opt;
2132 
2133 	TAILQ_FOREACH(opt, opts, link) {
2134 		if (strcmp(name, opt->name) != 0)
2135 			continue;
2136 		opt->seen = 1;
2137 		if (opt->value == NULL)
2138 			opt->len = strlen(value) + 1;
2139 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2140 			return (EINVAL);
2141 		return (0);
2142 	}
2143 	return (ENOENT);
2144 }
2145 
2146 /*
2147  * Find and copy a mount option.
2148  *
2149  * The size of the buffer has to be specified
2150  * in len, if it is not the same length as the
2151  * mount option, EINVAL is returned.
2152  * Returns ENOENT if the option is not found.
2153  */
2154 int
2155 vfs_copyopt(opts, name, dest, len)
2156 	struct vfsoptlist *opts;
2157 	const char *name;
2158 	void *dest;
2159 	int len;
2160 {
2161 	struct vfsopt *opt;
2162 
2163 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2164 
2165 	TAILQ_FOREACH(opt, opts, link) {
2166 		if (strcmp(name, opt->name) == 0) {
2167 			opt->seen = 1;
2168 			if (len != opt->len)
2169 				return (EINVAL);
2170 			bcopy(opt->value, dest, opt->len);
2171 			return (0);
2172 		}
2173 	}
2174 	return (ENOENT);
2175 }
2176 
2177 /*
2178  * This is a helper function for filesystems to traverse their
2179  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
2180  */
2181 
2182 struct vnode *
2183 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
2184 {
2185 	struct vnode *vp;
2186 
2187 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2188 
2189 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2190 	if ((*mvp)->v_yield++ == 500) {
2191 		MNT_IUNLOCK(mp);
2192 		(*mvp)->v_yield = 0;
2193 		uio_yield();
2194 		MNT_ILOCK(mp);
2195 	}
2196 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
2197 	while (vp != NULL && vp->v_type == VMARKER)
2198 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2199 
2200 	/* Check if we are done */
2201 	if (vp == NULL) {
2202 		__mnt_vnode_markerfree(mvp, mp);
2203 		return (NULL);
2204 	}
2205 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2206 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2207 	return (vp);
2208 }
2209 
2210 struct vnode *
2211 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
2212 {
2213 	struct vnode *vp;
2214 
2215 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2216 
2217 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2218 	while (vp != NULL && vp->v_type == VMARKER)
2219 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2220 
2221 	/* Check if we are done */
2222 	if (vp == NULL) {
2223 		*mvp = NULL;
2224 		return (NULL);
2225 	}
2226 	MNT_REF(mp);
2227 	MNT_IUNLOCK(mp);
2228 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
2229 				       M_VNODE_MARKER,
2230 				       M_WAITOK | M_ZERO);
2231 	MNT_ILOCK(mp);
2232 	(*mvp)->v_type = VMARKER;
2233 
2234 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2235 	while (vp != NULL && vp->v_type == VMARKER)
2236 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2237 
2238 	/* Check if we are done */
2239 	if (vp == NULL) {
2240 		MNT_IUNLOCK(mp);
2241 		free(*mvp, M_VNODE_MARKER);
2242 		MNT_ILOCK(mp);
2243 		*mvp = NULL;
2244 		MNT_REL(mp);
2245 		return (NULL);
2246 	}
2247 	(*mvp)->v_mount = mp;
2248 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2249 	return (vp);
2250 }
2251 
2252 
2253 void
2254 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
2255 {
2256 
2257 	if (*mvp == NULL)
2258 		return;
2259 
2260 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2261 
2262 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2263 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2264 	MNT_IUNLOCK(mp);
2265 	free(*mvp, M_VNODE_MARKER);
2266 	MNT_ILOCK(mp);
2267 	*mvp = NULL;
2268 	MNT_REL(mp);
2269 }
2270 
2271 
2272 int
2273 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2274 {
2275 	int error;
2276 
2277 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat);
2278 	if (sbp != &mp->mnt_stat)
2279 		*sbp = mp->mnt_stat;
2280 	return (error);
2281 }
2282 
2283 void
2284 vfs_mountedfrom(struct mount *mp, const char *from)
2285 {
2286 
2287 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2288 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2289 	    sizeof mp->mnt_stat.f_mntfromname);
2290 }
2291 
2292 /*
2293  * ---------------------------------------------------------------------
2294  * This is the api for building mount args and mounting filesystems from
2295  * inside the kernel.
2296  *
2297  * The API works by accumulation of individual args.  First error is
2298  * latched.
2299  *
2300  * XXX: should be documented in new manpage kernel_mount(9)
2301  */
2302 
2303 /* A memory allocation which must be freed when we are done */
2304 struct mntaarg {
2305 	SLIST_ENTRY(mntaarg)	next;
2306 };
2307 
2308 /* The header for the mount arguments */
2309 struct mntarg {
2310 	struct iovec *v;
2311 	int len;
2312 	int error;
2313 	SLIST_HEAD(, mntaarg)	list;
2314 };
2315 
2316 /*
2317  * Add a boolean argument.
2318  *
2319  * flag is the boolean value.
2320  * name must start with "no".
2321  */
2322 struct mntarg *
2323 mount_argb(struct mntarg *ma, int flag, const char *name)
2324 {
2325 
2326 	KASSERT(name[0] == 'n' && name[1] == 'o',
2327 	    ("mount_argb(...,%s): name must start with 'no'", name));
2328 
2329 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2330 }
2331 
2332 /*
2333  * Add an argument printf style
2334  */
2335 struct mntarg *
2336 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2337 {
2338 	va_list ap;
2339 	struct mntaarg *maa;
2340 	struct sbuf *sb;
2341 	int len;
2342 
2343 	if (ma == NULL) {
2344 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2345 		SLIST_INIT(&ma->list);
2346 	}
2347 	if (ma->error)
2348 		return (ma);
2349 
2350 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2351 	    M_MOUNT, M_WAITOK);
2352 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2353 	ma->v[ma->len].iov_len = strlen(name) + 1;
2354 	ma->len++;
2355 
2356 	sb = sbuf_new_auto();
2357 	va_start(ap, fmt);
2358 	sbuf_vprintf(sb, fmt, ap);
2359 	va_end(ap);
2360 	sbuf_finish(sb);
2361 	len = sbuf_len(sb) + 1;
2362 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2363 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2364 	bcopy(sbuf_data(sb), maa + 1, len);
2365 	sbuf_delete(sb);
2366 
2367 	ma->v[ma->len].iov_base = maa + 1;
2368 	ma->v[ma->len].iov_len = len;
2369 	ma->len++;
2370 
2371 	return (ma);
2372 }
2373 
2374 /*
2375  * Add an argument which is a userland string.
2376  */
2377 struct mntarg *
2378 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2379 {
2380 	struct mntaarg *maa;
2381 	char *tbuf;
2382 
2383 	if (val == NULL)
2384 		return (ma);
2385 	if (ma == NULL) {
2386 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2387 		SLIST_INIT(&ma->list);
2388 	}
2389 	if (ma->error)
2390 		return (ma);
2391 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2392 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2393 	tbuf = (void *)(maa + 1);
2394 	ma->error = copyinstr(val, tbuf, len, NULL);
2395 	return (mount_arg(ma, name, tbuf, -1));
2396 }
2397 
2398 /*
2399  * Plain argument.
2400  *
2401  * If length is -1, treat value as a C string.
2402  */
2403 struct mntarg *
2404 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2405 {
2406 
2407 	if (ma == NULL) {
2408 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2409 		SLIST_INIT(&ma->list);
2410 	}
2411 	if (ma->error)
2412 		return (ma);
2413 
2414 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2415 	    M_MOUNT, M_WAITOK);
2416 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2417 	ma->v[ma->len].iov_len = strlen(name) + 1;
2418 	ma->len++;
2419 
2420 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2421 	if (len < 0)
2422 		ma->v[ma->len].iov_len = strlen(val) + 1;
2423 	else
2424 		ma->v[ma->len].iov_len = len;
2425 	ma->len++;
2426 	return (ma);
2427 }
2428 
2429 /*
2430  * Free a mntarg structure
2431  */
2432 static void
2433 free_mntarg(struct mntarg *ma)
2434 {
2435 	struct mntaarg *maa;
2436 
2437 	while (!SLIST_EMPTY(&ma->list)) {
2438 		maa = SLIST_FIRST(&ma->list);
2439 		SLIST_REMOVE_HEAD(&ma->list, next);
2440 		free(maa, M_MOUNT);
2441 	}
2442 	free(ma->v, M_MOUNT);
2443 	free(ma, M_MOUNT);
2444 }
2445 
2446 /*
2447  * Mount a filesystem
2448  */
2449 int
2450 kernel_mount(struct mntarg *ma, int flags)
2451 {
2452 	struct uio auio;
2453 	int error;
2454 
2455 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2456 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2457 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2458 
2459 	auio.uio_iov = ma->v;
2460 	auio.uio_iovcnt = ma->len;
2461 	auio.uio_segflg = UIO_SYSSPACE;
2462 
2463 	error = ma->error;
2464 	if (!error)
2465 		error = vfs_donmount(curthread, flags, &auio);
2466 	free_mntarg(ma);
2467 	return (error);
2468 }
2469 
2470 /*
2471  * A printflike function to mount a filesystem.
2472  */
2473 int
2474 kernel_vmount(int flags, ...)
2475 {
2476 	struct mntarg *ma = NULL;
2477 	va_list ap;
2478 	const char *cp;
2479 	const void *vp;
2480 	int error;
2481 
2482 	va_start(ap, flags);
2483 	for (;;) {
2484 		cp = va_arg(ap, const char *);
2485 		if (cp == NULL)
2486 			break;
2487 		vp = va_arg(ap, const void *);
2488 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2489 	}
2490 	va_end(ap);
2491 
2492 	error = kernel_mount(ma, flags);
2493 	return (error);
2494 }
2495