xref: /freebsd/sys/kern/vfs_mount.c (revision a743df5c964d81a7c920cf257e87cb42ab993d58)
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 		AUDIT_ARG(text, fstype);
749 		mtx_lock(&Giant);	/* XXX ? */
750 		vfsp = vfs_byname_kld(fstype, td, &error);
751 		mtx_unlock(&Giant);
752 	}
753 	free(fstype, M_TEMP);
754 	if (error)
755 		return (error);
756 	if (vfsp == NULL)
757 		return (ENOENT);
758 	if (vfsp->vfc_vfsops->vfs_cmount == NULL)
759 		return (EOPNOTSUPP);
760 
761 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
762 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
763 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
764 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
765 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
766 
767 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
768 	return (error);
769 }
770 
771 
772 /*
773  * vfs_domount(): actually attempt a filesystem mount.
774  */
775 static int
776 vfs_domount(
777 	struct thread *td,	/* Flags common to all filesystems. */
778 	const char *fstype,	/* Filesystem type. */
779 	char *fspath,		/* Mount path. */
780 	int fsflags,		/* Flags common to all filesystems. */
781 	void *fsdata		/* Options local to the filesystem. */
782 	)
783 {
784 	struct vnode *vp;
785 	struct mount *mp;
786 	struct vfsconf *vfsp;
787 	struct export_args export;
788 	int error, flag = 0, kern_flag = 0;
789 	struct vattr va;
790 	struct nameidata nd;
791 
792 	mtx_assert(&Giant, MA_OWNED);
793 	/*
794 	 * Be ultra-paranoid about making sure the type and fspath
795 	 * variables will fit in our mp buffers, including the
796 	 * terminating NUL.
797 	 */
798 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
799 		return (ENAMETOOLONG);
800 
801 	if (jailed(td->td_ucred))
802 		return (EPERM);
803 	if (usermount == 0) {
804 		if ((error = suser(td)) != 0)
805 			return (error);
806 	}
807 
808 	/*
809 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
810 	 */
811 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
812 		if ((error = suser(td)) != 0)
813 			return (error);
814 	}
815 	/*
816 	 * Silently enforce MNT_NOSUID and MNT_USER for
817 	 * unprivileged users.
818 	 */
819 	if (suser(td) != 0)
820 		fsflags |= MNT_NOSUID | MNT_USER;
821 
822 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
823 	vfsp = NULL;
824 	if ((fsflags & MNT_UPDATE) == 0) {
825 		/* Don't try to load KLDs if we're mounting the root. */
826 		if (fsflags & MNT_ROOTFS)
827 			vfsp = vfs_byname(fstype);
828 		else
829 			vfsp = vfs_byname_kld(fstype, td, &error);
830 		if (vfsp == NULL)
831 			return (ENODEV);
832 	}
833 	/*
834 	 * Get vnode to be covered
835 	 */
836 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
837 	    fspath, td);
838 	if ((error = namei(&nd)) != 0)
839 		return (error);
840 	NDFREE(&nd, NDF_ONLY_PNBUF);
841 	vp = nd.ni_vp;
842 	if (fsflags & MNT_UPDATE) {
843 		if ((vp->v_vflag & VV_ROOT) == 0) {
844 			vput(vp);
845 			return (EINVAL);
846 		}
847 		mp = vp->v_mount;
848 		flag = mp->mnt_flag;
849 		kern_flag = mp->mnt_kern_flag;
850 		/*
851 		 * We only allow the filesystem to be reloaded if it
852 		 * is currently mounted read-only.
853 		 */
854 		if ((fsflags & MNT_RELOAD) &&
855 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
856 			vput(vp);
857 			return (EOPNOTSUPP);	/* Needs translation */
858 		}
859 		/*
860 		 * Only privileged root, or (if MNT_USER is set) the user that
861 		 * did the original mount is permitted to update it.
862 		 */
863 		error = vfs_suser(mp, td);
864 		if (error) {
865 			vput(vp);
866 			return (error);
867 		}
868 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
869 			vput(vp);
870 			return (EBUSY);
871 		}
872 		VI_LOCK(vp);
873 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
874 		    vp->v_mountedhere != NULL) {
875 			VI_UNLOCK(vp);
876 			vfs_unbusy(mp, td);
877 			vput(vp);
878 			return (EBUSY);
879 		}
880 		vp->v_iflag |= VI_MOUNT;
881 		VI_UNLOCK(vp);
882 		mp->mnt_flag |= fsflags &
883 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
884 		VOP_UNLOCK(vp, 0, td);
885 		mp->mnt_optnew = fsdata;
886 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
887 	} else {
888 		/*
889 		 * If the user is not root, ensure that they own the directory
890 		 * onto which we are attempting to mount.
891 		 */
892 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
893 		if (error) {
894 			vput(vp);
895 			return (error);
896 		}
897 		if (va.va_uid != td->td_ucred->cr_uid) {
898 			if ((error = suser(td)) != 0) {
899 				vput(vp);
900 				return (error);
901 			}
902 		}
903 		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
904 		if (error != 0) {
905 			vput(vp);
906 			return (error);
907 		}
908 		if (vp->v_type != VDIR) {
909 			vput(vp);
910 			return (ENOTDIR);
911 		}
912 		VI_LOCK(vp);
913 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
914 		    vp->v_mountedhere != NULL) {
915 			VI_UNLOCK(vp);
916 			vput(vp);
917 			return (EBUSY);
918 		}
919 		vp->v_iflag |= VI_MOUNT;
920 		VI_UNLOCK(vp);
921 
922 		/*
923 		 * Allocate and initialize the filesystem.
924 		 */
925 		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
926 		VOP_UNLOCK(vp, 0, td);
927 
928 		/* XXXMAC: pass to vfs_mount_alloc? */
929 		mp->mnt_optnew = fsdata;
930 	}
931 
932 	/*
933 	 * Set the mount level flags.
934 	 */
935 	if (fsflags & MNT_RDONLY)
936 		mp->mnt_flag |= MNT_RDONLY;
937 	mp->mnt_flag &=~ MNT_UPDATEMASK;
938 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
939 	/*
940 	 * Mount the filesystem.
941 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
942 	 * get.  No freeing of cn_pnbuf.
943 	 */
944         error = VFS_MOUNT(mp, td);
945 
946 	/*
947 	 * Process the export option only if we are
948 	 * updating mount options.
949 	 */
950 	if (!error && (fsflags & MNT_UPDATE)) {
951 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
952 		    sizeof(export)) == 0)
953 			error = vfs_export(mp, &export);
954 	}
955 
956 	if (!error) {
957 		if (mp->mnt_opt != NULL)
958 			vfs_freeopts(mp->mnt_opt);
959 		mp->mnt_opt = mp->mnt_optnew;
960 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
961 	}
962 	/*
963 	 * Prevent external consumers of mount options from reading
964 	 * mnt_optnew.
965 	*/
966 	mp->mnt_optnew = NULL;
967 	if (mp->mnt_flag & MNT_UPDATE) {
968 		mp->mnt_flag &=
969 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
970 		if (error) {
971 			mp->mnt_flag = flag;
972 			mp->mnt_kern_flag = kern_flag;
973 		}
974 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
975 			if (mp->mnt_syncer == NULL)
976 				error = vfs_allocate_syncvnode(mp);
977 		} else {
978 			if (mp->mnt_syncer != NULL)
979 				vrele(mp->mnt_syncer);
980 			mp->mnt_syncer = NULL;
981 		}
982 		vfs_unbusy(mp, td);
983 		VI_LOCK(vp);
984 		vp->v_iflag &= ~VI_MOUNT;
985 		VI_UNLOCK(vp);
986 		vrele(vp);
987 		return (error);
988 	}
989 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
990 	/*
991 	 * Put the new filesystem on the mount list after root.
992 	 */
993 	cache_purge(vp);
994 	if (!error) {
995 		struct vnode *newdp;
996 
997 		VI_LOCK(vp);
998 		vp->v_iflag &= ~VI_MOUNT;
999 		VI_UNLOCK(vp);
1000 		vp->v_mountedhere = mp;
1001 		mtx_lock(&mountlist_mtx);
1002 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1003 		mtx_unlock(&mountlist_mtx);
1004 		vfs_event_signal(NULL, VQ_MOUNT, 0);
1005 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1006 			panic("mount: lost mount");
1007 		mountcheckdirs(vp, newdp);
1008 		vput(newdp);
1009 		VOP_UNLOCK(vp, 0, td);
1010 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1011 			error = vfs_allocate_syncvnode(mp);
1012 		vfs_unbusy(mp, td);
1013 		if (error)
1014 			vrele(vp);
1015 	} else {
1016 		VI_LOCK(vp);
1017 		vp->v_iflag &= ~VI_MOUNT;
1018 		VI_UNLOCK(vp);
1019 		vfs_unbusy(mp, td);
1020 		vfs_mount_destroy(mp);
1021 		vput(vp);
1022 	}
1023 	return (error);
1024 }
1025 
1026 /*
1027  * ---------------------------------------------------------------------
1028  * Unmount a filesystem.
1029  *
1030  * Note: unmount takes a path to the vnode mounted on as argument,
1031  * not special file (as before).
1032  */
1033 #ifndef _SYS_SYSPROTO_H_
1034 struct unmount_args {
1035 	char	*path;
1036 	int	flags;
1037 };
1038 #endif
1039 /* ARGSUSED */
1040 int
1041 unmount(td, uap)
1042 	struct thread *td;
1043 	register struct unmount_args /* {
1044 		char *path;
1045 		int flags;
1046 	} */ *uap;
1047 {
1048 	struct mount *mp;
1049 	char *pathbuf;
1050 	int error, id0, id1;
1051 
1052 	if (jailed(td->td_ucred))
1053 		return (EPERM);
1054 	if (usermount == 0) {
1055 		if ((error = suser(td)) != 0)
1056 			return (error);
1057 	}
1058 
1059 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1060 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1061 	if (error) {
1062 		free(pathbuf, M_TEMP);
1063 		return (error);
1064 	}
1065 	AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1066 	if (uap->flags & MNT_BYFSID) {
1067 		/* Decode the filesystem ID. */
1068 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1069 			free(pathbuf, M_TEMP);
1070 			return (EINVAL);
1071 		}
1072 
1073 		mtx_lock(&mountlist_mtx);
1074 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1075 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1076 			    mp->mnt_stat.f_fsid.val[1] == id1)
1077 				break;
1078 		}
1079 		mtx_unlock(&mountlist_mtx);
1080 	} else {
1081 		mtx_lock(&mountlist_mtx);
1082 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1083 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1084 				break;
1085 		}
1086 		mtx_unlock(&mountlist_mtx);
1087 	}
1088 	free(pathbuf, M_TEMP);
1089 	if (mp == NULL) {
1090 		/*
1091 		 * Previously we returned ENOENT for a nonexistent path and
1092 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1093 		 * now, so in the !MNT_BYFSID case return the more likely
1094 		 * EINVAL for compatibility.
1095 		 */
1096 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1097 	}
1098 
1099 	/*
1100 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1101 	 * original mount is permitted to unmount this filesystem.
1102 	 */
1103 	error = vfs_suser(mp, td);
1104 	if (error)
1105 		return (error);
1106 
1107 	/*
1108 	 * Don't allow unmounting the root filesystem.
1109 	 */
1110 	if (mp->mnt_flag & MNT_ROOTFS)
1111 		return (EINVAL);
1112 	mtx_lock(&Giant);
1113 	error = dounmount(mp, uap->flags, td);
1114 	mtx_unlock(&Giant);
1115 	return (error);
1116 }
1117 
1118 /*
1119  * Do the actual filesystem unmount.
1120  */
1121 int
1122 dounmount(mp, flags, td)
1123 	struct mount *mp;
1124 	int flags;
1125 	struct thread *td;
1126 {
1127 	struct vnode *coveredvp, *fsrootvp;
1128 	int error;
1129 	int async_flag;
1130 
1131 	mtx_assert(&Giant, MA_OWNED);
1132 
1133 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1134 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY, td);
1135 	MNT_ILOCK(mp);
1136 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1137 		MNT_IUNLOCK(mp);
1138 		if (coveredvp)
1139 			VOP_UNLOCK(coveredvp, 0, td);
1140 		return (EBUSY);
1141 	}
1142 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1143 	/* Allow filesystems to detect that a forced unmount is in progress. */
1144 	if (flags & MNT_FORCE)
1145 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1146 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1147 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1148 	if (error) {
1149 		MNT_ILOCK(mp);
1150 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1151 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1152 			wakeup(mp);
1153 		MNT_IUNLOCK(mp);
1154 		if (coveredvp)
1155 			VOP_UNLOCK(coveredvp, 0, td);
1156 		return (error);
1157 	}
1158 	vn_start_write(NULL, &mp, V_WAIT);
1159 
1160 	if (mp->mnt_flag & MNT_EXPUBLIC)
1161 		vfs_setpublicfs(NULL, NULL, NULL);
1162 
1163 	vfs_msync(mp, MNT_WAIT);
1164 	async_flag = mp->mnt_flag & MNT_ASYNC;
1165 	mp->mnt_flag &= ~MNT_ASYNC;
1166 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1167 	if (mp->mnt_syncer != NULL)
1168 		vrele(mp->mnt_syncer);
1169 	/*
1170 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1171 	 * vnode to the covered vnode.  For non-forced unmounts we want
1172 	 * such references to cause an EBUSY error.
1173 	 */
1174 	if ((flags & MNT_FORCE) &&
1175 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1176 		if (mp->mnt_vnodecovered != NULL)
1177 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1178 		if (fsrootvp == rootvnode) {
1179 			vrele(rootvnode);
1180 			rootvnode = NULL;
1181 		}
1182 		vput(fsrootvp);
1183 	}
1184 	if (((mp->mnt_flag & MNT_RDONLY) ||
1185 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1186 	    (flags & MNT_FORCE)) {
1187 		error = VFS_UNMOUNT(mp, flags, td);
1188 	}
1189 	vn_finished_write(mp);
1190 	if (error) {
1191 		/* Undo cdir/rdir and rootvnode changes made above. */
1192 		if ((flags & MNT_FORCE) &&
1193 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1194 			if (mp->mnt_vnodecovered != NULL)
1195 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1196 			if (rootvnode == NULL) {
1197 				rootvnode = fsrootvp;
1198 				vref(rootvnode);
1199 			}
1200 			vput(fsrootvp);
1201 		}
1202 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1203 			(void) vfs_allocate_syncvnode(mp);
1204 		MNT_ILOCK(mp);
1205 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1206 		mp->mnt_flag |= async_flag;
1207 		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1208 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1209 			wakeup(mp);
1210 		MNT_IUNLOCK(mp);
1211 		if (coveredvp)
1212 			VOP_UNLOCK(coveredvp, 0, td);
1213 		return (error);
1214 	}
1215 	mtx_lock(&mountlist_mtx);
1216 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1217 	mtx_unlock(&mountlist_mtx);
1218 	if (coveredvp != NULL) {
1219 		coveredvp->v_mountedhere = NULL;
1220 		vput(coveredvp);
1221 	}
1222 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1223 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1224 	vfs_mount_destroy(mp);
1225 	return (0);
1226 }
1227 
1228 /*
1229  * ---------------------------------------------------------------------
1230  * Mounting of root filesystem
1231  *
1232  */
1233 
1234 struct root_hold_token {
1235 	const char 			*who;
1236 	LIST_ENTRY(root_hold_token)	list;
1237 };
1238 
1239 static LIST_HEAD(, root_hold_token)	root_holds =
1240     LIST_HEAD_INITIALIZER(&root_holds);
1241 
1242 struct root_hold_token *
1243 root_mount_hold(const char *identifier)
1244 {
1245 	struct root_hold_token *h;
1246 
1247 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1248 	h->who = identifier;
1249 	mtx_lock(&mountlist_mtx);
1250 	LIST_INSERT_HEAD(&root_holds, h, list);
1251 	mtx_unlock(&mountlist_mtx);
1252 	return (h);
1253 }
1254 
1255 void
1256 root_mount_rel(struct root_hold_token *h)
1257 {
1258 
1259 	mtx_lock(&mountlist_mtx);
1260 	LIST_REMOVE(h, list);
1261 	wakeup(&root_holds);
1262 	mtx_unlock(&mountlist_mtx);
1263 	free(h, M_DEVBUF);
1264 }
1265 
1266 static void
1267 root_mount_wait(void)
1268 {
1269 	struct root_hold_token *h;
1270 
1271 	for (;;) {
1272 		DROP_GIANT();
1273 		g_waitidle();
1274 		PICKUP_GIANT();
1275 		mtx_lock(&mountlist_mtx);
1276 		if (LIST_EMPTY(&root_holds)) {
1277 			mtx_unlock(&mountlist_mtx);
1278 			break;
1279 		}
1280 		printf("Root mount waiting for:");
1281 		LIST_FOREACH(h, &root_holds, list)
1282 			printf(" %s", h->who);
1283 		printf("\n");
1284 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1285 		    hz);
1286 	}
1287 }
1288 
1289 static void
1290 set_rootvnode(struct thread *td)
1291 {
1292 	struct proc *p;
1293 
1294 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1295 		panic("Cannot find root vnode");
1296 
1297 	p = td->td_proc;
1298 	FILEDESC_LOCK(p->p_fd);
1299 
1300 	if (p->p_fd->fd_cdir != NULL)
1301 		vrele(p->p_fd->fd_cdir);
1302 	p->p_fd->fd_cdir = rootvnode;
1303 	VREF(rootvnode);
1304 
1305 	if (p->p_fd->fd_rdir != NULL)
1306 		vrele(p->p_fd->fd_rdir);
1307 	p->p_fd->fd_rdir = rootvnode;
1308 	VREF(rootvnode);
1309 
1310 	FILEDESC_UNLOCK(p->p_fd);
1311 
1312 	VOP_UNLOCK(rootvnode, 0, td);
1313 }
1314 
1315 /*
1316  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1317  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1318  */
1319 
1320 static void
1321 devfs_first(void)
1322 {
1323 	struct thread *td = curthread;
1324 	struct vfsoptlist *opts;
1325 	struct vfsconf *vfsp;
1326 	struct mount *mp = NULL;
1327 	int error;
1328 
1329 	vfsp = vfs_byname("devfs");
1330 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1331 	if (vfsp == NULL)
1332 		return;
1333 
1334 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1335 
1336 	error = VFS_MOUNT(mp, td);
1337 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1338 	if (error)
1339 		return;
1340 
1341 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1342 	TAILQ_INIT(opts);
1343 	mp->mnt_opt = opts;
1344 
1345 	mtx_lock(&mountlist_mtx);
1346 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1347 	mtx_unlock(&mountlist_mtx);
1348 
1349 	set_rootvnode(td);
1350 
1351 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1352 	if (error)
1353 		printf("kern_symlink /dev -> / returns %d\n", error);
1354 }
1355 
1356 /*
1357  * Surgically move our devfs to be mounted on /dev.
1358  */
1359 
1360 static void
1361 devfs_fixup(struct thread *td)
1362 {
1363 	struct nameidata nd;
1364 	int error;
1365 	struct vnode *vp, *dvp;
1366 	struct mount *mp;
1367 
1368 	/* Remove our devfs mount from the mountlist and purge the cache */
1369 	mtx_lock(&mountlist_mtx);
1370 	mp = TAILQ_FIRST(&mountlist);
1371 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1372 	mtx_unlock(&mountlist_mtx);
1373 	cache_purgevfs(mp);
1374 
1375 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1376 	VI_LOCK(dvp);
1377 	dvp->v_iflag &= ~VI_MOUNT;
1378 	dvp->v_mountedhere = NULL;
1379 	VI_UNLOCK(dvp);
1380 
1381 	/* Set up the real rootvnode, and purge the cache */
1382 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1383 	set_rootvnode(td);
1384 	cache_purgevfs(rootvnode->v_mount);
1385 
1386 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1387 	error = namei(&nd);
1388 	if (error) {
1389 		printf("Lookup of /dev for devfs, error: %d\n", error);
1390 		return;
1391 	}
1392 	NDFREE(&nd, NDF_ONLY_PNBUF);
1393 	vp = nd.ni_vp;
1394 	if (vp->v_type != VDIR) {
1395 		vput(vp);
1396 	}
1397 	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1398 	if (error) {
1399 		vput(vp);
1400 	}
1401 	cache_purge(vp);
1402 	mp->mnt_vnodecovered = vp;
1403 	vp->v_mountedhere = mp;
1404 	mtx_lock(&mountlist_mtx);
1405 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1406 	mtx_unlock(&mountlist_mtx);
1407 	VOP_UNLOCK(vp, 0, td);
1408 	vput(dvp);
1409 	vfs_unbusy(mp, td);
1410 
1411 	/* Unlink the no longer needed /dev/dev -> / symlink */
1412 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1413 }
1414 
1415 /*
1416  * Report errors during filesystem mounting.
1417  */
1418 void
1419 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1420 {
1421 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1422 	va_list ap;
1423 	int error, len;
1424 	char *errmsg;
1425 
1426 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1427 	if (error || errmsg == NULL || len <= 0)
1428 		return;
1429 
1430 	va_start(ap, fmt);
1431 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1432 	va_end(ap);
1433 }
1434 
1435 /*
1436  * Find and mount the root filesystem
1437  */
1438 void
1439 vfs_mountroot(void)
1440 {
1441 	char *cp;
1442 	int error, i, asked = 0;
1443 
1444 	root_mount_wait();
1445 
1446 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1447 	    NULL, NULL, mount_init, mount_fini,
1448 	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1449 	devfs_first();
1450 
1451 	/*
1452 	 * We are booted with instructions to prompt for the root filesystem.
1453 	 */
1454 	if (boothowto & RB_ASKNAME) {
1455 		if (!vfs_mountroot_ask())
1456 			return;
1457 		asked = 1;
1458 	}
1459 
1460 	/*
1461 	 * The root filesystem information is compiled in, and we are
1462 	 * booted with instructions to use it.
1463 	 */
1464 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1465 		if (!vfs_mountroot_try(ctrootdevname))
1466 			return;
1467 		ctrootdevname = NULL;
1468 	}
1469 
1470 	/*
1471 	 * We've been given the generic "use CDROM as root" flag.  This is
1472 	 * necessary because one media may be used in many different
1473 	 * devices, so we need to search for them.
1474 	 */
1475 	if (boothowto & RB_CDROM) {
1476 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1477 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1478 				return;
1479 		}
1480 	}
1481 
1482 	/*
1483 	 * Try to use the value read by the loader from /etc/fstab, or
1484 	 * supplied via some other means.  This is the preferred
1485 	 * mechanism.
1486 	 */
1487 	cp = getenv("vfs.root.mountfrom");
1488 	if (cp != NULL) {
1489 		error = vfs_mountroot_try(cp);
1490 		freeenv(cp);
1491 		if (!error)
1492 			return;
1493 	}
1494 
1495 	/*
1496 	 * Try values that may have been computed by code during boot
1497 	 */
1498 	if (!vfs_mountroot_try(rootdevnames[0]))
1499 		return;
1500 	if (!vfs_mountroot_try(rootdevnames[1]))
1501 		return;
1502 
1503 	/*
1504 	 * If we (still) have a compiled-in default, try it.
1505 	 */
1506 	if (ctrootdevname != NULL)
1507 		if (!vfs_mountroot_try(ctrootdevname))
1508 			return;
1509 	/*
1510 	 * Everything so far has failed, prompt on the console if we haven't
1511 	 * already tried that.
1512 	 */
1513 	if (!asked)
1514 		if (!vfs_mountroot_ask())
1515 			return;
1516 
1517 	panic("Root mount failed, startup aborted.");
1518 }
1519 
1520 /*
1521  * Mount (mountfrom) as the root filesystem.
1522  */
1523 static int
1524 vfs_mountroot_try(const char *mountfrom)
1525 {
1526 	struct mount	*mp;
1527 	char		*vfsname, *path;
1528 	time_t		timebase;
1529 	int		error;
1530 	char		patt[32];
1531 
1532 	vfsname = NULL;
1533 	path    = NULL;
1534 	mp      = NULL;
1535 	error   = EINVAL;
1536 
1537 	if (mountfrom == NULL)
1538 		return (error);		/* don't complain */
1539 	printf("Trying to mount root from %s\n", mountfrom);
1540 
1541 	/* parse vfs name and path */
1542 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1543 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1544 	vfsname[0] = path[0] = 0;
1545 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1546 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1547 		goto out;
1548 
1549 	if (path[0] == '\0')
1550 		strcpy(path, ROOTNAME);
1551 
1552 	error = kernel_vmount(
1553 	    MNT_RDONLY | MNT_ROOTFS,
1554 	    "fstype", vfsname,
1555 	    "fspath", "/",
1556 	    "from", path,
1557 	    NULL);
1558 	if (error == 0) {
1559 		/*
1560 		 * We mount devfs prior to mounting the / FS, so the first
1561 		 * entry will typically be devfs.
1562 		 */
1563 		mp = TAILQ_FIRST(&mountlist);
1564 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1565 
1566 		/*
1567 		 * Iterate over all currently mounted file systems and use
1568 		 * the time stamp found to check and/or initialize the RTC.
1569 		 * Typically devfs has no time stamp and the only other FS
1570 		 * is the actual / FS.
1571 		 * Call inittodr() only once and pass it the largest of the
1572 		 * timestamps we encounter.
1573 		 */
1574 		timebase = 0;
1575 		do {
1576 			if (mp->mnt_time > timebase)
1577 				timebase = mp->mnt_time;
1578 			mp = TAILQ_NEXT(mp, mnt_list);
1579 		} while (mp != NULL);
1580 		inittodr(timebase);
1581 
1582 		devfs_fixup(curthread);
1583 	}
1584 out:
1585 	free(path, M_MOUNT);
1586 	free(vfsname, M_MOUNT);
1587 	return (error);
1588 }
1589 
1590 /*
1591  * ---------------------------------------------------------------------
1592  * Interactive root filesystem selection code.
1593  */
1594 
1595 static int
1596 vfs_mountroot_ask(void)
1597 {
1598 	char name[128];
1599 
1600 	for(;;) {
1601 		printf("\nManual root filesystem specification:\n");
1602 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1603 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1604 		printf("                       eg. ufs:da0s1a\n");
1605 #else
1606 		printf("                       eg. ufs:/dev/da0a\n");
1607 #endif
1608 		printf("  ?                  List valid disk boot devices\n");
1609 		printf("  <empty line>       Abort manual input\n");
1610 		printf("\nmountroot> ");
1611 		gets(name, sizeof(name), 1);
1612 		if (name[0] == '\0')
1613 			return (1);
1614 		if (name[0] == '?') {
1615 			printf("\nList of GEOM managed disk devices:\n  ");
1616 			g_dev_print();
1617 			continue;
1618 		}
1619 		if (!vfs_mountroot_try(name))
1620 			return (0);
1621 	}
1622 }
1623 
1624 /*
1625  * ---------------------------------------------------------------------
1626  * Functions for querying mount options/arguments from filesystems.
1627  */
1628 
1629 /*
1630  * Check that no unknown options are given
1631  */
1632 int
1633 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1634 {
1635 	struct vfsopt *opt;
1636 	const char **t, *p;
1637 
1638 
1639 	TAILQ_FOREACH(opt, opts, link) {
1640 		p = opt->name;
1641 		if (p[0] == 'n' && p[1] == 'o')
1642 			p += 2;
1643 		for(t = global_opts; *t != NULL; t++)
1644 			if (!strcmp(*t, p))
1645 				break;
1646 		if (*t != NULL)
1647 			continue;
1648 		for(t = legal; *t != NULL; t++)
1649 			if (!strcmp(*t, p))
1650 				break;
1651 		if (*t != NULL)
1652 			continue;
1653 		printf("mount option <%s> is unknown\n", p);
1654 		return (EINVAL);
1655 	}
1656 	return (0);
1657 }
1658 
1659 /*
1660  * Get a mount option by its name.
1661  *
1662  * Return 0 if the option was found, ENOENT otherwise.
1663  * If len is non-NULL it will be filled with the length
1664  * of the option. If buf is non-NULL, it will be filled
1665  * with the address of the option.
1666  */
1667 int
1668 vfs_getopt(opts, name, buf, len)
1669 	struct vfsoptlist *opts;
1670 	const char *name;
1671 	void **buf;
1672 	int *len;
1673 {
1674 	struct vfsopt *opt;
1675 
1676 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1677 
1678 	TAILQ_FOREACH(opt, opts, link) {
1679 		if (strcmp(name, opt->name) == 0) {
1680 			if (len != NULL)
1681 				*len = opt->len;
1682 			if (buf != NULL)
1683 				*buf = opt->value;
1684 			return (0);
1685 		}
1686 	}
1687 	return (ENOENT);
1688 }
1689 
1690 static int
1691 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1692 {
1693 	struct vfsopt *opt;
1694 	int i;
1695 
1696 	if (opts == NULL)
1697 		return (-1);
1698 
1699 	i = 0;
1700 	TAILQ_FOREACH(opt, opts, link) {
1701 		if (strcmp(name, opt->name) == 0)
1702 			return (i);
1703 		++i;
1704 	}
1705 	return (-1);
1706 }
1707 
1708 char *
1709 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1710 {
1711 	struct vfsopt *opt;
1712 
1713 	*error = 0;
1714 	TAILQ_FOREACH(opt, opts, link) {
1715 		if (strcmp(name, opt->name) != 0)
1716 			continue;
1717 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1718 			*error = EINVAL;
1719 			return (NULL);
1720 		}
1721 		return (opt->value);
1722 	}
1723 	return (NULL);
1724 }
1725 
1726 int
1727 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1728 {
1729 	struct vfsopt *opt;
1730 
1731 	TAILQ_FOREACH(opt, opts, link) {
1732 		if (strcmp(name, opt->name) == 0) {
1733 			if (w != NULL)
1734 				*w |= val;
1735 			return (1);
1736 		}
1737 	}
1738 	if (w != NULL)
1739 		*w &= ~val;
1740 	return (0);
1741 }
1742 
1743 int
1744 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1745 {
1746 	va_list ap;
1747 	struct vfsopt *opt;
1748 	int ret;
1749 
1750 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1751 
1752 	TAILQ_FOREACH(opt, opts, link) {
1753 		if (strcmp(name, opt->name) != 0)
1754 			continue;
1755 		if (((char *)opt->value)[opt->len - 1] != '\0')
1756 			return (0);
1757 		va_start(ap, fmt);
1758 		ret = vsscanf(opt->value, fmt, ap);
1759 		va_end(ap);
1760 		return (ret);
1761 	}
1762 	return (0);
1763 }
1764 
1765 /*
1766  * Find and copy a mount option.
1767  *
1768  * The size of the buffer has to be specified
1769  * in len, if it is not the same length as the
1770  * mount option, EINVAL is returned.
1771  * Returns ENOENT if the option is not found.
1772  */
1773 int
1774 vfs_copyopt(opts, name, dest, len)
1775 	struct vfsoptlist *opts;
1776 	const char *name;
1777 	void *dest;
1778 	int len;
1779 {
1780 	struct vfsopt *opt;
1781 
1782 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1783 
1784 	TAILQ_FOREACH(opt, opts, link) {
1785 		if (strcmp(name, opt->name) == 0) {
1786 			if (len != opt->len)
1787 				return (EINVAL);
1788 			bcopy(opt->value, dest, opt->len);
1789 			return (0);
1790 		}
1791 	}
1792 	return (ENOENT);
1793 }
1794 
1795 /*
1796  * This is a helper function for filesystems to traverse their
1797  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1798  */
1799 
1800 struct vnode *
1801 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1802 {
1803 	struct vnode *vp;
1804 
1805 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1806 
1807 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1808 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1809 	while (vp != NULL && vp->v_type == VMARKER)
1810 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1811 
1812 	/* Check if we are done */
1813 	if (vp == NULL) {
1814 		__mnt_vnode_markerfree(mvp, mp);
1815 		return (NULL);
1816 	}
1817 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1818 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1819 	return (vp);
1820 }
1821 
1822 struct vnode *
1823 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
1824 {
1825 	struct vnode *vp;
1826 
1827 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1828 
1829 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1830 	while (vp != NULL && vp->v_type == VMARKER)
1831 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1832 
1833 	/* Check if we are done */
1834 	if (vp == NULL) {
1835 		*mvp = NULL;
1836 		return (NULL);
1837 	}
1838 	mp->mnt_holdcnt++;
1839 	MNT_IUNLOCK(mp);
1840 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
1841 				       M_VNODE_MARKER,
1842 				       M_WAITOK | M_ZERO);
1843 	MNT_ILOCK(mp);
1844 	(*mvp)->v_type = VMARKER;
1845 
1846 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1847 	while (vp != NULL && vp->v_type == VMARKER)
1848 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1849 
1850 	/* Check if we are done */
1851 	if (vp == NULL) {
1852 		MNT_IUNLOCK(mp);
1853 		free(*mvp, M_VNODE_MARKER);
1854 		MNT_ILOCK(mp);
1855 		*mvp = NULL;
1856 		mp->mnt_holdcnt--;
1857 		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1858 			wakeup(&mp->mnt_holdcnt);
1859 		return (NULL);
1860 	}
1861 	mp->mnt_markercnt++;
1862 	(*mvp)->v_mount = mp;
1863 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1864 	return (vp);
1865 }
1866 
1867 
1868 void
1869 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
1870 {
1871 
1872 	if (*mvp == NULL)
1873 		return;
1874 
1875 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1876 
1877 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1878 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1879 	MNT_IUNLOCK(mp);
1880 	free(*mvp, M_VNODE_MARKER);
1881 	MNT_ILOCK(mp);
1882 	*mvp = NULL;
1883 
1884 	mp->mnt_markercnt--;
1885 	mp->mnt_holdcnt--;
1886 	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
1887 		wakeup(&mp->mnt_holdcnt);
1888 }
1889 
1890 
1891 int
1892 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
1893 {
1894 	int error;
1895 
1896 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
1897 	if (sbp != &mp->mnt_stat)
1898 		*sbp = mp->mnt_stat;
1899 	return (error);
1900 }
1901 
1902 void
1903 vfs_mountedfrom(struct mount *mp, const char *from)
1904 {
1905 
1906 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1907 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1908 	    sizeof mp->mnt_stat.f_mntfromname);
1909 }
1910 
1911 /*
1912  * ---------------------------------------------------------------------
1913  * This is the api for building mount args and mounting filesystems from
1914  * inside the kernel.
1915  *
1916  * The API works by accumulation of individual args.  First error is
1917  * latched.
1918  *
1919  * XXX: should be documented in new manpage kernel_mount(9)
1920  */
1921 
1922 /* A memory allocation which must be freed when we are done */
1923 struct mntaarg {
1924 	SLIST_ENTRY(mntaarg)	next;
1925 };
1926 
1927 /* The header for the mount arguments */
1928 struct mntarg {
1929 	struct iovec *v;
1930 	int len;
1931 	int error;
1932 	SLIST_HEAD(, mntaarg)	list;
1933 };
1934 
1935 /*
1936  * Add a boolean argument.
1937  *
1938  * flag is the boolean value.
1939  * name must start with "no".
1940  */
1941 struct mntarg *
1942 mount_argb(struct mntarg *ma, int flag, const char *name)
1943 {
1944 
1945 	KASSERT(name[0] == 'n' && name[1] == 'o',
1946 	    ("mount_argb(...,%s): name must start with 'no'", name));
1947 
1948 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1949 }
1950 
1951 /*
1952  * Add an argument printf style
1953  */
1954 struct mntarg *
1955 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1956 {
1957 	va_list ap;
1958 	struct mntaarg *maa;
1959 	struct sbuf *sb;
1960 	int len;
1961 
1962 	if (ma == NULL) {
1963 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1964 		SLIST_INIT(&ma->list);
1965 	}
1966 	if (ma->error)
1967 		return (ma);
1968 
1969 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1970 	    M_MOUNT, M_WAITOK);
1971 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1972 	ma->v[ma->len].iov_len = strlen(name) + 1;
1973 	ma->len++;
1974 
1975 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
1976 	va_start(ap, fmt);
1977 	sbuf_vprintf(sb, fmt, ap);
1978 	va_end(ap);
1979 	sbuf_finish(sb);
1980 	len = sbuf_len(sb) + 1;
1981 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1982 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1983 	bcopy(sbuf_data(sb), maa + 1, len);
1984 	sbuf_delete(sb);
1985 
1986 	ma->v[ma->len].iov_base = maa + 1;
1987 	ma->v[ma->len].iov_len = len;
1988 	ma->len++;
1989 
1990 	return (ma);
1991 }
1992 
1993 /*
1994  * Add an argument which is a userland string.
1995  */
1996 struct mntarg *
1997 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1998 {
1999 	struct mntaarg *maa;
2000 	char *tbuf;
2001 
2002 	if (val == NULL)
2003 		return (ma);
2004 	if (ma == NULL) {
2005 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2006 		SLIST_INIT(&ma->list);
2007 	}
2008 	if (ma->error)
2009 		return (ma);
2010 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2011 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2012 	tbuf = (void *)(maa + 1);
2013 	ma->error = copyinstr(val, tbuf, len, NULL);
2014 	return (mount_arg(ma, name, tbuf, -1));
2015 }
2016 
2017 /*
2018  * Plain argument.
2019  *
2020  * If length is -1, use printf.
2021  */
2022 struct mntarg *
2023 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2024 {
2025 
2026 	if (ma == NULL) {
2027 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2028 		SLIST_INIT(&ma->list);
2029 	}
2030 	if (ma->error)
2031 		return (ma);
2032 
2033 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2034 	    M_MOUNT, M_WAITOK);
2035 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2036 	ma->v[ma->len].iov_len = strlen(name) + 1;
2037 	ma->len++;
2038 
2039 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2040 	if (len < 0)
2041 		ma->v[ma->len].iov_len = strlen(val) + 1;
2042 	else
2043 		ma->v[ma->len].iov_len = len;
2044 	ma->len++;
2045 	return (ma);
2046 }
2047 
2048 /*
2049  * Free a mntarg structure
2050  */
2051 static void
2052 free_mntarg(struct mntarg *ma)
2053 {
2054 	struct mntaarg *maa;
2055 
2056 	while (!SLIST_EMPTY(&ma->list)) {
2057 		maa = SLIST_FIRST(&ma->list);
2058 		SLIST_REMOVE_HEAD(&ma->list, next);
2059 		free(maa, M_MOUNT);
2060 	}
2061 	free(ma->v, M_MOUNT);
2062 	free(ma, M_MOUNT);
2063 }
2064 
2065 /*
2066  * Mount a filesystem
2067  */
2068 int
2069 kernel_mount(struct mntarg *ma, int flags)
2070 {
2071 	struct uio auio;
2072 	int error;
2073 
2074 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2075 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2076 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2077 
2078 	auio.uio_iov = ma->v;
2079 	auio.uio_iovcnt = ma->len;
2080 	auio.uio_segflg = UIO_SYSSPACE;
2081 
2082 	error = ma->error;
2083 	if (!error)
2084 		error = vfs_donmount(curthread, flags, &auio);
2085 	free_mntarg(ma);
2086 	return (error);
2087 }
2088 
2089 /*
2090  * A printflike function to mount a filesystem.
2091  */
2092 int
2093 kernel_vmount(int flags, ...)
2094 {
2095 	struct mntarg *ma = NULL;
2096 	va_list ap;
2097 	const char *cp;
2098 	const void *vp;
2099 	int error;
2100 
2101 	va_start(ap, flags);
2102 	for (;;) {
2103 		cp = va_arg(ap, const char *);
2104 		if (cp == NULL)
2105 			break;
2106 		vp = va_arg(ap, const void *);
2107 		ma = mount_arg(ma, cp, vp, -1);
2108 	}
2109 	va_end(ap);
2110 
2111 	error = kernel_mount(ma, flags);
2112 	return (error);
2113 }
2114