xref: /freebsd/sys/kern/vfs_mount.c (revision 52ec752989b2e6d4e9a59a8ff25d8ff596d85e62)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * Copyright (c) 1999 Michael Smith
39  * All rights reserved.
40  * Copyright (c) 1999 Poul-Henning Kamp
41  * All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  */
64 
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
67 
68 #include <sys/param.h>
69 #include <sys/conf.h>
70 #include <sys/cons.h>
71 #include <sys/kernel.h>
72 #include <sys/linker.h>
73 #include <sys/mac.h>
74 #include <sys/malloc.h>
75 #include <sys/mount.h>
76 #include <sys/mutex.h>
77 #include <sys/namei.h>
78 #include <sys/proc.h>
79 #include <sys/filedesc.h>
80 #include <sys/reboot.h>
81 #include <sys/sysproto.h>
82 #include <sys/sx.h>
83 #include <sys/sysctl.h>
84 #include <sys/sysent.h>
85 #include <sys/systm.h>
86 #include <sys/vnode.h>
87 
88 #include <geom/geom.h>
89 
90 #include <machine/stdarg.h>
91 
92 #include "opt_rootdevname.h"
93 #include "opt_ddb.h"
94 #include "opt_mac.h"
95 
96 #ifdef DDB
97 #include <ddb/ddb.h>
98 #endif
99 
100 #define	ROOTNAME		"root_device"
101 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
102 
103 static void	checkdirs(struct vnode *olddp, struct vnode *newdp);
104 static void	gets(char *cp);
105 static int	vfs_domount(struct thread *td, const char *fstype,
106 		    char *fspath, int fsflags, void *fsdata, int compat);
107 static int	vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
108 		    const char *fspath, struct thread *td, struct mount **mpp);
109 static int	vfs_mountroot_ask(void);
110 static int	vfs_mountroot_try(char *mountfrom);
111 static int	vfs_nmount(struct thread *td, int fsflags, struct uio *fsopts);
112 
113 static int	usermount = 0;	/* if 1, non-root can mount fs. */
114 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
115 
116 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
117 
118 /* List of mounted filesystems. */
119 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
120 
121 /* For any iteration/modification of mountlist */
122 struct mtx mountlist_mtx;
123 
124 /*
125  * The vnode of the system's root (/ in the filesystem, without chroot
126  * active.)
127  */
128 struct vnode	*rootvnode;
129 
130 /*
131  * The root filesystem is detailed in the kernel environment variable
132  * vfs.root.mountfrom, which is expected to be in the general format
133  *
134  * <vfsname>:[<path>]
135  * vfsname   := the name of a VFS known to the kernel and capable
136  *              of being mounted as root
137  * path      := disk device name or other data used by the filesystem
138  *              to locate its physical store
139  */
140 
141 /*
142  * The root specifiers we will try if RB_CDROM is specified.
143  */
144 static char *cdrom_rootdevnames[] = {
145 	"cd9660:cd0",
146 	"cd9660:acd0",
147 	NULL
148 };
149 
150 /* legacy find-root code */
151 char		*rootdevnames[2] = {NULL, NULL};
152 static int	setrootbyname(char *name);
153 dev_t		rootdev = NODEV;
154 
155 /*
156  * Has to be dynamic as the value of rootdev can change; however, it can't
157  * change after the root is mounted, so a user process can't access this
158  * sysctl until after the value is unchangeable.
159  */
160 static int
161 sysctl_rootdev(SYSCTL_HANDLER_ARGS)
162 {
163 	int error;
164 
165 	/* _RD prevents this from happening. */
166 	KASSERT(req->newptr == NULL, ("Attempt to change root device name"));
167 
168 	if (rootdev != NODEV)
169 		error = sysctl_handle_string(oidp, rootdev->si_name, 0, req);
170 	else
171 		error = sysctl_handle_string(oidp, "", 0, req);
172 
173 	return (error);
174 }
175 
176 SYSCTL_PROC(_kern, OID_AUTO, rootdev, CTLTYPE_STRING | CTLFLAG_RD,
177     0, 0, sysctl_rootdev, "A", "Root file system device");
178 
179 /* Remove one mount option. */
180 static void
181 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
182 {
183 
184 	TAILQ_REMOVE(opts, opt, link);
185 	free(opt->name, M_MOUNT);
186 	if (opt->value != NULL)
187 		free(opt->value, M_MOUNT);
188 #ifdef INVARIANTS
189 	else if (opt->len != 0)
190 		panic("%s: mount option with NULL value but length != 0",
191 		    __func__);
192 #endif
193 	free(opt, M_MOUNT);
194 }
195 
196 /* Release all resources related to the mount options. */
197 static void
198 vfs_freeopts(struct vfsoptlist *opts)
199 {
200 	struct vfsopt *opt;
201 
202 	while (!TAILQ_EMPTY(opts)) {
203 		opt = TAILQ_FIRST(opts);
204 		vfs_freeopt(opts, opt);
205 	}
206 	free(opts, M_MOUNT);
207 }
208 
209 /*
210  * If a mount option is specified several times,
211  * (with or without the "no" prefix) only keep
212  * the last occurence of it.
213  */
214 static void
215 vfs_sanitizeopts(struct vfsoptlist *opts)
216 {
217 	struct vfsopt *opt, *opt2, *tmp;
218 	int noopt;
219 
220 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
221 		if (strncmp(opt->name, "no", 2) == 0)
222 			noopt = 1;
223 		else
224 			noopt = 0;
225 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
226 		while (opt2 != NULL) {
227 			if (strcmp(opt2->name, opt->name) == 0 ||
228 			    (noopt && strcmp(opt->name + 2, opt2->name) == 0) ||
229 			    (!noopt && strncmp(opt2->name, "no", 2) == 0 &&
230 			    strcmp(opt2->name + 2, opt->name) == 0)) {
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 = optlen;
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->value = malloc(optlen, M_MOUNT, M_WAITOK);
298 			if (auio->uio_segflg == UIO_SYSSPACE) {
299 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
300 				    optlen);
301 			} else {
302 				error = copyin(auio->uio_iov[i + 1].iov_base,
303 				    opt->value, optlen);
304 				if (error)
305 					goto bad;
306 			}
307 		}
308 	}
309 	vfs_sanitizeopts(opts);
310 	*options = opts;
311 	return (0);
312 bad:
313 	vfs_freeopts(opts);
314 	return (error);
315 }
316 
317 /*
318  * Merge the old mount options with the new ones passed
319  * in the MNT_UPDATE case.
320  */
321 static void
322 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
323 {
324 	struct vfsopt *opt, *opt2, *new;
325 
326 	TAILQ_FOREACH(opt, opts, link) {
327 		/*
328 		 * Check that this option hasn't been redefined
329 		 * nor cancelled with a "no" mount option.
330 		 */
331 		opt2 = TAILQ_FIRST(toopts);
332 		while (opt2 != NULL) {
333 			if (strcmp(opt2->name, opt->name) == 0)
334 				goto next;
335 			if (strncmp(opt2->name, "no", 2) == 0 &&
336 			    strcmp(opt2->name + 2, opt->name) == 0) {
337 				vfs_freeopt(toopts, opt2);
338 				goto next;
339 			}
340 			opt2 = TAILQ_NEXT(opt2, link);
341 		}
342 		/* We want this option, duplicate it. */
343 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
344 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
345 		strcpy(new->name, opt->name);
346 		if (opt->len != 0) {
347 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
348 			bcopy(opt->value, new->value, opt->len);
349 		} else {
350 			new->value = NULL;
351 		}
352 		new->len = opt->len;
353 		TAILQ_INSERT_TAIL(toopts, new, link);
354 next:
355 		continue;
356 	}
357 }
358 
359 /*
360  * New mount API.
361  */
362 int
363 nmount(td, uap)
364 	struct thread *td;
365 	struct nmount_args /* {
366 		struct iovec *iovp;
367 		unsigned int iovcnt;
368 		int flags;
369 	} */ *uap;
370 {
371 	struct uio auio;
372 	struct iovec *iov, *needfree;
373 	struct iovec aiov[UIO_SMALLIOV];
374 	unsigned int i;
375 	int error;
376 	u_int iovlen, iovcnt;
377 
378 	iovcnt = uap->iovcnt;
379 	iovlen = iovcnt * sizeof (struct iovec);
380 	/*
381 	 * Check that we have an even number of iovec's
382 	 * and that we have at least two options.
383 	 */
384 	if ((iovcnt & 1) || (iovcnt < 4) || (iovcnt > UIO_MAXIOV))
385 		return (EINVAL);
386 
387 	if (iovcnt > UIO_SMALLIOV) {
388 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
389 		needfree = iov;
390 	} else {
391 		iov = aiov;
392 		needfree = NULL;
393 	}
394 	auio.uio_iov = iov;
395 	auio.uio_iovcnt = iovcnt;
396 	auio.uio_segflg = UIO_USERSPACE;
397 	if ((error = copyin(uap->iovp, iov, iovlen)))
398 		goto finish;
399 
400 	for (i = 0; i < iovcnt; i++) {
401 		if (iov->iov_len > MMAXOPTIONLEN) {
402 			error = EINVAL;
403 			goto finish;
404 		}
405 		iov++;
406 	}
407 	error = vfs_nmount(td, uap->flags, &auio);
408 finish:
409 	if (needfree != NULL)
410 		free(needfree, M_TEMP);
411 	return (error);
412 }
413 
414 int
415 kernel_mount(iovp, iovcnt, flags)
416 	struct iovec *iovp;
417 	unsigned int iovcnt;
418 	int flags;
419 {
420 	struct uio auio;
421 	int error;
422 
423 	/*
424 	 * Check that we have an even number of iovec's
425 	 * and that we have at least two options.
426 	 */
427 	if ((iovcnt & 1) || (iovcnt < 4))
428 		return (EINVAL);
429 
430 	auio.uio_iov = iovp;
431 	auio.uio_iovcnt = iovcnt;
432 	auio.uio_segflg = UIO_SYSSPACE;
433 
434 	error = vfs_nmount(curthread, flags, &auio);
435 	return (error);
436 }
437 
438 int
439 kernel_vmount(int flags, ...)
440 {
441 	struct iovec *iovp;
442 	struct uio auio;
443 	va_list ap;
444 	unsigned int iovcnt, iovlen, len;
445 	const char *cp;
446 	char *buf, *pos;
447 	size_t n;
448 	int error, i;
449 
450 	len = 0;
451 	va_start(ap, flags);
452 	for (iovcnt = 0; (cp = va_arg(ap, const char *)) != NULL; iovcnt++)
453 		len += strlen(cp) + 1;
454 	va_end(ap);
455 
456 	if (iovcnt < 4 || iovcnt & 1)
457 		return (EINVAL);
458 
459 	iovlen = iovcnt * sizeof (struct iovec);
460 	MALLOC(iovp, struct iovec *, iovlen, M_MOUNT, M_WAITOK);
461 	MALLOC(buf, char *, len, M_MOUNT, M_WAITOK);
462 	pos = buf;
463 	va_start(ap, flags);
464 	for (i = 0; i < iovcnt; i++) {
465 		cp = va_arg(ap, const char *);
466 		copystr(cp, pos, len - (pos - buf), &n);
467 		iovp[i].iov_base = pos;
468 		iovp[i].iov_len = n;
469 		pos += n;
470 	}
471 	va_end(ap);
472 
473 	auio.uio_iov = iovp;
474 	auio.uio_iovcnt = iovcnt;
475 	auio.uio_segflg = UIO_SYSSPACE;
476 
477 	error = vfs_nmount(curthread, flags, &auio);
478 	FREE(iovp, M_MOUNT);
479 	FREE(buf, M_MOUNT);
480 	return (error);
481 }
482 
483 /*
484  * Allocate and initialize the mount point struct.
485  */
486 static int
487 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
488     const char *fspath, struct thread *td, struct mount **mpp)
489 {
490 	struct mount *mp;
491 
492 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
493 	TAILQ_INIT(&mp->mnt_nvnodelist);
494 	TAILQ_INIT(&mp->mnt_reservedvnlist);
495 	mp->mnt_nvnodelistsize = 0;
496 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
497 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
498 	vfs_busy(mp, LK_NOWAIT, 0, td);
499 	mp->mnt_op = vfsp->vfc_vfsops;
500 	mp->mnt_vfc = vfsp;
501 	vfsp->vfc_refcount++;
502 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
503 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
504 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
505 	mp->mnt_vnodecovered = vp;
506 	mp->mnt_cred = crdup(td->td_ucred);
507 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
508 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
509 	mp->mnt_iosize_max = DFLTPHYS;
510 #ifdef MAC
511 	mac_init_mount(mp);
512 	mac_create_mount(td->td_ucred, mp);
513 #endif
514 	*mpp = mp;
515 	return (0);
516 }
517 
518 /*
519  * Destroy the mount struct previously allocated by vfs_mount_alloc().
520  */
521 void
522 vfs_mount_destroy(struct mount *mp, struct thread *td)
523 {
524 
525 	mp->mnt_vfc->vfc_refcount--;
526 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
527 		panic("unmount: dangling vnode");
528 	vfs_unbusy(mp,td);
529 	lockdestroy(&mp->mnt_lock);
530 	mtx_destroy(&mp->mnt_mtx);
531 	if (mp->mnt_kern_flag & MNTK_MWAIT)
532 		wakeup(mp);
533 #ifdef MAC
534 	mac_destroy_mount(mp);
535 #endif
536 	if (mp->mnt_opt != NULL)
537 		vfs_freeopts(mp->mnt_opt);
538 	crfree(mp->mnt_cred);
539 	free(mp, M_MOUNT);
540 }
541 
542 static int
543 vfs_nmount(struct thread *td, int fsflags, struct uio *fsoptions)
544 {
545 	struct vfsoptlist *optlist;
546 	char *fstype, *fspath;
547 	int error, fstypelen, fspathlen;
548 
549 	error = vfs_buildopts(fsoptions, &optlist);
550 	if (error)
551 		return (error);
552 
553 	/*
554 	 * We need these two options before the others,
555 	 * and they are mandatory for any filesystem.
556 	 * Ensure they are NUL terminated as well.
557 	 */
558 	fstypelen = 0;
559 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
560 	if (error || fstype[fstypelen - 1] != '\0') {
561 		error = EINVAL;
562 		goto bail;
563 	}
564 	fspathlen = 0;
565 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
566 	if (error || fspath[fspathlen - 1] != '\0') {
567 		error = EINVAL;
568 		goto bail;
569 	}
570 
571 	/*
572 	 * Be ultra-paranoid about making sure the type and fspath
573 	 * variables will fit in our mp buffers, including the
574 	 * terminating NUL.
575 	 */
576 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
577 		error = ENAMETOOLONG;
578 		goto bail;
579 	}
580 
581 	error = vfs_domount(td, fstype, fspath, fsflags, optlist, 0);
582 bail:
583 	if (error)
584 		vfs_freeopts(optlist);
585 	return (error);
586 }
587 
588 /*
589  * Old mount API.
590  */
591 #ifndef _SYS_SYSPROTO_H_
592 struct mount_args {
593 	char	*type;
594 	char	*path;
595 	int	flags;
596 	caddr_t	data;
597 };
598 #endif
599 /* ARGSUSED */
600 int
601 mount(td, uap)
602 	struct thread *td;
603 	struct mount_args /* {
604 		char *type;
605 		char *path;
606 		int flags;
607 		caddr_t data;
608 	} */ *uap;
609 {
610 	char *fstype;
611 	char *fspath;
612 	int error;
613 
614 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
615 	fspath = malloc(MNAMELEN, M_TEMP, M_WAITOK);
616 
617 	/*
618 	 * vfs_mount() actually takes a kernel string for `type' and
619 	 * `path' now, so extract them.
620 	 */
621 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
622 	if (error == 0)
623 		error = copyinstr(uap->path, fspath, MNAMELEN, NULL);
624 	if (error == 0)
625 		error = vfs_domount(td, fstype, fspath, uap->flags,
626 		    uap->data, 1);
627 	free(fstype, M_TEMP);
628 	free(fspath, M_TEMP);
629 	return (error);
630 }
631 
632 /*
633  * vfs_mount(): actually attempt a filesystem mount.
634  *
635  * This routine is designed to be a "generic" entry point for routines
636  * that wish to mount a filesystem. All parameters except `fsdata' are
637  * pointers into kernel space. `fsdata' is currently still a pointer
638  * into userspace.
639  */
640 int
641 vfs_mount(td, fstype, fspath, fsflags, fsdata)
642 	struct thread *td;
643 	const char *fstype;
644 	char *fspath;
645 	int fsflags;
646 	void *fsdata;
647 {
648 
649 	return (vfs_domount(td, fstype, fspath, fsflags, fsdata, 1));
650 }
651 
652 /*
653  * vfs_domount(): actually attempt a filesystem mount.
654  */
655 static int
656 vfs_domount(
657 	struct thread *td,	/* Flags common to all filesystems. */
658 	const char *fstype,	/* Filesystem type. */
659 	char *fspath,		/* Mount path. */
660 	int fsflags,		/* Flags common to all filesystems. */
661 	void *fsdata,		/* Options local to the filesystem. */
662 	int compat		/* Invocation from compat syscall. */
663 	)
664 {
665 	linker_file_t lf;
666 	struct vnode *vp;
667 	struct mount *mp;
668 	struct vfsconf *vfsp;
669 	int error, flag = 0, kern_flag = 0;
670 	struct vattr va;
671 	struct nameidata nd;
672 
673 	/*
674 	 * Be ultra-paranoid about making sure the type and fspath
675 	 * variables will fit in our mp buffers, including the
676 	 * terminating NUL.
677 	 */
678 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
679 		return (ENAMETOOLONG);
680 
681 	if (usermount == 0) {
682 		error = suser(td);
683 		if (error)
684 			return (error);
685 	}
686 	/*
687 	 * Do not allow NFS export by non-root users.
688 	 */
689 	if (fsflags & MNT_EXPORTED) {
690 		error = suser(td);
691 		if (error)
692 			return (error);
693 	}
694 	/*
695 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users.
696 	 */
697 	if (suser(td))
698 		fsflags |= MNT_NOSUID | MNT_NODEV;
699 	/*
700 	 * Get vnode to be covered
701 	 */
702 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
703 	if ((error = namei(&nd)) != 0)
704 		return (error);
705 	NDFREE(&nd, NDF_ONLY_PNBUF);
706 	vp = nd.ni_vp;
707 	if (fsflags & MNT_UPDATE) {
708 		if ((vp->v_vflag & VV_ROOT) == 0) {
709 			vput(vp);
710 			return (EINVAL);
711 		}
712 		mp = vp->v_mount;
713 		flag = mp->mnt_flag;
714 		kern_flag = mp->mnt_kern_flag;
715 		/*
716 		 * We only allow the filesystem to be reloaded if it
717 		 * is currently mounted read-only.
718 		 */
719 		if ((fsflags & MNT_RELOAD) &&
720 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
721 			vput(vp);
722 			return (EOPNOTSUPP);	/* Needs translation */
723 		}
724 		/*
725 		 * Only root, or the user that did the original mount is
726 		 * permitted to update it.
727 		 */
728 		if (mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
729 			error = suser(td);
730 			if (error) {
731 				vput(vp);
732 				return (error);
733 			}
734 		}
735 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
736 			vput(vp);
737 			return (EBUSY);
738 		}
739 		VI_LOCK(vp);
740 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
741 		    vp->v_mountedhere != NULL) {
742 			VI_UNLOCK(vp);
743 			vfs_unbusy(mp, td);
744 			vput(vp);
745 			return (EBUSY);
746 		}
747 		vp->v_iflag |= VI_MOUNT;
748 		VI_UNLOCK(vp);
749 		mp->mnt_flag |= fsflags &
750 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT);
751 		VOP_UNLOCK(vp, 0, td);
752 		if (compat == 0) {
753 			mp->mnt_optnew = fsdata;
754 			vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
755 		}
756 		goto update;
757 	}
758 	/*
759 	 * If the user is not root, ensure that they own the directory
760 	 * onto which we are attempting to mount.
761 	 */
762 	error = VOP_GETATTR(vp, &va, td->td_ucred, td);
763 	if (error) {
764 		vput(vp);
765 		return (error);
766 	}
767 	if (va.va_uid != td->td_ucred->cr_uid) {
768 		error = suser(td);
769 		if (error) {
770 			vput(vp);
771 			return (error);
772 		}
773 	}
774 	if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) {
775 		vput(vp);
776 		return (error);
777 	}
778 	if (vp->v_type != VDIR) {
779 		vput(vp);
780 		return (ENOTDIR);
781 	}
782 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
783 		if (!strcmp(vfsp->vfc_name, fstype))
784 			break;
785 	if (vfsp == NULL) {
786 		/* Only load modules for root (very important!). */
787 		error = suser(td);
788 		if (error) {
789 			vput(vp);
790 			return (error);
791 		}
792 		error = securelevel_gt(td->td_ucred, 0);
793 		if (error) {
794 			vput(vp);
795 			return (error);
796 		}
797 		error = linker_load_module(NULL, fstype, NULL, NULL, &lf);
798 		if (error || lf == NULL) {
799 			vput(vp);
800 			if (lf == NULL)
801 				error = ENODEV;
802 			return (error);
803 		}
804 		lf->userrefs++;
805 		/* Look up again to see if the VFS was loaded. */
806 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
807 			if (!strcmp(vfsp->vfc_name, fstype))
808 				break;
809 		if (vfsp == NULL) {
810 			lf->userrefs--;
811 			linker_file_unload(lf);
812 			vput(vp);
813 			return (ENODEV);
814 		}
815 	}
816 	VI_LOCK(vp);
817 	if ((vp->v_iflag & VI_MOUNT) != 0 ||
818 	    vp->v_mountedhere != NULL) {
819 		VI_UNLOCK(vp);
820 		vput(vp);
821 		return (EBUSY);
822 	}
823 	vp->v_iflag |= VI_MOUNT;
824 	VI_UNLOCK(vp);
825 
826 	/*
827 	 * Allocate and initialize the filesystem.
828 	 */
829 	error = vfs_mount_alloc(vp, vfsp, fspath, td, &mp);
830 	if (error) {
831 		vput(vp);
832 		return (error);
833 	}
834 	VOP_UNLOCK(vp, 0, td);
835 
836 	/* XXXMAC: pass to vfs_mount_alloc? */
837 	if (compat == 0)
838 		mp->mnt_optnew = fsdata;
839 update:
840 	/*
841 	 * Check if the fs implements the type VFS_[N]MOUNT()
842 	 * function we are looking for.
843 	 */
844 	if ((compat == 0) == (mp->mnt_op->vfs_mount != NULL)) {
845 		printf("%s doesn't support the %s mount syscall\n",
846 		    mp->mnt_vfc->vfc_name, compat? "old" : "new");
847 		VI_LOCK(vp);
848 		vp->v_iflag &= ~VI_MOUNT;
849 		VI_UNLOCK(vp);
850 		if (mp->mnt_flag & MNT_UPDATE)
851 			vfs_unbusy(mp, td);
852 		else
853 			vfs_mount_destroy(mp, td);
854 		vrele(vp);
855 		return (EOPNOTSUPP);
856 	}
857 
858 	/*
859 	 * Set the mount level flags.
860 	 */
861 	if (fsflags & MNT_RDONLY)
862 		mp->mnt_flag |= MNT_RDONLY;
863 	else if (mp->mnt_flag & MNT_RDONLY)
864 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
865 	mp->mnt_flag &=~ MNT_UPDATEMASK;
866 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE);
867 	/*
868 	 * Mount the filesystem.
869 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
870 	 * get.  No freeing of cn_pnbuf.
871 	 */
872 	error = compat? VFS_MOUNT(mp, fspath, fsdata, &nd, td) :
873 	    VFS_NMOUNT(mp, &nd, td);
874 	if (!error) {
875 		if (mp->mnt_opt != NULL)
876 			vfs_freeopts(mp->mnt_opt);
877 		mp->mnt_opt = mp->mnt_optnew;
878 	}
879 	/*
880 	 * Prevent external consumers of mount options from reading
881 	 * mnt_optnew.
882 	*/
883 	mp->mnt_optnew = NULL;
884 	if (mp->mnt_flag & MNT_UPDATE) {
885 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
886 			mp->mnt_flag &= ~MNT_RDONLY;
887 		mp->mnt_flag &=~
888 		    (MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
889 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
890 		if (error) {
891 			mp->mnt_flag = flag;
892 			mp->mnt_kern_flag = kern_flag;
893 		}
894 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
895 			if (mp->mnt_syncer == NULL)
896 				error = vfs_allocate_syncvnode(mp);
897 		} else {
898 			if (mp->mnt_syncer != NULL)
899 				vrele(mp->mnt_syncer);
900 			mp->mnt_syncer = NULL;
901 		}
902 		vfs_unbusy(mp, td);
903 		VI_LOCK(vp);
904 		vp->v_iflag &= ~VI_MOUNT;
905 		VI_UNLOCK(vp);
906 		vrele(vp);
907 		return (error);
908 	}
909 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
910 	/*
911 	 * Put the new filesystem on the mount list after root.
912 	 */
913 	cache_purge(vp);
914 	if (!error) {
915 		struct vnode *newdp;
916 
917 		VI_LOCK(vp);
918 		vp->v_iflag &= ~VI_MOUNT;
919 		VI_UNLOCK(vp);
920 		vp->v_mountedhere = mp;
921 		mtx_lock(&mountlist_mtx);
922 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
923 		mtx_unlock(&mountlist_mtx);
924 		if (VFS_ROOT(mp, &newdp))
925 			panic("mount: lost mount");
926 		checkdirs(vp, newdp);
927 		vput(newdp);
928 		VOP_UNLOCK(vp, 0, td);
929 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
930 			error = vfs_allocate_syncvnode(mp);
931 		vfs_unbusy(mp, td);
932 		if ((error = VFS_START(mp, 0, td)) != 0)
933 			vrele(vp);
934 	} else {
935 		VI_LOCK(vp);
936 		vp->v_iflag &= ~VI_MOUNT;
937 		VI_UNLOCK(vp);
938 		vfs_mount_destroy(mp, td);
939 		vput(vp);
940 	}
941 	return (error);
942 }
943 
944 /*
945  * Scan all active processes to see if any of them have a current
946  * or root directory of `olddp'. If so, replace them with the new
947  * mount point.
948  */
949 static void
950 checkdirs(olddp, newdp)
951 	struct vnode *olddp, *newdp;
952 {
953 	struct filedesc *fdp;
954 	struct proc *p;
955 	int nrele;
956 
957 	if (vrefcnt(olddp) == 1)
958 		return;
959 	sx_slock(&allproc_lock);
960 	LIST_FOREACH(p, &allproc, p_list) {
961 		mtx_lock(&fdesc_mtx);
962 		fdp = p->p_fd;
963 		if (fdp == NULL) {
964 			mtx_unlock(&fdesc_mtx);
965 			continue;
966 		}
967 		nrele = 0;
968 		FILEDESC_LOCK(fdp);
969 		if (fdp->fd_cdir == olddp) {
970 			VREF(newdp);
971 			fdp->fd_cdir = newdp;
972 			nrele++;
973 		}
974 		if (fdp->fd_rdir == olddp) {
975 			VREF(newdp);
976 			fdp->fd_rdir = newdp;
977 			nrele++;
978 		}
979 		FILEDESC_UNLOCK(fdp);
980 		mtx_unlock(&fdesc_mtx);
981 		while (nrele--)
982 			vrele(olddp);
983 	}
984 	sx_sunlock(&allproc_lock);
985 	if (rootvnode == olddp) {
986 		vrele(rootvnode);
987 		VREF(newdp);
988 		rootvnode = newdp;
989 	}
990 }
991 
992 /*
993  * Unmount a filesystem.
994  *
995  * Note: unmount takes a path to the vnode mounted on as argument,
996  * not special file (as before).
997  */
998 #ifndef _SYS_SYSPROTO_H_
999 struct unmount_args {
1000 	char	*path;
1001 	int	flags;
1002 };
1003 #endif
1004 /* ARGSUSED */
1005 int
1006 unmount(td, uap)
1007 	struct thread *td;
1008 	register struct unmount_args /* {
1009 		char *path;
1010 		int flags;
1011 	} */ *uap;
1012 {
1013 	struct mount *mp;
1014 	char *pathbuf;
1015 	int error, id0, id1;
1016 
1017 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1018 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1019 	if (error) {
1020 		free(pathbuf, M_TEMP);
1021 		return (error);
1022 	}
1023 	if (uap->flags & MNT_BYFSID) {
1024 		/* Decode the filesystem ID. */
1025 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1026 			free(pathbuf, M_TEMP);
1027 			return (EINVAL);
1028 		}
1029 
1030 		mtx_lock(&mountlist_mtx);
1031 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list)
1032 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1033 			    mp->mnt_stat.f_fsid.val[1] == id1)
1034 				break;
1035 		mtx_unlock(&mountlist_mtx);
1036 	} else {
1037 		mtx_lock(&mountlist_mtx);
1038 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list)
1039 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1040 				break;
1041 		mtx_unlock(&mountlist_mtx);
1042 	}
1043 	free(pathbuf, M_TEMP);
1044 	if (mp == NULL) {
1045 		/*
1046 		 * Previously we returned ENOENT for a nonexistent path and
1047 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1048 		 * now, so in the !MNT_BYFSID case return the more likely
1049 		 * EINVAL for compatibility.
1050 		 */
1051 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1052 	}
1053 
1054 	/*
1055 	 * Only root, or the user that did the original mount is
1056 	 * permitted to unmount this filesystem.
1057 	 */
1058 	if (mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
1059 		error = suser(td);
1060 		if (error)
1061 			return (error);
1062 	}
1063 
1064 	/*
1065 	 * Don't allow unmounting the root filesystem.
1066 	 */
1067 	if (mp->mnt_flag & MNT_ROOTFS)
1068 		return (EINVAL);
1069 	return (dounmount(mp, uap->flags, td));
1070 }
1071 
1072 /*
1073  * Do the actual filesystem unmount.
1074  */
1075 int
1076 dounmount(mp, flags, td)
1077 	struct mount *mp;
1078 	int flags;
1079 	struct thread *td;
1080 {
1081 	struct vnode *coveredvp, *fsrootvp;
1082 	int error;
1083 	int async_flag;
1084 
1085 	mtx_lock(&mountlist_mtx);
1086 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1087 		mtx_unlock(&mountlist_mtx);
1088 		return (EBUSY);
1089 	}
1090 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1091 	/* Allow filesystems to detect that a forced unmount is in progress. */
1092 	if (flags & MNT_FORCE)
1093 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1094 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1095 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td);
1096 	if (error) {
1097 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1098 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1099 			wakeup(mp);
1100 		return (error);
1101 	}
1102 	vn_start_write(NULL, &mp, V_WAIT);
1103 
1104 	if (mp->mnt_flag & MNT_EXPUBLIC)
1105 		vfs_setpublicfs(NULL, NULL, NULL);
1106 
1107 	vfs_msync(mp, MNT_WAIT);
1108 	async_flag = mp->mnt_flag & MNT_ASYNC;
1109 	mp->mnt_flag &=~ MNT_ASYNC;
1110 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1111 	if (mp->mnt_syncer != NULL)
1112 		vrele(mp->mnt_syncer);
1113 	/*
1114 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1115 	 * vnode to the covered vnode.  For non-forced unmounts we want
1116 	 * such references to cause an EBUSY error.
1117 	 */
1118 	if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp) == 0) {
1119 		if (mp->mnt_vnodecovered != NULL)
1120 			checkdirs(fsrootvp, mp->mnt_vnodecovered);
1121 		if (fsrootvp == rootvnode) {
1122 			vrele(rootvnode);
1123 			rootvnode = NULL;
1124 		}
1125 		vput(fsrootvp);
1126 	}
1127 	if (((mp->mnt_flag & MNT_RDONLY) ||
1128 	     (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) ||
1129 	    (flags & MNT_FORCE)) {
1130 		error = VFS_UNMOUNT(mp, flags, td);
1131 	}
1132 	vn_finished_write(mp);
1133 	if (error) {
1134 		/* Undo cdir/rdir and rootvnode changes made above. */
1135 		if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp) == 0) {
1136 			if (mp->mnt_vnodecovered != NULL)
1137 				checkdirs(mp->mnt_vnodecovered, fsrootvp);
1138 			if (rootvnode == NULL) {
1139 				rootvnode = fsrootvp;
1140 				vref(rootvnode);
1141 			}
1142 			vput(fsrootvp);
1143 		}
1144 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1145 			(void) vfs_allocate_syncvnode(mp);
1146 		mtx_lock(&mountlist_mtx);
1147 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1148 		mp->mnt_flag |= async_flag;
1149 		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK,
1150 		    &mountlist_mtx, td);
1151 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1152 			wakeup(mp);
1153 		return (error);
1154 	}
1155 	mtx_lock(&mountlist_mtx);
1156 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1157 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1158 		coveredvp->v_mountedhere = NULL;
1159 	mtx_unlock(&mountlist_mtx);
1160 	vfs_mount_destroy(mp, td);
1161 	if (coveredvp != NULL)
1162 		vrele(coveredvp);
1163 	return (0);
1164 }
1165 
1166 /*
1167  * Lookup a filesystem type, and if found allocate and initialize
1168  * a mount structure for it.
1169  *
1170  * Devname is usually updated by mount(8) after booting.
1171  */
1172 int
1173 vfs_rootmountalloc(fstypename, devname, mpp)
1174 	char *fstypename;
1175 	char *devname;
1176 	struct mount **mpp;
1177 {
1178 	struct thread *td = curthread;	/* XXX */
1179 	struct vfsconf *vfsp;
1180 	struct mount *mp;
1181 	int error;
1182 
1183 	if (fstypename == NULL)
1184 		return (ENODEV);
1185 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1186 		if (!strcmp(vfsp->vfc_name, fstypename))
1187 			break;
1188 	if (vfsp == NULL)
1189 		return (ENODEV);
1190 	error = vfs_mount_alloc(NULLVP, vfsp, "/", td, &mp);
1191 	if (error)
1192 		return (error);
1193 	mp->mnt_flag |= MNT_RDONLY | MNT_ROOTFS;
1194 	strlcpy(mp->mnt_stat.f_mntfromname, devname, MNAMELEN);
1195 	*mpp = mp;
1196 	return (0);
1197 }
1198 
1199 /*
1200  * Find and mount the root filesystem
1201  */
1202 void
1203 vfs_mountroot(void)
1204 {
1205 	char *cp;
1206 	int error, i;
1207 
1208 	g_waitidle();
1209 
1210 	/*
1211 	 * The root filesystem information is compiled in, and we are
1212 	 * booted with instructions to use it.
1213 	 */
1214 #ifdef ROOTDEVNAME
1215 	if ((boothowto & RB_DFLTROOT) && !vfs_mountroot_try(ROOTDEVNAME))
1216 		return;
1217 #endif
1218 	/*
1219 	 * We are booted with instructions to prompt for the root filesystem,
1220 	 * or to use the compiled-in default when it doesn't exist.
1221 	 */
1222 	if (boothowto & (RB_DFLTROOT | RB_ASKNAME)) {
1223 		if (!vfs_mountroot_ask())
1224 			return;
1225 	}
1226 
1227 	/*
1228 	 * We've been given the generic "use CDROM as root" flag.  This is
1229 	 * necessary because one media may be used in many different
1230 	 * devices, so we need to search for them.
1231 	 */
1232 	if (boothowto & RB_CDROM) {
1233 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1234 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1235 				return;
1236 		}
1237 	}
1238 
1239 	/*
1240 	 * Try to use the value read by the loader from /etc/fstab, or
1241 	 * supplied via some other means.  This is the preferred
1242 	 * mechanism.
1243 	 */
1244 	if ((cp = getenv("vfs.root.mountfrom")) != NULL) {
1245 		error = vfs_mountroot_try(cp);
1246 		freeenv(cp);
1247 		if (!error)
1248 			return;
1249 	}
1250 
1251 	/*
1252 	 * Try values that may have been computed by the machine-dependant
1253 	 * legacy code.
1254 	 */
1255 	if (!vfs_mountroot_try(rootdevnames[0]))
1256 		return;
1257 	if (!vfs_mountroot_try(rootdevnames[1]))
1258 		return;
1259 
1260 	/*
1261 	 * If we have a compiled-in default, and haven't already tried it, try
1262 	 * it now.
1263 	 */
1264 #ifdef ROOTDEVNAME
1265 	if (!(boothowto & RB_DFLTROOT))
1266 		if (!vfs_mountroot_try(ROOTDEVNAME))
1267 			return;
1268 #endif
1269 
1270 	/*
1271 	 * Everything so far has failed, prompt on the console if we haven't
1272 	 * already tried that.
1273 	 */
1274 	if (!(boothowto & (RB_DFLTROOT | RB_ASKNAME)) && !vfs_mountroot_ask())
1275 		return;
1276 	panic("Root mount failed, startup aborted.");
1277 }
1278 
1279 /*
1280  * Mount (mountfrom) as the root filesystem.
1281  */
1282 static int
1283 vfs_mountroot_try(char *mountfrom)
1284 {
1285         struct mount	*mp;
1286 	char		*vfsname, *path;
1287 	const char	*devname;
1288 	int		error;
1289 	char		patt[32];
1290 	int		s;
1291 
1292 	vfsname = NULL;
1293 	path    = NULL;
1294 	mp      = NULL;
1295 	error   = EINVAL;
1296 
1297 	if (mountfrom == NULL)
1298 		return(error);		/* don't complain */
1299 
1300 	s = splcam();			/* Overkill, but annoying without it */
1301 	printf("Mounting root from %s\n", mountfrom);
1302 	splx(s);
1303 
1304 	/* parse vfs name and path */
1305 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1306 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1307 	vfsname[0] = path[0] = 0;
1308 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1309 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1310 		goto done;
1311 
1312 	/* allocate a root mount */
1313 	error = vfs_rootmountalloc(vfsname, path[0] != 0 ? path : ROOTNAME,
1314 				   &mp);
1315 	if (error != 0) {
1316 		printf("Can't allocate root mount for filesystem '%s': %d\n",
1317 		       vfsname, error);
1318 		goto done;
1319 	}
1320 
1321 	/* do our best to set rootdev */
1322 	if ((path[0] != 0) && setrootbyname(path))
1323 		printf("setrootbyname failed\n");
1324 
1325 	/* If the root device is a type "memory disk", mount RW */
1326 	if (rootdev != NODEV && devsw(rootdev) != NULL) {
1327 		devname = devtoname(rootdev);
1328 		if (devname[0] == 'm' && devname[1] == 'd')
1329 			mp->mnt_flag &= ~MNT_RDONLY;
1330 	}
1331 
1332 	error = VFS_MOUNT(mp, NULL, NULL, NULL, curthread);
1333 
1334 done:
1335 	if (vfsname != NULL)
1336 		free(vfsname, M_MOUNT);
1337 	if (path != NULL)
1338 		free(path, M_MOUNT);
1339 	if (error != 0) {
1340 		if (mp != NULL)
1341 			vfs_mount_destroy(mp, curthread);
1342 		printf("Root mount failed: %d\n", error);
1343 	} else {
1344 
1345 		/* register with list of mounted filesystems */
1346 		mtx_lock(&mountlist_mtx);
1347 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1348 		mtx_unlock(&mountlist_mtx);
1349 
1350 		/* sanity check system clock against root fs timestamp */
1351 		inittodr(mp->mnt_time);
1352 		vfs_unbusy(mp, curthread);
1353 		error = VFS_START(mp, 0, curthread);
1354 	}
1355 	return(error);
1356 }
1357 
1358 /*
1359  * Spin prompting on the console for a suitable root filesystem
1360  */
1361 static int
1362 vfs_mountroot_ask(void)
1363 {
1364 	char name[128];
1365 
1366 	for(;;) {
1367 		printf("\nManual root filesystem specification:\n");
1368 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1369 #if defined(__i386__) || defined(__ia64__)
1370 		printf("                       eg. ufs:da0s1a\n");
1371 #else
1372 		printf("                       eg. ufs:/dev/da0a\n");
1373 #endif
1374 		printf("  ?                  List valid disk boot devices\n");
1375 		printf("  <empty line>       Abort manual input\n");
1376 		printf("\nmountroot> ");
1377 		gets(name);
1378 		if (name[0] == 0)
1379 			return(1);
1380 		if (name[0] == '?') {
1381 			printf("\nList of GEOM managed disk devices:\n  ");
1382 			g_dev_print();
1383 			continue;
1384 		}
1385 		if (!vfs_mountroot_try(name))
1386 			return(0);
1387 	}
1388 }
1389 
1390 /*
1391  * Local helper function for vfs_mountroot_ask.
1392  */
1393 static void
1394 gets(char *cp)
1395 {
1396 	char *lp;
1397 	int c;
1398 
1399 	lp = cp;
1400 	for (;;) {
1401 		printf("%c", c = cngetc() & 0177);
1402 		switch (c) {
1403 		case -1:
1404 		case '\n':
1405 		case '\r':
1406 			*lp++ = '\0';
1407 			return;
1408 		case '\b':
1409 		case '\177':
1410 			if (lp > cp) {
1411 				printf(" \b");
1412 				lp--;
1413 			}
1414 			continue;
1415 		case '#':
1416 			lp--;
1417 			if (lp < cp)
1418 				lp = cp;
1419 			continue;
1420 		case '@':
1421 		case 'u' & 037:
1422 			lp = cp;
1423 			printf("%c", '\n');
1424 			continue;
1425 		default:
1426 			*lp++ = c;
1427 		}
1428 	}
1429 }
1430 
1431 /*
1432  * Convert a given name to the dev_t of the disk-like device
1433  * it refers to.
1434  */
1435 dev_t
1436 getdiskbyname(char *name) {
1437 	char *cp;
1438 	dev_t dev;
1439 
1440 	cp = name;
1441 	if (!bcmp(cp, "/dev/", 5))
1442 		cp += 5;
1443 
1444 	dev = NODEV;
1445 	EVENTHANDLER_INVOKE(dev_clone, cp, strlen(cp), &dev);
1446 	return (dev);
1447 }
1448 
1449 /*
1450  * Set rootdev to match (name), given that we expect it to
1451  * refer to a disk-like device.
1452  */
1453 static int
1454 setrootbyname(char *name)
1455 {
1456 	dev_t diskdev;
1457 
1458 	diskdev = getdiskbyname(name);
1459 	if (diskdev != NODEV) {
1460 		rootdev = diskdev;
1461 		return (0);
1462 	}
1463 
1464 	return (1);
1465 }
1466 
1467 /* Show the dev_t for a disk specified by name */
1468 #ifdef DDB
1469 DB_SHOW_COMMAND(disk, db_getdiskbyname)
1470 {
1471 	dev_t dev;
1472 
1473 	if (modif[0] == '\0') {
1474 		db_error("usage: show disk/devicename");
1475 		return;
1476 	}
1477 	dev = getdiskbyname(modif);
1478 	if (dev != NODEV)
1479 		db_printf("dev_t = %p\n", dev);
1480 	else
1481 		db_printf("No disk device matched.\n");
1482 }
1483 #endif
1484 
1485 /*
1486  * Get a mount option by its name.
1487  *
1488  * Return 0 if the option was found, ENOENT otherwise.
1489  * If len is non-NULL it will be filled with the length
1490  * of the option. If buf is non-NULL, it will be filled
1491  * with the address of the option.
1492  */
1493 int
1494 vfs_getopt(opts, name, buf, len)
1495 	struct vfsoptlist *opts;
1496 	const char *name;
1497 	void **buf;
1498 	int *len;
1499 {
1500 	struct vfsopt *opt;
1501 
1502 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1503 
1504 	TAILQ_FOREACH(opt, opts, link) {
1505 		if (strcmp(name, opt->name) == 0) {
1506 			if (len != NULL)
1507 				*len = opt->len;
1508 			if (buf != NULL)
1509 				*buf = opt->value;
1510 			return (0);
1511 		}
1512 	}
1513 	return (ENOENT);
1514 }
1515 
1516 /*
1517  * Find and copy a mount option.
1518  *
1519  * The size of the buffer has to be specified
1520  * in len, if it is not the same length as the
1521  * mount option, EINVAL is returned.
1522  * Returns ENOENT if the option is not found.
1523  */
1524 int
1525 vfs_copyopt(opts, name, dest, len)
1526 	struct vfsoptlist *opts;
1527 	const char *name;
1528 	void *dest;
1529 	int len;
1530 {
1531 	struct vfsopt *opt;
1532 
1533 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1534 
1535 	TAILQ_FOREACH(opt, opts, link) {
1536 		if (strcmp(name, opt->name) == 0) {
1537 			if (len != opt->len)
1538 				return (EINVAL);
1539 			bcopy(opt->value, dest, opt->len);
1540 			return (0);
1541 		}
1542 	}
1543 	return (ENOENT);
1544 }
1545