xref: /freebsd/sys/kern/vfs_mount.c (revision d2387d42b8da231a5b95cbc313825fb2aadf26f6)
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/jail.h>
72 #include <sys/kernel.h>
73 #include <sys/linker.h>
74 #include <sys/mac.h>
75 #include <sys/malloc.h>
76 #include <sys/mount.h>
77 #include <sys/mutex.h>
78 #include <sys/namei.h>
79 #include <sys/proc.h>
80 #include <sys/filedesc.h>
81 #include <sys/reboot.h>
82 #include <sys/sysproto.h>
83 #include <sys/sx.h>
84 #include <sys/sysctl.h>
85 #include <sys/sysent.h>
86 #include <sys/systm.h>
87 #include <sys/vnode.h>
88 
89 #include <geom/geom.h>
90 
91 #include <machine/stdarg.h>
92 
93 #include "opt_rootdevname.h"
94 #include "opt_ddb.h"
95 #include "opt_mac.h"
96 
97 #ifdef DDB
98 #include <ddb/ddb.h>
99 #endif
100 
101 #define	ROOTNAME		"root_device"
102 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
103 
104 static void	checkdirs(struct vnode *olddp, struct vnode *newdp);
105 static void	gets(char *cp);
106 static int	vfs_domount(struct thread *td, const char *fstype,
107 		    char *fspath, int fsflags, void *fsdata, int compat);
108 static int	vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
109 		    const char *fspath, struct thread *td, struct mount **mpp);
110 static int	vfs_mountroot_ask(void);
111 static int	vfs_mountroot_try(char *mountfrom);
112 
113 static int	usermount = 0;
114 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
115     "Unprivileged users may mount and unmount file systems");
116 
117 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
118 
119 /* List of mounted filesystems. */
120 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
121 
122 /* For any iteration/modification of mountlist */
123 struct mtx mountlist_mtx;
124 
125 /*
126  * The vnode of the system's root (/ in the filesystem, without chroot
127  * active.)
128  */
129 struct vnode	*rootvnode;
130 
131 /*
132  * The root filesystem is detailed in the kernel environment variable
133  * vfs.root.mountfrom, which is expected to be in the general format
134  *
135  * <vfsname>:[<path>]
136  * vfsname   := the name of a VFS known to the kernel and capable
137  *              of being mounted as root
138  * path      := disk device name or other data used by the filesystem
139  *              to locate its physical store
140  */
141 
142 /*
143  * The root specifiers we will try if RB_CDROM is specified.
144  */
145 static char *cdrom_rootdevnames[] = {
146 	"cd9660:cd0",
147 	"cd9660:acd0",
148 	NULL
149 };
150 
151 /* legacy find-root code */
152 char		*rootdevnames[2] = {NULL, NULL};
153 static int	setrootbyname(char *name);
154 dev_t		rootdev = NODEV;
155 
156 /*
157  * Has to be dynamic as the value of rootdev can change; however, it can't
158  * change after the root is mounted, so a user process can't access this
159  * sysctl until after the value is unchangeable.
160  */
161 static int
162 sysctl_rootdev(SYSCTL_HANDLER_ARGS)
163 {
164 	int error;
165 
166 	/* _RD prevents this from happening. */
167 	KASSERT(req->newptr == NULL, ("Attempt to change root device name"));
168 
169 	if (rootdev != NODEV)
170 		error = sysctl_handle_string(oidp, rootdev->si_name, 0, req);
171 	else
172 		error = sysctl_handle_string(oidp, "", 0, req);
173 
174 	return (error);
175 }
176 
177 SYSCTL_PROC(_kern, OID_AUTO, rootdev, CTLTYPE_STRING | CTLFLAG_RD,
178     0, 0, sysctl_rootdev, "A", "Root file system device");
179 
180 /* Remove one mount option. */
181 static void
182 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
183 {
184 
185 	TAILQ_REMOVE(opts, opt, link);
186 	free(opt->name, M_MOUNT);
187 	if (opt->value != NULL)
188 		free(opt->value, M_MOUNT);
189 #ifdef INVARIANTS
190 	else if (opt->len != 0)
191 		panic("%s: mount option with NULL value but length != 0",
192 		    __func__);
193 #endif
194 	free(opt, M_MOUNT);
195 }
196 
197 /* Release all resources related to the mount options. */
198 static void
199 vfs_freeopts(struct vfsoptlist *opts)
200 {
201 	struct vfsopt *opt;
202 
203 	while (!TAILQ_EMPTY(opts)) {
204 		opt = TAILQ_FIRST(opts);
205 		vfs_freeopt(opts, opt);
206 	}
207 	free(opts, M_MOUNT);
208 }
209 
210 /*
211  * Check if options are equal (with or without the "no" prefix).
212  */
213 static int
214 vfs_equalopts(const char *opt1, const char *opt2)
215 {
216 
217 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
218 	if (strcmp(opt1, opt2) == 0)
219 		return (1);
220 	/* "noopt" vs. "opt" */
221 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
222 		return (1);
223 	/* "opt" vs. "noopt" */
224 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
225 		return (1);
226 	return (0);
227 }
228 
229 /*
230  * If a mount option is specified several times,
231  * (with or without the "no" prefix) only keep
232  * the last occurence of it.
233  */
234 static void
235 vfs_sanitizeopts(struct vfsoptlist *opts)
236 {
237 	struct vfsopt *opt, *opt2, *tmp;
238 
239 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
240 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
241 		while (opt2 != NULL) {
242 			if (vfs_equalopts(opt->name, opt2->name)) {
243 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
244 				vfs_freeopt(opts, opt2);
245 				opt2 = tmp;
246 			} else {
247 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
248 			}
249 		}
250 	}
251 }
252 
253 /*
254  * Build a linked list of mount options from a struct uio.
255  */
256 static int
257 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
258 {
259 	struct vfsoptlist *opts;
260 	struct vfsopt *opt;
261 	size_t memused;
262 	unsigned int i, iovcnt;
263 	int error, namelen, optlen;
264 
265 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
266 	TAILQ_INIT(opts);
267 	memused = 0;
268 	iovcnt = auio->uio_iovcnt;
269 	for (i = 0; i < iovcnt; i += 2) {
270 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
271 		namelen = auio->uio_iov[i].iov_len;
272 		optlen = auio->uio_iov[i + 1].iov_len;
273 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
274 		opt->value = NULL;
275 		opt->len = optlen;
276 
277 		/*
278 		 * Do this early, so jumps to "bad" will free the current
279 		 * option.
280 		 */
281 		TAILQ_INSERT_TAIL(opts, opt, link);
282 		memused += sizeof(struct vfsopt) + optlen + namelen;
283 
284 		/*
285 		 * Avoid consuming too much memory, and attempts to overflow
286 		 * memused.
287 		 */
288 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
289 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
290 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
291 			error = EINVAL;
292 			goto bad;
293 		}
294 
295 		if (auio->uio_segflg == UIO_SYSSPACE) {
296 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
297 		} else {
298 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
299 			    namelen);
300 			if (error)
301 				goto bad;
302 		}
303 		/* Ensure names are null-terminated strings. */
304 		if (opt->name[namelen - 1] != '\0') {
305 			error = EINVAL;
306 			goto bad;
307 		}
308 		if (optlen != 0) {
309 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
310 			if (auio->uio_segflg == UIO_SYSSPACE) {
311 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
312 				    optlen);
313 			} else {
314 				error = copyin(auio->uio_iov[i + 1].iov_base,
315 				    opt->value, optlen);
316 				if (error)
317 					goto bad;
318 			}
319 		}
320 	}
321 	vfs_sanitizeopts(opts);
322 	*options = opts;
323 	return (0);
324 bad:
325 	vfs_freeopts(opts);
326 	return (error);
327 }
328 
329 /*
330  * Merge the old mount options with the new ones passed
331  * in the MNT_UPDATE case.
332  */
333 static void
334 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
335 {
336 	struct vfsopt *opt, *opt2, *new;
337 
338 	TAILQ_FOREACH(opt, opts, link) {
339 		/*
340 		 * Check that this option hasn't been redefined
341 		 * nor cancelled with a "no" mount option.
342 		 */
343 		opt2 = TAILQ_FIRST(toopts);
344 		while (opt2 != NULL) {
345 			if (strcmp(opt2->name, opt->name) == 0)
346 				goto next;
347 			if (strncmp(opt2->name, "no", 2) == 0 &&
348 			    strcmp(opt2->name + 2, opt->name) == 0) {
349 				vfs_freeopt(toopts, opt2);
350 				goto next;
351 			}
352 			opt2 = TAILQ_NEXT(opt2, link);
353 		}
354 		/* We want this option, duplicate it. */
355 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
356 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
357 		strcpy(new->name, opt->name);
358 		if (opt->len != 0) {
359 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
360 			bcopy(opt->value, new->value, opt->len);
361 		} else {
362 			new->value = NULL;
363 		}
364 		new->len = opt->len;
365 		TAILQ_INSERT_TAIL(toopts, new, link);
366 next:
367 		continue;
368 	}
369 }
370 
371 /*
372  * New mount API.
373  */
374 int
375 nmount(td, uap)
376 	struct thread *td;
377 	struct nmount_args /* {
378 		struct iovec *iovp;
379 		unsigned int iovcnt;
380 		int flags;
381 	} */ *uap;
382 {
383 	struct uio auio;
384 	struct iovec *iov, *needfree;
385 	struct iovec aiov[UIO_SMALLIOV];
386 	unsigned int i;
387 	int error;
388 	u_int iovlen, iovcnt;
389 
390 	iovcnt = uap->iovcnt;
391 	iovlen = iovcnt * sizeof (struct iovec);
392 	/*
393 	 * Check that we have an even number of iovec's
394 	 * and that we have at least two options.
395 	 */
396 	if ((iovcnt & 1) || (iovcnt < 4) || (iovcnt > UIO_MAXIOV))
397 		return (EINVAL);
398 
399 	if (iovcnt > UIO_SMALLIOV) {
400 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
401 		needfree = iov;
402 	} else {
403 		iov = aiov;
404 		needfree = NULL;
405 	}
406 	auio.uio_iov = iov;
407 	auio.uio_iovcnt = iovcnt;
408 	auio.uio_segflg = UIO_USERSPACE;
409 	if ((error = copyin(uap->iovp, iov, iovlen)))
410 		goto finish;
411 
412 	for (i = 0; i < iovcnt; i++) {
413 		if (iov->iov_len > MMAXOPTIONLEN) {
414 			error = EINVAL;
415 			goto finish;
416 		}
417 		iov++;
418 	}
419 	error = vfs_nmount(td, uap->flags, &auio);
420 finish:
421 	if (needfree != NULL)
422 		free(needfree, M_TEMP);
423 	return (error);
424 }
425 
426 int
427 kernel_mount(iovp, iovcnt, flags)
428 	struct iovec *iovp;
429 	unsigned int iovcnt;
430 	int flags;
431 {
432 	struct uio auio;
433 	int error;
434 
435 	/*
436 	 * Check that we have an even number of iovec's
437 	 * and that we have at least two options.
438 	 */
439 	if ((iovcnt & 1) || (iovcnt < 4))
440 		return (EINVAL);
441 
442 	auio.uio_iov = iovp;
443 	auio.uio_iovcnt = iovcnt;
444 	auio.uio_segflg = UIO_SYSSPACE;
445 
446 	error = vfs_nmount(curthread, flags, &auio);
447 	return (error);
448 }
449 
450 int
451 kernel_vmount(int flags, ...)
452 {
453 	struct iovec *iovp;
454 	struct uio auio;
455 	va_list ap;
456 	unsigned int iovcnt, iovlen, len;
457 	const char *cp;
458 	char *buf, *pos;
459 	size_t n;
460 	int error, i;
461 
462 	len = 0;
463 	va_start(ap, flags);
464 	for (iovcnt = 0; (cp = va_arg(ap, const char *)) != NULL; iovcnt++)
465 		len += strlen(cp) + 1;
466 	va_end(ap);
467 
468 	if (iovcnt < 4 || iovcnt & 1)
469 		return (EINVAL);
470 
471 	iovlen = iovcnt * sizeof (struct iovec);
472 	MALLOC(iovp, struct iovec *, iovlen, M_MOUNT, M_WAITOK);
473 	MALLOC(buf, char *, len, M_MOUNT, M_WAITOK);
474 	pos = buf;
475 	va_start(ap, flags);
476 	for (i = 0; i < iovcnt; i++) {
477 		cp = va_arg(ap, const char *);
478 		copystr(cp, pos, len - (pos - buf), &n);
479 		iovp[i].iov_base = pos;
480 		iovp[i].iov_len = n;
481 		pos += n;
482 	}
483 	va_end(ap);
484 
485 	auio.uio_iov = iovp;
486 	auio.uio_iovcnt = iovcnt;
487 	auio.uio_segflg = UIO_SYSSPACE;
488 
489 	error = vfs_nmount(curthread, flags, &auio);
490 	FREE(iovp, M_MOUNT);
491 	FREE(buf, M_MOUNT);
492 	return (error);
493 }
494 
495 /*
496  * Allocate and initialize the mount point struct.
497  */
498 static int
499 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
500     const char *fspath, struct thread *td, struct mount **mpp)
501 {
502 	struct mount *mp;
503 
504 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
505 	TAILQ_INIT(&mp->mnt_nvnodelist);
506 	mp->mnt_nvnodelistsize = 0;
507 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
508 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
509 	vfs_busy(mp, LK_NOWAIT, 0, td);
510 	mp->mnt_op = vfsp->vfc_vfsops;
511 	mp->mnt_vfc = vfsp;
512 	vfsp->vfc_refcount++;
513 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
514 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
515 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
516 	mp->mnt_vnodecovered = vp;
517 	mp->mnt_cred = crdup(td->td_ucred);
518 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
519 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
520 	mp->mnt_iosize_max = DFLTPHYS;
521 #ifdef MAC
522 	mac_init_mount(mp);
523 	mac_create_mount(td->td_ucred, mp);
524 #endif
525 	*mpp = mp;
526 	return (0);
527 }
528 
529 /*
530  * Destroy the mount struct previously allocated by vfs_mount_alloc().
531  */
532 void
533 vfs_mount_destroy(struct mount *mp, struct thread *td)
534 {
535 
536 	mp->mnt_vfc->vfc_refcount--;
537 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
538 		panic("unmount: dangling vnode");
539 	vfs_unbusy(mp,td);
540 	lockdestroy(&mp->mnt_lock);
541 	mtx_destroy(&mp->mnt_mtx);
542 	if (mp->mnt_kern_flag & MNTK_MWAIT)
543 		wakeup(mp);
544 #ifdef MAC
545 	mac_destroy_mount(mp);
546 #endif
547 	if (mp->mnt_opt != NULL)
548 		vfs_freeopts(mp->mnt_opt);
549 	crfree(mp->mnt_cred);
550 	free(mp, M_MOUNT);
551 }
552 
553 int
554 vfs_nmount(struct thread *td, int fsflags, struct uio *fsoptions)
555 {
556 	struct vfsoptlist *optlist;
557 	char *fstype, *fspath;
558 	int error, fstypelen, fspathlen;
559 
560 	error = vfs_buildopts(fsoptions, &optlist);
561 	if (error)
562 		return (error);
563 
564 	/*
565 	 * We need these two options before the others,
566 	 * and they are mandatory for any filesystem.
567 	 * Ensure they are NUL terminated as well.
568 	 */
569 	fstypelen = 0;
570 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
571 	if (error || fstype[fstypelen - 1] != '\0') {
572 		error = EINVAL;
573 		goto bail;
574 	}
575 	fspathlen = 0;
576 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
577 	if (error || fspath[fspathlen - 1] != '\0') {
578 		error = EINVAL;
579 		goto bail;
580 	}
581 
582 	/*
583 	 * Be ultra-paranoid about making sure the type and fspath
584 	 * variables will fit in our mp buffers, including the
585 	 * terminating NUL.
586 	 */
587 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
588 		error = ENAMETOOLONG;
589 		goto bail;
590 	}
591 
592 	error = vfs_domount(td, fstype, fspath, fsflags, optlist, 0);
593 bail:
594 	if (error)
595 		vfs_freeopts(optlist);
596 	return (error);
597 }
598 
599 /*
600  * Old mount API.
601  */
602 #ifndef _SYS_SYSPROTO_H_
603 struct mount_args {
604 	char	*type;
605 	char	*path;
606 	int	flags;
607 	caddr_t	data;
608 };
609 #endif
610 /* ARGSUSED */
611 int
612 mount(td, uap)
613 	struct thread *td;
614 	struct mount_args /* {
615 		char *type;
616 		char *path;
617 		int flags;
618 		caddr_t data;
619 	} */ *uap;
620 {
621 	char *fstype;
622 	char *fspath;
623 	int error;
624 
625 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
626 	fspath = malloc(MNAMELEN, M_TEMP, M_WAITOK);
627 
628 	/*
629 	 * vfs_mount() actually takes a kernel string for `type' and
630 	 * `path' now, so extract them.
631 	 */
632 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
633 	if (error == 0)
634 		error = copyinstr(uap->path, fspath, MNAMELEN, NULL);
635 	if (error == 0)
636 		error = vfs_domount(td, fstype, fspath, uap->flags,
637 		    uap->data, 1);
638 	free(fstype, M_TEMP);
639 	free(fspath, M_TEMP);
640 	return (error);
641 }
642 
643 /*
644  * vfs_mount(): actually attempt a filesystem mount.
645  *
646  * This routine is designed to be a "generic" entry point for routines
647  * that wish to mount a filesystem. All parameters except `fsdata' are
648  * pointers into kernel space. `fsdata' is currently still a pointer
649  * into userspace.
650  */
651 int
652 vfs_mount(td, fstype, fspath, fsflags, fsdata)
653 	struct thread *td;
654 	const char *fstype;
655 	char *fspath;
656 	int fsflags;
657 	void *fsdata;
658 {
659 
660 	return (vfs_domount(td, fstype, fspath, fsflags, fsdata, 1));
661 }
662 
663 /*
664  * vfs_domount(): actually attempt a filesystem mount.
665  */
666 static int
667 vfs_domount(
668 	struct thread *td,	/* Flags common to all filesystems. */
669 	const char *fstype,	/* Filesystem type. */
670 	char *fspath,		/* Mount path. */
671 	int fsflags,		/* Flags common to all filesystems. */
672 	void *fsdata,		/* Options local to the filesystem. */
673 	int compat		/* Invocation from compat syscall. */
674 	)
675 {
676 	linker_file_t lf;
677 	struct vnode *vp;
678 	struct mount *mp;
679 	struct vfsconf *vfsp;
680 	int error, flag = 0, kern_flag = 0;
681 	struct vattr va;
682 	struct nameidata nd;
683 
684 	/*
685 	 * Be ultra-paranoid about making sure the type and fspath
686 	 * variables will fit in our mp buffers, including the
687 	 * terminating NUL.
688 	 */
689 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
690 		return (ENAMETOOLONG);
691 
692 	if (jailed(td->td_ucred))
693 		return (EPERM);
694 	if (usermount == 0) {
695 		if ((error = suser(td)) != 0)
696 			return (error);
697 	}
698 
699 	/*
700 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
701 	 */
702 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
703 		if ((error = suser(td)) != 0)
704 			return (error);
705 	}
706 	/*
707 	 * Silently enforce MNT_NODEV, MNT_NOSUID and MNT_USER for
708 	 * unprivileged users.
709 	 */
710 	if (suser(td) != 0)
711 		fsflags |= MNT_NODEV | MNT_NOSUID | MNT_USER;
712 	/*
713 	 * Get vnode to be covered
714 	 */
715 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
716 	if ((error = namei(&nd)) != 0)
717 		return (error);
718 	NDFREE(&nd, NDF_ONLY_PNBUF);
719 	vp = nd.ni_vp;
720 	if (fsflags & MNT_UPDATE) {
721 		if ((vp->v_vflag & VV_ROOT) == 0) {
722 			vput(vp);
723 			return (EINVAL);
724 		}
725 		mp = vp->v_mount;
726 		flag = mp->mnt_flag;
727 		kern_flag = mp->mnt_kern_flag;
728 		/*
729 		 * We only allow the filesystem to be reloaded if it
730 		 * is currently mounted read-only.
731 		 */
732 		if ((fsflags & MNT_RELOAD) &&
733 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
734 			vput(vp);
735 			return (EOPNOTSUPP);	/* Needs translation */
736 		}
737 		/*
738 		 * Only privileged root, or (if MNT_USER is set) the user that
739 		 * did the original mount is permitted to update it.
740 		 */
741 		if ((mp->mnt_flag & MNT_USER) == 0 ||
742 		    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
743 			if ((error = suser(td)) != 0) {
744 				vput(vp);
745 				return (error);
746 			}
747 		}
748 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
749 			vput(vp);
750 			return (EBUSY);
751 		}
752 		VI_LOCK(vp);
753 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
754 		    vp->v_mountedhere != NULL) {
755 			VI_UNLOCK(vp);
756 			vfs_unbusy(mp, td);
757 			vput(vp);
758 			return (EBUSY);
759 		}
760 		vp->v_iflag |= VI_MOUNT;
761 		VI_UNLOCK(vp);
762 		mp->mnt_flag |= fsflags &
763 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT);
764 		VOP_UNLOCK(vp, 0, td);
765 		if (compat == 0) {
766 			mp->mnt_optnew = fsdata;
767 			vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
768 		}
769 		goto update;
770 	}
771 	/*
772 	 * If the user is not root, ensure that they own the directory
773 	 * onto which we are attempting to mount.
774 	 */
775 	error = VOP_GETATTR(vp, &va, td->td_ucred, td);
776 	if (error) {
777 		vput(vp);
778 		return (error);
779 	}
780 	if (va.va_uid != td->td_ucred->cr_uid) {
781 		if ((error = suser(td)) != 0) {
782 			vput(vp);
783 			return (error);
784 		}
785 	}
786 	if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) {
787 		vput(vp);
788 		return (error);
789 	}
790 	if (vp->v_type != VDIR) {
791 		vput(vp);
792 		return (ENOTDIR);
793 	}
794 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
795 		if (strcmp(vfsp->vfc_name, fstype) == 0)
796 			break;
797 	if (vfsp == NULL) {
798 		/* Only load modules for root (very important!). */
799 		if ((error = suser(td)) != 0) {
800 			vput(vp);
801 			return (error);
802 		}
803 		error = securelevel_gt(td->td_ucred, 0);
804 		if (error) {
805 			vput(vp);
806 			return (error);
807 		}
808 		error = linker_load_module(NULL, fstype, NULL, NULL, &lf);
809 		if (error || lf == NULL) {
810 			vput(vp);
811 			if (lf == NULL)
812 				error = ENODEV;
813 			return (error);
814 		}
815 		lf->userrefs++;
816 		/* Look up again to see if the VFS was loaded. */
817 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
818 			if (strcmp(vfsp->vfc_name, fstype) == 0)
819 				break;
820 		if (vfsp == NULL) {
821 			lf->userrefs--;
822 			linker_file_unload(lf);
823 			vput(vp);
824 			return (ENODEV);
825 		}
826 	}
827 	VI_LOCK(vp);
828 	if ((vp->v_iflag & VI_MOUNT) != 0 ||
829 	    vp->v_mountedhere != NULL) {
830 		VI_UNLOCK(vp);
831 		vput(vp);
832 		return (EBUSY);
833 	}
834 	vp->v_iflag |= VI_MOUNT;
835 	VI_UNLOCK(vp);
836 
837 	/*
838 	 * Allocate and initialize the filesystem.
839 	 */
840 	error = vfs_mount_alloc(vp, vfsp, fspath, td, &mp);
841 	if (error) {
842 		vput(vp);
843 		return (error);
844 	}
845 	VOP_UNLOCK(vp, 0, td);
846 
847 	/* XXXMAC: pass to vfs_mount_alloc? */
848 	if (compat == 0)
849 		mp->mnt_optnew = fsdata;
850 update:
851 	/*
852 	 * Check if the fs implements the type VFS_[N]MOUNT()
853 	 * function we are looking for.
854 	 */
855 	if ((compat == 0) == (mp->mnt_op->vfs_mount != NULL)) {
856 		printf("%s doesn't support the %s mount syscall\n",
857 		    mp->mnt_vfc->vfc_name, compat ? "old" : "new");
858 		VI_LOCK(vp);
859 		vp->v_iflag &= ~VI_MOUNT;
860 		VI_UNLOCK(vp);
861 		if (mp->mnt_flag & MNT_UPDATE)
862 			vfs_unbusy(mp, td);
863 		else
864 			vfs_mount_destroy(mp, td);
865 		vrele(vp);
866 		return (EOPNOTSUPP);
867 	}
868 
869 	/*
870 	 * Set the mount level flags.
871 	 */
872 	if (fsflags & MNT_RDONLY)
873 		mp->mnt_flag |= MNT_RDONLY;
874 	else if (mp->mnt_flag & MNT_RDONLY)
875 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
876 	mp->mnt_flag &=~ MNT_UPDATEMASK;
877 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE);
878 	/*
879 	 * Mount the filesystem.
880 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
881 	 * get.  No freeing of cn_pnbuf.
882 	 */
883 	error = compat ? VFS_MOUNT(mp, fspath, fsdata, &nd, td) :
884 	    VFS_NMOUNT(mp, &nd, td);
885 	if (!error) {
886 		if (mp->mnt_opt != NULL)
887 			vfs_freeopts(mp->mnt_opt);
888 		mp->mnt_opt = mp->mnt_optnew;
889 	}
890 	/*
891 	 * Prevent external consumers of mount options from reading
892 	 * mnt_optnew.
893 	*/
894 	mp->mnt_optnew = NULL;
895 	if (mp->mnt_flag & MNT_UPDATE) {
896 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
897 			mp->mnt_flag &= ~MNT_RDONLY;
898 		mp->mnt_flag &=
899 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
900 		mp->mnt_kern_flag &= ~MNTK_WANTRDWR;
901 		if (error) {
902 			mp->mnt_flag = flag;
903 			mp->mnt_kern_flag = kern_flag;
904 		}
905 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
906 			if (mp->mnt_syncer == NULL)
907 				error = vfs_allocate_syncvnode(mp);
908 		} else {
909 			if (mp->mnt_syncer != NULL)
910 				vrele(mp->mnt_syncer);
911 			mp->mnt_syncer = NULL;
912 		}
913 		vfs_unbusy(mp, td);
914 		VI_LOCK(vp);
915 		vp->v_iflag &= ~VI_MOUNT;
916 		VI_UNLOCK(vp);
917 		vrele(vp);
918 		return (error);
919 	}
920 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
921 	/*
922 	 * Put the new filesystem on the mount list after root.
923 	 */
924 	cache_purge(vp);
925 	if (!error) {
926 		struct vnode *newdp;
927 
928 		VI_LOCK(vp);
929 		vp->v_iflag &= ~VI_MOUNT;
930 		VI_UNLOCK(vp);
931 		vp->v_mountedhere = mp;
932 		mtx_lock(&mountlist_mtx);
933 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
934 		mtx_unlock(&mountlist_mtx);
935 		if (VFS_ROOT(mp, &newdp))
936 			panic("mount: lost mount");
937 		checkdirs(vp, newdp);
938 		vput(newdp);
939 		VOP_UNLOCK(vp, 0, td);
940 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
941 			error = vfs_allocate_syncvnode(mp);
942 		vfs_unbusy(mp, td);
943 		if (error || (error = VFS_START(mp, 0, td)) != 0)
944 			vrele(vp);
945 	} else {
946 		VI_LOCK(vp);
947 		vp->v_iflag &= ~VI_MOUNT;
948 		VI_UNLOCK(vp);
949 		vfs_mount_destroy(mp, td);
950 		vput(vp);
951 	}
952 	return (error);
953 }
954 
955 /*
956  * Scan all active processes to see if any of them have a current
957  * or root directory of `olddp'. If so, replace them with the new
958  * mount point.
959  */
960 static void
961 checkdirs(olddp, newdp)
962 	struct vnode *olddp, *newdp;
963 {
964 	struct filedesc *fdp;
965 	struct proc *p;
966 	int nrele;
967 
968 	if (vrefcnt(olddp) == 1)
969 		return;
970 	sx_slock(&allproc_lock);
971 	LIST_FOREACH(p, &allproc, p_list) {
972 		mtx_lock(&fdesc_mtx);
973 		fdp = p->p_fd;
974 		if (fdp == NULL) {
975 			mtx_unlock(&fdesc_mtx);
976 			continue;
977 		}
978 		nrele = 0;
979 		FILEDESC_LOCK(fdp);
980 		if (fdp->fd_cdir == olddp) {
981 			VREF(newdp);
982 			fdp->fd_cdir = newdp;
983 			nrele++;
984 		}
985 		if (fdp->fd_rdir == olddp) {
986 			VREF(newdp);
987 			fdp->fd_rdir = newdp;
988 			nrele++;
989 		}
990 		FILEDESC_UNLOCK(fdp);
991 		mtx_unlock(&fdesc_mtx);
992 		while (nrele--)
993 			vrele(olddp);
994 	}
995 	sx_sunlock(&allproc_lock);
996 	if (rootvnode == olddp) {
997 		vrele(rootvnode);
998 		VREF(newdp);
999 		rootvnode = newdp;
1000 	}
1001 }
1002 
1003 /*
1004  * Unmount a filesystem.
1005  *
1006  * Note: unmount takes a path to the vnode mounted on as argument,
1007  * not special file (as before).
1008  */
1009 #ifndef _SYS_SYSPROTO_H_
1010 struct unmount_args {
1011 	char	*path;
1012 	int	flags;
1013 };
1014 #endif
1015 /* ARGSUSED */
1016 int
1017 unmount(td, uap)
1018 	struct thread *td;
1019 	register struct unmount_args /* {
1020 		char *path;
1021 		int flags;
1022 	} */ *uap;
1023 {
1024 	struct mount *mp;
1025 	char *pathbuf;
1026 	int error, id0, id1;
1027 
1028 	if (jailed(td->td_ucred))
1029 		return (EPERM);
1030 	if (usermount == 0) {
1031 		if ((error = suser(td)) != 0)
1032 			return (error);
1033 	}
1034 
1035 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1036 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1037 	if (error) {
1038 		free(pathbuf, M_TEMP);
1039 		return (error);
1040 	}
1041 	if (uap->flags & MNT_BYFSID) {
1042 		/* Decode the filesystem ID. */
1043 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1044 			free(pathbuf, M_TEMP);
1045 			return (EINVAL);
1046 		}
1047 
1048 		mtx_lock(&mountlist_mtx);
1049 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1050 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1051 			    mp->mnt_stat.f_fsid.val[1] == id1)
1052 				break;
1053 		}
1054 		mtx_unlock(&mountlist_mtx);
1055 	} else {
1056 		mtx_lock(&mountlist_mtx);
1057 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1058 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1059 				break;
1060 		}
1061 		mtx_unlock(&mountlist_mtx);
1062 	}
1063 	free(pathbuf, M_TEMP);
1064 	if (mp == NULL) {
1065 		/*
1066 		 * Previously we returned ENOENT for a nonexistent path and
1067 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1068 		 * now, so in the !MNT_BYFSID case return the more likely
1069 		 * EINVAL for compatibility.
1070 		 */
1071 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1072 	}
1073 
1074 	/*
1075 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1076 	 * original mount is permitted to unmount this filesystem.
1077 	 */
1078 	if ((mp->mnt_flag & MNT_USER) == 0 ||
1079 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
1080 		if ((error = suser(td)) != 0)
1081 			return (error);
1082 	}
1083 
1084 	/*
1085 	 * Don't allow unmounting the root filesystem.
1086 	 */
1087 	if (mp->mnt_flag & MNT_ROOTFS)
1088 		return (EINVAL);
1089 	return (dounmount(mp, uap->flags, td));
1090 }
1091 
1092 /*
1093  * Do the actual filesystem unmount.
1094  */
1095 int
1096 dounmount(mp, flags, td)
1097 	struct mount *mp;
1098 	int flags;
1099 	struct thread *td;
1100 {
1101 	struct vnode *coveredvp, *fsrootvp;
1102 	int error;
1103 	int async_flag;
1104 
1105 	mtx_lock(&mountlist_mtx);
1106 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1107 		mtx_unlock(&mountlist_mtx);
1108 		return (EBUSY);
1109 	}
1110 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1111 	/* Allow filesystems to detect that a forced unmount is in progress. */
1112 	if (flags & MNT_FORCE)
1113 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1114 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1115 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td);
1116 	if (error) {
1117 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1118 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1119 			wakeup(mp);
1120 		return (error);
1121 	}
1122 	vn_start_write(NULL, &mp, V_WAIT);
1123 
1124 	if (mp->mnt_flag & MNT_EXPUBLIC)
1125 		vfs_setpublicfs(NULL, NULL, NULL);
1126 
1127 	vfs_msync(mp, MNT_WAIT);
1128 	async_flag = mp->mnt_flag & MNT_ASYNC;
1129 	mp->mnt_flag &= ~MNT_ASYNC;
1130 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1131 	if (mp->mnt_syncer != NULL)
1132 		vrele(mp->mnt_syncer);
1133 	/*
1134 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1135 	 * vnode to the covered vnode.  For non-forced unmounts we want
1136 	 * such references to cause an EBUSY error.
1137 	 */
1138 	if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp) == 0) {
1139 		if (mp->mnt_vnodecovered != NULL)
1140 			checkdirs(fsrootvp, mp->mnt_vnodecovered);
1141 		if (fsrootvp == rootvnode) {
1142 			vrele(rootvnode);
1143 			rootvnode = NULL;
1144 		}
1145 		vput(fsrootvp);
1146 	}
1147 	if (((mp->mnt_flag & MNT_RDONLY) ||
1148 	     (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) ||
1149 	    (flags & MNT_FORCE)) {
1150 		error = VFS_UNMOUNT(mp, flags, td);
1151 	}
1152 	vn_finished_write(mp);
1153 	if (error) {
1154 		/* Undo cdir/rdir and rootvnode changes made above. */
1155 		if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp) == 0) {
1156 			if (mp->mnt_vnodecovered != NULL)
1157 				checkdirs(mp->mnt_vnodecovered, fsrootvp);
1158 			if (rootvnode == NULL) {
1159 				rootvnode = fsrootvp;
1160 				vref(rootvnode);
1161 			}
1162 			vput(fsrootvp);
1163 		}
1164 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1165 			(void) vfs_allocate_syncvnode(mp);
1166 		mtx_lock(&mountlist_mtx);
1167 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1168 		mp->mnt_flag |= async_flag;
1169 		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK,
1170 		    &mountlist_mtx, td);
1171 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1172 			wakeup(mp);
1173 		return (error);
1174 	}
1175 	mtx_lock(&mountlist_mtx);
1176 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1177 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1178 		coveredvp->v_mountedhere = NULL;
1179 	mtx_unlock(&mountlist_mtx);
1180 	vfs_mount_destroy(mp, td);
1181 	if (coveredvp != NULL)
1182 		vrele(coveredvp);
1183 	return (0);
1184 }
1185 
1186 /*
1187  * Lookup a filesystem type, and if found allocate and initialize
1188  * a mount structure for it.
1189  *
1190  * Devname is usually updated by mount(8) after booting.
1191  */
1192 int
1193 vfs_rootmountalloc(fstypename, devname, mpp)
1194 	char *fstypename;
1195 	char *devname;
1196 	struct mount **mpp;
1197 {
1198 	struct thread *td = curthread;	/* XXX */
1199 	struct vfsconf *vfsp;
1200 	struct mount *mp;
1201 	int error;
1202 
1203 	if (fstypename == NULL)
1204 		return (ENODEV);
1205 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1206 		if (!strcmp(vfsp->vfc_name, fstypename))
1207 			break;
1208 	if (vfsp == NULL)
1209 		return (ENODEV);
1210 	error = vfs_mount_alloc(NULLVP, vfsp, "/", td, &mp);
1211 	if (error)
1212 		return (error);
1213 	mp->mnt_flag |= MNT_RDONLY | MNT_ROOTFS;
1214 	strlcpy(mp->mnt_stat.f_mntfromname, devname, MNAMELEN);
1215 	*mpp = mp;
1216 	return (0);
1217 }
1218 
1219 /*
1220  * Find and mount the root filesystem
1221  */
1222 void
1223 vfs_mountroot(void)
1224 {
1225 	char *cp;
1226 	int error, i;
1227 
1228 	g_waitidle();
1229 
1230 	/*
1231 	 * The root filesystem information is compiled in, and we are
1232 	 * booted with instructions to use it.
1233 	 */
1234 #ifdef ROOTDEVNAME
1235 	if ((boothowto & RB_DFLTROOT) && !vfs_mountroot_try(ROOTDEVNAME))
1236 		return;
1237 #endif
1238 	/*
1239 	 * We are booted with instructions to prompt for the root filesystem,
1240 	 * or to use the compiled-in default when it doesn't exist.
1241 	 */
1242 	if (boothowto & (RB_DFLTROOT | RB_ASKNAME)) {
1243 		if (!vfs_mountroot_ask())
1244 			return;
1245 	}
1246 
1247 	/*
1248 	 * We've been given the generic "use CDROM as root" flag.  This is
1249 	 * necessary because one media may be used in many different
1250 	 * devices, so we need to search for them.
1251 	 */
1252 	if (boothowto & RB_CDROM) {
1253 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1254 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1255 				return;
1256 		}
1257 	}
1258 
1259 	/*
1260 	 * Try to use the value read by the loader from /etc/fstab, or
1261 	 * supplied via some other means.  This is the preferred
1262 	 * mechanism.
1263 	 */
1264 	if ((cp = getenv("vfs.root.mountfrom")) != NULL) {
1265 		error = vfs_mountroot_try(cp);
1266 		freeenv(cp);
1267 		if (!error)
1268 			return;
1269 	}
1270 
1271 	/*
1272 	 * Try values that may have been computed by the machine-dependant
1273 	 * legacy code.
1274 	 */
1275 	if (!vfs_mountroot_try(rootdevnames[0]))
1276 		return;
1277 	if (!vfs_mountroot_try(rootdevnames[1]))
1278 		return;
1279 
1280 	/*
1281 	 * If we have a compiled-in default, and haven't already tried it, try
1282 	 * it now.
1283 	 */
1284 #ifdef ROOTDEVNAME
1285 	if (!(boothowto & RB_DFLTROOT))
1286 		if (!vfs_mountroot_try(ROOTDEVNAME))
1287 			return;
1288 #endif
1289 
1290 	/*
1291 	 * Everything so far has failed, prompt on the console if we haven't
1292 	 * already tried that.
1293 	 */
1294 	if (!(boothowto & (RB_DFLTROOT | RB_ASKNAME)) && !vfs_mountroot_ask())
1295 		return;
1296 	panic("Root mount failed, startup aborted.");
1297 }
1298 
1299 /*
1300  * Mount (mountfrom) as the root filesystem.
1301  */
1302 static int
1303 vfs_mountroot_try(char *mountfrom)
1304 {
1305         struct mount	*mp;
1306 	char		*vfsname, *path;
1307 	const char	*devname;
1308 	int		error;
1309 	char		patt[32];
1310 	int		s;
1311 
1312 	vfsname = NULL;
1313 	path    = NULL;
1314 	mp      = NULL;
1315 	error   = EINVAL;
1316 
1317 	if (mountfrom == NULL)
1318 		return (error);		/* don't complain */
1319 
1320 	s = splcam();			/* Overkill, but annoying without it */
1321 	printf("Mounting root from %s\n", mountfrom);
1322 	splx(s);
1323 
1324 	/* parse vfs name and path */
1325 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1326 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1327 	vfsname[0] = path[0] = 0;
1328 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1329 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1330 		goto done;
1331 
1332 	/* allocate a root mount */
1333 	error = vfs_rootmountalloc(vfsname, path[0] != 0 ? path : ROOTNAME,
1334 	    &mp);
1335 	if (error != 0) {
1336 		printf("Can't allocate root mount for filesystem '%s': %d\n",
1337 		       vfsname, error);
1338 		goto done;
1339 	}
1340 
1341 	/* do our best to set rootdev */
1342 	if (path[0] != '\0' && setrootbyname(path))
1343 		printf("setrootbyname failed\n");
1344 
1345 	/* If the root device is a type "memory disk", mount RW */
1346 	if (rootdev != NODEV && devsw(rootdev) != NULL) {
1347 		devname = devtoname(rootdev);
1348 		if (devname[0] == 'm' && devname[1] == 'd')
1349 			mp->mnt_flag &= ~MNT_RDONLY;
1350 	}
1351 
1352 	error = VFS_MOUNT(mp, NULL, NULL, NULL, curthread);
1353 
1354 done:
1355 	if (vfsname != NULL)
1356 		free(vfsname, M_MOUNT);
1357 	if (path != NULL)
1358 		free(path, M_MOUNT);
1359 	if (error != 0) {
1360 		if (mp != NULL)
1361 			vfs_mount_destroy(mp, curthread);
1362 		printf("Root mount failed: %d\n", error);
1363 	} else {
1364 
1365 		/* register with list of mounted filesystems */
1366 		mtx_lock(&mountlist_mtx);
1367 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1368 		mtx_unlock(&mountlist_mtx);
1369 
1370 		/* sanity check system clock against root fs timestamp */
1371 		inittodr(mp->mnt_time);
1372 		vfs_unbusy(mp, curthread);
1373 		error = VFS_START(mp, 0, curthread);
1374 	}
1375 	return (error);
1376 }
1377 
1378 /*
1379  * Spin prompting on the console for a suitable root filesystem
1380  */
1381 static int
1382 vfs_mountroot_ask(void)
1383 {
1384 	char name[128];
1385 
1386 	for(;;) {
1387 		printf("\nManual root filesystem specification:\n");
1388 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1389 #if defined(__i386__) || defined(__ia64__)
1390 		printf("                       eg. ufs:da0s1a\n");
1391 #else
1392 		printf("                       eg. ufs:/dev/da0a\n");
1393 #endif
1394 		printf("  ?                  List valid disk boot devices\n");
1395 		printf("  <empty line>       Abort manual input\n");
1396 		printf("\nmountroot> ");
1397 		gets(name);
1398 		if (name[0] == '\0')
1399 			return (1);
1400 		if (name[0] == '?') {
1401 			printf("\nList of GEOM managed disk devices:\n  ");
1402 			g_dev_print();
1403 			continue;
1404 		}
1405 		if (!vfs_mountroot_try(name))
1406 			return (0);
1407 	}
1408 }
1409 
1410 /*
1411  * Local helper function for vfs_mountroot_ask.
1412  */
1413 static void
1414 gets(char *cp)
1415 {
1416 	char *lp;
1417 	int c;
1418 
1419 	lp = cp;
1420 	for (;;) {
1421 		printf("%c", c = cngetc() & 0177);
1422 		switch (c) {
1423 		case -1:
1424 		case '\n':
1425 		case '\r':
1426 			*lp++ = '\0';
1427 			return;
1428 		case '\b':
1429 		case '\177':
1430 			if (lp > cp) {
1431 				printf(" \b");
1432 				lp--;
1433 			}
1434 			continue;
1435 		case '#':
1436 			lp--;
1437 			if (lp < cp)
1438 				lp = cp;
1439 			continue;
1440 		case '@':
1441 		case 'u' & 037:
1442 			lp = cp;
1443 			printf("%c", '\n');
1444 			continue;
1445 		default:
1446 			*lp++ = c;
1447 		}
1448 	}
1449 }
1450 
1451 /*
1452  * Convert a given name to the dev_t of the disk-like device
1453  * it refers to.
1454  */
1455 dev_t
1456 getdiskbyname(char *name) {
1457 	char *cp;
1458 	dev_t dev;
1459 
1460 	cp = name;
1461 	if (!bcmp(cp, "/dev/", 5))
1462 		cp += 5;
1463 
1464 	dev = NODEV;
1465 	EVENTHANDLER_INVOKE(dev_clone, cp, strlen(cp), &dev);
1466 	return (dev);
1467 }
1468 
1469 /*
1470  * Set rootdev to match (name), given that we expect it to
1471  * refer to a disk-like device.
1472  */
1473 static int
1474 setrootbyname(char *name)
1475 {
1476 	dev_t diskdev;
1477 
1478 	diskdev = getdiskbyname(name);
1479 	if (diskdev != NODEV) {
1480 		rootdev = diskdev;
1481 		return (0);
1482 	}
1483 
1484 	return (1);
1485 }
1486 
1487 /* Show the dev_t for a disk specified by name */
1488 #ifdef DDB
1489 DB_SHOW_COMMAND(disk, db_getdiskbyname)
1490 {
1491 	dev_t dev;
1492 
1493 	if (modif[0] == '\0') {
1494 		db_error("usage: show disk/devicename");
1495 		return;
1496 	}
1497 	dev = getdiskbyname(modif);
1498 	if (dev != NODEV)
1499 		db_printf("dev_t = %p\n", dev);
1500 	else
1501 		db_printf("No disk device matched.\n");
1502 }
1503 #endif
1504 
1505 /*
1506  * Get a mount option by its name.
1507  *
1508  * Return 0 if the option was found, ENOENT otherwise.
1509  * If len is non-NULL it will be filled with the length
1510  * of the option. If buf is non-NULL, it will be filled
1511  * with the address of the option.
1512  */
1513 int
1514 vfs_getopt(opts, name, buf, len)
1515 	struct vfsoptlist *opts;
1516 	const char *name;
1517 	void **buf;
1518 	int *len;
1519 {
1520 	struct vfsopt *opt;
1521 
1522 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1523 
1524 	TAILQ_FOREACH(opt, opts, link) {
1525 		if (strcmp(name, opt->name) == 0) {
1526 			if (len != NULL)
1527 				*len = opt->len;
1528 			if (buf != NULL)
1529 				*buf = opt->value;
1530 			return (0);
1531 		}
1532 	}
1533 	return (ENOENT);
1534 }
1535 
1536 /*
1537  * Find and copy a mount option.
1538  *
1539  * The size of the buffer has to be specified
1540  * in len, if it is not the same length as the
1541  * mount option, EINVAL is returned.
1542  * Returns ENOENT if the option is not found.
1543  */
1544 int
1545 vfs_copyopt(opts, name, dest, len)
1546 	struct vfsoptlist *opts;
1547 	const char *name;
1548 	void *dest;
1549 	int len;
1550 {
1551 	struct vfsopt *opt;
1552 
1553 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1554 
1555 	TAILQ_FOREACH(opt, opts, link) {
1556 		if (strcmp(name, opt->name) == 0) {
1557 			if (len != opt->len)
1558 				return (EINVAL);
1559 			bcopy(opt->value, dest, opt->len);
1560 			return (0);
1561 		}
1562 	}
1563 	return (ENOENT);
1564 }
1565