xref: /freebsd/sys/kern/vfs_mount.c (revision 6949461a180c90c17d6f0adb3ff1ef76e3415bc3)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * Copyright (c) 1999 Michael Smith
35  * All rights reserved.
36  * Copyright (c) 1999 Poul-Henning Kamp
37  * All rights reserved.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  *
48  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  */
60 
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
63 
64 #include <sys/param.h>
65 #include <sys/conf.h>
66 #include <sys/cons.h>
67 #include <sys/jail.h>
68 #include <sys/kernel.h>
69 #include <sys/linker.h>
70 #include <sys/mac.h>
71 #include <sys/malloc.h>
72 #include <sys/mount.h>
73 #include <sys/mutex.h>
74 #include <sys/namei.h>
75 #include <sys/proc.h>
76 #include <sys/filedesc.h>
77 #include <sys/reboot.h>
78 #include <sys/sysproto.h>
79 #include <sys/sx.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysent.h>
82 #include <sys/systm.h>
83 #include <sys/vnode.h>
84 
85 #include <geom/geom.h>
86 
87 #include <machine/stdarg.h>
88 
89 #include "opt_rootdevname.h"
90 #include "opt_ddb.h"
91 #include "opt_mac.h"
92 
93 #ifdef DDB
94 #include <ddb/ddb.h>
95 #endif
96 
97 #define	ROOTNAME		"root_device"
98 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
99 
100 static void	checkdirs(struct vnode *olddp, struct vnode *newdp);
101 static struct cdev *getdiskbyname(char *_name);
102 static void	gets(char *cp);
103 static int	vfs_domount(struct thread *td, const char *fstype,
104 		    char *fspath, int fsflags, void *fsdata, int compat);
105 static int	vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
106 		    const char *fspath, struct thread *td, struct mount **mpp);
107 static int	vfs_mountroot_ask(void);
108 static int	vfs_mountroot_try(const char *mountfrom);
109 static int	vfs_donmount(struct thread *td, int fsflags,
110 		    struct uio *fsoptions);
111 
112 static int	usermount = 0;
113 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
114     "Unprivileged users may mount and unmount file systems");
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 struct cdev *rootdev = NULL;
153 #ifdef ROOTDEVNAME
154 const char	*ctrootdevname = ROOTDEVNAME;
155 #else
156 const char	*ctrootdevname = NULL;
157 #endif
158 
159 /*
160  * Has to be dynamic as the value of rootdev can change; however, it can't
161  * change after the root is mounted, so a user process can't access this
162  * sysctl until after the value is unchangeable.
163  */
164 static int
165 sysctl_rootdev(SYSCTL_HANDLER_ARGS)
166 {
167 	int error;
168 
169 	/* _RD prevents this from happening. */
170 	KASSERT(req->newptr == NULL, ("Attempt to change root device name"));
171 
172 	if (rootdev != NULL)
173 		error = sysctl_handle_string(oidp, rootdev->si_name, 0, req);
174 	else
175 		error = sysctl_handle_string(oidp, "", 0, req);
176 
177 	return (error);
178 }
179 
180 SYSCTL_PROC(_kern, OID_AUTO, rootdev, CTLTYPE_STRING | CTLFLAG_RD,
181     0, 0, sysctl_rootdev, "A", "Root file system device");
182 
183 /* Remove one mount option. */
184 static void
185 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
186 {
187 
188 	TAILQ_REMOVE(opts, opt, link);
189 	free(opt->name, M_MOUNT);
190 	if (opt->value != NULL)
191 		free(opt->value, M_MOUNT);
192 #ifdef INVARIANTS
193 	else if (opt->len != 0)
194 		panic("%s: mount option with NULL value but length != 0",
195 		    __func__);
196 #endif
197 	free(opt, M_MOUNT);
198 }
199 
200 /* Release all resources related to the mount options. */
201 static void
202 vfs_freeopts(struct vfsoptlist *opts)
203 {
204 	struct vfsopt *opt;
205 
206 	while (!TAILQ_EMPTY(opts)) {
207 		opt = TAILQ_FIRST(opts);
208 		vfs_freeopt(opts, opt);
209 	}
210 	free(opts, M_MOUNT);
211 }
212 
213 /*
214  * Check if options are equal (with or without the "no" prefix).
215  */
216 static int
217 vfs_equalopts(const char *opt1, const char *opt2)
218 {
219 
220 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
221 	if (strcmp(opt1, opt2) == 0)
222 		return (1);
223 	/* "noopt" vs. "opt" */
224 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
225 		return (1);
226 	/* "opt" vs. "noopt" */
227 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
228 		return (1);
229 	return (0);
230 }
231 
232 /*
233  * If a mount option is specified several times,
234  * (with or without the "no" prefix) only keep
235  * the last occurence of it.
236  */
237 static void
238 vfs_sanitizeopts(struct vfsoptlist *opts)
239 {
240 	struct vfsopt *opt, *opt2, *tmp;
241 
242 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
243 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
244 		while (opt2 != NULL) {
245 			if (vfs_equalopts(opt->name, opt2->name)) {
246 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
247 				vfs_freeopt(opts, opt2);
248 				opt2 = tmp;
249 			} else {
250 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
251 			}
252 		}
253 	}
254 }
255 
256 /*
257  * Build a linked list of mount options from a struct uio.
258  */
259 static int
260 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
261 {
262 	struct vfsoptlist *opts;
263 	struct vfsopt *opt;
264 	size_t memused;
265 	unsigned int i, iovcnt;
266 	int error, namelen, optlen;
267 
268 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
269 	TAILQ_INIT(opts);
270 	memused = 0;
271 	iovcnt = auio->uio_iovcnt;
272 	for (i = 0; i < iovcnt; i += 2) {
273 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
274 		namelen = auio->uio_iov[i].iov_len;
275 		optlen = auio->uio_iov[i + 1].iov_len;
276 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
277 		opt->value = NULL;
278 		opt->len = 0;
279 
280 		/*
281 		 * Do this early, so jumps to "bad" will free the current
282 		 * option.
283 		 */
284 		TAILQ_INSERT_TAIL(opts, opt, link);
285 		memused += sizeof(struct vfsopt) + optlen + namelen;
286 
287 		/*
288 		 * Avoid consuming too much memory, and attempts to overflow
289 		 * memused.
290 		 */
291 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
292 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
293 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
294 			error = EINVAL;
295 			goto bad;
296 		}
297 
298 		if (auio->uio_segflg == UIO_SYSSPACE) {
299 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
300 		} else {
301 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
302 			    namelen);
303 			if (error)
304 				goto bad;
305 		}
306 		/* Ensure names are null-terminated strings. */
307 		if (opt->name[namelen - 1] != '\0') {
308 			error = EINVAL;
309 			goto bad;
310 		}
311 		if (optlen != 0) {
312 			opt->len = optlen;
313 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
314 			if (auio->uio_segflg == UIO_SYSSPACE) {
315 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
316 				    optlen);
317 			} else {
318 				error = copyin(auio->uio_iov[i + 1].iov_base,
319 				    opt->value, optlen);
320 				if (error)
321 					goto bad;
322 			}
323 		}
324 	}
325 	vfs_sanitizeopts(opts);
326 	*options = opts;
327 	return (0);
328 bad:
329 	vfs_freeopts(opts);
330 	return (error);
331 }
332 
333 /*
334  * Merge the old mount options with the new ones passed
335  * in the MNT_UPDATE case.
336  */
337 static void
338 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
339 {
340 	struct vfsopt *opt, *opt2, *new;
341 
342 	TAILQ_FOREACH(opt, opts, link) {
343 		/*
344 		 * Check that this option hasn't been redefined
345 		 * nor cancelled with a "no" mount option.
346 		 */
347 		opt2 = TAILQ_FIRST(toopts);
348 		while (opt2 != NULL) {
349 			if (strcmp(opt2->name, opt->name) == 0)
350 				goto next;
351 			if (strncmp(opt2->name, "no", 2) == 0 &&
352 			    strcmp(opt2->name + 2, opt->name) == 0) {
353 				vfs_freeopt(toopts, opt2);
354 				goto next;
355 			}
356 			opt2 = TAILQ_NEXT(opt2, link);
357 		}
358 		/* We want this option, duplicate it. */
359 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
360 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
361 		strcpy(new->name, opt->name);
362 		if (opt->len != 0) {
363 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
364 			bcopy(opt->value, new->value, opt->len);
365 		} else {
366 			new->value = NULL;
367 		}
368 		new->len = opt->len;
369 		TAILQ_INSERT_TAIL(toopts, new, link);
370 next:
371 		continue;
372 	}
373 }
374 
375 /*
376  * New mount API.
377  */
378 int
379 nmount(td, uap)
380 	struct thread *td;
381 	struct nmount_args /* {
382 		struct iovec *iovp;
383 		unsigned int iovcnt;
384 		int flags;
385 	} */ *uap;
386 {
387 	struct uio *auio;
388 	struct iovec *iov;
389 	unsigned int i;
390 	int error;
391 	u_int iovcnt;
392 
393 	iovcnt = uap->iovcnt;
394 	/*
395 	 * Check that we have an even number of iovec's
396 	 * and that we have at least two options.
397 	 */
398 	if ((iovcnt & 1) || (iovcnt < 4))
399 		return (EINVAL);
400 	error = copyinuio(uap->iovp, iovcnt, &auio);
401 	if (error)
402 		return (error);
403 	iov = auio->uio_iov;
404 	for (i = 0; i < iovcnt; i++) {
405 		if (iov->iov_len > MMAXOPTIONLEN) {
406 			free(auio, M_IOV);
407 			return (EINVAL);
408 		}
409 		iov++;
410 	}
411 	error = vfs_donmount(td, uap->flags, auio);
412 	free(auio, M_IOV);
413 	return (error);
414 }
415 
416 int
417 kernel_mount(struct iovec *iovp, u_int iovcnt, int flags)
418 {
419 	struct uio auio;
420 	int error;
421 
422 	/*
423 	 * Check that we have an even number of iovec's
424 	 * and that we have at least two options.
425 	 */
426 	if ((iovcnt & 1) || (iovcnt < 4))
427 		return (EINVAL);
428 
429 	auio.uio_iov = iovp;
430 	auio.uio_iovcnt = iovcnt;
431 	auio.uio_segflg = UIO_SYSSPACE;
432 
433 	error = vfs_donmount(curthread, flags, &auio);
434 	return (error);
435 }
436 
437 int
438 kernel_vmount(int flags, ...)
439 {
440 	struct iovec *iovp;
441 	struct uio auio;
442 	va_list ap;
443 	u_int iovcnt, iovlen, len;
444 	const char *cp;
445 	char *buf, *pos;
446 	size_t n;
447 	int error, i;
448 
449 	len = 0;
450 	va_start(ap, flags);
451 	for (iovcnt = 0; (cp = va_arg(ap, const char *)) != NULL; iovcnt++)
452 		len += strlen(cp) + 1;
453 	va_end(ap);
454 
455 	if (iovcnt < 4 || iovcnt & 1)
456 		return (EINVAL);
457 
458 	iovlen = iovcnt * sizeof (struct iovec);
459 	MALLOC(iovp, struct iovec *, iovlen, M_MOUNT, M_WAITOK);
460 	MALLOC(buf, char *, len, M_MOUNT, M_WAITOK);
461 	pos = buf;
462 	va_start(ap, flags);
463 	for (i = 0; i < iovcnt; i++) {
464 		cp = va_arg(ap, const char *);
465 		copystr(cp, pos, len - (pos - buf), &n);
466 		iovp[i].iov_base = pos;
467 		iovp[i].iov_len = n;
468 		pos += n;
469 	}
470 	va_end(ap);
471 
472 	auio.uio_iov = iovp;
473 	auio.uio_iovcnt = iovcnt;
474 	auio.uio_segflg = UIO_SYSSPACE;
475 
476 	error = vfs_donmount(curthread, flags, &auio);
477 	FREE(iovp, M_MOUNT);
478 	FREE(buf, M_MOUNT);
479 	return (error);
480 }
481 
482 /*
483  * Allocate and initialize the mount point struct.
484  */
485 static int
486 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
487     const char *fspath, struct thread *td, struct mount **mpp)
488 {
489 	struct mount *mp;
490 
491 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
492 	TAILQ_INIT(&mp->mnt_nvnodelist);
493 	mp->mnt_nvnodelistsize = 0;
494 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
495 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
496 	vfs_busy(mp, LK_NOWAIT, 0, td);
497 	mp->mnt_op = vfsp->vfc_vfsops;
498 	mp->mnt_vfc = vfsp;
499 	vfsp->vfc_refcount++;
500 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
501 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
502 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
503 	mp->mnt_vnodecovered = vp;
504 	mp->mnt_cred = crdup(td->td_ucred);
505 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
506 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
507 	mp->mnt_iosize_max = DFLTPHYS;
508 #ifdef MAC
509 	mac_init_mount(mp);
510 	mac_create_mount(td->td_ucred, mp);
511 #endif
512 	*mpp = mp;
513 	return (0);
514 }
515 
516 /*
517  * Destroy the mount struct previously allocated by vfs_mount_alloc().
518  */
519 void
520 vfs_mount_destroy(struct mount *mp, struct thread *td)
521 {
522 
523 	mp->mnt_vfc->vfc_refcount--;
524 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
525 		panic("unmount: dangling vnode");
526 	vfs_unbusy(mp,td);
527 	lockdestroy(&mp->mnt_lock);
528 	mtx_destroy(&mp->mnt_mtx);
529 	if (mp->mnt_kern_flag & MNTK_MWAIT)
530 		wakeup(mp);
531 #ifdef MAC
532 	mac_destroy_mount(mp);
533 #endif
534 	if (mp->mnt_opt != NULL)
535 		vfs_freeopts(mp->mnt_opt);
536 	crfree(mp->mnt_cred);
537 	free(mp, M_MOUNT);
538 }
539 
540 static int
541 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
542 {
543 	struct vfsoptlist *optlist;
544 	char *fstype, *fspath;
545 	int error, fstypelen, fspathlen;
546 
547 	error = vfs_buildopts(fsoptions, &optlist);
548 	if (error)
549 		return (error);
550 
551 	/*
552 	 * We need these two options before the others,
553 	 * and they are mandatory for any filesystem.
554 	 * Ensure they are NUL terminated as well.
555 	 */
556 	fstypelen = 0;
557 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
558 	if (error || fstype[fstypelen - 1] != '\0') {
559 		error = EINVAL;
560 		goto bail;
561 	}
562 	fspathlen = 0;
563 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
564 	if (error || fspath[fspathlen - 1] != '\0') {
565 		error = EINVAL;
566 		goto bail;
567 	}
568 
569 	/*
570 	 * Be ultra-paranoid about making sure the type and fspath
571 	 * variables will fit in our mp buffers, including the
572 	 * terminating NUL.
573 	 */
574 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
575 		error = ENAMETOOLONG;
576 		goto bail;
577 	}
578 
579 	error = vfs_domount(td, fstype, fspath, fsflags, optlist, 0);
580 bail:
581 	if (error)
582 		vfs_freeopts(optlist);
583 	return (error);
584 }
585 
586 /*
587  * Old mount API.
588  */
589 #ifndef _SYS_SYSPROTO_H_
590 struct mount_args {
591 	char	*type;
592 	char	*path;
593 	int	flags;
594 	caddr_t	data;
595 };
596 #endif
597 /* ARGSUSED */
598 int
599 mount(td, uap)
600 	struct thread *td;
601 	struct mount_args /* {
602 		char *type;
603 		char *path;
604 		int flags;
605 		caddr_t data;
606 	} */ *uap;
607 {
608 	char *fstype;
609 	char *fspath;
610 	int error;
611 
612 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
613 	fspath = malloc(MNAMELEN, M_TEMP, M_WAITOK);
614 
615 	/*
616 	 * vfs_mount() actually takes a kernel string for `type' and
617 	 * `path' now, so extract them.
618 	 */
619 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
620 	if (error == 0)
621 		error = copyinstr(uap->path, fspath, MNAMELEN, NULL);
622 	if (error == 0)
623 		error = vfs_domount(td, fstype, fspath, uap->flags,
624 		    uap->data, 1);
625 	free(fstype, M_TEMP);
626 	free(fspath, M_TEMP);
627 	return (error);
628 }
629 
630 /*
631  * vfs_mount(): actually attempt a filesystem mount.
632  *
633  * This routine is designed to be a "generic" entry point for routines
634  * that wish to mount a filesystem. All parameters except `fsdata' are
635  * pointers into kernel space. `fsdata' is currently still a pointer
636  * into userspace.
637  */
638 int
639 vfs_mount(td, fstype, fspath, fsflags, fsdata)
640 	struct thread *td;
641 	const char *fstype;
642 	char *fspath;
643 	int fsflags;
644 	void *fsdata;
645 {
646 
647 	return (vfs_domount(td, fstype, fspath, fsflags, fsdata, 1));
648 }
649 
650 /*
651  * vfs_domount(): actually attempt a filesystem mount.
652  */
653 static int
654 vfs_domount(
655 	struct thread *td,	/* Flags common to all filesystems. */
656 	const char *fstype,	/* Filesystem type. */
657 	char *fspath,		/* Mount path. */
658 	int fsflags,		/* Flags common to all filesystems. */
659 	void *fsdata,		/* Options local to the filesystem. */
660 	int compat		/* Invocation from compat syscall. */
661 	)
662 {
663 	linker_file_t lf;
664 	struct vnode *vp;
665 	struct mount *mp;
666 	struct vfsconf *vfsp;
667 	int error, flag = 0, kern_flag = 0;
668 	struct vattr va;
669 	struct nameidata nd;
670 
671 	/*
672 	 * Be ultra-paranoid about making sure the type and fspath
673 	 * variables will fit in our mp buffers, including the
674 	 * terminating NUL.
675 	 */
676 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
677 		return (ENAMETOOLONG);
678 
679 	if (jailed(td->td_ucred))
680 		return (EPERM);
681 	if (usermount == 0) {
682 		if ((error = suser(td)) != 0)
683 			return (error);
684 	}
685 
686 	/*
687 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
688 	 */
689 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
690 		if ((error = suser(td)) != 0)
691 			return (error);
692 	}
693 	/*
694 	 * Silently enforce MNT_NODEV, MNT_NOSUID and MNT_USER for
695 	 * unprivileged users.
696 	 */
697 	if (suser(td) != 0)
698 		fsflags |= MNT_NODEV | MNT_NOSUID | MNT_USER;
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 privileged root, or (if MNT_USER is set) the user that
726 		 * did the original mount is permitted to update it.
727 		 */
728 		error = vfs_suser(mp, td);
729 		if (error) {
730 			vput(vp);
731 			return (error);
732 		}
733 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
734 			vput(vp);
735 			return (EBUSY);
736 		}
737 		VI_LOCK(vp);
738 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
739 		    vp->v_mountedhere != NULL) {
740 			VI_UNLOCK(vp);
741 			vfs_unbusy(mp, td);
742 			vput(vp);
743 			return (EBUSY);
744 		}
745 		vp->v_iflag |= VI_MOUNT;
746 		VI_UNLOCK(vp);
747 		mp->mnt_flag |= fsflags &
748 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT);
749 		VOP_UNLOCK(vp, 0, td);
750 		if (compat == 0) {
751 			mp->mnt_optnew = fsdata;
752 			vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
753 		}
754 	} else {
755 		/*
756 		 * If the user is not root, ensure that they own the directory
757 		 * onto which we are attempting to mount.
758 		 */
759 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
760 		if (error) {
761 			vput(vp);
762 			return (error);
763 		}
764 		if (va.va_uid != td->td_ucred->cr_uid) {
765 			if ((error = suser(td)) != 0) {
766 				vput(vp);
767 				return (error);
768 			}
769 		}
770 		if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) {
771 			vput(vp);
772 			return (error);
773 		}
774 		if (vp->v_type != VDIR) {
775 			vput(vp);
776 			return (ENOTDIR);
777 		}
778 		vfsp = vfs_byname(fstype);
779 		if (vfsp == NULL) {
780 			/* Only load modules for root (very important!). */
781 			if ((error = suser(td)) != 0) {
782 				vput(vp);
783 				return (error);
784 			}
785 			error = securelevel_gt(td->td_ucred, 0);
786 			if (error) {
787 				vput(vp);
788 				return (error);
789 			}
790 			error = linker_load_module(NULL, fstype, NULL, NULL, &lf);
791 			if (error || lf == NULL) {
792 				vput(vp);
793 				if (lf == NULL)
794 					error = ENODEV;
795 				return (error);
796 			}
797 			lf->userrefs++;
798 			/* Look up again to see if the VFS was loaded. */
799 			vfsp = vfs_byname(fstype);
800 			if (vfsp == NULL) {
801 				lf->userrefs--;
802 				linker_file_unload(lf, LINKER_UNLOAD_FORCE);
803 				vput(vp);
804 				return (ENODEV);
805 			}
806 		}
807 		VI_LOCK(vp);
808 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
809 		    vp->v_mountedhere != NULL) {
810 			VI_UNLOCK(vp);
811 			vput(vp);
812 			return (EBUSY);
813 		}
814 		vp->v_iflag |= VI_MOUNT;
815 		VI_UNLOCK(vp);
816 
817 		/*
818 		 * Allocate and initialize the filesystem.
819 		 */
820 		error = vfs_mount_alloc(vp, vfsp, fspath, td, &mp);
821 		if (error) {
822 			vput(vp);
823 			return (error);
824 		}
825 		VOP_UNLOCK(vp, 0, td);
826 
827 		/* XXXMAC: pass to vfs_mount_alloc? */
828 		if (compat == 0)
829 			mp->mnt_optnew = fsdata;
830 	}
831 	/*
832 	 * Check if the fs implements the type VFS_[O]MOUNT()
833 	 * function we are looking for.
834 	 */
835 	if ((compat == 0) == (mp->mnt_op->vfs_omount != NULL)) {
836 		printf("%s doesn't support the %s mount syscall\n",
837 		    mp->mnt_vfc->vfc_name, compat ? "old" : "new");
838 		VI_LOCK(vp);
839 		vp->v_iflag &= ~VI_MOUNT;
840 		VI_UNLOCK(vp);
841 		if (mp->mnt_flag & MNT_UPDATE)
842 			vfs_unbusy(mp, td);
843 		else
844 			vfs_mount_destroy(mp, td);
845 		vrele(vp);
846 		return (EOPNOTSUPP);
847 	}
848 
849 	/*
850 	 * Set the mount level flags.
851 	 */
852 	if (fsflags & MNT_RDONLY)
853 		mp->mnt_flag |= MNT_RDONLY;
854 	else if (mp->mnt_flag & MNT_RDONLY)
855 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
856 	mp->mnt_flag &=~ MNT_UPDATEMASK;
857 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE);
858 	/*
859 	 * Mount the filesystem.
860 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
861 	 * get.  No freeing of cn_pnbuf.
862 	 */
863 	if (compat)
864 	    error = VFS_OMOUNT(mp, fspath, fsdata, td);
865 	else
866 	    error = VFS_MOUNT(mp, td);
867 	if (!error) {
868 		if (mp->mnt_opt != NULL)
869 			vfs_freeopts(mp->mnt_opt);
870 		mp->mnt_opt = mp->mnt_optnew;
871 	}
872 	/*
873 	 * Prevent external consumers of mount options from reading
874 	 * mnt_optnew.
875 	*/
876 	mp->mnt_optnew = NULL;
877 	if (mp->mnt_flag & MNT_UPDATE) {
878 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
879 			mp->mnt_flag &= ~MNT_RDONLY;
880 		mp->mnt_flag &=
881 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
882 		mp->mnt_kern_flag &= ~MNTK_WANTRDWR;
883 		if (error) {
884 			mp->mnt_flag = flag;
885 			mp->mnt_kern_flag = kern_flag;
886 		}
887 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
888 			if (mp->mnt_syncer == NULL)
889 				error = vfs_allocate_syncvnode(mp);
890 		} else {
891 			if (mp->mnt_syncer != NULL)
892 				vrele(mp->mnt_syncer);
893 			mp->mnt_syncer = NULL;
894 		}
895 		vfs_unbusy(mp, td);
896 		VI_LOCK(vp);
897 		vp->v_iflag &= ~VI_MOUNT;
898 		VI_UNLOCK(vp);
899 		vrele(vp);
900 		return (error);
901 	}
902 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
903 	/*
904 	 * Put the new filesystem on the mount list after root.
905 	 */
906 	cache_purge(vp);
907 	if (!error) {
908 		struct vnode *newdp;
909 
910 		VI_LOCK(vp);
911 		vp->v_iflag &= ~VI_MOUNT;
912 		VI_UNLOCK(vp);
913 		vp->v_mountedhere = mp;
914 		mtx_lock(&mountlist_mtx);
915 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
916 		mtx_unlock(&mountlist_mtx);
917 		vfs_event_signal(NULL, VQ_MOUNT, 0);
918 		if (VFS_ROOT(mp, &newdp, td))
919 			panic("mount: lost mount");
920 		checkdirs(vp, newdp);
921 		vput(newdp);
922 		VOP_UNLOCK(vp, 0, td);
923 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
924 			error = vfs_allocate_syncvnode(mp);
925 		vfs_unbusy(mp, td);
926 		if (error || (error = VFS_START(mp, 0, td)) != 0)
927 			vrele(vp);
928 	} else {
929 		VI_LOCK(vp);
930 		vp->v_iflag &= ~VI_MOUNT;
931 		VI_UNLOCK(vp);
932 		vfs_mount_destroy(mp, td);
933 		vput(vp);
934 	}
935 	return (error);
936 }
937 
938 /*
939  * Scan all active processes to see if any of them have a current
940  * or root directory of `olddp'. If so, replace them with the new
941  * mount point.
942  */
943 static void
944 checkdirs(olddp, newdp)
945 	struct vnode *olddp, *newdp;
946 {
947 	struct filedesc *fdp;
948 	struct proc *p;
949 	int nrele;
950 
951 	if (vrefcnt(olddp) == 1)
952 		return;
953 	sx_slock(&allproc_lock);
954 	LIST_FOREACH(p, &allproc, p_list) {
955 		mtx_lock(&fdesc_mtx);
956 		fdp = p->p_fd;
957 		if (fdp == NULL) {
958 			mtx_unlock(&fdesc_mtx);
959 			continue;
960 		}
961 		nrele = 0;
962 		FILEDESC_LOCK(fdp);
963 		if (fdp->fd_cdir == olddp) {
964 			VREF(newdp);
965 			fdp->fd_cdir = newdp;
966 			nrele++;
967 		}
968 		if (fdp->fd_rdir == olddp) {
969 			VREF(newdp);
970 			fdp->fd_rdir = newdp;
971 			nrele++;
972 		}
973 		FILEDESC_UNLOCK(fdp);
974 		mtx_unlock(&fdesc_mtx);
975 		while (nrele--)
976 			vrele(olddp);
977 	}
978 	sx_sunlock(&allproc_lock);
979 	if (rootvnode == olddp) {
980 		vrele(rootvnode);
981 		VREF(newdp);
982 		rootvnode = newdp;
983 	}
984 }
985 
986 /*
987  * Unmount a filesystem.
988  *
989  * Note: unmount takes a path to the vnode mounted on as argument,
990  * not special file (as before).
991  */
992 #ifndef _SYS_SYSPROTO_H_
993 struct unmount_args {
994 	char	*path;
995 	int	flags;
996 };
997 #endif
998 /* ARGSUSED */
999 int
1000 unmount(td, uap)
1001 	struct thread *td;
1002 	register struct unmount_args /* {
1003 		char *path;
1004 		int flags;
1005 	} */ *uap;
1006 {
1007 	struct mount *mp;
1008 	char *pathbuf;
1009 	int error, id0, id1;
1010 
1011 	if (jailed(td->td_ucred))
1012 		return (EPERM);
1013 	if (usermount == 0) {
1014 		if ((error = suser(td)) != 0)
1015 			return (error);
1016 	}
1017 
1018 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1019 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1020 	if (error) {
1021 		free(pathbuf, M_TEMP);
1022 		return (error);
1023 	}
1024 	if (uap->flags & MNT_BYFSID) {
1025 		/* Decode the filesystem ID. */
1026 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1027 			free(pathbuf, M_TEMP);
1028 			return (EINVAL);
1029 		}
1030 
1031 		mtx_lock(&mountlist_mtx);
1032 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1033 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1034 			    mp->mnt_stat.f_fsid.val[1] == id1)
1035 				break;
1036 		}
1037 		mtx_unlock(&mountlist_mtx);
1038 	} else {
1039 		mtx_lock(&mountlist_mtx);
1040 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1041 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1042 				break;
1043 		}
1044 		mtx_unlock(&mountlist_mtx);
1045 	}
1046 	free(pathbuf, M_TEMP);
1047 	if (mp == NULL) {
1048 		/*
1049 		 * Previously we returned ENOENT for a nonexistent path and
1050 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1051 		 * now, so in the !MNT_BYFSID case return the more likely
1052 		 * EINVAL for compatibility.
1053 		 */
1054 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1055 	}
1056 
1057 	/*
1058 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1059 	 * original mount is permitted to unmount this filesystem.
1060 	 */
1061 	error = vfs_suser(mp, td);
1062 	if (error)
1063 		return (error);
1064 
1065 	/*
1066 	 * Don't allow unmounting the root filesystem.
1067 	 */
1068 	if (mp->mnt_flag & MNT_ROOTFS)
1069 		return (EINVAL);
1070 	return (dounmount(mp, uap->flags, td));
1071 }
1072 
1073 /*
1074  * Do the actual filesystem unmount.
1075  */
1076 int
1077 dounmount(mp, flags, td)
1078 	struct mount *mp;
1079 	int flags;
1080 	struct thread *td;
1081 {
1082 	struct vnode *coveredvp, *fsrootvp;
1083 	int error;
1084 	int async_flag;
1085 
1086 	mtx_lock(&mountlist_mtx);
1087 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1088 		mtx_unlock(&mountlist_mtx);
1089 		return (EBUSY);
1090 	}
1091 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1092 	/* Allow filesystems to detect that a forced unmount is in progress. */
1093 	if (flags & MNT_FORCE)
1094 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1095 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1096 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td);
1097 	if (error) {
1098 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1099 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1100 			wakeup(mp);
1101 		return (error);
1102 	}
1103 	vn_start_write(NULL, &mp, V_WAIT);
1104 
1105 	if (mp->mnt_flag & MNT_EXPUBLIC)
1106 		vfs_setpublicfs(NULL, NULL, NULL);
1107 
1108 	vfs_msync(mp, MNT_WAIT);
1109 	async_flag = mp->mnt_flag & MNT_ASYNC;
1110 	mp->mnt_flag &= ~MNT_ASYNC;
1111 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1112 	if (mp->mnt_syncer != NULL)
1113 		vrele(mp->mnt_syncer);
1114 	/*
1115 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1116 	 * vnode to the covered vnode.  For non-forced unmounts we want
1117 	 * such references to cause an EBUSY error.
1118 	 */
1119 	if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) {
1120 		if (mp->mnt_vnodecovered != NULL)
1121 			checkdirs(fsrootvp, mp->mnt_vnodecovered);
1122 		if (fsrootvp == rootvnode) {
1123 			vrele(rootvnode);
1124 			rootvnode = NULL;
1125 		}
1126 		vput(fsrootvp);
1127 	}
1128 	if (((mp->mnt_flag & MNT_RDONLY) ||
1129 	     (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) ||
1130 	    (flags & MNT_FORCE)) {
1131 		error = VFS_UNMOUNT(mp, flags, td);
1132 	}
1133 	vn_finished_write(mp);
1134 	if (error) {
1135 		/* Undo cdir/rdir and rootvnode changes made above. */
1136 		if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) {
1137 			if (mp->mnt_vnodecovered != NULL)
1138 				checkdirs(mp->mnt_vnodecovered, fsrootvp);
1139 			if (rootvnode == NULL) {
1140 				rootvnode = fsrootvp;
1141 				vref(rootvnode);
1142 			}
1143 			vput(fsrootvp);
1144 		}
1145 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1146 			(void) vfs_allocate_syncvnode(mp);
1147 		mtx_lock(&mountlist_mtx);
1148 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1149 		mp->mnt_flag |= async_flag;
1150 		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK,
1151 		    &mountlist_mtx, td);
1152 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1153 			wakeup(mp);
1154 		return (error);
1155 	}
1156 	mtx_lock(&mountlist_mtx);
1157 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1158 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1159 		coveredvp->v_mountedhere = NULL;
1160 	mtx_unlock(&mountlist_mtx);
1161 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1162 	vfs_mount_destroy(mp, td);
1163 	if (coveredvp != NULL)
1164 		vrele(coveredvp);
1165 	return (0);
1166 }
1167 
1168 /*
1169  * Lookup a filesystem type, and if found allocate and initialize
1170  * a mount structure for it.
1171  *
1172  * Devname is usually updated by mount(8) after booting.
1173  */
1174 int
1175 vfs_rootmountalloc(fstypename, devname, mpp)
1176 	char *fstypename;
1177 	char *devname;
1178 	struct mount **mpp;
1179 {
1180 	struct thread *td = curthread;	/* XXX */
1181 	struct vfsconf *vfsp;
1182 	struct mount *mp;
1183 	int error;
1184 
1185 	if (fstypename == NULL)
1186 		return (ENODEV);
1187 	vfsp = vfs_byname(fstypename);
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, asked = 0;
1207 
1208 
1209 	/*
1210 	 * Wait for GEOM to settle down
1211 	 */
1212 	DROP_GIANT();
1213 	g_waitidle();
1214 	PICKUP_GIANT();
1215 
1216 	/*
1217 	 * We are booted with instructions to prompt for the root filesystem.
1218 	 */
1219 	if (boothowto & RB_ASKNAME) {
1220 		if (!vfs_mountroot_ask())
1221 			return;
1222 		asked = 1;
1223 	}
1224 
1225 	/*
1226 	 * The root filesystem information is compiled in, and we are
1227 	 * booted with instructions to use it.
1228 	 */
1229 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1230 		if (!vfs_mountroot_try(ctrootdevname))
1231 			return;
1232 		ctrootdevname = NULL;
1233 	}
1234 
1235 	/*
1236 	 * We've been given the generic "use CDROM as root" flag.  This is
1237 	 * necessary because one media may be used in many different
1238 	 * devices, so we need to search for them.
1239 	 */
1240 	if (boothowto & RB_CDROM) {
1241 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1242 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1243 				return;
1244 		}
1245 	}
1246 
1247 	/*
1248 	 * Try to use the value read by the loader from /etc/fstab, or
1249 	 * supplied via some other means.  This is the preferred
1250 	 * mechanism.
1251 	 */
1252 	cp = getenv("vfs.root.mountfrom");
1253 	if (cp != NULL) {
1254 		error = vfs_mountroot_try(cp);
1255 		freeenv(cp);
1256 		if (!error)
1257 			return;
1258 	}
1259 
1260 	/*
1261 	 * Try values that may have been computed by code during boot
1262 	 */
1263 	if (!vfs_mountroot_try(rootdevnames[0]))
1264 		return;
1265 	if (!vfs_mountroot_try(rootdevnames[1]))
1266 		return;
1267 
1268 	/*
1269 	 * If we (still) have a compiled-in default, try it.
1270 	 */
1271 	if (ctrootdevname != NULL)
1272 		if (!vfs_mountroot_try(ctrootdevname))
1273 			return;
1274 
1275 	/*
1276 	 * Everything so far has failed, prompt on the console if we haven't
1277 	 * already tried that.
1278 	 */
1279 	if (!asked)
1280 		if (!vfs_mountroot_ask())
1281 			return;
1282 	panic("Root mount failed, startup aborted.");
1283 }
1284 
1285 /*
1286  * Mount (mountfrom) as the root filesystem.
1287  */
1288 static int
1289 vfs_mountroot_try(const char *mountfrom)
1290 {
1291         struct mount	*mp;
1292 	char		*vfsname, *path;
1293 	int		error;
1294 	char		patt[32];
1295 	int		s;
1296 
1297 	vfsname = NULL;
1298 	path    = NULL;
1299 	mp      = NULL;
1300 	error   = EINVAL;
1301 
1302 	if (mountfrom == NULL)
1303 		return (error);		/* don't complain */
1304 
1305 	s = splcam();			/* Overkill, but annoying without it */
1306 	printf("Mounting root from %s\n", mountfrom);
1307 	splx(s);
1308 
1309 	/* parse vfs name and path */
1310 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1311 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1312 	vfsname[0] = path[0] = 0;
1313 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1314 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1315 		goto done;
1316 
1317 	/* allocate a root mount */
1318 	error = vfs_rootmountalloc(vfsname, path[0] != 0 ? path : ROOTNAME,
1319 	    &mp);
1320 	if (error != 0) {
1321 		printf("Can't allocate root mount for filesystem '%s': %d\n",
1322 		       vfsname, error);
1323 		goto done;
1324 	}
1325 
1326 	/*
1327 	 * do our best to set rootdev
1328 	 * XXX: This does not belong here!
1329 	 */
1330 	if (path[0] != '\0') {
1331 		struct cdev *diskdev;
1332 		diskdev = getdiskbyname(path);
1333 		if (diskdev != NULL)
1334 			rootdev = diskdev;
1335 		else
1336 			printf("setrootbyname failed\n");
1337 	}
1338 
1339 	error = VFS_OMOUNT(mp, path, NULL, curthread);
1340 
1341 done:
1342 	if (vfsname != NULL)
1343 		free(vfsname, M_MOUNT);
1344 	if (path != NULL)
1345 		free(path, M_MOUNT);
1346 	if (error != 0) {
1347 		if (mp != NULL)
1348 			vfs_mount_destroy(mp, curthread);
1349 		printf("Root mount failed: %d\n", error);
1350 	} else {
1351 
1352 		/* register with list of mounted filesystems */
1353 		mtx_lock(&mountlist_mtx);
1354 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1355 		mtx_unlock(&mountlist_mtx);
1356 
1357 		/* sanity check system clock against root fs timestamp */
1358 		inittodr(mp->mnt_time);
1359 		vfs_unbusy(mp, curthread);
1360 		error = VFS_START(mp, 0, curthread);
1361 	}
1362 	return (error);
1363 }
1364 
1365 /*
1366  * Spin prompting on the console for a suitable root filesystem
1367  */
1368 static int
1369 vfs_mountroot_ask(void)
1370 {
1371 	char name[128];
1372 
1373 	for(;;) {
1374 		printf("\nManual root filesystem specification:\n");
1375 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1376 #if defined(__i386__) || defined(__ia64__)
1377 		printf("                       eg. ufs:da0s1a\n");
1378 #else
1379 		printf("                       eg. ufs:/dev/da0a\n");
1380 #endif
1381 		printf("  ?                  List valid disk boot devices\n");
1382 		printf("  <empty line>       Abort manual input\n");
1383 		printf("\nmountroot> ");
1384 		gets(name);
1385 		if (name[0] == '\0')
1386 			return (1);
1387 		if (name[0] == '?') {
1388 			printf("\nList of GEOM managed disk devices:\n  ");
1389 			g_dev_print();
1390 			continue;
1391 		}
1392 		if (!vfs_mountroot_try(name))
1393 			return (0);
1394 	}
1395 }
1396 
1397 /*
1398  * Local helper function for vfs_mountroot_ask.
1399  */
1400 static void
1401 gets(char *cp)
1402 {
1403 	char *lp;
1404 	int c;
1405 
1406 	lp = cp;
1407 	for (;;) {
1408 		printf("%c", c = cngetc() & 0177);
1409 		switch (c) {
1410 		case -1:
1411 		case '\n':
1412 		case '\r':
1413 			*lp++ = '\0';
1414 			return;
1415 		case '\b':
1416 		case '\177':
1417 			if (lp > cp) {
1418 				printf(" \b");
1419 				lp--;
1420 			}
1421 			continue;
1422 		case '#':
1423 			lp--;
1424 			if (lp < cp)
1425 				lp = cp;
1426 			continue;
1427 		case '@':
1428 		case 'u' & 037:
1429 			lp = cp;
1430 			printf("%c", '\n');
1431 			continue;
1432 		default:
1433 			*lp++ = c;
1434 		}
1435 	}
1436 }
1437 
1438 /*
1439  * Convert a given name to the cdev pointer of the device, which is probably
1440  * but not by definition, a disk.  Mount a DEVFS (on nothing), look the name
1441  * up, extract the cdev from the vnode and unmount it again.  Unfortunately
1442  * we cannot use the vnode directly (because we unmount the DEVFS again)
1443  * so the filesystems still have to do the bdevvp() stunt.
1444  */
1445 static struct cdev *
1446 getdiskbyname(char *name)
1447 {
1448 	char *cp = name;
1449 	struct cdev *dev = NULL;
1450 	struct thread *td = curthread;
1451 	struct vfsconf *vfsp;
1452 	struct mount *mp = NULL;
1453 	struct vnode *vroot = NULL;
1454 	struct nameidata nid;
1455 	int error;
1456 
1457 	if (!bcmp(cp, "/dev/", 5))
1458 		cp += 5;
1459 
1460 	do {
1461 		vfsp = vfs_byname("devfs");
1462 		if (vfsp == NULL)
1463 			break;
1464 		error = vfs_mount_alloc(NULLVP, vfsp, "/dev", td, &mp);
1465 		if (error)
1466 			break;
1467 		mp->mnt_flag |= MNT_RDONLY;
1468 
1469 		error = VFS_MOUNT(mp, curthread);
1470 		if (error)
1471 			break;
1472 		VFS_START(mp, 0, td);
1473 		VFS_ROOT(mp, &vroot, td);
1474 		VOP_UNLOCK(vroot, 0, td);
1475 
1476 		NDINIT(&nid, LOOKUP, NOCACHE|FOLLOW,
1477 		    UIO_SYSSPACE, cp, curthread);
1478 		nid.ni_startdir = vroot;
1479 		nid.ni_pathlen = strlen(cp);
1480 		nid.ni_cnd.cn_cred = curthread->td_ucred;
1481 		nid.ni_cnd.cn_nameptr = cp;
1482 
1483 		error = lookup(&nid);
1484 		if (error)
1485 			break;
1486 		if (nid.ni_vp->v_type != VCHR)
1487 			dev = NULL;
1488 		else
1489 			dev = nid.ni_vp->v_rdev;
1490 		NDFREE(&nid, 0);
1491 	} while (0);
1492 
1493 	if (vroot != NULL)
1494 		VFS_UNMOUNT(mp, 0, td);
1495 	if (mp != NULL)
1496 		vfs_mount_destroy(mp, td);
1497   	return (dev);
1498 }
1499 
1500 /* Show the struct cdev *for a disk specified by name */
1501 #ifdef DDB
1502 DB_SHOW_COMMAND(disk, db_getdiskbyname)
1503 {
1504 	struct cdev *dev;
1505 
1506 	if (modif[0] == '\0') {
1507 		db_error("usage: show disk/devicename");
1508 		return;
1509 	}
1510 	dev = getdiskbyname(modif);
1511 	if (dev != NULL)
1512 		db_printf("struct cdev *= %p\n", dev);
1513 	else
1514 		db_printf("No disk device matched.\n");
1515 }
1516 #endif
1517 
1518 /*
1519  * Get a mount option by its name.
1520  *
1521  * Return 0 if the option was found, ENOENT otherwise.
1522  * If len is non-NULL it will be filled with the length
1523  * of the option. If buf is non-NULL, it will be filled
1524  * with the address of the option.
1525  */
1526 int
1527 vfs_getopt(opts, name, buf, len)
1528 	struct vfsoptlist *opts;
1529 	const char *name;
1530 	void **buf;
1531 	int *len;
1532 {
1533 	struct vfsopt *opt;
1534 
1535 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1536 
1537 	TAILQ_FOREACH(opt, opts, link) {
1538 		if (strcmp(name, opt->name) == 0) {
1539 			if (len != NULL)
1540 				*len = opt->len;
1541 			if (buf != NULL)
1542 				*buf = opt->value;
1543 			return (0);
1544 		}
1545 	}
1546 	return (ENOENT);
1547 }
1548 
1549 /*
1550  * Find and copy a mount option.
1551  *
1552  * The size of the buffer has to be specified
1553  * in len, if it is not the same length as the
1554  * mount option, EINVAL is returned.
1555  * Returns ENOENT if the option is not found.
1556  */
1557 int
1558 vfs_copyopt(opts, name, dest, len)
1559 	struct vfsoptlist *opts;
1560 	const char *name;
1561 	void *dest;
1562 	int len;
1563 {
1564 	struct vfsopt *opt;
1565 
1566 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1567 
1568 	TAILQ_FOREACH(opt, opts, link) {
1569 		if (strcmp(name, opt->name) == 0) {
1570 			if (len != opt->len)
1571 				return (EINVAL);
1572 			bcopy(opt->value, dest, opt->len);
1573 			return (0);
1574 		}
1575 	}
1576 	return (ENOENT);
1577 }
1578 
1579 
1580 /*
1581  * This is a helper function for filesystems to traverse their
1582  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1583  */
1584 
1585 struct vnode *
1586 __mnt_vnode_next(struct vnode **nvp, struct mount *mp)
1587 {
1588 	struct vnode *vp;
1589 
1590 	mtx_assert(&mp->mnt_mtx, MA_OWNED);
1591 	vp = *nvp;
1592 	/* Check if we are done */
1593 	if (vp == NULL)
1594 		return (NULL);
1595 	/* If our next vnode is no longer ours, start over */
1596 	if (vp->v_mount != mp)
1597 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1598 	/* Save pointer to next vnode in list */
1599 	if (vp != NULL)
1600 		*nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1601 	else
1602 		*nvp = NULL;
1603 	return (vp);
1604 }
1605