xref: /freebsd/sys/kern/vfs_mount.c (revision 65a311fcb2f44fce7eb05160d3198cefed5c27f9)
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 void	gets(char *cp);
102 static int	vfs_domount(struct thread *td, const char *fstype,
103 		    char *fspath, int fsflags, void *fsdata, int compat);
104 static int	vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
105 		    const char *fspath, struct thread *td, struct mount **mpp);
106 static int	vfs_mountroot_ask(void);
107 static int	vfs_mountroot_try(char *mountfrom);
108 
109 static int	usermount = 0;
110 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
111     "Unprivileged users may mount and unmount file systems");
112 
113 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
114 
115 /* List of mounted filesystems. */
116 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
117 
118 /* For any iteration/modification of mountlist */
119 struct mtx mountlist_mtx;
120 
121 /*
122  * The vnode of the system's root (/ in the filesystem, without chroot
123  * active.)
124  */
125 struct vnode	*rootvnode;
126 
127 /*
128  * The root filesystem is detailed in the kernel environment variable
129  * vfs.root.mountfrom, which is expected to be in the general format
130  *
131  * <vfsname>:[<path>]
132  * vfsname   := the name of a VFS known to the kernel and capable
133  *              of being mounted as root
134  * path      := disk device name or other data used by the filesystem
135  *              to locate its physical store
136  */
137 
138 /*
139  * The root specifiers we will try if RB_CDROM is specified.
140  */
141 static char *cdrom_rootdevnames[] = {
142 	"cd9660:cd0",
143 	"cd9660:acd0",
144 	NULL
145 };
146 
147 /* legacy find-root code */
148 char		*rootdevnames[2] = {NULL, NULL};
149 static int	setrootbyname(char *name);
150 struct cdev *rootdev = NULL;
151 
152 /*
153  * Has to be dynamic as the value of rootdev can change; however, it can't
154  * change after the root is mounted, so a user process can't access this
155  * sysctl until after the value is unchangeable.
156  */
157 static int
158 sysctl_rootdev(SYSCTL_HANDLER_ARGS)
159 {
160 	int error;
161 
162 	/* _RD prevents this from happening. */
163 	KASSERT(req->newptr == NULL, ("Attempt to change root device name"));
164 
165 	if (rootdev != NULL)
166 		error = sysctl_handle_string(oidp, rootdev->si_name, 0, req);
167 	else
168 		error = sysctl_handle_string(oidp, "", 0, req);
169 
170 	return (error);
171 }
172 
173 SYSCTL_PROC(_kern, OID_AUTO, rootdev, CTLTYPE_STRING | CTLFLAG_RD,
174     0, 0, sysctl_rootdev, "A", "Root file system device");
175 
176 /* Remove one mount option. */
177 static void
178 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
179 {
180 
181 	TAILQ_REMOVE(opts, opt, link);
182 	free(opt->name, M_MOUNT);
183 	if (opt->value != NULL)
184 		free(opt->value, M_MOUNT);
185 #ifdef INVARIANTS
186 	else if (opt->len != 0)
187 		panic("%s: mount option with NULL value but length != 0",
188 		    __func__);
189 #endif
190 	free(opt, M_MOUNT);
191 }
192 
193 /* Release all resources related to the mount options. */
194 static void
195 vfs_freeopts(struct vfsoptlist *opts)
196 {
197 	struct vfsopt *opt;
198 
199 	while (!TAILQ_EMPTY(opts)) {
200 		opt = TAILQ_FIRST(opts);
201 		vfs_freeopt(opts, opt);
202 	}
203 	free(opts, M_MOUNT);
204 }
205 
206 /*
207  * Check if options are equal (with or without the "no" prefix).
208  */
209 static int
210 vfs_equalopts(const char *opt1, const char *opt2)
211 {
212 
213 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
214 	if (strcmp(opt1, opt2) == 0)
215 		return (1);
216 	/* "noopt" vs. "opt" */
217 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
218 		return (1);
219 	/* "opt" vs. "noopt" */
220 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
221 		return (1);
222 	return (0);
223 }
224 
225 /*
226  * If a mount option is specified several times,
227  * (with or without the "no" prefix) only keep
228  * the last occurence of it.
229  */
230 static void
231 vfs_sanitizeopts(struct vfsoptlist *opts)
232 {
233 	struct vfsopt *opt, *opt2, *tmp;
234 
235 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
236 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
237 		while (opt2 != NULL) {
238 			if (vfs_equalopts(opt->name, opt2->name)) {
239 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
240 				vfs_freeopt(opts, opt2);
241 				opt2 = tmp;
242 			} else {
243 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
244 			}
245 		}
246 	}
247 }
248 
249 /*
250  * Build a linked list of mount options from a struct uio.
251  */
252 static int
253 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
254 {
255 	struct vfsoptlist *opts;
256 	struct vfsopt *opt;
257 	size_t memused;
258 	unsigned int i, iovcnt;
259 	int error, namelen, optlen;
260 
261 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
262 	TAILQ_INIT(opts);
263 	memused = 0;
264 	iovcnt = auio->uio_iovcnt;
265 	for (i = 0; i < iovcnt; i += 2) {
266 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
267 		namelen = auio->uio_iov[i].iov_len;
268 		optlen = auio->uio_iov[i + 1].iov_len;
269 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
270 		opt->value = NULL;
271 		opt->len = optlen;
272 
273 		/*
274 		 * Do this early, so jumps to "bad" will free the current
275 		 * option.
276 		 */
277 		TAILQ_INSERT_TAIL(opts, opt, link);
278 		memused += sizeof(struct vfsopt) + optlen + namelen;
279 
280 		/*
281 		 * Avoid consuming too much memory, and attempts to overflow
282 		 * memused.
283 		 */
284 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
285 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
286 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
287 			error = EINVAL;
288 			goto bad;
289 		}
290 
291 		if (auio->uio_segflg == UIO_SYSSPACE) {
292 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
293 		} else {
294 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
295 			    namelen);
296 			if (error)
297 				goto bad;
298 		}
299 		/* Ensure names are null-terminated strings. */
300 		if (opt->name[namelen - 1] != '\0') {
301 			error = EINVAL;
302 			goto bad;
303 		}
304 		if (optlen != 0) {
305 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
306 			if (auio->uio_segflg == UIO_SYSSPACE) {
307 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
308 				    optlen);
309 			} else {
310 				error = copyin(auio->uio_iov[i + 1].iov_base,
311 				    opt->value, optlen);
312 				if (error)
313 					goto bad;
314 			}
315 		}
316 	}
317 	vfs_sanitizeopts(opts);
318 	*options = opts;
319 	return (0);
320 bad:
321 	vfs_freeopts(opts);
322 	return (error);
323 }
324 
325 /*
326  * Merge the old mount options with the new ones passed
327  * in the MNT_UPDATE case.
328  */
329 static void
330 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
331 {
332 	struct vfsopt *opt, *opt2, *new;
333 
334 	TAILQ_FOREACH(opt, opts, link) {
335 		/*
336 		 * Check that this option hasn't been redefined
337 		 * nor cancelled with a "no" mount option.
338 		 */
339 		opt2 = TAILQ_FIRST(toopts);
340 		while (opt2 != NULL) {
341 			if (strcmp(opt2->name, opt->name) == 0)
342 				goto next;
343 			if (strncmp(opt2->name, "no", 2) == 0 &&
344 			    strcmp(opt2->name + 2, opt->name) == 0) {
345 				vfs_freeopt(toopts, opt2);
346 				goto next;
347 			}
348 			opt2 = TAILQ_NEXT(opt2, link);
349 		}
350 		/* We want this option, duplicate it. */
351 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
352 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
353 		strcpy(new->name, opt->name);
354 		if (opt->len != 0) {
355 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
356 			bcopy(opt->value, new->value, opt->len);
357 		} else {
358 			new->value = NULL;
359 		}
360 		new->len = opt->len;
361 		TAILQ_INSERT_TAIL(toopts, new, link);
362 next:
363 		continue;
364 	}
365 }
366 
367 /*
368  * New mount API.
369  */
370 int
371 nmount(td, uap)
372 	struct thread *td;
373 	struct nmount_args /* {
374 		struct iovec *iovp;
375 		unsigned int iovcnt;
376 		int flags;
377 	} */ *uap;
378 {
379 	struct uio *auio;
380 	struct iovec *iov;
381 	unsigned int i;
382 	int error;
383 	u_int iovcnt;
384 
385 	iovcnt = uap->iovcnt;
386 	/*
387 	 * Check that we have an even number of iovec's
388 	 * and that we have at least two options.
389 	 */
390 	if ((iovcnt & 1) || (iovcnt < 4))
391 		return (EINVAL);
392 	error = copyinuio(uap->iovp, iovcnt, &auio);
393 	if (error)
394 		return (error);
395 	iov = auio->uio_iov;
396 	for (i = 0; i < iovcnt; i++) {
397 		if (iov->iov_len > MMAXOPTIONLEN) {
398 			free(auio, M_IOV);
399 			return (EINVAL);
400 		}
401 		iov++;
402 	}
403 	error = vfs_nmount(td, uap->flags, auio);
404 	free(auio, M_IOV);
405 	return (error);
406 }
407 
408 int
409 kernel_mount(iovp, iovcnt, flags)
410 	struct iovec *iovp;
411 	unsigned int iovcnt;
412 	int flags;
413 {
414 	struct uio auio;
415 	int error;
416 
417 	/*
418 	 * Check that we have an even number of iovec's
419 	 * and that we have at least two options.
420 	 */
421 	if ((iovcnt & 1) || (iovcnt < 4))
422 		return (EINVAL);
423 
424 	auio.uio_iov = iovp;
425 	auio.uio_iovcnt = iovcnt;
426 	auio.uio_segflg = UIO_SYSSPACE;
427 
428 	error = vfs_nmount(curthread, flags, &auio);
429 	return (error);
430 }
431 
432 int
433 kernel_vmount(int flags, ...)
434 {
435 	struct iovec *iovp;
436 	struct uio auio;
437 	va_list ap;
438 	unsigned int iovcnt, iovlen, len;
439 	const char *cp;
440 	char *buf, *pos;
441 	size_t n;
442 	int error, i;
443 
444 	len = 0;
445 	va_start(ap, flags);
446 	for (iovcnt = 0; (cp = va_arg(ap, const char *)) != NULL; iovcnt++)
447 		len += strlen(cp) + 1;
448 	va_end(ap);
449 
450 	if (iovcnt < 4 || iovcnt & 1)
451 		return (EINVAL);
452 
453 	iovlen = iovcnt * sizeof (struct iovec);
454 	MALLOC(iovp, struct iovec *, iovlen, M_MOUNT, M_WAITOK);
455 	MALLOC(buf, char *, len, M_MOUNT, M_WAITOK);
456 	pos = buf;
457 	va_start(ap, flags);
458 	for (i = 0; i < iovcnt; i++) {
459 		cp = va_arg(ap, const char *);
460 		copystr(cp, pos, len - (pos - buf), &n);
461 		iovp[i].iov_base = pos;
462 		iovp[i].iov_len = n;
463 		pos += n;
464 	}
465 	va_end(ap);
466 
467 	auio.uio_iov = iovp;
468 	auio.uio_iovcnt = iovcnt;
469 	auio.uio_segflg = UIO_SYSSPACE;
470 
471 	error = vfs_nmount(curthread, flags, &auio);
472 	FREE(iovp, M_MOUNT);
473 	FREE(buf, M_MOUNT);
474 	return (error);
475 }
476 
477 /*
478  * Allocate and initialize the mount point struct.
479  */
480 static int
481 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
482     const char *fspath, struct thread *td, struct mount **mpp)
483 {
484 	struct mount *mp;
485 
486 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
487 	TAILQ_INIT(&mp->mnt_nvnodelist);
488 	mp->mnt_nvnodelistsize = 0;
489 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
490 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
491 	vfs_busy(mp, LK_NOWAIT, 0, td);
492 	mp->mnt_op = vfsp->vfc_vfsops;
493 	mp->mnt_vfc = vfsp;
494 	vfsp->vfc_refcount++;
495 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
496 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
497 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
498 	mp->mnt_vnodecovered = vp;
499 	mp->mnt_cred = crdup(td->td_ucred);
500 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
501 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
502 	mp->mnt_iosize_max = DFLTPHYS;
503 #ifdef MAC
504 	mac_init_mount(mp);
505 	mac_create_mount(td->td_ucred, mp);
506 #endif
507 	*mpp = mp;
508 	return (0);
509 }
510 
511 /*
512  * Destroy the mount struct previously allocated by vfs_mount_alloc().
513  */
514 void
515 vfs_mount_destroy(struct mount *mp, struct thread *td)
516 {
517 
518 	mp->mnt_vfc->vfc_refcount--;
519 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
520 		panic("unmount: dangling vnode");
521 	vfs_unbusy(mp,td);
522 	lockdestroy(&mp->mnt_lock);
523 	mtx_destroy(&mp->mnt_mtx);
524 	if (mp->mnt_kern_flag & MNTK_MWAIT)
525 		wakeup(mp);
526 #ifdef MAC
527 	mac_destroy_mount(mp);
528 #endif
529 	if (mp->mnt_opt != NULL)
530 		vfs_freeopts(mp->mnt_opt);
531 	crfree(mp->mnt_cred);
532 	free(mp, M_MOUNT);
533 }
534 
535 int
536 vfs_nmount(struct thread *td, int fsflags, struct uio *fsoptions)
537 {
538 	struct vfsoptlist *optlist;
539 	char *fstype, *fspath;
540 	int error, fstypelen, fspathlen;
541 
542 	error = vfs_buildopts(fsoptions, &optlist);
543 	if (error)
544 		return (error);
545 
546 	/*
547 	 * We need these two options before the others,
548 	 * and they are mandatory for any filesystem.
549 	 * Ensure they are NUL terminated as well.
550 	 */
551 	fstypelen = 0;
552 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
553 	if (error || fstype[fstypelen - 1] != '\0') {
554 		error = EINVAL;
555 		goto bail;
556 	}
557 	fspathlen = 0;
558 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
559 	if (error || fspath[fspathlen - 1] != '\0') {
560 		error = EINVAL;
561 		goto bail;
562 	}
563 
564 	/*
565 	 * Be ultra-paranoid about making sure the type and fspath
566 	 * variables will fit in our mp buffers, including the
567 	 * terminating NUL.
568 	 */
569 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
570 		error = ENAMETOOLONG;
571 		goto bail;
572 	}
573 
574 	error = vfs_domount(td, fstype, fspath, fsflags, optlist, 0);
575 bail:
576 	if (error)
577 		vfs_freeopts(optlist);
578 	return (error);
579 }
580 
581 /*
582  * Old mount API.
583  */
584 #ifndef _SYS_SYSPROTO_H_
585 struct mount_args {
586 	char	*type;
587 	char	*path;
588 	int	flags;
589 	caddr_t	data;
590 };
591 #endif
592 /* ARGSUSED */
593 int
594 mount(td, uap)
595 	struct thread *td;
596 	struct mount_args /* {
597 		char *type;
598 		char *path;
599 		int flags;
600 		caddr_t data;
601 	} */ *uap;
602 {
603 	char *fstype;
604 	char *fspath;
605 	int error;
606 
607 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
608 	fspath = malloc(MNAMELEN, M_TEMP, M_WAITOK);
609 
610 	/*
611 	 * vfs_mount() actually takes a kernel string for `type' and
612 	 * `path' now, so extract them.
613 	 */
614 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
615 	if (error == 0)
616 		error = copyinstr(uap->path, fspath, MNAMELEN, NULL);
617 	if (error == 0)
618 		error = vfs_domount(td, fstype, fspath, uap->flags,
619 		    uap->data, 1);
620 	free(fstype, M_TEMP);
621 	free(fspath, M_TEMP);
622 	return (error);
623 }
624 
625 /*
626  * vfs_mount(): actually attempt a filesystem mount.
627  *
628  * This routine is designed to be a "generic" entry point for routines
629  * that wish to mount a filesystem. All parameters except `fsdata' are
630  * pointers into kernel space. `fsdata' is currently still a pointer
631  * into userspace.
632  */
633 int
634 vfs_mount(td, fstype, fspath, fsflags, fsdata)
635 	struct thread *td;
636 	const char *fstype;
637 	char *fspath;
638 	int fsflags;
639 	void *fsdata;
640 {
641 
642 	return (vfs_domount(td, fstype, fspath, fsflags, fsdata, 1));
643 }
644 
645 /*
646  * vfs_domount(): actually attempt a filesystem mount.
647  */
648 static int
649 vfs_domount(
650 	struct thread *td,	/* Flags common to all filesystems. */
651 	const char *fstype,	/* Filesystem type. */
652 	char *fspath,		/* Mount path. */
653 	int fsflags,		/* Flags common to all filesystems. */
654 	void *fsdata,		/* Options local to the filesystem. */
655 	int compat		/* Invocation from compat syscall. */
656 	)
657 {
658 	linker_file_t lf;
659 	struct vnode *vp;
660 	struct mount *mp;
661 	struct vfsconf *vfsp;
662 	int error, flag = 0, kern_flag = 0;
663 	struct vattr va;
664 	struct nameidata nd;
665 
666 	/*
667 	 * Be ultra-paranoid about making sure the type and fspath
668 	 * variables will fit in our mp buffers, including the
669 	 * terminating NUL.
670 	 */
671 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
672 		return (ENAMETOOLONG);
673 
674 	if (jailed(td->td_ucred))
675 		return (EPERM);
676 	if (usermount == 0) {
677 		if ((error = suser(td)) != 0)
678 			return (error);
679 	}
680 
681 	/*
682 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
683 	 */
684 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
685 		if ((error = suser(td)) != 0)
686 			return (error);
687 	}
688 	/*
689 	 * Silently enforce MNT_NODEV, MNT_NOSUID and MNT_USER for
690 	 * unprivileged users.
691 	 */
692 	if (suser(td) != 0)
693 		fsflags |= MNT_NODEV | MNT_NOSUID | MNT_USER;
694 	/*
695 	 * Get vnode to be covered
696 	 */
697 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
698 	if ((error = namei(&nd)) != 0)
699 		return (error);
700 	NDFREE(&nd, NDF_ONLY_PNBUF);
701 	vp = nd.ni_vp;
702 	if (fsflags & MNT_UPDATE) {
703 		if ((vp->v_vflag & VV_ROOT) == 0) {
704 			vput(vp);
705 			return (EINVAL);
706 		}
707 		mp = vp->v_mount;
708 		flag = mp->mnt_flag;
709 		kern_flag = mp->mnt_kern_flag;
710 		/*
711 		 * We only allow the filesystem to be reloaded if it
712 		 * is currently mounted read-only.
713 		 */
714 		if ((fsflags & MNT_RELOAD) &&
715 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
716 			vput(vp);
717 			return (EOPNOTSUPP);	/* Needs translation */
718 		}
719 		/*
720 		 * Only privileged root, or (if MNT_USER is set) the user that
721 		 * did the original mount is permitted to update it.
722 		 */
723 		error = vfs_suser(mp, td);
724 		if (error) {
725 			vput(vp);
726 			return (error);
727 		}
728 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
729 			vput(vp);
730 			return (EBUSY);
731 		}
732 		VI_LOCK(vp);
733 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
734 		    vp->v_mountedhere != NULL) {
735 			VI_UNLOCK(vp);
736 			vfs_unbusy(mp, td);
737 			vput(vp);
738 			return (EBUSY);
739 		}
740 		vp->v_iflag |= VI_MOUNT;
741 		VI_UNLOCK(vp);
742 		mp->mnt_flag |= fsflags &
743 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT);
744 		VOP_UNLOCK(vp, 0, td);
745 		if (compat == 0) {
746 			mp->mnt_optnew = fsdata;
747 			vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
748 		}
749 		goto update;
750 	}
751 	/*
752 	 * If the user is not root, ensure that they own the directory
753 	 * onto which we are attempting to mount.
754 	 */
755 	error = VOP_GETATTR(vp, &va, td->td_ucred, td);
756 	if (error) {
757 		vput(vp);
758 		return (error);
759 	}
760 	if (va.va_uid != td->td_ucred->cr_uid) {
761 		if ((error = suser(td)) != 0) {
762 			vput(vp);
763 			return (error);
764 		}
765 	}
766 	if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) {
767 		vput(vp);
768 		return (error);
769 	}
770 	if (vp->v_type != VDIR) {
771 		vput(vp);
772 		return (ENOTDIR);
773 	}
774 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
775 		if (strcmp(vfsp->vfc_name, fstype) == 0)
776 			break;
777 	if (vfsp == NULL) {
778 		/* Only load modules for root (very important!). */
779 		if ((error = suser(td)) != 0) {
780 			vput(vp);
781 			return (error);
782 		}
783 		error = securelevel_gt(td->td_ucred, 0);
784 		if (error) {
785 			vput(vp);
786 			return (error);
787 		}
788 		error = linker_load_module(NULL, fstype, NULL, NULL, &lf);
789 		if (error || lf == NULL) {
790 			vput(vp);
791 			if (lf == NULL)
792 				error = ENODEV;
793 			return (error);
794 		}
795 		lf->userrefs++;
796 		/* Look up again to see if the VFS was loaded. */
797 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
798 			if (strcmp(vfsp->vfc_name, fstype) == 0)
799 				break;
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 update:
831 	/*
832 	 * Check if the fs implements the type VFS_[N]MOUNT()
833 	 * function we are looking for.
834 	 */
835 	if ((compat == 0) == (mp->mnt_op->vfs_mount != 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 	error = compat ? VFS_MOUNT(mp, fspath, fsdata, &nd, td) :
864 	    VFS_NMOUNT(mp, &nd, td);
865 	if (!error) {
866 		if (mp->mnt_opt != NULL)
867 			vfs_freeopts(mp->mnt_opt);
868 		mp->mnt_opt = mp->mnt_optnew;
869 	}
870 	/*
871 	 * Prevent external consumers of mount options from reading
872 	 * mnt_optnew.
873 	*/
874 	mp->mnt_optnew = NULL;
875 	if (mp->mnt_flag & MNT_UPDATE) {
876 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
877 			mp->mnt_flag &= ~MNT_RDONLY;
878 		mp->mnt_flag &=
879 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
880 		mp->mnt_kern_flag &= ~MNTK_WANTRDWR;
881 		if (error) {
882 			mp->mnt_flag = flag;
883 			mp->mnt_kern_flag = kern_flag;
884 		}
885 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
886 			if (mp->mnt_syncer == NULL)
887 				error = vfs_allocate_syncvnode(mp);
888 		} else {
889 			if (mp->mnt_syncer != NULL)
890 				vrele(mp->mnt_syncer);
891 			mp->mnt_syncer = NULL;
892 		}
893 		vfs_unbusy(mp, td);
894 		VI_LOCK(vp);
895 		vp->v_iflag &= ~VI_MOUNT;
896 		VI_UNLOCK(vp);
897 		vrele(vp);
898 		return (error);
899 	}
900 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
901 	/*
902 	 * Put the new filesystem on the mount list after root.
903 	 */
904 	cache_purge(vp);
905 	if (!error) {
906 		struct vnode *newdp;
907 
908 		VI_LOCK(vp);
909 		vp->v_iflag &= ~VI_MOUNT;
910 		VI_UNLOCK(vp);
911 		vp->v_mountedhere = mp;
912 		mtx_lock(&mountlist_mtx);
913 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
914 		mtx_unlock(&mountlist_mtx);
915 		vfs_event_signal(NULL, VQ_MOUNT, 0);
916 		if (VFS_ROOT(mp, &newdp, td))
917 			panic("mount: lost mount");
918 		checkdirs(vp, newdp);
919 		vput(newdp);
920 		VOP_UNLOCK(vp, 0, td);
921 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
922 			error = vfs_allocate_syncvnode(mp);
923 		vfs_unbusy(mp, td);
924 		if (error || (error = VFS_START(mp, 0, td)) != 0)
925 			vrele(vp);
926 	} else {
927 		VI_LOCK(vp);
928 		vp->v_iflag &= ~VI_MOUNT;
929 		VI_UNLOCK(vp);
930 		vfs_mount_destroy(mp, td);
931 		vput(vp);
932 	}
933 	return (error);
934 }
935 
936 /*
937  * Scan all active processes to see if any of them have a current
938  * or root directory of `olddp'. If so, replace them with the new
939  * mount point.
940  */
941 static void
942 checkdirs(olddp, newdp)
943 	struct vnode *olddp, *newdp;
944 {
945 	struct filedesc *fdp;
946 	struct proc *p;
947 	int nrele;
948 
949 	if (vrefcnt(olddp) == 1)
950 		return;
951 	sx_slock(&allproc_lock);
952 	LIST_FOREACH(p, &allproc, p_list) {
953 		mtx_lock(&fdesc_mtx);
954 		fdp = p->p_fd;
955 		if (fdp == NULL) {
956 			mtx_unlock(&fdesc_mtx);
957 			continue;
958 		}
959 		nrele = 0;
960 		FILEDESC_LOCK(fdp);
961 		if (fdp->fd_cdir == olddp) {
962 			VREF(newdp);
963 			fdp->fd_cdir = newdp;
964 			nrele++;
965 		}
966 		if (fdp->fd_rdir == olddp) {
967 			VREF(newdp);
968 			fdp->fd_rdir = newdp;
969 			nrele++;
970 		}
971 		FILEDESC_UNLOCK(fdp);
972 		mtx_unlock(&fdesc_mtx);
973 		while (nrele--)
974 			vrele(olddp);
975 	}
976 	sx_sunlock(&allproc_lock);
977 	if (rootvnode == olddp) {
978 		vrele(rootvnode);
979 		VREF(newdp);
980 		rootvnode = newdp;
981 	}
982 }
983 
984 /*
985  * Unmount a filesystem.
986  *
987  * Note: unmount takes a path to the vnode mounted on as argument,
988  * not special file (as before).
989  */
990 #ifndef _SYS_SYSPROTO_H_
991 struct unmount_args {
992 	char	*path;
993 	int	flags;
994 };
995 #endif
996 /* ARGSUSED */
997 int
998 unmount(td, uap)
999 	struct thread *td;
1000 	register struct unmount_args /* {
1001 		char *path;
1002 		int flags;
1003 	} */ *uap;
1004 {
1005 	struct mount *mp;
1006 	char *pathbuf;
1007 	int error, id0, id1;
1008 
1009 	if (jailed(td->td_ucred))
1010 		return (EPERM);
1011 	if (usermount == 0) {
1012 		if ((error = suser(td)) != 0)
1013 			return (error);
1014 	}
1015 
1016 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1017 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1018 	if (error) {
1019 		free(pathbuf, M_TEMP);
1020 		return (error);
1021 	}
1022 	if (uap->flags & MNT_BYFSID) {
1023 		/* Decode the filesystem ID. */
1024 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1025 			free(pathbuf, M_TEMP);
1026 			return (EINVAL);
1027 		}
1028 
1029 		mtx_lock(&mountlist_mtx);
1030 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1031 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1032 			    mp->mnt_stat.f_fsid.val[1] == id1)
1033 				break;
1034 		}
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 		}
1042 		mtx_unlock(&mountlist_mtx);
1043 	}
1044 	free(pathbuf, M_TEMP);
1045 	if (mp == NULL) {
1046 		/*
1047 		 * Previously we returned ENOENT for a nonexistent path and
1048 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1049 		 * now, so in the !MNT_BYFSID case return the more likely
1050 		 * EINVAL for compatibility.
1051 		 */
1052 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1053 	}
1054 
1055 	/*
1056 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1057 	 * original mount is permitted to unmount this filesystem.
1058 	 */
1059 	error = vfs_suser(mp, td);
1060 	if (error)
1061 		return (error);
1062 
1063 	/*
1064 	 * Don't allow unmounting the root filesystem.
1065 	 */
1066 	if (mp->mnt_flag & MNT_ROOTFS)
1067 		return (EINVAL);
1068 	return (dounmount(mp, uap->flags, td));
1069 }
1070 
1071 /*
1072  * Do the actual filesystem unmount.
1073  */
1074 int
1075 dounmount(mp, flags, td)
1076 	struct mount *mp;
1077 	int flags;
1078 	struct thread *td;
1079 {
1080 	struct vnode *coveredvp, *fsrootvp;
1081 	int error;
1082 	int async_flag;
1083 
1084 	mtx_lock(&mountlist_mtx);
1085 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1086 		mtx_unlock(&mountlist_mtx);
1087 		return (EBUSY);
1088 	}
1089 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1090 	/* Allow filesystems to detect that a forced unmount is in progress. */
1091 	if (flags & MNT_FORCE)
1092 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1093 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1094 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td);
1095 	if (error) {
1096 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1097 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1098 			wakeup(mp);
1099 		return (error);
1100 	}
1101 	vn_start_write(NULL, &mp, V_WAIT);
1102 
1103 	if (mp->mnt_flag & MNT_EXPUBLIC)
1104 		vfs_setpublicfs(NULL, NULL, NULL);
1105 
1106 	vfs_msync(mp, MNT_WAIT);
1107 	async_flag = mp->mnt_flag & MNT_ASYNC;
1108 	mp->mnt_flag &= ~MNT_ASYNC;
1109 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1110 	if (mp->mnt_syncer != NULL)
1111 		vrele(mp->mnt_syncer);
1112 	/*
1113 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1114 	 * vnode to the covered vnode.  For non-forced unmounts we want
1115 	 * such references to cause an EBUSY error.
1116 	 */
1117 	if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) {
1118 		if (mp->mnt_vnodecovered != NULL)
1119 			checkdirs(fsrootvp, mp->mnt_vnodecovered);
1120 		if (fsrootvp == rootvnode) {
1121 			vrele(rootvnode);
1122 			rootvnode = NULL;
1123 		}
1124 		vput(fsrootvp);
1125 	}
1126 	if (((mp->mnt_flag & MNT_RDONLY) ||
1127 	     (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) ||
1128 	    (flags & MNT_FORCE)) {
1129 		error = VFS_UNMOUNT(mp, flags, td);
1130 	}
1131 	vn_finished_write(mp);
1132 	if (error) {
1133 		/* Undo cdir/rdir and rootvnode changes made above. */
1134 		if ((flags & MNT_FORCE) && VFS_ROOT(mp, &fsrootvp, td) == 0) {
1135 			if (mp->mnt_vnodecovered != NULL)
1136 				checkdirs(mp->mnt_vnodecovered, fsrootvp);
1137 			if (rootvnode == NULL) {
1138 				rootvnode = fsrootvp;
1139 				vref(rootvnode);
1140 			}
1141 			vput(fsrootvp);
1142 		}
1143 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1144 			(void) vfs_allocate_syncvnode(mp);
1145 		mtx_lock(&mountlist_mtx);
1146 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1147 		mp->mnt_flag |= async_flag;
1148 		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK,
1149 		    &mountlist_mtx, td);
1150 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1151 			wakeup(mp);
1152 		return (error);
1153 	}
1154 	mtx_lock(&mountlist_mtx);
1155 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1156 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1157 		coveredvp->v_mountedhere = NULL;
1158 	mtx_unlock(&mountlist_mtx);
1159 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
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 != NULL && 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 cdev pointer of the device, which is probably
1433  * but not by definition, a disk.  Mount a DEVFS (on nothing), look the name
1434  * up, extract the cdev from the vnode and unmount it again.  Unfortunately
1435  * we cannot use the vnode directly (because we unmount the DEVFS again)
1436  * so the filesystems still have to do the bdevvp() stunt.
1437  */
1438 struct cdev *
1439 getdiskbyname(char *name)
1440 {
1441 	char *cp = name;
1442 	struct cdev *dev = NULL;
1443 	struct thread *td = curthread;
1444 	struct vfsconf *vfsp;
1445 	struct mount *mp = NULL;
1446 	struct vnode *vroot = NULL;
1447 	struct nameidata nid;
1448 	int error;
1449 
1450 	if (!bcmp(cp, "/dev/", 5))
1451 		cp += 5;
1452 
1453 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1454 		if (!strcmp(vfsp->vfc_name, "devfs"))
1455 			break;
1456 	do {
1457 		if (vfsp == NULL)
1458 			break;
1459 		error = vfs_mount_alloc(NULLVP, vfsp, "/dev", td, &mp);
1460 		if (error)
1461 			break;
1462 		mp->mnt_flag |= MNT_RDONLY;
1463 
1464 		error = VFS_NMOUNT(mp, NULL, curthread);
1465 		if (error)
1466 			break;
1467 		VFS_START(mp, 0, td);
1468 		VFS_ROOT(mp, &vroot, td);
1469 		VOP_UNLOCK(vroot, 0, td);
1470 
1471 		NDINIT(&nid, LOOKUP, NOCACHE|FOLLOW,
1472 		    UIO_SYSSPACE, cp, curthread);
1473 		nid.ni_startdir = vroot;
1474 		nid.ni_pathlen = strlen(cp);
1475 		nid.ni_cnd.cn_cred = curthread->td_ucred;
1476 		nid.ni_cnd.cn_nameptr = cp;
1477 
1478 		error = lookup(&nid);
1479 		if (error)
1480 			break;
1481 		dev = vn_todev (nid.ni_vp);
1482 		NDFREE(&nid, 0);
1483 	} while (0);
1484 
1485 	if (vroot != NULL)
1486 		VFS_UNMOUNT(mp, 0, td);
1487 	if (mp != NULL)
1488 		vfs_mount_destroy(mp, td);
1489   	return (dev);
1490 }
1491 
1492 /*
1493  * Set rootdev to match (name), given that we expect it to
1494  * refer to a disk-like device.
1495  */
1496 static int
1497 setrootbyname(char *name)
1498 {
1499 	struct cdev *diskdev;
1500 
1501 	diskdev = getdiskbyname(name);
1502 	if (diskdev != NULL) {
1503 		rootdev = diskdev;
1504 		return (0);
1505 	}
1506 
1507 	return (1);
1508 }
1509 
1510 /* Show the struct cdev *for a disk specified by name */
1511 #ifdef DDB
1512 DB_SHOW_COMMAND(disk, db_getdiskbyname)
1513 {
1514 	struct cdev *dev;
1515 
1516 	if (modif[0] == '\0') {
1517 		db_error("usage: show disk/devicename");
1518 		return;
1519 	}
1520 	dev = getdiskbyname(modif);
1521 	if (dev != NULL)
1522 		db_printf("struct cdev *= %p\n", dev);
1523 	else
1524 		db_printf("No disk device matched.\n");
1525 }
1526 #endif
1527 
1528 /*
1529  * Get a mount option by its name.
1530  *
1531  * Return 0 if the option was found, ENOENT otherwise.
1532  * If len is non-NULL it will be filled with the length
1533  * of the option. If buf is non-NULL, it will be filled
1534  * with the address of the option.
1535  */
1536 int
1537 vfs_getopt(opts, name, buf, len)
1538 	struct vfsoptlist *opts;
1539 	const char *name;
1540 	void **buf;
1541 	int *len;
1542 {
1543 	struct vfsopt *opt;
1544 
1545 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1546 
1547 	TAILQ_FOREACH(opt, opts, link) {
1548 		if (strcmp(name, opt->name) == 0) {
1549 			if (len != NULL)
1550 				*len = opt->len;
1551 			if (buf != NULL)
1552 				*buf = opt->value;
1553 			return (0);
1554 		}
1555 	}
1556 	return (ENOENT);
1557 }
1558 
1559 /*
1560  * Find and copy a mount option.
1561  *
1562  * The size of the buffer has to be specified
1563  * in len, if it is not the same length as the
1564  * mount option, EINVAL is returned.
1565  * Returns ENOENT if the option is not found.
1566  */
1567 int
1568 vfs_copyopt(opts, name, dest, len)
1569 	struct vfsoptlist *opts;
1570 	const char *name;
1571 	void *dest;
1572 	int len;
1573 {
1574 	struct vfsopt *opt;
1575 
1576 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1577 
1578 	TAILQ_FOREACH(opt, opts, link) {
1579 		if (strcmp(name, opt->name) == 0) {
1580 			if (len != opt->len)
1581 				return (EINVAL);
1582 			bcopy(opt->value, dest, opt->len);
1583 			return (0);
1584 		}
1585 	}
1586 	return (ENOENT);
1587 }
1588 
1589 
1590 /*
1591  * This is a helper function for filesystems to traverse their
1592  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1593  */
1594 
1595 struct vnode *
1596 __mnt_vnode_next(struct vnode **nvp, struct mount *mp)
1597 {
1598 	struct vnode *vp;
1599 
1600 	mtx_assert(&mp->mnt_mtx, MA_OWNED);
1601 	vp = *nvp;
1602 	/* Check if we are done */
1603 	if (vp == NULL)
1604 		return (NULL);
1605 	/* If our next vnode is no longer ours, start over */
1606 	if (vp->v_mount != mp)
1607 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1608 	/* Save pointer to next vnode in list */
1609 	if (vp != NULL)
1610 		*nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1611 	else
1612 		*nvp = NULL;
1613 	return (vp);
1614 }
1615