xref: /freebsd/sys/kern/vfs_mountroot.c (revision c9dbb1cc52b063bbd9ab078a7afc89a8696da659)
1 /*-
2  * Copyright (c) 2010 Marcel Moolenaar
3  * Copyright (c) 1999-2004 Poul-Henning Kamp
4  * Copyright (c) 1999 Michael Smith
5  * Copyright (c) 1989, 1993
6  *      The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 4. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 #include "opt_rootdevname.h"
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include <sys/param.h>
44 #include <sys/conf.h>
45 #include <sys/cons.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mdioctl.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/filedesc.h>
57 #include <sys/reboot.h>
58 #include <sys/sbuf.h>
59 #include <sys/stat.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/systm.h>
66 #include <sys/vnode.h>
67 
68 #include <geom/geom.h>
69 
70 /*
71  * The root filesystem is detailed in the kernel environment variable
72  * vfs.root.mountfrom, which is expected to be in the general format
73  *
74  * <vfsname>:[<path>][	<vfsname>:[<path>] ...]
75  * vfsname   := the name of a VFS known to the kernel and capable
76  *              of being mounted as root
77  * path      := disk device name or other data used by the filesystem
78  *              to locate its physical store
79  *
80  * If the environment variable vfs.root.mountfrom is a space separated list,
81  * each list element is tried in turn and the root filesystem will be mounted
82  * from the first one that suceeds.
83  *
84  * The environment variable vfs.root.mountfrom.options is a comma delimited
85  * set of string mount options.  These mount options must be parseable
86  * by nmount() in the kernel.
87  */
88 
89 static int parse_mount(char **);
90 static struct mntarg *parse_mountroot_options(struct mntarg *, const char *);
91 
92 /*
93  * The vnode of the system's root (/ in the filesystem, without chroot
94  * active.)
95  */
96 struct vnode *rootvnode;
97 
98 /*
99  * Mount of the system's /dev.
100  */
101 struct mount *rootdevmp;
102 
103 char *rootdevnames[2] = {NULL, NULL};
104 
105 struct mtx root_holds_mtx;
106 MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF);
107 
108 struct root_hold_token {
109 	const char			*who;
110 	LIST_ENTRY(root_hold_token)	list;
111 };
112 
113 static LIST_HEAD(, root_hold_token)	root_holds =
114     LIST_HEAD_INITIALIZER(root_holds);
115 
116 enum action {
117 	A_CONTINUE,
118 	A_PANIC,
119 	A_REBOOT,
120 	A_RETRY
121 };
122 
123 static enum action root_mount_onfail = A_CONTINUE;
124 
125 static int root_mount_mddev;
126 static int root_mount_complete;
127 
128 /* By default wait up to 3 seconds for devices to appear. */
129 static int root_mount_timeout = 3;
130 TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout);
131 
132 struct root_hold_token *
133 root_mount_hold(const char *identifier)
134 {
135 	struct root_hold_token *h;
136 
137 	if (root_mounted())
138 		return (NULL);
139 
140 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
141 	h->who = identifier;
142 	mtx_lock(&root_holds_mtx);
143 	LIST_INSERT_HEAD(&root_holds, h, list);
144 	mtx_unlock(&root_holds_mtx);
145 	return (h);
146 }
147 
148 void
149 root_mount_rel(struct root_hold_token *h)
150 {
151 
152 	if (h == NULL)
153 		return;
154 	mtx_lock(&root_holds_mtx);
155 	LIST_REMOVE(h, list);
156 	wakeup(&root_holds);
157 	mtx_unlock(&root_holds_mtx);
158 	free(h, M_DEVBUF);
159 }
160 
161 int
162 root_mounted(void)
163 {
164 
165 	/* No mutex is acquired here because int stores are atomic. */
166 	return (root_mount_complete);
167 }
168 
169 void
170 root_mount_wait(void)
171 {
172 
173 	/*
174 	 * Panic on an obvious deadlock - the function can't be called from
175 	 * a thread which is doing the whole SYSINIT stuff.
176 	 */
177 	KASSERT(curthread->td_proc->p_pid != 0,
178 	    ("root_mount_wait: cannot be called from the swapper thread"));
179 	mtx_lock(&root_holds_mtx);
180 	while (!root_mount_complete) {
181 		msleep(&root_mount_complete, &root_holds_mtx, PZERO, "rootwait",
182 		    hz);
183 	}
184 	mtx_unlock(&root_holds_mtx);
185 }
186 
187 static void
188 set_rootvnode(void)
189 {
190 	struct proc *p;
191 
192 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
193 		panic("Cannot find root vnode");
194 
195 	VOP_UNLOCK(rootvnode, 0);
196 
197 	p = curthread->td_proc;
198 	FILEDESC_XLOCK(p->p_fd);
199 
200 	if (p->p_fd->fd_cdir != NULL)
201 		vrele(p->p_fd->fd_cdir);
202 	p->p_fd->fd_cdir = rootvnode;
203 	VREF(rootvnode);
204 
205 	if (p->p_fd->fd_rdir != NULL)
206 		vrele(p->p_fd->fd_rdir);
207 	p->p_fd->fd_rdir = rootvnode;
208 	VREF(rootvnode);
209 
210 	FILEDESC_XUNLOCK(p->p_fd);
211 }
212 
213 static int
214 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
215 {
216 	struct vfsoptlist *opts;
217 	struct vfsconf *vfsp;
218 	struct mount *mp;
219 	int error;
220 
221 	*mpp = NULL;
222 
223 	if (rootdevmp != NULL) {
224 		/*
225 		 * Already have /dev; this happens during rerooting.
226 		 */
227 		error = vfs_busy(rootdevmp, 0);
228 		if (error != 0)
229 			return (error);
230 		*mpp = rootdevmp;
231 	} else {
232 		vfsp = vfs_byname("devfs");
233 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
234 		if (vfsp == NULL)
235 			return (ENOENT);
236 
237 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
238 
239 		error = VFS_MOUNT(mp);
240 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
241 		if (error)
242 			return (error);
243 
244 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
245 		TAILQ_INIT(opts);
246 		mp->mnt_opt = opts;
247 
248 		mtx_lock(&mountlist_mtx);
249 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
250 		mtx_unlock(&mountlist_mtx);
251 
252 		*mpp = mp;
253 		rootdevmp = mp;
254 	}
255 
256 	set_rootvnode();
257 
258 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
259 	if (error)
260 		printf("kern_symlink /dev -> / returns %d\n", error);
261 
262 	return (error);
263 }
264 
265 static void
266 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
267 {
268 	struct nameidata nd;
269 	struct mount *mporoot, *mpnroot;
270 	struct vnode *vp, *vporoot, *vpdevfs;
271 	char *fspath;
272 	int error;
273 
274 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
275 
276 	/* Shuffle the mountlist. */
277 	mtx_lock(&mountlist_mtx);
278 	mporoot = TAILQ_FIRST(&mountlist);
279 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
280 	if (mporoot != mpdevfs) {
281 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
282 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
283 	}
284 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
285 	mtx_unlock(&mountlist_mtx);
286 
287 	cache_purgevfs(mporoot);
288 	if (mporoot != mpdevfs)
289 		cache_purgevfs(mpdevfs);
290 
291 	VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot);
292 
293 	VI_LOCK(vporoot);
294 	vporoot->v_iflag &= ~VI_MOUNT;
295 	VI_UNLOCK(vporoot);
296 	vporoot->v_mountedhere = NULL;
297 	mporoot->mnt_flag &= ~MNT_ROOTFS;
298 	mporoot->mnt_vnodecovered = NULL;
299 	vput(vporoot);
300 
301 	/* Set up the new rootvnode, and purge the cache */
302 	mpnroot->mnt_vnodecovered = NULL;
303 	set_rootvnode();
304 	cache_purgevfs(rootvnode->v_mount);
305 
306 	if (mporoot != mpdevfs) {
307 		/* Remount old root under /.mount or /mnt */
308 		fspath = "/.mount";
309 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
310 		    fspath, td);
311 		error = namei(&nd);
312 		if (error) {
313 			NDFREE(&nd, NDF_ONLY_PNBUF);
314 			fspath = "/mnt";
315 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
316 			    fspath, td);
317 			error = namei(&nd);
318 		}
319 		if (!error) {
320 			vp = nd.ni_vp;
321 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
322 			if (!error)
323 				error = vinvalbuf(vp, V_SAVE, 0, 0);
324 			if (!error) {
325 				cache_purge(vp);
326 				mporoot->mnt_vnodecovered = vp;
327 				vp->v_mountedhere = mporoot;
328 				strlcpy(mporoot->mnt_stat.f_mntonname,
329 				    fspath, MNAMELEN);
330 				VOP_UNLOCK(vp, 0);
331 			} else
332 				vput(vp);
333 		}
334 		NDFREE(&nd, NDF_ONLY_PNBUF);
335 
336 		if (error && bootverbose)
337 			printf("mountroot: unable to remount previous root "
338 			    "under /.mount or /mnt (error %d).\n", error);
339 	}
340 
341 	/* Remount devfs under /dev */
342 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
343 	error = namei(&nd);
344 	if (!error) {
345 		vp = nd.ni_vp;
346 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
347 		if (!error)
348 			error = vinvalbuf(vp, V_SAVE, 0, 0);
349 		if (!error) {
350 			vpdevfs = mpdevfs->mnt_vnodecovered;
351 			if (vpdevfs != NULL) {
352 				cache_purge(vpdevfs);
353 				vpdevfs->v_mountedhere = NULL;
354 				vrele(vpdevfs);
355 			}
356 			mpdevfs->mnt_vnodecovered = vp;
357 			vp->v_mountedhere = mpdevfs;
358 			VOP_UNLOCK(vp, 0);
359 		} else
360 			vput(vp);
361 	}
362 	if (error && bootverbose)
363 		printf("mountroot: unable to remount devfs under /dev "
364 		    "(error %d).\n", error);
365 	NDFREE(&nd, NDF_ONLY_PNBUF);
366 
367 	if (mporoot == mpdevfs) {
368 		vfs_unbusy(mpdevfs);
369 		/* Unlink the no longer needed /dev/dev -> / symlink */
370 		error = kern_unlinkat(td, AT_FDCWD, "/dev/dev",
371 		    UIO_SYSSPACE, 0);
372 		if (error && bootverbose)
373 			printf("mountroot: unable to unlink /dev/dev "
374 			    "(error %d)\n", error);
375 	}
376 }
377 
378 /*
379  * Configuration parser.
380  */
381 
382 /* Parser character classes. */
383 #define	CC_WHITESPACE		-1
384 #define	CC_NONWHITESPACE	-2
385 
386 /* Parse errors. */
387 #define	PE_EOF			-1
388 #define	PE_EOL			-2
389 
390 static __inline int
391 parse_peek(char **conf)
392 {
393 
394 	return (**conf);
395 }
396 
397 static __inline void
398 parse_poke(char **conf, int c)
399 {
400 
401 	**conf = c;
402 }
403 
404 static __inline void
405 parse_advance(char **conf)
406 {
407 
408 	(*conf)++;
409 }
410 
411 static int
412 parse_skipto(char **conf, int mc)
413 {
414 	int c, match;
415 
416 	while (1) {
417 		c = parse_peek(conf);
418 		if (c == 0)
419 			return (PE_EOF);
420 		switch (mc) {
421 		case CC_WHITESPACE:
422 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
423 			break;
424 		case CC_NONWHITESPACE:
425 			if (c == '\n')
426 				return (PE_EOL);
427 			match = (c != ' ' && c != '\t') ? 1 : 0;
428 			break;
429 		default:
430 			match = (c == mc) ? 1 : 0;
431 			break;
432 		}
433 		if (match)
434 			break;
435 		parse_advance(conf);
436 	}
437 	return (0);
438 }
439 
440 static int
441 parse_token(char **conf, char **tok)
442 {
443 	char *p;
444 	size_t len;
445 	int error;
446 
447 	*tok = NULL;
448 	error = parse_skipto(conf, CC_NONWHITESPACE);
449 	if (error)
450 		return (error);
451 	p = *conf;
452 	error = parse_skipto(conf, CC_WHITESPACE);
453 	len = *conf - p;
454 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
455 	bcopy(p, *tok, len);
456 	return (0);
457 }
458 
459 static void
460 parse_dir_ask_printenv(const char *var)
461 {
462 	char *val;
463 
464 	val = kern_getenv(var);
465 	if (val != NULL) {
466 		printf("  %s=%s\n", var, val);
467 		freeenv(val);
468 	}
469 }
470 
471 static int
472 parse_dir_ask(char **conf)
473 {
474 	char name[80];
475 	char *mnt;
476 	int error;
477 
478 	printf("\nLoader variables:\n");
479 	parse_dir_ask_printenv("vfs.root.mountfrom");
480 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
481 
482 	printf("\nManual root filesystem specification:\n");
483 	printf("  <fstype>:<device> [options]\n");
484 	printf("      Mount <device> using filesystem <fstype>\n");
485 	printf("      and with the specified (optional) option list.\n");
486 	printf("\n");
487 	printf("    eg. ufs:/dev/da0s1a\n");
488 	printf("        zfs:tank\n");
489 	printf("        cd9660:/dev/acd0 ro\n");
490 	printf("          (which is equivalent to: ");
491 	printf("mount -t cd9660 -o ro /dev/acd0 /)\n");
492 	printf("\n");
493 	printf("  ?               List valid disk boot devices\n");
494 	printf("  .               Yield 1 second (for background tasks)\n");
495 	printf("  <empty line>    Abort manual input\n");
496 
497 	do {
498 		error = EINVAL;
499 		printf("\nmountroot> ");
500 		cngets(name, sizeof(name), GETS_ECHO);
501 		if (name[0] == '\0')
502 			break;
503 		if (name[0] == '?' && name[1] == '\0') {
504 			printf("\nList of GEOM managed disk devices:\n  ");
505 			g_dev_print();
506 			continue;
507 		}
508 		if (name[0] == '.' && name[1] == '\0') {
509 			pause("rmask", hz);
510 			continue;
511 		}
512 		mnt = name;
513 		error = parse_mount(&mnt);
514 		if (error == -1)
515 			printf("Invalid file system specification.\n");
516 	} while (error != 0);
517 
518 	return (error);
519 }
520 
521 static int
522 parse_dir_md(char **conf)
523 {
524 	struct stat sb;
525 	struct thread *td;
526 	struct md_ioctl *mdio;
527 	char *path, *tok;
528 	int error, fd, len;
529 
530 	td = curthread;
531 
532 	error = parse_token(conf, &tok);
533 	if (error)
534 		return (error);
535 
536 	len = strlen(tok);
537 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
538 	path = (void *)(mdio + 1);
539 	bcopy(tok, path, len);
540 	free(tok, M_TEMP);
541 
542 	/* Get file status. */
543 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
544 	if (error)
545 		goto out;
546 
547 	/* Open /dev/mdctl so that we can attach/detach. */
548 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
549 	    O_RDWR, 0);
550 	if (error)
551 		goto out;
552 
553 	fd = td->td_retval[0];
554 	mdio->md_version = MDIOVERSION;
555 	mdio->md_type = MD_VNODE;
556 
557 	if (root_mount_mddev != -1) {
558 		mdio->md_unit = root_mount_mddev;
559 		DROP_GIANT();
560 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
561 		PICKUP_GIANT();
562 		/* Ignore errors. We don't care. */
563 		root_mount_mddev = -1;
564 	}
565 
566 	mdio->md_file = (void *)(mdio + 1);
567 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
568 	mdio->md_mediasize = sb.st_size;
569 	mdio->md_unit = 0;
570 	DROP_GIANT();
571 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
572 	PICKUP_GIANT();
573 	if (error)
574 		goto out;
575 
576 	if (mdio->md_unit > 9) {
577 		printf("rootmount: too many md units\n");
578 		mdio->md_file = NULL;
579 		mdio->md_options = 0;
580 		mdio->md_mediasize = 0;
581 		DROP_GIANT();
582 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
583 		PICKUP_GIANT();
584 		/* Ignore errors. We don't care. */
585 		error = ERANGE;
586 		goto out;
587 	}
588 
589 	root_mount_mddev = mdio->md_unit;
590 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
591 
592 	error = kern_close(td, fd);
593 
594  out:
595 	free(mdio, M_TEMP);
596 	return (error);
597 }
598 
599 static int
600 parse_dir_onfail(char **conf)
601 {
602 	char *action;
603 	int error;
604 
605 	error = parse_token(conf, &action);
606 	if (error)
607 		return (error);
608 
609 	if (!strcmp(action, "continue"))
610 		root_mount_onfail = A_CONTINUE;
611 	else if (!strcmp(action, "panic"))
612 		root_mount_onfail = A_PANIC;
613 	else if (!strcmp(action, "reboot"))
614 		root_mount_onfail = A_REBOOT;
615 	else if (!strcmp(action, "retry"))
616 		root_mount_onfail = A_RETRY;
617 	else {
618 		printf("rootmount: %s: unknown action\n", action);
619 		error = EINVAL;
620 	}
621 
622 	free(action, M_TEMP);
623 	return (0);
624 }
625 
626 static int
627 parse_dir_timeout(char **conf)
628 {
629 	char *tok, *endtok;
630 	long secs;
631 	int error;
632 
633 	error = parse_token(conf, &tok);
634 	if (error)
635 		return (error);
636 
637 	secs = strtol(tok, &endtok, 0);
638 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
639 	if (!error)
640 		root_mount_timeout = secs;
641 	free(tok, M_TEMP);
642 	return (error);
643 }
644 
645 static int
646 parse_directive(char **conf)
647 {
648 	char *dir;
649 	int error;
650 
651 	error = parse_token(conf, &dir);
652 	if (error)
653 		return (error);
654 
655 	if (strcmp(dir, ".ask") == 0)
656 		error = parse_dir_ask(conf);
657 	else if (strcmp(dir, ".md") == 0)
658 		error = parse_dir_md(conf);
659 	else if (strcmp(dir, ".onfail") == 0)
660 		error = parse_dir_onfail(conf);
661 	else if (strcmp(dir, ".timeout") == 0)
662 		error = parse_dir_timeout(conf);
663 	else {
664 		printf("mountroot: invalid directive `%s'\n", dir);
665 		/* Ignore the rest of the line. */
666 		(void)parse_skipto(conf, '\n');
667 		error = EINVAL;
668 	}
669 	free(dir, M_TEMP);
670 	return (error);
671 }
672 
673 static int
674 parse_mount_dev_present(const char *dev)
675 {
676 	struct nameidata nd;
677 	int error;
678 
679 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
680 	error = namei(&nd);
681 	if (!error)
682 		vput(nd.ni_vp);
683 	NDFREE(&nd, NDF_ONLY_PNBUF);
684 	return (error != 0) ? 0 : 1;
685 }
686 
687 #define	ERRMSGL	255
688 static int
689 parse_mount(char **conf)
690 {
691 	char *errmsg;
692 	struct mntarg *ma;
693 	char *dev, *fs, *opts, *tok;
694 	int delay, error, timeout;
695 
696 	error = parse_token(conf, &tok);
697 	if (error)
698 		return (error);
699 	fs = tok;
700 	error = parse_skipto(&tok, ':');
701 	if (error) {
702 		free(fs, M_TEMP);
703 		return (error);
704 	}
705 	parse_poke(&tok, '\0');
706 	parse_advance(&tok);
707 	dev = tok;
708 
709 	if (root_mount_mddev != -1) {
710 		/* Handle substitution for the md unit number. */
711 		tok = strstr(dev, "md#");
712 		if (tok != NULL)
713 			tok[2] = '0' + root_mount_mddev;
714 	}
715 
716 	/* Parse options. */
717 	error = parse_token(conf, &tok);
718 	opts = (error == 0) ? tok : NULL;
719 
720 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
721 	    (opts != NULL) ? opts : "");
722 
723 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
724 
725 	if (vfs_byname(fs) == NULL) {
726 		strlcpy(errmsg, "unknown file system", ERRMSGL);
727 		error = ENOENT;
728 		goto out;
729 	}
730 
731 	if (strcmp(fs, "zfs") != 0 && strstr(fs, "nfs") == NULL &&
732 	    dev[0] != '\0' && !parse_mount_dev_present(dev)) {
733 		printf("mountroot: waiting for device %s ...\n", dev);
734 		delay = hz / 10;
735 		timeout = root_mount_timeout * hz;
736 		do {
737 			pause("rmdev", delay);
738 			timeout -= delay;
739 		} while (timeout > 0 && !parse_mount_dev_present(dev));
740 		if (timeout <= 0) {
741 			error = ENODEV;
742 			goto out;
743 		}
744 	}
745 
746 	ma = NULL;
747 	ma = mount_arg(ma, "fstype", fs, -1);
748 	ma = mount_arg(ma, "fspath", "/", -1);
749 	ma = mount_arg(ma, "from", dev, -1);
750 	ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
751 	ma = mount_arg(ma, "ro", NULL, 0);
752 	ma = parse_mountroot_options(ma, opts);
753 	error = kernel_mount(ma, MNT_ROOTFS);
754 
755  out:
756 	if (error) {
757 		printf("Mounting from %s:%s failed with error %d",
758 		    fs, dev, error);
759 		if (errmsg[0] != '\0')
760 			printf(": %s", errmsg);
761 		printf(".\n");
762 	}
763 	free(fs, M_TEMP);
764 	free(errmsg, M_TEMP);
765 	if (opts != NULL)
766 		free(opts, M_TEMP);
767 	/* kernel_mount can return -1 on error. */
768 	return ((error < 0) ? EDOOFUS : error);
769 }
770 #undef ERRMSGL
771 
772 static int
773 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
774 {
775 	struct mount *mp;
776 	char *conf;
777 	int error;
778 
779 	root_mount_mddev = -1;
780 
781 retry:
782 	conf = sbuf_data(sb);
783 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
784 	error = (mp == NULL) ? 0 : EDOOFUS;
785 	root_mount_onfail = A_CONTINUE;
786 	while (mp == NULL) {
787 		error = parse_skipto(&conf, CC_NONWHITESPACE);
788 		if (error == PE_EOL) {
789 			parse_advance(&conf);
790 			continue;
791 		}
792 		if (error < 0)
793 			break;
794 		switch (parse_peek(&conf)) {
795 		case '#':
796 			error = parse_skipto(&conf, '\n');
797 			break;
798 		case '.':
799 			error = parse_directive(&conf);
800 			break;
801 		default:
802 			error = parse_mount(&conf);
803 			if (error == -1) {
804 				printf("mountroot: invalid file system "
805 				    "specification.\n");
806 				error = 0;
807 			}
808 			break;
809 		}
810 		if (error < 0)
811 			break;
812 		/* Ignore any trailing garbage on the line. */
813 		if (parse_peek(&conf) != '\n') {
814 			printf("mountroot: advancing to next directive...\n");
815 			(void)parse_skipto(&conf, '\n');
816 		}
817 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
818 	}
819 	if (mp != NULL)
820 		return (0);
821 
822 	/*
823 	 * We failed to mount (a new) root.
824 	 */
825 	switch (root_mount_onfail) {
826 	case A_CONTINUE:
827 		break;
828 	case A_PANIC:
829 		panic("mountroot: unable to (re-)mount root.");
830 		/* NOTREACHED */
831 	case A_RETRY:
832 		goto retry;
833 	case A_REBOOT:
834 		kern_reboot(RB_NOSYNC);
835 		/* NOTREACHED */
836 	}
837 
838 	return (error);
839 }
840 
841 static void
842 vfs_mountroot_conf0(struct sbuf *sb)
843 {
844 	char *s, *tok, *mnt, *opt;
845 	int error;
846 
847 	sbuf_printf(sb, ".onfail panic\n");
848 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
849 	if (boothowto & RB_ASKNAME)
850 		sbuf_printf(sb, ".ask\n");
851 #ifdef ROOTDEVNAME
852 	if (boothowto & RB_DFLTROOT)
853 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
854 #endif
855 	if (boothowto & RB_CDROM) {
856 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
857 		sbuf_printf(sb, ".timeout 0\n");
858 		sbuf_printf(sb, "cd9660:/dev/acd0 ro\n");
859 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
860 	}
861 	s = kern_getenv("vfs.root.mountfrom");
862 	if (s != NULL) {
863 		opt = kern_getenv("vfs.root.mountfrom.options");
864 		tok = s;
865 		error = parse_token(&tok, &mnt);
866 		while (!error) {
867 			sbuf_printf(sb, "%s %s\n", mnt,
868 			    (opt != NULL) ? opt : "");
869 			free(mnt, M_TEMP);
870 			error = parse_token(&tok, &mnt);
871 		}
872 		if (opt != NULL)
873 			freeenv(opt);
874 		freeenv(s);
875 	}
876 	if (rootdevnames[0] != NULL)
877 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
878 	if (rootdevnames[1] != NULL)
879 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
880 #ifdef ROOTDEVNAME
881 	if (!(boothowto & RB_DFLTROOT))
882 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
883 #endif
884 	if (!(boothowto & RB_ASKNAME))
885 		sbuf_printf(sb, ".ask\n");
886 }
887 
888 static int
889 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
890 {
891 	static char buf[128];
892 	struct nameidata nd;
893 	off_t ofs;
894 	ssize_t resid;
895 	int error, flags, len;
896 
897 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
898 	flags = FREAD;
899 	error = vn_open(&nd, &flags, 0, NULL);
900 	if (error)
901 		return (error);
902 
903 	NDFREE(&nd, NDF_ONLY_PNBUF);
904 	ofs = 0;
905 	len = sizeof(buf) - 1;
906 	while (1) {
907 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
908 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
909 		    NOCRED, &resid, td);
910 		if (error)
911 			break;
912 		if (resid == len)
913 			break;
914 		buf[len - resid] = 0;
915 		sbuf_printf(sb, "%s", buf);
916 		ofs += len - resid;
917 	}
918 
919 	VOP_UNLOCK(nd.ni_vp, 0);
920 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
921 	return (error);
922 }
923 
924 static void
925 vfs_mountroot_wait(void)
926 {
927 	struct root_hold_token *h;
928 	struct timeval lastfail;
929 	int curfail;
930 
931 	curfail = 0;
932 	while (1) {
933 		DROP_GIANT();
934 		g_waitidle();
935 		PICKUP_GIANT();
936 		mtx_lock(&root_holds_mtx);
937 		if (LIST_EMPTY(&root_holds)) {
938 			mtx_unlock(&root_holds_mtx);
939 			break;
940 		}
941 		if (ppsratecheck(&lastfail, &curfail, 1)) {
942 			printf("Root mount waiting for:");
943 			LIST_FOREACH(h, &root_holds, list)
944 				printf(" %s", h->who);
945 			printf("\n");
946 		}
947 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
948 		    hz);
949 	}
950 }
951 
952 void
953 vfs_mountroot(void)
954 {
955 	struct mount *mp;
956 	struct sbuf *sb;
957 	struct thread *td;
958 	time_t timebase;
959 	int error;
960 
961 	td = curthread;
962 
963 	vfs_mountroot_wait();
964 
965 	sb = sbuf_new_auto();
966 	vfs_mountroot_conf0(sb);
967 	sbuf_finish(sb);
968 
969 	error = vfs_mountroot_devfs(td, &mp);
970 	while (!error) {
971 		error = vfs_mountroot_parse(sb, mp);
972 		if (!error) {
973 			vfs_mountroot_shuffle(td, mp);
974 			sbuf_clear(sb);
975 			error = vfs_mountroot_readconf(td, sb);
976 			sbuf_finish(sb);
977 		}
978 	}
979 
980 	sbuf_delete(sb);
981 
982 	/*
983 	 * Iterate over all currently mounted file systems and use
984 	 * the time stamp found to check and/or initialize the RTC.
985 	 * Call inittodr() only once and pass it the largest of the
986 	 * timestamps we encounter.
987 	 */
988 	timebase = 0;
989 	mtx_lock(&mountlist_mtx);
990 	mp = TAILQ_FIRST(&mountlist);
991 	while (mp != NULL) {
992 		if (mp->mnt_time > timebase)
993 			timebase = mp->mnt_time;
994 		mp = TAILQ_NEXT(mp, mnt_list);
995 	}
996 	mtx_unlock(&mountlist_mtx);
997 	inittodr(timebase);
998 
999 	/* Keep prison0's root in sync with the global rootvnode. */
1000 	mtx_lock(&prison0.pr_mtx);
1001 	prison0.pr_root = rootvnode;
1002 	vref(prison0.pr_root);
1003 	mtx_unlock(&prison0.pr_mtx);
1004 
1005 	mtx_lock(&root_holds_mtx);
1006 	atomic_store_rel_int(&root_mount_complete, 1);
1007 	wakeup(&root_mount_complete);
1008 	mtx_unlock(&root_holds_mtx);
1009 
1010 	EVENTHANDLER_INVOKE(mountroot);
1011 }
1012 
1013 static struct mntarg *
1014 parse_mountroot_options(struct mntarg *ma, const char *options)
1015 {
1016 	char *p;
1017 	char *name, *name_arg;
1018 	char *val, *val_arg;
1019 	char *opts;
1020 
1021 	if (options == NULL || options[0] == '\0')
1022 		return (ma);
1023 
1024 	p = opts = strdup(options, M_MOUNT);
1025 	if (opts == NULL) {
1026 		return (ma);
1027 	}
1028 
1029 	while((name = strsep(&p, ",")) != NULL) {
1030 		if (name[0] == '\0')
1031 			break;
1032 
1033 		val = strchr(name, '=');
1034 		if (val != NULL) {
1035 			*val = '\0';
1036 			++val;
1037 		}
1038 		if( strcmp(name, "rw") == 0 ||
1039 		    strcmp(name, "noro") == 0) {
1040 			/*
1041 			 * The first time we mount the root file system,
1042 			 * we need to mount 'ro', so We need to ignore
1043 			 * 'rw' and 'noro' mount options.
1044 			 */
1045 			continue;
1046 		}
1047 		name_arg = strdup(name, M_MOUNT);
1048 		val_arg = NULL;
1049 		if (val != NULL)
1050 			val_arg = strdup(val, M_MOUNT);
1051 
1052 		ma = mount_arg(ma, name_arg, val_arg,
1053 		    (val_arg != NULL ? -1 : 0));
1054 	}
1055 	free(opts, M_MOUNT);
1056 	return (ma);
1057 }
1058