xref: /freebsd/sys/kern/vfs_mountroot.c (revision 48c5129f93c5eb5419c87b08e4677d51513f1dc0)
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 char *rootdevnames[2] = {NULL, NULL};
99 
100 struct root_hold_token {
101 	const char			*who;
102 	LIST_ENTRY(root_hold_token)	list;
103 };
104 
105 static LIST_HEAD(, root_hold_token)	root_holds =
106     LIST_HEAD_INITIALIZER(root_holds);
107 
108 enum action {
109 	A_CONTINUE,
110 	A_PANIC,
111 	A_REBOOT,
112 	A_RETRY
113 };
114 
115 static enum action root_mount_onfail = A_CONTINUE;
116 
117 static int root_mount_mddev;
118 static int root_mount_complete;
119 
120 /* By default wait up to 3 seconds for devices to appear. */
121 static int root_mount_timeout = 3;
122 
123 struct root_hold_token *
124 root_mount_hold(const char *identifier)
125 {
126 	struct root_hold_token *h;
127 
128 	if (root_mounted())
129 		return (NULL);
130 
131 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
132 	h->who = identifier;
133 	mtx_lock(&mountlist_mtx);
134 	LIST_INSERT_HEAD(&root_holds, h, list);
135 	mtx_unlock(&mountlist_mtx);
136 	return (h);
137 }
138 
139 void
140 root_mount_rel(struct root_hold_token *h)
141 {
142 
143 	if (h == NULL)
144 		return;
145 	mtx_lock(&mountlist_mtx);
146 	LIST_REMOVE(h, list);
147 	wakeup(&root_holds);
148 	mtx_unlock(&mountlist_mtx);
149 	free(h, M_DEVBUF);
150 }
151 
152 int
153 root_mounted(void)
154 {
155 
156 	/* No mutex is acquired here because int stores are atomic. */
157 	return (root_mount_complete);
158 }
159 
160 void
161 root_mount_wait(void)
162 {
163 
164 	/*
165 	 * Panic on an obvious deadlock - the function can't be called from
166 	 * a thread which is doing the whole SYSINIT stuff.
167 	 */
168 	KASSERT(curthread->td_proc->p_pid != 0,
169 	    ("root_mount_wait: cannot be called from the swapper thread"));
170 	mtx_lock(&mountlist_mtx);
171 	while (!root_mount_complete) {
172 		msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
173 		    hz);
174 	}
175 	mtx_unlock(&mountlist_mtx);
176 }
177 
178 static void
179 set_rootvnode(void)
180 {
181 	struct proc *p;
182 
183 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
184 		panic("Cannot find root vnode");
185 
186 	VOP_UNLOCK(rootvnode, 0);
187 
188 	p = curthread->td_proc;
189 	FILEDESC_XLOCK(p->p_fd);
190 
191 	if (p->p_fd->fd_cdir != NULL)
192 		vrele(p->p_fd->fd_cdir);
193 	p->p_fd->fd_cdir = rootvnode;
194 	VREF(rootvnode);
195 
196 	if (p->p_fd->fd_rdir != NULL)
197 		vrele(p->p_fd->fd_rdir);
198 	p->p_fd->fd_rdir = rootvnode;
199 	VREF(rootvnode);
200 
201 	FILEDESC_XUNLOCK(p->p_fd);
202 
203 	EVENTHANDLER_INVOKE(mountroot);
204 }
205 
206 static int
207 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
208 {
209 	struct vfsoptlist *opts;
210 	struct vfsconf *vfsp;
211 	struct mount *mp;
212 	int error;
213 
214 	*mpp = NULL;
215 
216 	vfsp = vfs_byname("devfs");
217 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
218 	if (vfsp == NULL)
219 		return (ENOENT);
220 
221 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
222 
223 	error = VFS_MOUNT(mp);
224 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
225 	if (error)
226 		return (error);
227 
228 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
229 	TAILQ_INIT(opts);
230 	mp->mnt_opt = opts;
231 
232 	mtx_lock(&mountlist_mtx);
233 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
234 	mtx_unlock(&mountlist_mtx);
235 
236 	*mpp = mp;
237 	set_rootvnode();
238 
239 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
240 	if (error)
241 		printf("kern_symlink /dev -> / returns %d\n", error);
242 
243 	return (error);
244 }
245 
246 static int
247 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
248 {
249 	struct nameidata nd;
250 	struct mount *mporoot, *mpnroot;
251 	struct vnode *vp, *vporoot, *vpdevfs;
252 	char *fspath;
253 	int error;
254 
255 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
256 
257 	/* Shuffle the mountlist. */
258 	mtx_lock(&mountlist_mtx);
259 	mporoot = TAILQ_FIRST(&mountlist);
260 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
261 	if (mporoot != mpdevfs) {
262 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
263 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
264 	}
265 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
266 	mtx_unlock(&mountlist_mtx);
267 
268 	cache_purgevfs(mporoot);
269 	if (mporoot != mpdevfs)
270 		cache_purgevfs(mpdevfs);
271 
272 	VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot);
273 
274 	VI_LOCK(vporoot);
275 	vporoot->v_iflag &= ~VI_MOUNT;
276 	VI_UNLOCK(vporoot);
277 	vporoot->v_mountedhere = NULL;
278 	mporoot->mnt_flag &= ~MNT_ROOTFS;
279 	mporoot->mnt_vnodecovered = NULL;
280 	vput(vporoot);
281 
282 	/* Set up the new rootvnode, and purge the cache */
283 	mpnroot->mnt_vnodecovered = NULL;
284 	set_rootvnode();
285 	cache_purgevfs(rootvnode->v_mount);
286 
287 	if (mporoot != mpdevfs) {
288 		/* Remount old root under /.mount or /mnt */
289 		fspath = "/.mount";
290 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
291 		    fspath, td);
292 		error = namei(&nd);
293 		if (error) {
294 			NDFREE(&nd, NDF_ONLY_PNBUF);
295 			fspath = "/mnt";
296 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
297 			    fspath, td);
298 			error = namei(&nd);
299 		}
300 		if (!error) {
301 			vp = nd.ni_vp;
302 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
303 			if (!error)
304 				error = vinvalbuf(vp, V_SAVE, 0, 0);
305 			if (!error) {
306 				cache_purge(vp);
307 				mporoot->mnt_vnodecovered = vp;
308 				vp->v_mountedhere = mporoot;
309 				strlcpy(mporoot->mnt_stat.f_mntonname,
310 				    fspath, MNAMELEN);
311 				VOP_UNLOCK(vp, 0);
312 			} else
313 				vput(vp);
314 		}
315 		NDFREE(&nd, NDF_ONLY_PNBUF);
316 
317 		if (error && bootverbose)
318 			printf("mountroot: unable to remount previous root "
319 			    "under /.mount or /mnt (error %d).\n", error);
320 	}
321 
322 	/* Remount devfs under /dev */
323 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
324 	error = namei(&nd);
325 	if (!error) {
326 		vp = nd.ni_vp;
327 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
328 		if (!error)
329 			error = vinvalbuf(vp, V_SAVE, 0, 0);
330 		if (!error) {
331 			vpdevfs = mpdevfs->mnt_vnodecovered;
332 			if (vpdevfs != NULL) {
333 				cache_purge(vpdevfs);
334 				vpdevfs->v_mountedhere = NULL;
335 				vrele(vpdevfs);
336 			}
337 			mpdevfs->mnt_vnodecovered = vp;
338 			vp->v_mountedhere = mpdevfs;
339 			VOP_UNLOCK(vp, 0);
340 		} else
341 			vput(vp);
342 	}
343 	if (error && bootverbose)
344 		printf("mountroot: unable to remount devfs under /dev "
345 		    "(error %d).\n", error);
346 	NDFREE(&nd, NDF_ONLY_PNBUF);
347 
348 	if (mporoot == mpdevfs) {
349 		vfs_unbusy(mpdevfs);
350 		/* Unlink the no longer needed /dev/dev -> / symlink */
351 		error = kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
352 		if (error && bootverbose)
353 			printf("mountroot: unable to unlink /dev/dev "
354 			    "(error %d)\n", error);
355 	}
356 
357 	return (0);
358 }
359 
360 /*
361  * Configuration parser.
362  */
363 
364 /* Parser character classes. */
365 #define	CC_WHITESPACE		-1
366 #define	CC_NONWHITESPACE	-2
367 
368 /* Parse errors. */
369 #define	PE_EOF			-1
370 #define	PE_EOL			-2
371 
372 static __inline int
373 parse_peek(char **conf)
374 {
375 
376 	return (**conf);
377 }
378 
379 static __inline void
380 parse_poke(char **conf, int c)
381 {
382 
383 	**conf = c;
384 }
385 
386 static __inline void
387 parse_advance(char **conf)
388 {
389 
390 	(*conf)++;
391 }
392 
393 static __inline int
394 parse_isspace(int c)
395 {
396 
397 	return ((c == ' ' || c == '\t' || c == '\n') ? 1 : 0);
398 }
399 
400 static int
401 parse_skipto(char **conf, int mc)
402 {
403 	int c, match;
404 
405 	while (1) {
406 		c = parse_peek(conf);
407 		if (c == 0)
408 			return (PE_EOF);
409 		switch (mc) {
410 		case CC_WHITESPACE:
411 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
412 			break;
413 		case CC_NONWHITESPACE:
414 			if (c == '\n')
415 				return (PE_EOL);
416 			match = (c != ' ' && c != '\t') ? 1 : 0;
417 			break;
418 		default:
419 			match = (c == mc) ? 1 : 0;
420 			break;
421 		}
422 		if (match)
423 			break;
424 		parse_advance(conf);
425 	}
426 	return (0);
427 }
428 
429 static int
430 parse_token(char **conf, char **tok)
431 {
432 	char *p;
433 	size_t len;
434 	int error;
435 
436 	*tok = NULL;
437 	error = parse_skipto(conf, CC_NONWHITESPACE);
438 	if (error)
439 		return (error);
440 	p = *conf;
441 	error = parse_skipto(conf, CC_WHITESPACE);
442 	len = *conf - p;
443 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
444 	bcopy(p, *tok, len);
445 	return (0);
446 }
447 
448 static void
449 parse_dir_ask_printenv(const char *var)
450 {
451 	char *val;
452 
453 	val = getenv(var);
454 	if (val != NULL) {
455 		printf("  %s=%s\n", var, val);
456 		freeenv(val);
457 	}
458 }
459 
460 static int
461 parse_dir_ask(char **conf)
462 {
463 	char name[80];
464 	char *mnt;
465 	int error;
466 
467 	printf("\nLoader variables:\n");
468 	parse_dir_ask_printenv("vfs.root.mountfrom");
469 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
470 
471 	printf("\nManual root filesystem specification:\n");
472 	printf("  <fstype>:<device> [options]\n");
473 	printf("      Mount <device> using filesystem <fstype>\n");
474 	printf("      and with the specified (optional) option list.\n");
475 	printf("\n");
476 	printf("    eg. ufs:/dev/da0s1a\n");
477 	printf("        zfs:tank\n");
478 	printf("        cd9660:/dev/acd0 ro\n");
479 	printf("          (which is equivalent to: ");
480 	printf("mount -t cd9660 -o ro /dev/acd0 /)\n");
481 	printf("\n");
482 	printf("  ?               List valid disk boot devices\n");
483 	printf("  .               Yield 1 second (for background tasks)\n");
484 	printf("  <empty line>    Abort manual input\n");
485 
486 	do {
487 		error = EINVAL;
488 		printf("\nmountroot> ");
489 		cngets(name, sizeof(name), GETS_ECHO);
490 		if (name[0] == '\0')
491 			break;
492 		if (name[0] == '?' && name[1] == '\0') {
493 			printf("\nList of GEOM managed disk devices:\n  ");
494 			g_dev_print();
495 			continue;
496 		}
497 		if (name[0] == '.' && name[1] == '\0') {
498 			pause("rmask", hz);
499 			continue;
500 		}
501 		mnt = name;
502 		error = parse_mount(&mnt);
503 		if (error == -1)
504 			printf("Invalid file system specification.\n");
505 	} while (error != 0);
506 
507 	return (error);
508 }
509 
510 static int
511 parse_dir_md(char **conf)
512 {
513 	struct stat sb;
514 	struct thread *td;
515 	struct md_ioctl *mdio;
516 	char *path, *tok;
517 	int error, fd, len;
518 
519 	td = curthread;
520 
521 	error = parse_token(conf, &tok);
522 	if (error)
523 		return (error);
524 
525 	len = strlen(tok);
526 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
527 	path = (void *)(mdio + 1);
528 	bcopy(tok, path, len);
529 	free(tok, M_TEMP);
530 
531 	/* Get file status. */
532 	error = kern_stat(td, path, UIO_SYSSPACE, &sb);
533 	if (error)
534 		goto out;
535 
536 	/* Open /dev/mdctl so that we can attach/detach. */
537 	error = kern_open(td, "/dev/" MDCTL_NAME, UIO_SYSSPACE, O_RDWR, 0);
538 	if (error)
539 		goto out;
540 
541 	fd = td->td_retval[0];
542 	mdio->md_version = MDIOVERSION;
543 	mdio->md_type = MD_VNODE;
544 
545 	if (root_mount_mddev != -1) {
546 		mdio->md_unit = root_mount_mddev;
547 		DROP_GIANT();
548 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
549 		PICKUP_GIANT();
550 		/* Ignore errors. We don't care. */
551 		root_mount_mddev = -1;
552 	}
553 
554 	mdio->md_file = (void *)(mdio + 1);
555 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
556 	mdio->md_mediasize = sb.st_size;
557 	mdio->md_unit = 0;
558 	DROP_GIANT();
559 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
560 	PICKUP_GIANT();
561 	if (error)
562 		goto out;
563 
564 	if (mdio->md_unit > 9) {
565 		printf("rootmount: too many md units\n");
566 		mdio->md_file = NULL;
567 		mdio->md_options = 0;
568 		mdio->md_mediasize = 0;
569 		DROP_GIANT();
570 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
571 		PICKUP_GIANT();
572 		/* Ignore errors. We don't care. */
573 		error = ERANGE;
574 		goto out;
575 	}
576 
577 	root_mount_mddev = mdio->md_unit;
578 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
579 
580 	error = kern_close(td, fd);
581 
582  out:
583 	free(mdio, M_TEMP);
584 	return (error);
585 }
586 
587 static int
588 parse_dir_onfail(char **conf)
589 {
590 	char *action;
591 	int error;
592 
593 	error = parse_token(conf, &action);
594 	if (error)
595 		return (error);
596 
597 	if (!strcmp(action, "continue"))
598 		root_mount_onfail = A_CONTINUE;
599 	else if (!strcmp(action, "panic"))
600 		root_mount_onfail = A_PANIC;
601 	else if (!strcmp(action, "reboot"))
602 		root_mount_onfail = A_REBOOT;
603 	else if (!strcmp(action, "retry"))
604 		root_mount_onfail = A_RETRY;
605 	else {
606 		printf("rootmount: %s: unknown action\n", action);
607 		error = EINVAL;
608 	}
609 
610 	free(action, M_TEMP);
611 	return (0);
612 }
613 
614 static int
615 parse_dir_timeout(char **conf)
616 {
617 	char *tok, *endtok;
618 	long secs;
619 	int error;
620 
621 	error = parse_token(conf, &tok);
622 	if (error)
623 		return (error);
624 
625 	secs = strtol(tok, &endtok, 0);
626 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
627 	if (!error)
628 		root_mount_timeout = secs;
629 	free(tok, M_TEMP);
630 	return (error);
631 }
632 
633 static int
634 parse_directive(char **conf)
635 {
636 	char *dir;
637 	int error;
638 
639 	error = parse_token(conf, &dir);
640 	if (error)
641 		return (error);
642 
643 	if (strcmp(dir, ".ask") == 0)
644 		error = parse_dir_ask(conf);
645 	else if (strcmp(dir, ".md") == 0)
646 		error = parse_dir_md(conf);
647 	else if (strcmp(dir, ".onfail") == 0)
648 		error = parse_dir_onfail(conf);
649 	else if (strcmp(dir, ".timeout") == 0)
650 		error = parse_dir_timeout(conf);
651 	else {
652 		printf("mountroot: invalid directive `%s'\n", dir);
653 		/* Ignore the rest of the line. */
654 		(void)parse_skipto(conf, '\n');
655 		error = EINVAL;
656 	}
657 	free(dir, M_TEMP);
658 	return (error);
659 }
660 
661 static int
662 parse_mount_dev_present(const char *dev)
663 {
664 	struct nameidata nd;
665 	int error;
666 
667 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
668 	error = namei(&nd);
669 	if (!error)
670 		vput(nd.ni_vp);
671 	NDFREE(&nd, NDF_ONLY_PNBUF);
672 	return (error != 0) ? 0 : 1;
673 }
674 
675 static int
676 parse_mount(char **conf)
677 {
678 	char errmsg[255];
679 	struct mntarg *ma;
680 	char *dev, *fs, *opts, *tok;
681 	int delay, error, timeout;
682 
683 	error = parse_token(conf, &tok);
684 	if (error)
685 		return (error);
686 	fs = tok;
687 	error = parse_skipto(&tok, ':');
688 	if (error) {
689 		free(fs, M_TEMP);
690 		return (error);
691 	}
692 	parse_poke(&tok, '\0');
693 	parse_advance(&tok);
694 	dev = tok;
695 
696 	if (root_mount_mddev != -1) {
697 		/* Handle substitution for the md unit number. */
698 		tok = strstr(dev, "md#");
699 		if (tok != NULL)
700 			tok[2] = '0' + root_mount_mddev;
701 	}
702 
703 	/* Parse options. */
704 	error = parse_token(conf, &tok);
705 	opts = (error == 0) ? tok : NULL;
706 
707 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
708 	    (opts != NULL) ? opts : "");
709 
710 	bzero(errmsg, sizeof(errmsg));
711 
712 	if (vfs_byname(fs) == NULL) {
713 		strlcpy(errmsg, "unknown file system", sizeof(errmsg));
714 		error = ENOENT;
715 		goto out;
716 	}
717 
718 	if (strcmp(fs, "zfs") != 0 && dev[0] != '\0' &&
719 	    !parse_mount_dev_present(dev)) {
720 		printf("mountroot: waiting for device %s ...\n", dev);
721 		delay = hz / 10;
722 		timeout = root_mount_timeout * hz;
723 		do {
724 			pause("rmdev", delay);
725 			timeout -= delay;
726 		} while (timeout > 0 && !parse_mount_dev_present(dev));
727 		if (timeout <= 0) {
728 			error = ENODEV;
729 			goto out;
730 		}
731 	}
732 
733 	ma = NULL;
734 	ma = mount_arg(ma, "fstype", fs, -1);
735 	ma = mount_arg(ma, "fspath", "/", -1);
736 	ma = mount_arg(ma, "from", dev, -1);
737 	ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg));
738 	ma = mount_arg(ma, "ro", NULL, 0);
739 	ma = parse_mountroot_options(ma, opts);
740 	error = kernel_mount(ma, MNT_ROOTFS);
741 
742  out:
743 	if (error) {
744 		printf("Mounting from %s:%s failed with error %d",
745 		    fs, dev, error);
746 		if (errmsg[0] != '\0')
747 			printf(": %s", errmsg);
748 		printf(".\n");
749 	}
750 	free(fs, M_TEMP);
751 	if (opts != NULL)
752 		free(opts, M_TEMP);
753 	/* kernel_mount can return -1 on error. */
754 	return ((error < 0) ? EDOOFUS : error);
755 }
756 
757 static int
758 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
759 {
760 	struct mount *mp;
761 	char *conf;
762 	int error;
763 
764 	root_mount_mddev = -1;
765 
766 retry:
767 	conf = sbuf_data(sb);
768 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
769 	error = (mp == NULL) ? 0 : EDOOFUS;
770 	root_mount_onfail = A_CONTINUE;
771 	while (mp == NULL) {
772 		error = parse_skipto(&conf, CC_NONWHITESPACE);
773 		if (error == PE_EOL) {
774 			parse_advance(&conf);
775 			continue;
776 		}
777 		if (error < 0)
778 			break;
779 		switch (parse_peek(&conf)) {
780 		case '#':
781 			error = parse_skipto(&conf, '\n');
782 			break;
783 		case '.':
784 			error = parse_directive(&conf);
785 			break;
786 		default:
787 			error = parse_mount(&conf);
788 			break;
789 		}
790 		if (error < 0)
791 			break;
792 		/* Ignore any trailing garbage on the line. */
793 		if (parse_peek(&conf) != '\n') {
794 			printf("mountroot: advancing to next directive...\n");
795 			(void)parse_skipto(&conf, '\n');
796 		}
797 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
798 	}
799 	if (mp != NULL)
800 		return (0);
801 
802 	/*
803 	 * We failed to mount (a new) root.
804 	 */
805 	switch (root_mount_onfail) {
806 	case A_CONTINUE:
807 		break;
808 	case A_PANIC:
809 		panic("mountroot: unable to (re-)mount root.");
810 		/* NOTREACHED */
811 	case A_RETRY:
812 		goto retry;
813 	case A_REBOOT:
814 		kern_reboot(RB_NOSYNC);
815 		/* NOTREACHED */
816 	}
817 
818 	return (error);
819 }
820 
821 static void
822 vfs_mountroot_conf0(struct sbuf *sb)
823 {
824 	char *s, *tok, *mnt, *opt;
825 	int error;
826 
827 	sbuf_printf(sb, ".onfail panic\n");
828 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
829 	if (boothowto & RB_ASKNAME)
830 		sbuf_printf(sb, ".ask\n");
831 #ifdef ROOTDEVNAME
832 	if (boothowto & RB_DFLTROOT)
833 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
834 #endif
835 	if (boothowto & RB_CDROM) {
836 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
837 		sbuf_printf(sb, ".timeout 0\n");
838 		sbuf_printf(sb, "cd9660:/dev/acd0 ro\n");
839 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
840 	}
841 	s = getenv("vfs.root.mountfrom");
842 	if (s != NULL) {
843 		opt = getenv("vfs.root.mountfrom.options");
844 		tok = s;
845 		error = parse_token(&tok, &mnt);
846 		while (!error) {
847 			sbuf_printf(sb, "%s %s\n", mnt,
848 			    (opt != NULL) ? opt : "");
849 			free(mnt, M_TEMP);
850 			error = parse_token(&tok, &mnt);
851 		}
852 		if (opt != NULL)
853 			freeenv(opt);
854 		freeenv(s);
855 	}
856 	if (rootdevnames[0] != NULL)
857 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
858 	if (rootdevnames[1] != NULL)
859 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
860 #ifdef ROOTDEVNAME
861 	if (!(boothowto & RB_DFLTROOT))
862 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
863 #endif
864 	if (!(boothowto & RB_ASKNAME))
865 		sbuf_printf(sb, ".ask\n");
866 }
867 
868 static int
869 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
870 {
871 	static char buf[128];
872 	struct nameidata nd;
873 	off_t ofs;
874 	ssize_t resid;
875 	int error, flags, len;
876 
877 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
878 	flags = FREAD;
879 	error = vn_open(&nd, &flags, 0, NULL);
880 	if (error)
881 		return (error);
882 
883 	NDFREE(&nd, NDF_ONLY_PNBUF);
884 	ofs = 0;
885 	len = sizeof(buf) - 1;
886 	while (1) {
887 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
888 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
889 		    NOCRED, &resid, td);
890 		if (error)
891 			break;
892 		if (resid == len)
893 			break;
894 		buf[len - resid] = 0;
895 		sbuf_printf(sb, "%s", buf);
896 		ofs += len - resid;
897 	}
898 
899 	VOP_UNLOCK(nd.ni_vp, 0);
900 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
901 	return (error);
902 }
903 
904 static void
905 vfs_mountroot_wait(void)
906 {
907 	struct root_hold_token *h;
908 	struct timeval lastfail;
909 	int curfail;
910 
911 	curfail = 0;
912 	while (1) {
913 		DROP_GIANT();
914 		g_waitidle();
915 		PICKUP_GIANT();
916 		mtx_lock(&mountlist_mtx);
917 		if (LIST_EMPTY(&root_holds)) {
918 			mtx_unlock(&mountlist_mtx);
919 			break;
920 		}
921 		if (ppsratecheck(&lastfail, &curfail, 1)) {
922 			printf("Root mount waiting for:");
923 			LIST_FOREACH(h, &root_holds, list)
924 				printf(" %s", h->who);
925 			printf("\n");
926 		}
927 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
928 		    hz);
929 	}
930 }
931 
932 void
933 vfs_mountroot(void)
934 {
935 	struct mount *mp;
936 	struct sbuf *sb;
937 	struct thread *td;
938 	time_t timebase;
939 	int error;
940 
941 	td = curthread;
942 
943 	vfs_mountroot_wait();
944 
945 	sb = sbuf_new_auto();
946 	vfs_mountroot_conf0(sb);
947 	sbuf_finish(sb);
948 
949 	error = vfs_mountroot_devfs(td, &mp);
950 	while (!error) {
951 		error = vfs_mountroot_parse(sb, mp);
952 		if (!error) {
953 			error = vfs_mountroot_shuffle(td, mp);
954 			if (!error) {
955 				sbuf_clear(sb);
956 				error = vfs_mountroot_readconf(td, sb);
957 				sbuf_finish(sb);
958 			}
959 		}
960 	}
961 
962 	sbuf_delete(sb);
963 
964 	/*
965 	 * Iterate over all currently mounted file systems and use
966 	 * the time stamp found to check and/or initialize the RTC.
967 	 * Call inittodr() only once and pass it the largest of the
968 	 * timestamps we encounter.
969 	 */
970 	timebase = 0;
971 	mtx_lock(&mountlist_mtx);
972 	mp = TAILQ_FIRST(&mountlist);
973 	while (mp != NULL) {
974 		if (mp->mnt_time > timebase)
975 			timebase = mp->mnt_time;
976 		mp = TAILQ_NEXT(mp, mnt_list);
977 	}
978 	mtx_unlock(&mountlist_mtx);
979 	inittodr(timebase);
980 
981 	/* Keep prison0's root in sync with the global rootvnode. */
982 	mtx_lock(&prison0.pr_mtx);
983 	prison0.pr_root = rootvnode;
984 	vref(prison0.pr_root);
985 	mtx_unlock(&prison0.pr_mtx);
986 
987 	mtx_lock(&mountlist_mtx);
988 	atomic_store_rel_int(&root_mount_complete, 1);
989 	wakeup(&root_mount_complete);
990 	mtx_unlock(&mountlist_mtx);
991 }
992 
993 static struct mntarg *
994 parse_mountroot_options(struct mntarg *ma, const char *options)
995 {
996 	char *p;
997 	char *name, *name_arg;
998 	char *val, *val_arg;
999 	char *opts;
1000 
1001 	if (options == NULL || options[0] == '\0')
1002 		return (ma);
1003 
1004 	p = opts = strdup(options, M_MOUNT);
1005 	if (opts == NULL) {
1006 		return (ma);
1007 	}
1008 
1009 	while((name = strsep(&p, ",")) != NULL) {
1010 		if (name[0] == '\0')
1011 			break;
1012 
1013 		val = strchr(name, '=');
1014 		if (val != NULL) {
1015 			*val = '\0';
1016 			++val;
1017 		}
1018 		if( strcmp(name, "rw") == 0 ||
1019 		    strcmp(name, "noro") == 0) {
1020 			/*
1021 			 * The first time we mount the root file system,
1022 			 * we need to mount 'ro', so We need to ignore
1023 			 * 'rw' and 'noro' mount options.
1024 			 */
1025 			continue;
1026 		}
1027 		name_arg = strdup(name, M_MOUNT);
1028 		val_arg = NULL;
1029 		if (val != NULL)
1030 			val_arg = strdup(val, M_MOUNT);
1031 
1032 		ma = mount_arg(ma, name_arg, val_arg,
1033 		    (val_arg != NULL ? -1 : 0));
1034 	}
1035 	free(opts, M_MOUNT);
1036 	return (ma);
1037 }
1038