xref: /freebsd/sys/kern/vfs_mountroot.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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 
204 static int
205 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
206 {
207 	struct vfsoptlist *opts;
208 	struct vfsconf *vfsp;
209 	struct mount *mp;
210 	int error;
211 
212 	*mpp = NULL;
213 
214 	vfsp = vfs_byname("devfs");
215 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
216 	if (vfsp == NULL)
217 		return (ENOENT);
218 
219 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
220 
221 	error = VFS_MOUNT(mp);
222 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
223 	if (error)
224 		return (error);
225 
226 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
227 	TAILQ_INIT(opts);
228 	mp->mnt_opt = opts;
229 
230 	mtx_lock(&mountlist_mtx);
231 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
232 	mtx_unlock(&mountlist_mtx);
233 
234 	*mpp = mp;
235 	set_rootvnode();
236 
237 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
238 	if (error)
239 		printf("kern_symlink /dev -> / returns %d\n", error);
240 
241 	return (error);
242 }
243 
244 static int
245 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
246 {
247 	struct nameidata nd;
248 	struct mount *mporoot, *mpnroot;
249 	struct vnode *vp, *vporoot, *vpdevfs;
250 	char *fspath;
251 	int error;
252 
253 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
254 
255 	/* Shuffle the mountlist. */
256 	mtx_lock(&mountlist_mtx);
257 	mporoot = TAILQ_FIRST(&mountlist);
258 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
259 	if (mporoot != mpdevfs) {
260 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
261 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
262 	}
263 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
264 	mtx_unlock(&mountlist_mtx);
265 
266 	cache_purgevfs(mporoot);
267 	if (mporoot != mpdevfs)
268 		cache_purgevfs(mpdevfs);
269 
270 	VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot);
271 
272 	VI_LOCK(vporoot);
273 	vporoot->v_iflag &= ~VI_MOUNT;
274 	VI_UNLOCK(vporoot);
275 	vporoot->v_mountedhere = NULL;
276 	mporoot->mnt_flag &= ~MNT_ROOTFS;
277 	mporoot->mnt_vnodecovered = NULL;
278 	vput(vporoot);
279 
280 	/* Set up the new rootvnode, and purge the cache */
281 	mpnroot->mnt_vnodecovered = NULL;
282 	set_rootvnode();
283 	cache_purgevfs(rootvnode->v_mount);
284 
285 	if (mporoot != mpdevfs) {
286 		/* Remount old root under /.mount or /mnt */
287 		fspath = "/.mount";
288 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
289 		    fspath, td);
290 		error = namei(&nd);
291 		if (error) {
292 			NDFREE(&nd, NDF_ONLY_PNBUF);
293 			fspath = "/mnt";
294 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
295 			    fspath, td);
296 			error = namei(&nd);
297 		}
298 		if (!error) {
299 			vp = nd.ni_vp;
300 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
301 			if (!error)
302 				error = vinvalbuf(vp, V_SAVE, 0, 0);
303 			if (!error) {
304 				cache_purge(vp);
305 				mporoot->mnt_vnodecovered = vp;
306 				vp->v_mountedhere = mporoot;
307 				strlcpy(mporoot->mnt_stat.f_mntonname,
308 				    fspath, MNAMELEN);
309 				VOP_UNLOCK(vp, 0);
310 			} else
311 				vput(vp);
312 		}
313 		NDFREE(&nd, NDF_ONLY_PNBUF);
314 
315 		if (error && bootverbose)
316 			printf("mountroot: unable to remount previous root "
317 			    "under /.mount or /mnt (error %d).\n", error);
318 	}
319 
320 	/* Remount devfs under /dev */
321 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
322 	error = namei(&nd);
323 	if (!error) {
324 		vp = nd.ni_vp;
325 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
326 		if (!error)
327 			error = vinvalbuf(vp, V_SAVE, 0, 0);
328 		if (!error) {
329 			vpdevfs = mpdevfs->mnt_vnodecovered;
330 			if (vpdevfs != NULL) {
331 				cache_purge(vpdevfs);
332 				vpdevfs->v_mountedhere = NULL;
333 				vrele(vpdevfs);
334 			}
335 			mpdevfs->mnt_vnodecovered = vp;
336 			vp->v_mountedhere = mpdevfs;
337 			VOP_UNLOCK(vp, 0);
338 		} else
339 			vput(vp);
340 	}
341 	if (error && bootverbose)
342 		printf("mountroot: unable to remount devfs under /dev "
343 		    "(error %d).\n", error);
344 	NDFREE(&nd, NDF_ONLY_PNBUF);
345 
346 	if (mporoot == mpdevfs) {
347 		vfs_unbusy(mpdevfs);
348 		/* Unlink the no longer needed /dev/dev -> / symlink */
349 		error = kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
350 		if (error && bootverbose)
351 			printf("mountroot: unable to unlink /dev/dev "
352 			    "(error %d)\n", error);
353 	}
354 
355 	return (0);
356 }
357 
358 /*
359  * Configuration parser.
360  */
361 
362 /* Parser character classes. */
363 #define	CC_WHITESPACE		-1
364 #define	CC_NONWHITESPACE	-2
365 
366 /* Parse errors. */
367 #define	PE_EOF			-1
368 #define	PE_EOL			-2
369 
370 static __inline int
371 parse_peek(char **conf)
372 {
373 
374 	return (**conf);
375 }
376 
377 static __inline void
378 parse_poke(char **conf, int c)
379 {
380 
381 	**conf = c;
382 }
383 
384 static __inline void
385 parse_advance(char **conf)
386 {
387 
388 	(*conf)++;
389 }
390 
391 static __inline int
392 parse_isspace(int c)
393 {
394 
395 	return ((c == ' ' || c == '\t' || c == '\n') ? 1 : 0);
396 }
397 
398 static int
399 parse_skipto(char **conf, int mc)
400 {
401 	int c, match;
402 
403 	while (1) {
404 		c = parse_peek(conf);
405 		if (c == 0)
406 			return (PE_EOF);
407 		switch (mc) {
408 		case CC_WHITESPACE:
409 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
410 			break;
411 		case CC_NONWHITESPACE:
412 			if (c == '\n')
413 				return (PE_EOL);
414 			match = (c != ' ' && c != '\t') ? 1 : 0;
415 			break;
416 		default:
417 			match = (c == mc) ? 1 : 0;
418 			break;
419 		}
420 		if (match)
421 			break;
422 		parse_advance(conf);
423 	}
424 	return (0);
425 }
426 
427 static int
428 parse_token(char **conf, char **tok)
429 {
430 	char *p;
431 	size_t len;
432 	int error;
433 
434 	*tok = NULL;
435 	error = parse_skipto(conf, CC_NONWHITESPACE);
436 	if (error)
437 		return (error);
438 	p = *conf;
439 	error = parse_skipto(conf, CC_WHITESPACE);
440 	len = *conf - p;
441 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
442 	bcopy(p, *tok, len);
443 	return (0);
444 }
445 
446 static void
447 parse_dir_ask_printenv(const char *var)
448 {
449 	char *val;
450 
451 	val = getenv(var);
452 	if (val != NULL) {
453 		printf("  %s=%s\n", var, val);
454 		freeenv(val);
455 	}
456 }
457 
458 static int
459 parse_dir_ask(char **conf)
460 {
461 	char name[80];
462 	char *mnt;
463 	int error;
464 
465 	printf("\nLoader variables:\n");
466 	parse_dir_ask_printenv("vfs.root.mountfrom");
467 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
468 
469 	printf("\nManual root filesystem specification:\n");
470 	printf("  <fstype>:<device> [options]\n");
471 	printf("      Mount <device> using filesystem <fstype>\n");
472 	printf("      and with the specified (optional) option list.\n");
473 	printf("\n");
474 	printf("    eg. ufs:/dev/da0s1a\n");
475 	printf("        zfs:tank\n");
476 	printf("        cd9660:/dev/acd0 ro\n");
477 	printf("          (which is equivalent to: ");
478 	printf("mount -t cd9660 -o ro /dev/acd0 /)\n");
479 	printf("\n");
480 	printf("  ?               List valid disk boot devices\n");
481 	printf("  .               Yield 1 second (for background tasks)\n");
482 	printf("  <empty line>    Abort manual input\n");
483 
484 	do {
485 		error = EINVAL;
486 		printf("\nmountroot> ");
487 		cngets(name, sizeof(name), GETS_ECHO);
488 		if (name[0] == '\0')
489 			break;
490 		if (name[0] == '?' && name[1] == '\0') {
491 			printf("\nList of GEOM managed disk devices:\n  ");
492 			g_dev_print();
493 			continue;
494 		}
495 		if (name[0] == '.' && name[1] == '\0') {
496 			pause("rmask", hz);
497 			continue;
498 		}
499 		mnt = name;
500 		error = parse_mount(&mnt);
501 		if (error == -1)
502 			printf("Invalid file system specification.\n");
503 	} while (error != 0);
504 
505 	return (error);
506 }
507 
508 static int
509 parse_dir_md(char **conf)
510 {
511 	struct stat sb;
512 	struct thread *td;
513 	struct md_ioctl *mdio;
514 	char *path, *tok;
515 	int error, fd, len;
516 
517 	td = curthread;
518 
519 	error = parse_token(conf, &tok);
520 	if (error)
521 		return (error);
522 
523 	len = strlen(tok);
524 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
525 	path = (void *)(mdio + 1);
526 	bcopy(tok, path, len);
527 	free(tok, M_TEMP);
528 
529 	/* Get file status. */
530 	error = kern_stat(td, path, UIO_SYSSPACE, &sb);
531 	if (error)
532 		goto out;
533 
534 	/* Open /dev/mdctl so that we can attach/detach. */
535 	error = kern_open(td, "/dev/" MDCTL_NAME, UIO_SYSSPACE, O_RDWR, 0);
536 	if (error)
537 		goto out;
538 
539 	fd = td->td_retval[0];
540 	mdio->md_version = MDIOVERSION;
541 	mdio->md_type = MD_VNODE;
542 
543 	if (root_mount_mddev != -1) {
544 		mdio->md_unit = root_mount_mddev;
545 		DROP_GIANT();
546 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
547 		PICKUP_GIANT();
548 		/* Ignore errors. We don't care. */
549 		root_mount_mddev = -1;
550 	}
551 
552 	mdio->md_file = (void *)(mdio + 1);
553 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
554 	mdio->md_mediasize = sb.st_size;
555 	mdio->md_unit = 0;
556 	DROP_GIANT();
557 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
558 	PICKUP_GIANT();
559 	if (error)
560 		goto out;
561 
562 	if (mdio->md_unit > 9) {
563 		printf("rootmount: too many md units\n");
564 		mdio->md_file = NULL;
565 		mdio->md_options = 0;
566 		mdio->md_mediasize = 0;
567 		DROP_GIANT();
568 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
569 		PICKUP_GIANT();
570 		/* Ignore errors. We don't care. */
571 		error = ERANGE;
572 		goto out;
573 	}
574 
575 	root_mount_mddev = mdio->md_unit;
576 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
577 
578 	error = kern_close(td, fd);
579 
580  out:
581 	free(mdio, M_TEMP);
582 	return (error);
583 }
584 
585 static int
586 parse_dir_onfail(char **conf)
587 {
588 	char *action;
589 	int error;
590 
591 	error = parse_token(conf, &action);
592 	if (error)
593 		return (error);
594 
595 	if (!strcmp(action, "continue"))
596 		root_mount_onfail = A_CONTINUE;
597 	else if (!strcmp(action, "panic"))
598 		root_mount_onfail = A_PANIC;
599 	else if (!strcmp(action, "reboot"))
600 		root_mount_onfail = A_REBOOT;
601 	else if (!strcmp(action, "retry"))
602 		root_mount_onfail = A_RETRY;
603 	else {
604 		printf("rootmount: %s: unknown action\n", action);
605 		error = EINVAL;
606 	}
607 
608 	free(action, M_TEMP);
609 	return (0);
610 }
611 
612 static int
613 parse_dir_timeout(char **conf)
614 {
615 	char *tok, *endtok;
616 	long secs;
617 	int error;
618 
619 	error = parse_token(conf, &tok);
620 	if (error)
621 		return (error);
622 
623 	secs = strtol(tok, &endtok, 0);
624 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
625 	if (!error)
626 		root_mount_timeout = secs;
627 	free(tok, M_TEMP);
628 	return (error);
629 }
630 
631 static int
632 parse_directive(char **conf)
633 {
634 	char *dir;
635 	int error;
636 
637 	error = parse_token(conf, &dir);
638 	if (error)
639 		return (error);
640 
641 	if (strcmp(dir, ".ask") == 0)
642 		error = parse_dir_ask(conf);
643 	else if (strcmp(dir, ".md") == 0)
644 		error = parse_dir_md(conf);
645 	else if (strcmp(dir, ".onfail") == 0)
646 		error = parse_dir_onfail(conf);
647 	else if (strcmp(dir, ".timeout") == 0)
648 		error = parse_dir_timeout(conf);
649 	else {
650 		printf("mountroot: invalid directive `%s'\n", dir);
651 		/* Ignore the rest of the line. */
652 		(void)parse_skipto(conf, '\n');
653 		error = EINVAL;
654 	}
655 	free(dir, M_TEMP);
656 	return (error);
657 }
658 
659 static int
660 parse_mount_dev_present(const char *dev)
661 {
662 	struct nameidata nd;
663 	int error;
664 
665 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
666 	error = namei(&nd);
667 	if (!error)
668 		vput(nd.ni_vp);
669 	NDFREE(&nd, NDF_ONLY_PNBUF);
670 	return (error != 0) ? 0 : 1;
671 }
672 
673 #define	ERRMSGL	255
674 static int
675 parse_mount(char **conf)
676 {
677 	char *errmsg;
678 	struct mntarg *ma;
679 	char *dev, *fs, *opts, *tok;
680 	int delay, error, timeout;
681 
682 	error = parse_token(conf, &tok);
683 	if (error)
684 		return (error);
685 	fs = tok;
686 	error = parse_skipto(&tok, ':');
687 	if (error) {
688 		free(fs, M_TEMP);
689 		return (error);
690 	}
691 	parse_poke(&tok, '\0');
692 	parse_advance(&tok);
693 	dev = tok;
694 
695 	if (root_mount_mddev != -1) {
696 		/* Handle substitution for the md unit number. */
697 		tok = strstr(dev, "md#");
698 		if (tok != NULL)
699 			tok[2] = '0' + root_mount_mddev;
700 	}
701 
702 	/* Parse options. */
703 	error = parse_token(conf, &tok);
704 	opts = (error == 0) ? tok : NULL;
705 
706 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
707 	    (opts != NULL) ? opts : "");
708 
709 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
710 
711 	if (vfs_byname(fs) == NULL) {
712 		strlcpy(errmsg, "unknown file system", sizeof(errmsg));
713 		error = ENOENT;
714 		goto out;
715 	}
716 
717 	if (strcmp(fs, "zfs") != 0 && dev[0] != '\0' &&
718 	    !parse_mount_dev_present(dev)) {
719 		printf("mountroot: waiting for device %s ...\n", dev);
720 		delay = hz / 10;
721 		timeout = root_mount_timeout * hz;
722 		do {
723 			pause("rmdev", delay);
724 			timeout -= delay;
725 		} while (timeout > 0 && !parse_mount_dev_present(dev));
726 		if (timeout <= 0) {
727 			error = ENODEV;
728 			goto out;
729 		}
730 	}
731 
732 	ma = NULL;
733 	ma = mount_arg(ma, "fstype", fs, -1);
734 	ma = mount_arg(ma, "fspath", "/", -1);
735 	ma = mount_arg(ma, "from", dev, -1);
736 	ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
737 	ma = mount_arg(ma, "ro", NULL, 0);
738 	ma = parse_mountroot_options(ma, opts);
739 	error = kernel_mount(ma, MNT_ROOTFS);
740 
741  out:
742 	if (error) {
743 		printf("Mounting from %s:%s failed with error %d",
744 		    fs, dev, error);
745 		if (errmsg[0] != '\0')
746 			printf(": %s", errmsg);
747 		printf(".\n");
748 	}
749 	free(fs, M_TEMP);
750 	free(errmsg, 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 #undef ERRMSGL
757 
758 static int
759 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
760 {
761 	struct mount *mp;
762 	char *conf;
763 	int error;
764 
765 	root_mount_mddev = -1;
766 
767 retry:
768 	conf = sbuf_data(sb);
769 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
770 	error = (mp == NULL) ? 0 : EDOOFUS;
771 	root_mount_onfail = A_CONTINUE;
772 	while (mp == NULL) {
773 		error = parse_skipto(&conf, CC_NONWHITESPACE);
774 		if (error == PE_EOL) {
775 			parse_advance(&conf);
776 			continue;
777 		}
778 		if (error < 0)
779 			break;
780 		switch (parse_peek(&conf)) {
781 		case '#':
782 			error = parse_skipto(&conf, '\n');
783 			break;
784 		case '.':
785 			error = parse_directive(&conf);
786 			break;
787 		default:
788 			error = parse_mount(&conf);
789 			break;
790 		}
791 		if (error < 0)
792 			break;
793 		/* Ignore any trailing garbage on the line. */
794 		if (parse_peek(&conf) != '\n') {
795 			printf("mountroot: advancing to next directive...\n");
796 			(void)parse_skipto(&conf, '\n');
797 		}
798 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
799 	}
800 	if (mp != NULL)
801 		return (0);
802 
803 	/*
804 	 * We failed to mount (a new) root.
805 	 */
806 	switch (root_mount_onfail) {
807 	case A_CONTINUE:
808 		break;
809 	case A_PANIC:
810 		panic("mountroot: unable to (re-)mount root.");
811 		/* NOTREACHED */
812 	case A_RETRY:
813 		goto retry;
814 	case A_REBOOT:
815 		kern_reboot(RB_NOSYNC);
816 		/* NOTREACHED */
817 	}
818 
819 	return (error);
820 }
821 
822 static void
823 vfs_mountroot_conf0(struct sbuf *sb)
824 {
825 	char *s, *tok, *mnt, *opt;
826 	int error;
827 
828 	sbuf_printf(sb, ".onfail panic\n");
829 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
830 	if (boothowto & RB_ASKNAME)
831 		sbuf_printf(sb, ".ask\n");
832 #ifdef ROOTDEVNAME
833 	if (boothowto & RB_DFLTROOT)
834 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
835 #endif
836 	if (boothowto & RB_CDROM) {
837 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
838 		sbuf_printf(sb, ".timeout 0\n");
839 		sbuf_printf(sb, "cd9660:/dev/acd0 ro\n");
840 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
841 	}
842 	s = getenv("vfs.root.mountfrom");
843 	if (s != NULL) {
844 		opt = getenv("vfs.root.mountfrom.options");
845 		tok = s;
846 		error = parse_token(&tok, &mnt);
847 		while (!error) {
848 			sbuf_printf(sb, "%s %s\n", mnt,
849 			    (opt != NULL) ? opt : "");
850 			free(mnt, M_TEMP);
851 			error = parse_token(&tok, &mnt);
852 		}
853 		if (opt != NULL)
854 			freeenv(opt);
855 		freeenv(s);
856 	}
857 	if (rootdevnames[0] != NULL)
858 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
859 	if (rootdevnames[1] != NULL)
860 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
861 #ifdef ROOTDEVNAME
862 	if (!(boothowto & RB_DFLTROOT))
863 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
864 #endif
865 	if (!(boothowto & RB_ASKNAME))
866 		sbuf_printf(sb, ".ask\n");
867 }
868 
869 static int
870 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
871 {
872 	static char buf[128];
873 	struct nameidata nd;
874 	off_t ofs;
875 	ssize_t resid;
876 	int error, flags, len;
877 
878 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
879 	flags = FREAD;
880 	error = vn_open(&nd, &flags, 0, NULL);
881 	if (error)
882 		return (error);
883 
884 	NDFREE(&nd, NDF_ONLY_PNBUF);
885 	ofs = 0;
886 	len = sizeof(buf) - 1;
887 	while (1) {
888 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
889 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
890 		    NOCRED, &resid, td);
891 		if (error)
892 			break;
893 		if (resid == len)
894 			break;
895 		buf[len - resid] = 0;
896 		sbuf_printf(sb, "%s", buf);
897 		ofs += len - resid;
898 	}
899 
900 	VOP_UNLOCK(nd.ni_vp, 0);
901 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
902 	return (error);
903 }
904 
905 static void
906 vfs_mountroot_wait(void)
907 {
908 	struct root_hold_token *h;
909 	struct timeval lastfail;
910 	int curfail;
911 
912 	curfail = 0;
913 	while (1) {
914 		DROP_GIANT();
915 		g_waitidle();
916 		PICKUP_GIANT();
917 		mtx_lock(&mountlist_mtx);
918 		if (LIST_EMPTY(&root_holds)) {
919 			mtx_unlock(&mountlist_mtx);
920 			break;
921 		}
922 		if (ppsratecheck(&lastfail, &curfail, 1)) {
923 			printf("Root mount waiting for:");
924 			LIST_FOREACH(h, &root_holds, list)
925 				printf(" %s", h->who);
926 			printf("\n");
927 		}
928 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
929 		    hz);
930 	}
931 }
932 
933 void
934 vfs_mountroot(void)
935 {
936 	struct mount *mp;
937 	struct sbuf *sb;
938 	struct thread *td;
939 	time_t timebase;
940 	int error;
941 
942 	td = curthread;
943 
944 	vfs_mountroot_wait();
945 
946 	sb = sbuf_new_auto();
947 	vfs_mountroot_conf0(sb);
948 	sbuf_finish(sb);
949 
950 	error = vfs_mountroot_devfs(td, &mp);
951 	while (!error) {
952 		error = vfs_mountroot_parse(sb, mp);
953 		if (!error) {
954 			error = vfs_mountroot_shuffle(td, mp);
955 			if (!error) {
956 				sbuf_clear(sb);
957 				error = vfs_mountroot_readconf(td, sb);
958 				sbuf_finish(sb);
959 			}
960 		}
961 	}
962 
963 	sbuf_delete(sb);
964 
965 	/*
966 	 * Iterate over all currently mounted file systems and use
967 	 * the time stamp found to check and/or initialize the RTC.
968 	 * Call inittodr() only once and pass it the largest of the
969 	 * timestamps we encounter.
970 	 */
971 	timebase = 0;
972 	mtx_lock(&mountlist_mtx);
973 	mp = TAILQ_FIRST(&mountlist);
974 	while (mp != NULL) {
975 		if (mp->mnt_time > timebase)
976 			timebase = mp->mnt_time;
977 		mp = TAILQ_NEXT(mp, mnt_list);
978 	}
979 	mtx_unlock(&mountlist_mtx);
980 	inittodr(timebase);
981 
982 	/* Keep prison0's root in sync with the global rootvnode. */
983 	mtx_lock(&prison0.pr_mtx);
984 	prison0.pr_root = rootvnode;
985 	vref(prison0.pr_root);
986 	mtx_unlock(&prison0.pr_mtx);
987 
988 	mtx_lock(&mountlist_mtx);
989 	atomic_store_rel_int(&root_mount_complete, 1);
990 	wakeup(&root_mount_complete);
991 	mtx_unlock(&mountlist_mtx);
992 
993 	EVENTHANDLER_INVOKE(mountroot);
994 }
995 
996 static struct mntarg *
997 parse_mountroot_options(struct mntarg *ma, const char *options)
998 {
999 	char *p;
1000 	char *name, *name_arg;
1001 	char *val, *val_arg;
1002 	char *opts;
1003 
1004 	if (options == NULL || options[0] == '\0')
1005 		return (ma);
1006 
1007 	p = opts = strdup(options, M_MOUNT);
1008 	if (opts == NULL) {
1009 		return (ma);
1010 	}
1011 
1012 	while((name = strsep(&p, ",")) != NULL) {
1013 		if (name[0] == '\0')
1014 			break;
1015 
1016 		val = strchr(name, '=');
1017 		if (val != NULL) {
1018 			*val = '\0';
1019 			++val;
1020 		}
1021 		if( strcmp(name, "rw") == 0 ||
1022 		    strcmp(name, "noro") == 0) {
1023 			/*
1024 			 * The first time we mount the root file system,
1025 			 * we need to mount 'ro', so We need to ignore
1026 			 * 'rw' and 'noro' mount options.
1027 			 */
1028 			continue;
1029 		}
1030 		name_arg = strdup(name, M_MOUNT);
1031 		val_arg = NULL;
1032 		if (val != NULL)
1033 			val_arg = strdup(val, M_MOUNT);
1034 
1035 		ma = mount_arg(ma, name_arg, val_arg,
1036 		    (val_arg != NULL ? -1 : 0));
1037 	}
1038 	free(opts, M_MOUNT);
1039 	return (ma);
1040 }
1041