xref: /freebsd/sys/kern/vfs_mountroot.c (revision 9fc5c47fa5c7fa58d61245f0408611943e613164)
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 	vfsp = vfs_byname("devfs");
224 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
225 	if (vfsp == NULL)
226 		return (ENOENT);
227 
228 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
229 
230 	error = VFS_MOUNT(mp);
231 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
232 	if (error)
233 		return (error);
234 
235 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
236 	TAILQ_INIT(opts);
237 	mp->mnt_opt = opts;
238 
239 	mtx_lock(&mountlist_mtx);
240 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
241 	mtx_unlock(&mountlist_mtx);
242 
243 	*mpp = mp;
244 	rootdevmp = mp;
245 	set_rootvnode();
246 
247 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
248 	if (error)
249 		printf("kern_symlink /dev -> / returns %d\n", error);
250 
251 	return (error);
252 }
253 
254 static int
255 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
256 {
257 	struct nameidata nd;
258 	struct mount *mporoot, *mpnroot;
259 	struct vnode *vp, *vporoot, *vpdevfs;
260 	char *fspath;
261 	int error;
262 
263 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
264 
265 	/* Shuffle the mountlist. */
266 	mtx_lock(&mountlist_mtx);
267 	mporoot = TAILQ_FIRST(&mountlist);
268 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
269 	if (mporoot != mpdevfs) {
270 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
271 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
272 	}
273 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
274 	mtx_unlock(&mountlist_mtx);
275 
276 	cache_purgevfs(mporoot);
277 	if (mporoot != mpdevfs)
278 		cache_purgevfs(mpdevfs);
279 
280 	VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot);
281 
282 	VI_LOCK(vporoot);
283 	vporoot->v_iflag &= ~VI_MOUNT;
284 	VI_UNLOCK(vporoot);
285 	vporoot->v_mountedhere = NULL;
286 	mporoot->mnt_flag &= ~MNT_ROOTFS;
287 	mporoot->mnt_vnodecovered = NULL;
288 	vput(vporoot);
289 
290 	/* Set up the new rootvnode, and purge the cache */
291 	mpnroot->mnt_vnodecovered = NULL;
292 	set_rootvnode();
293 	cache_purgevfs(rootvnode->v_mount);
294 
295 	if (mporoot != mpdevfs) {
296 		/* Remount old root under /.mount or /mnt */
297 		fspath = "/.mount";
298 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
299 		    fspath, td);
300 		error = namei(&nd);
301 		if (error) {
302 			NDFREE(&nd, NDF_ONLY_PNBUF);
303 			fspath = "/mnt";
304 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
305 			    fspath, td);
306 			error = namei(&nd);
307 		}
308 		if (!error) {
309 			vp = nd.ni_vp;
310 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
311 			if (!error)
312 				error = vinvalbuf(vp, V_SAVE, 0, 0);
313 			if (!error) {
314 				cache_purge(vp);
315 				mporoot->mnt_vnodecovered = vp;
316 				vp->v_mountedhere = mporoot;
317 				strlcpy(mporoot->mnt_stat.f_mntonname,
318 				    fspath, MNAMELEN);
319 				VOP_UNLOCK(vp, 0);
320 			} else
321 				vput(vp);
322 		}
323 		NDFREE(&nd, NDF_ONLY_PNBUF);
324 
325 		if (error && bootverbose)
326 			printf("mountroot: unable to remount previous root "
327 			    "under /.mount or /mnt (error %d).\n", error);
328 	}
329 
330 	/* Remount devfs under /dev */
331 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
332 	error = namei(&nd);
333 	if (!error) {
334 		vp = nd.ni_vp;
335 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
336 		if (!error)
337 			error = vinvalbuf(vp, V_SAVE, 0, 0);
338 		if (!error) {
339 			vpdevfs = mpdevfs->mnt_vnodecovered;
340 			if (vpdevfs != NULL) {
341 				cache_purge(vpdevfs);
342 				vpdevfs->v_mountedhere = NULL;
343 				vrele(vpdevfs);
344 			}
345 			mpdevfs->mnt_vnodecovered = vp;
346 			vp->v_mountedhere = mpdevfs;
347 			VOP_UNLOCK(vp, 0);
348 		} else
349 			vput(vp);
350 	}
351 	if (error && bootverbose)
352 		printf("mountroot: unable to remount devfs under /dev "
353 		    "(error %d).\n", error);
354 	NDFREE(&nd, NDF_ONLY_PNBUF);
355 
356 	if (mporoot == mpdevfs) {
357 		vfs_unbusy(mpdevfs);
358 		/* Unlink the no longer needed /dev/dev -> / symlink */
359 		error = kern_unlinkat(td, AT_FDCWD, "/dev/dev",
360 		    UIO_SYSSPACE, 0);
361 		if (error && bootverbose)
362 			printf("mountroot: unable to unlink /dev/dev "
363 			    "(error %d)\n", error);
364 	}
365 
366 	return (0);
367 }
368 
369 /*
370  * Configuration parser.
371  */
372 
373 /* Parser character classes. */
374 #define	CC_WHITESPACE		-1
375 #define	CC_NONWHITESPACE	-2
376 
377 /* Parse errors. */
378 #define	PE_EOF			-1
379 #define	PE_EOL			-2
380 
381 static __inline int
382 parse_peek(char **conf)
383 {
384 
385 	return (**conf);
386 }
387 
388 static __inline void
389 parse_poke(char **conf, int c)
390 {
391 
392 	**conf = c;
393 }
394 
395 static __inline void
396 parse_advance(char **conf)
397 {
398 
399 	(*conf)++;
400 }
401 
402 static int
403 parse_skipto(char **conf, int mc)
404 {
405 	int c, match;
406 
407 	while (1) {
408 		c = parse_peek(conf);
409 		if (c == 0)
410 			return (PE_EOF);
411 		switch (mc) {
412 		case CC_WHITESPACE:
413 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
414 			break;
415 		case CC_NONWHITESPACE:
416 			if (c == '\n')
417 				return (PE_EOL);
418 			match = (c != ' ' && c != '\t') ? 1 : 0;
419 			break;
420 		default:
421 			match = (c == mc) ? 1 : 0;
422 			break;
423 		}
424 		if (match)
425 			break;
426 		parse_advance(conf);
427 	}
428 	return (0);
429 }
430 
431 static int
432 parse_token(char **conf, char **tok)
433 {
434 	char *p;
435 	size_t len;
436 	int error;
437 
438 	*tok = NULL;
439 	error = parse_skipto(conf, CC_NONWHITESPACE);
440 	if (error)
441 		return (error);
442 	p = *conf;
443 	error = parse_skipto(conf, CC_WHITESPACE);
444 	len = *conf - p;
445 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
446 	bcopy(p, *tok, len);
447 	return (0);
448 }
449 
450 static void
451 parse_dir_ask_printenv(const char *var)
452 {
453 	char *val;
454 
455 	val = kern_getenv(var);
456 	if (val != NULL) {
457 		printf("  %s=%s\n", var, val);
458 		freeenv(val);
459 	}
460 }
461 
462 static int
463 parse_dir_ask(char **conf)
464 {
465 	char name[80];
466 	char *mnt;
467 	int error;
468 
469 	printf("\nLoader variables:\n");
470 	parse_dir_ask_printenv("vfs.root.mountfrom");
471 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
472 
473 	printf("\nManual root filesystem specification:\n");
474 	printf("  <fstype>:<device> [options]\n");
475 	printf("      Mount <device> using filesystem <fstype>\n");
476 	printf("      and with the specified (optional) option list.\n");
477 	printf("\n");
478 	printf("    eg. ufs:/dev/da0s1a\n");
479 	printf("        zfs:tank\n");
480 	printf("        cd9660:/dev/acd0 ro\n");
481 	printf("          (which is equivalent to: ");
482 	printf("mount -t cd9660 -o ro /dev/acd0 /)\n");
483 	printf("\n");
484 	printf("  ?               List valid disk boot devices\n");
485 	printf("  .               Yield 1 second (for background tasks)\n");
486 	printf("  <empty line>    Abort manual input\n");
487 
488 	do {
489 		error = EINVAL;
490 		printf("\nmountroot> ");
491 		cngets(name, sizeof(name), GETS_ECHO);
492 		if (name[0] == '\0')
493 			break;
494 		if (name[0] == '?' && name[1] == '\0') {
495 			printf("\nList of GEOM managed disk devices:\n  ");
496 			g_dev_print();
497 			continue;
498 		}
499 		if (name[0] == '.' && name[1] == '\0') {
500 			pause("rmask", hz);
501 			continue;
502 		}
503 		mnt = name;
504 		error = parse_mount(&mnt);
505 		if (error == -1)
506 			printf("Invalid file system specification.\n");
507 	} while (error != 0);
508 
509 	return (error);
510 }
511 
512 static int
513 parse_dir_md(char **conf)
514 {
515 	struct stat sb;
516 	struct thread *td;
517 	struct md_ioctl *mdio;
518 	char *path, *tok;
519 	int error, fd, len;
520 
521 	td = curthread;
522 
523 	error = parse_token(conf, &tok);
524 	if (error)
525 		return (error);
526 
527 	len = strlen(tok);
528 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
529 	path = (void *)(mdio + 1);
530 	bcopy(tok, path, len);
531 	free(tok, M_TEMP);
532 
533 	/* Get file status. */
534 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
535 	if (error)
536 		goto out;
537 
538 	/* Open /dev/mdctl so that we can attach/detach. */
539 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
540 	    O_RDWR, 0);
541 	if (error)
542 		goto out;
543 
544 	fd = td->td_retval[0];
545 	mdio->md_version = MDIOVERSION;
546 	mdio->md_type = MD_VNODE;
547 
548 	if (root_mount_mddev != -1) {
549 		mdio->md_unit = root_mount_mddev;
550 		DROP_GIANT();
551 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
552 		PICKUP_GIANT();
553 		/* Ignore errors. We don't care. */
554 		root_mount_mddev = -1;
555 	}
556 
557 	mdio->md_file = (void *)(mdio + 1);
558 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
559 	mdio->md_mediasize = sb.st_size;
560 	mdio->md_unit = 0;
561 	DROP_GIANT();
562 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
563 	PICKUP_GIANT();
564 	if (error)
565 		goto out;
566 
567 	if (mdio->md_unit > 9) {
568 		printf("rootmount: too many md units\n");
569 		mdio->md_file = NULL;
570 		mdio->md_options = 0;
571 		mdio->md_mediasize = 0;
572 		DROP_GIANT();
573 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
574 		PICKUP_GIANT();
575 		/* Ignore errors. We don't care. */
576 		error = ERANGE;
577 		goto out;
578 	}
579 
580 	root_mount_mddev = mdio->md_unit;
581 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
582 
583 	error = kern_close(td, fd);
584 
585  out:
586 	free(mdio, M_TEMP);
587 	return (error);
588 }
589 
590 static int
591 parse_dir_onfail(char **conf)
592 {
593 	char *action;
594 	int error;
595 
596 	error = parse_token(conf, &action);
597 	if (error)
598 		return (error);
599 
600 	if (!strcmp(action, "continue"))
601 		root_mount_onfail = A_CONTINUE;
602 	else if (!strcmp(action, "panic"))
603 		root_mount_onfail = A_PANIC;
604 	else if (!strcmp(action, "reboot"))
605 		root_mount_onfail = A_REBOOT;
606 	else if (!strcmp(action, "retry"))
607 		root_mount_onfail = A_RETRY;
608 	else {
609 		printf("rootmount: %s: unknown action\n", action);
610 		error = EINVAL;
611 	}
612 
613 	free(action, M_TEMP);
614 	return (0);
615 }
616 
617 static int
618 parse_dir_timeout(char **conf)
619 {
620 	char *tok, *endtok;
621 	long secs;
622 	int error;
623 
624 	error = parse_token(conf, &tok);
625 	if (error)
626 		return (error);
627 
628 	secs = strtol(tok, &endtok, 0);
629 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
630 	if (!error)
631 		root_mount_timeout = secs;
632 	free(tok, M_TEMP);
633 	return (error);
634 }
635 
636 static int
637 parse_directive(char **conf)
638 {
639 	char *dir;
640 	int error;
641 
642 	error = parse_token(conf, &dir);
643 	if (error)
644 		return (error);
645 
646 	if (strcmp(dir, ".ask") == 0)
647 		error = parse_dir_ask(conf);
648 	else if (strcmp(dir, ".md") == 0)
649 		error = parse_dir_md(conf);
650 	else if (strcmp(dir, ".onfail") == 0)
651 		error = parse_dir_onfail(conf);
652 	else if (strcmp(dir, ".timeout") == 0)
653 		error = parse_dir_timeout(conf);
654 	else {
655 		printf("mountroot: invalid directive `%s'\n", dir);
656 		/* Ignore the rest of the line. */
657 		(void)parse_skipto(conf, '\n');
658 		error = EINVAL;
659 	}
660 	free(dir, M_TEMP);
661 	return (error);
662 }
663 
664 static int
665 parse_mount_dev_present(const char *dev)
666 {
667 	struct nameidata nd;
668 	int error;
669 
670 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
671 	error = namei(&nd);
672 	if (!error)
673 		vput(nd.ni_vp);
674 	NDFREE(&nd, NDF_ONLY_PNBUF);
675 	return (error != 0) ? 0 : 1;
676 }
677 
678 #define	ERRMSGL	255
679 static int
680 parse_mount(char **conf)
681 {
682 	char *errmsg;
683 	struct mntarg *ma;
684 	char *dev, *fs, *opts, *tok;
685 	int delay, error, timeout;
686 
687 	error = parse_token(conf, &tok);
688 	if (error)
689 		return (error);
690 	fs = tok;
691 	error = parse_skipto(&tok, ':');
692 	if (error) {
693 		free(fs, M_TEMP);
694 		return (error);
695 	}
696 	parse_poke(&tok, '\0');
697 	parse_advance(&tok);
698 	dev = tok;
699 
700 	if (root_mount_mddev != -1) {
701 		/* Handle substitution for the md unit number. */
702 		tok = strstr(dev, "md#");
703 		if (tok != NULL)
704 			tok[2] = '0' + root_mount_mddev;
705 	}
706 
707 	/* Parse options. */
708 	error = parse_token(conf, &tok);
709 	opts = (error == 0) ? tok : NULL;
710 
711 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
712 	    (opts != NULL) ? opts : "");
713 
714 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
715 
716 	if (vfs_byname(fs) == NULL) {
717 		strlcpy(errmsg, "unknown file system", ERRMSGL);
718 		error = ENOENT;
719 		goto out;
720 	}
721 
722 	if (strcmp(fs, "zfs") != 0 && strstr(fs, "nfs") == NULL &&
723 	    dev[0] != '\0' && !parse_mount_dev_present(dev)) {
724 		printf("mountroot: waiting for device %s ...\n", dev);
725 		delay = hz / 10;
726 		timeout = root_mount_timeout * hz;
727 		do {
728 			pause("rmdev", delay);
729 			timeout -= delay;
730 		} while (timeout > 0 && !parse_mount_dev_present(dev));
731 		if (timeout <= 0) {
732 			error = ENODEV;
733 			goto out;
734 		}
735 	}
736 
737 	ma = NULL;
738 	ma = mount_arg(ma, "fstype", fs, -1);
739 	ma = mount_arg(ma, "fspath", "/", -1);
740 	ma = mount_arg(ma, "from", dev, -1);
741 	ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
742 	ma = mount_arg(ma, "ro", NULL, 0);
743 	ma = parse_mountroot_options(ma, opts);
744 	error = kernel_mount(ma, MNT_ROOTFS);
745 
746  out:
747 	if (error) {
748 		printf("Mounting from %s:%s failed with error %d",
749 		    fs, dev, error);
750 		if (errmsg[0] != '\0')
751 			printf(": %s", errmsg);
752 		printf(".\n");
753 	}
754 	free(fs, M_TEMP);
755 	free(errmsg, M_TEMP);
756 	if (opts != NULL)
757 		free(opts, M_TEMP);
758 	/* kernel_mount can return -1 on error. */
759 	return ((error < 0) ? EDOOFUS : error);
760 }
761 #undef ERRMSGL
762 
763 static int
764 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
765 {
766 	struct mount *mp;
767 	char *conf;
768 	int error;
769 
770 	root_mount_mddev = -1;
771 
772 retry:
773 	conf = sbuf_data(sb);
774 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
775 	error = (mp == NULL) ? 0 : EDOOFUS;
776 	root_mount_onfail = A_CONTINUE;
777 	while (mp == NULL) {
778 		error = parse_skipto(&conf, CC_NONWHITESPACE);
779 		if (error == PE_EOL) {
780 			parse_advance(&conf);
781 			continue;
782 		}
783 		if (error < 0)
784 			break;
785 		switch (parse_peek(&conf)) {
786 		case '#':
787 			error = parse_skipto(&conf, '\n');
788 			break;
789 		case '.':
790 			error = parse_directive(&conf);
791 			break;
792 		default:
793 			error = parse_mount(&conf);
794 			if (error == -1) {
795 				printf("mountroot: invalid file system "
796 				    "specification.\n");
797 				error = 0;
798 			}
799 			break;
800 		}
801 		if (error < 0)
802 			break;
803 		/* Ignore any trailing garbage on the line. */
804 		if (parse_peek(&conf) != '\n') {
805 			printf("mountroot: advancing to next directive...\n");
806 			(void)parse_skipto(&conf, '\n');
807 		}
808 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
809 	}
810 	if (mp != NULL)
811 		return (0);
812 
813 	/*
814 	 * We failed to mount (a new) root.
815 	 */
816 	switch (root_mount_onfail) {
817 	case A_CONTINUE:
818 		break;
819 	case A_PANIC:
820 		panic("mountroot: unable to (re-)mount root.");
821 		/* NOTREACHED */
822 	case A_RETRY:
823 		goto retry;
824 	case A_REBOOT:
825 		kern_reboot(RB_NOSYNC);
826 		/* NOTREACHED */
827 	}
828 
829 	return (error);
830 }
831 
832 static void
833 vfs_mountroot_conf0(struct sbuf *sb)
834 {
835 	char *s, *tok, *mnt, *opt;
836 	int error;
837 
838 	sbuf_printf(sb, ".onfail panic\n");
839 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
840 	if (boothowto & RB_ASKNAME)
841 		sbuf_printf(sb, ".ask\n");
842 #ifdef ROOTDEVNAME
843 	if (boothowto & RB_DFLTROOT)
844 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
845 #endif
846 	if (boothowto & RB_CDROM) {
847 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
848 		sbuf_printf(sb, ".timeout 0\n");
849 		sbuf_printf(sb, "cd9660:/dev/acd0 ro\n");
850 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
851 	}
852 	s = kern_getenv("vfs.root.mountfrom");
853 	if (s != NULL) {
854 		opt = kern_getenv("vfs.root.mountfrom.options");
855 		tok = s;
856 		error = parse_token(&tok, &mnt);
857 		while (!error) {
858 			sbuf_printf(sb, "%s %s\n", mnt,
859 			    (opt != NULL) ? opt : "");
860 			free(mnt, M_TEMP);
861 			error = parse_token(&tok, &mnt);
862 		}
863 		if (opt != NULL)
864 			freeenv(opt);
865 		freeenv(s);
866 	}
867 	if (rootdevnames[0] != NULL)
868 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
869 	if (rootdevnames[1] != NULL)
870 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
871 #ifdef ROOTDEVNAME
872 	if (!(boothowto & RB_DFLTROOT))
873 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
874 #endif
875 	if (!(boothowto & RB_ASKNAME))
876 		sbuf_printf(sb, ".ask\n");
877 }
878 
879 static int
880 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
881 {
882 	static char buf[128];
883 	struct nameidata nd;
884 	off_t ofs;
885 	ssize_t resid;
886 	int error, flags, len;
887 
888 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
889 	flags = FREAD;
890 	error = vn_open(&nd, &flags, 0, NULL);
891 	if (error)
892 		return (error);
893 
894 	NDFREE(&nd, NDF_ONLY_PNBUF);
895 	ofs = 0;
896 	len = sizeof(buf) - 1;
897 	while (1) {
898 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
899 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
900 		    NOCRED, &resid, td);
901 		if (error)
902 			break;
903 		if (resid == len)
904 			break;
905 		buf[len - resid] = 0;
906 		sbuf_printf(sb, "%s", buf);
907 		ofs += len - resid;
908 	}
909 
910 	VOP_UNLOCK(nd.ni_vp, 0);
911 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
912 	return (error);
913 }
914 
915 static void
916 vfs_mountroot_wait(void)
917 {
918 	struct root_hold_token *h;
919 	struct timeval lastfail;
920 	int curfail;
921 
922 	curfail = 0;
923 	while (1) {
924 		DROP_GIANT();
925 		g_waitidle();
926 		PICKUP_GIANT();
927 		mtx_lock(&root_holds_mtx);
928 		if (LIST_EMPTY(&root_holds)) {
929 			mtx_unlock(&root_holds_mtx);
930 			break;
931 		}
932 		if (ppsratecheck(&lastfail, &curfail, 1)) {
933 			printf("Root mount waiting for:");
934 			LIST_FOREACH(h, &root_holds, list)
935 				printf(" %s", h->who);
936 			printf("\n");
937 		}
938 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
939 		    hz);
940 	}
941 }
942 
943 void
944 vfs_mountroot(void)
945 {
946 	struct mount *mp;
947 	struct sbuf *sb;
948 	struct thread *td;
949 	time_t timebase;
950 	int error;
951 
952 	td = curthread;
953 
954 	vfs_mountroot_wait();
955 
956 	sb = sbuf_new_auto();
957 	vfs_mountroot_conf0(sb);
958 	sbuf_finish(sb);
959 
960 	error = vfs_mountroot_devfs(td, &mp);
961 	while (!error) {
962 		error = vfs_mountroot_parse(sb, mp);
963 		if (!error) {
964 			error = vfs_mountroot_shuffle(td, mp);
965 			if (!error) {
966 				sbuf_clear(sb);
967 				error = vfs_mountroot_readconf(td, sb);
968 				sbuf_finish(sb);
969 			}
970 		}
971 	}
972 
973 	sbuf_delete(sb);
974 
975 	/*
976 	 * Iterate over all currently mounted file systems and use
977 	 * the time stamp found to check and/or initialize the RTC.
978 	 * Call inittodr() only once and pass it the largest of the
979 	 * timestamps we encounter.
980 	 */
981 	timebase = 0;
982 	mtx_lock(&mountlist_mtx);
983 	mp = TAILQ_FIRST(&mountlist);
984 	while (mp != NULL) {
985 		if (mp->mnt_time > timebase)
986 			timebase = mp->mnt_time;
987 		mp = TAILQ_NEXT(mp, mnt_list);
988 	}
989 	mtx_unlock(&mountlist_mtx);
990 	inittodr(timebase);
991 
992 	/* Keep prison0's root in sync with the global rootvnode. */
993 	mtx_lock(&prison0.pr_mtx);
994 	prison0.pr_root = rootvnode;
995 	vref(prison0.pr_root);
996 	mtx_unlock(&prison0.pr_mtx);
997 
998 	mtx_lock(&root_holds_mtx);
999 	atomic_store_rel_int(&root_mount_complete, 1);
1000 	wakeup(&root_mount_complete);
1001 	mtx_unlock(&root_holds_mtx);
1002 
1003 	EVENTHANDLER_INVOKE(mountroot);
1004 }
1005 
1006 static struct mntarg *
1007 parse_mountroot_options(struct mntarg *ma, const char *options)
1008 {
1009 	char *p;
1010 	char *name, *name_arg;
1011 	char *val, *val_arg;
1012 	char *opts;
1013 
1014 	if (options == NULL || options[0] == '\0')
1015 		return (ma);
1016 
1017 	p = opts = strdup(options, M_MOUNT);
1018 	if (opts == NULL) {
1019 		return (ma);
1020 	}
1021 
1022 	while((name = strsep(&p, ",")) != NULL) {
1023 		if (name[0] == '\0')
1024 			break;
1025 
1026 		val = strchr(name, '=');
1027 		if (val != NULL) {
1028 			*val = '\0';
1029 			++val;
1030 		}
1031 		if( strcmp(name, "rw") == 0 ||
1032 		    strcmp(name, "noro") == 0) {
1033 			/*
1034 			 * The first time we mount the root file system,
1035 			 * we need to mount 'ro', so We need to ignore
1036 			 * 'rw' and 'noro' mount options.
1037 			 */
1038 			continue;
1039 		}
1040 		name_arg = strdup(name, M_MOUNT);
1041 		val_arg = NULL;
1042 		if (val != NULL)
1043 			val_arg = strdup(val, M_MOUNT);
1044 
1045 		ma = mount_arg(ma, name_arg, val_arg,
1046 		    (val_arg != NULL ? -1 : 0));
1047 	}
1048 	free(opts, M_MOUNT);
1049 	return (ma);
1050 }
1051