xref: /freebsd/sys/kern/vfs_mountroot.c (revision 48c779cdecb5f803e5fe5d761987e976ca9609db)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2010 Marcel Moolenaar
5  * Copyright (c) 1999-2004 Poul-Henning Kamp
6  * Copyright (c) 1999 Michael Smith
7  * Copyright (c) 1989, 1993
8  *      The Regents of the University of California.  All rights reserved.
9  * (c) UNIX System Laboratories, Inc.
10  * All or some portions of this file are derived from material licensed
11  * to the University of California by American Telephone and Telegraph
12  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
13  * the permission of UNIX System Laboratories, Inc.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  */
39 
40 #include "opt_rootdevname.h"
41 
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
44 
45 #include <sys/param.h>
46 #include <sys/conf.h>
47 #include <sys/cons.h>
48 #include <sys/eventhandler.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/malloc.h>
53 #include <sys/mdioctl.h>
54 #include <sys/mount.h>
55 #include <sys/mutex.h>
56 #include <sys/namei.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/filedesc.h>
60 #include <sys/reboot.h>
61 #include <sys/sbuf.h>
62 #include <sys/stat.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysproto.h>
65 #include <sys/sx.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/systm.h>
69 #include <sys/vnode.h>
70 
71 #include <geom/geom.h>
72 
73 /*
74  * The root filesystem is detailed in the kernel environment variable
75  * vfs.root.mountfrom, which is expected to be in the general format
76  *
77  * <vfsname>:[<path>][	<vfsname>:[<path>] ...]
78  * vfsname   := the name of a VFS known to the kernel and capable
79  *              of being mounted as root
80  * path      := disk device name or other data used by the filesystem
81  *              to locate its physical store
82  *
83  * If the environment variable vfs.root.mountfrom is a space separated list,
84  * each list element is tried in turn and the root filesystem will be mounted
85  * from the first one that succeeds.
86  *
87  * The environment variable vfs.root.mountfrom.options is a comma delimited
88  * set of string mount options.  These mount options must be parseable
89  * by nmount() in the kernel.
90  */
91 
92 static int parse_mount(char **);
93 static struct mntarg *parse_mountroot_options(struct mntarg *, const char *);
94 static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS);
95 static void vfs_mountroot_wait(void);
96 static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev);
97 
98 /*
99  * The vnode of the system's root (/ in the filesystem, without chroot
100  * active.)
101  */
102 struct vnode *rootvnode;
103 
104 /*
105  * Mount of the system's /dev.
106  */
107 struct mount *rootdevmp;
108 
109 char *rootdevnames[2] = {NULL, NULL};
110 
111 struct mtx root_holds_mtx;
112 MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF);
113 
114 struct root_hold_token {
115 	const char			*who;
116 	LIST_ENTRY(root_hold_token)	list;
117 };
118 
119 static LIST_HEAD(, root_hold_token)	root_holds =
120     LIST_HEAD_INITIALIZER(root_holds);
121 
122 enum action {
123 	A_CONTINUE,
124 	A_PANIC,
125 	A_REBOOT,
126 	A_RETRY
127 };
128 
129 static enum action root_mount_onfail = A_CONTINUE;
130 
131 static int root_mount_mddev;
132 static int root_mount_complete;
133 
134 /* By default wait up to 3 seconds for devices to appear. */
135 static int root_mount_timeout = 3;
136 TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout);
137 
138 static int root_mount_always_wait = 0;
139 SYSCTL_INT(_vfs, OID_AUTO, root_mount_always_wait, CTLFLAG_RDTUN,
140     &root_mount_always_wait, 0,
141     "Wait for root mount holds even if the root device already exists");
142 
143 SYSCTL_PROC(_vfs, OID_AUTO, root_mount_hold,
144     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
145     NULL, 0, sysctl_vfs_root_mount_hold, "A",
146     "List of root mount hold tokens");
147 
148 static int
149 sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS)
150 {
151 	struct sbuf sb;
152 	struct root_hold_token *h;
153 	int error;
154 
155 	sbuf_new(&sb, NULL, 256, SBUF_AUTOEXTEND | SBUF_INCLUDENUL);
156 
157 	mtx_lock(&root_holds_mtx);
158 	LIST_FOREACH(h, &root_holds, list) {
159 		if (h != LIST_FIRST(&root_holds))
160 			sbuf_putc(&sb, ' ');
161 		sbuf_printf(&sb, "%s", h->who);
162 	}
163 	mtx_unlock(&root_holds_mtx);
164 
165 	error = sbuf_finish(&sb);
166 	if (error == 0)
167 		error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
168 	sbuf_delete(&sb);
169 	return (error);
170 }
171 
172 struct root_hold_token *
173 root_mount_hold(const char *identifier)
174 {
175 	struct root_hold_token *h;
176 
177 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
178 	h->who = identifier;
179 	mtx_lock(&root_holds_mtx);
180 	TSHOLD("root mount");
181 	LIST_INSERT_HEAD(&root_holds, h, list);
182 	mtx_unlock(&root_holds_mtx);
183 	return (h);
184 }
185 
186 void
187 root_mount_rel(struct root_hold_token *h)
188 {
189 
190 	if (h == NULL)
191 		return;
192 
193 	mtx_lock(&root_holds_mtx);
194 	LIST_REMOVE(h, list);
195 	TSRELEASE("root mount");
196 	wakeup(&root_holds);
197 	mtx_unlock(&root_holds_mtx);
198 	free(h, M_DEVBUF);
199 }
200 
201 int
202 root_mounted(void)
203 {
204 
205 	/* No mutex is acquired here because int stores are atomic. */
206 	return (root_mount_complete);
207 }
208 
209 static void
210 set_rootvnode(void)
211 {
212 	struct proc *p;
213 
214 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
215 		panic("set_rootvnode: Cannot find root vnode");
216 
217 	VOP_UNLOCK(rootvnode, 0);
218 
219 	p = curthread->td_proc;
220 	FILEDESC_XLOCK(p->p_fd);
221 
222 	if (p->p_fd->fd_cdir != NULL)
223 		vrele(p->p_fd->fd_cdir);
224 	p->p_fd->fd_cdir = rootvnode;
225 	VREF(rootvnode);
226 
227 	if (p->p_fd->fd_rdir != NULL)
228 		vrele(p->p_fd->fd_rdir);
229 	p->p_fd->fd_rdir = rootvnode;
230 	VREF(rootvnode);
231 
232 	FILEDESC_XUNLOCK(p->p_fd);
233 }
234 
235 static int
236 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
237 {
238 	struct vfsoptlist *opts;
239 	struct vfsconf *vfsp;
240 	struct mount *mp;
241 	int error;
242 
243 	*mpp = NULL;
244 
245 	if (rootdevmp != NULL) {
246 		/*
247 		 * Already have /dev; this happens during rerooting.
248 		 */
249 		error = vfs_busy(rootdevmp, 0);
250 		if (error != 0)
251 			return (error);
252 		*mpp = rootdevmp;
253 	} else {
254 		vfsp = vfs_byname("devfs");
255 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
256 		if (vfsp == NULL)
257 			return (ENOENT);
258 
259 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
260 
261 		error = VFS_MOUNT(mp);
262 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
263 		if (error)
264 			return (error);
265 
266 		error = VFS_STATFS(mp, &mp->mnt_stat);
267 		KASSERT(error == 0, ("VFS_STATFS(devfs) failed %d", error));
268 		if (error)
269 			return (error);
270 
271 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
272 		TAILQ_INIT(opts);
273 		mp->mnt_opt = opts;
274 
275 		mtx_lock(&mountlist_mtx);
276 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
277 		mtx_unlock(&mountlist_mtx);
278 
279 		*mpp = mp;
280 		rootdevmp = mp;
281 		vfs_op_exit(mp);
282 	}
283 
284 	set_rootvnode();
285 
286 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
287 	if (error)
288 		printf("kern_symlink /dev -> / returns %d\n", error);
289 
290 	return (error);
291 }
292 
293 static void
294 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
295 {
296 	struct nameidata nd;
297 	struct mount *mporoot, *mpnroot;
298 	struct vnode *vp, *vporoot, *vpdevfs;
299 	char *fspath;
300 	int error;
301 
302 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
303 
304 	/* Shuffle the mountlist. */
305 	mtx_lock(&mountlist_mtx);
306 	mporoot = TAILQ_FIRST(&mountlist);
307 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
308 	if (mporoot != mpdevfs) {
309 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
310 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
311 	}
312 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
313 	mtx_unlock(&mountlist_mtx);
314 
315 	cache_purgevfs(mporoot, true);
316 	if (mporoot != mpdevfs)
317 		cache_purgevfs(mpdevfs, true);
318 
319 	if (VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot))
320 		panic("vfs_mountroot_shuffle: Cannot find root vnode");
321 
322 	VI_LOCK(vporoot);
323 	vporoot->v_iflag &= ~VI_MOUNT;
324 	VI_UNLOCK(vporoot);
325 	vporoot->v_mountedhere = NULL;
326 	mporoot->mnt_flag &= ~MNT_ROOTFS;
327 	mporoot->mnt_vnodecovered = NULL;
328 	vput(vporoot);
329 
330 	/* Set up the new rootvnode, and purge the cache */
331 	mpnroot->mnt_vnodecovered = NULL;
332 	set_rootvnode();
333 	cache_purgevfs(rootvnode->v_mount, true);
334 
335 	if (mporoot != mpdevfs) {
336 		/* Remount old root under /.mount or /mnt */
337 		fspath = "/.mount";
338 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
339 		    fspath, td);
340 		error = namei(&nd);
341 		if (error) {
342 			NDFREE(&nd, NDF_ONLY_PNBUF);
343 			fspath = "/mnt";
344 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
345 			    fspath, td);
346 			error = namei(&nd);
347 		}
348 		if (!error) {
349 			vp = nd.ni_vp;
350 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
351 			if (!error)
352 				error = vinvalbuf(vp, V_SAVE, 0, 0);
353 			if (!error) {
354 				cache_purge(vp);
355 				mporoot->mnt_vnodecovered = vp;
356 				vp->v_mountedhere = mporoot;
357 				strlcpy(mporoot->mnt_stat.f_mntonname,
358 				    fspath, MNAMELEN);
359 				VOP_UNLOCK(vp, 0);
360 			} else
361 				vput(vp);
362 		}
363 		NDFREE(&nd, NDF_ONLY_PNBUF);
364 
365 		if (error)
366 			printf("mountroot: unable to remount previous root "
367 			    "under /.mount or /mnt (error %d)\n", error);
368 	}
369 
370 	/* Remount devfs under /dev */
371 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
372 	error = namei(&nd);
373 	if (!error) {
374 		vp = nd.ni_vp;
375 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
376 		if (!error)
377 			error = vinvalbuf(vp, V_SAVE, 0, 0);
378 		if (!error) {
379 			vpdevfs = mpdevfs->mnt_vnodecovered;
380 			if (vpdevfs != NULL) {
381 				cache_purge(vpdevfs);
382 				vpdevfs->v_mountedhere = NULL;
383 				vrele(vpdevfs);
384 			}
385 			mpdevfs->mnt_vnodecovered = vp;
386 			vp->v_mountedhere = mpdevfs;
387 			VOP_UNLOCK(vp, 0);
388 		} else
389 			vput(vp);
390 	}
391 	if (error)
392 		printf("mountroot: unable to remount devfs under /dev "
393 		    "(error %d)\n", error);
394 	NDFREE(&nd, NDF_ONLY_PNBUF);
395 
396 	if (mporoot == mpdevfs) {
397 		vfs_unbusy(mpdevfs);
398 		/* Unlink the no longer needed /dev/dev -> / symlink */
399 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
400 		    UIO_SYSSPACE, 0, 0);
401 		if (error)
402 			printf("mountroot: unable to unlink /dev/dev "
403 			    "(error %d)\n", error);
404 	}
405 }
406 
407 /*
408  * Configuration parser.
409  */
410 
411 /* Parser character classes. */
412 #define	CC_WHITESPACE		-1
413 #define	CC_NONWHITESPACE	-2
414 
415 /* Parse errors. */
416 #define	PE_EOF			-1
417 #define	PE_EOL			-2
418 
419 static __inline int
420 parse_peek(char **conf)
421 {
422 
423 	return (**conf);
424 }
425 
426 static __inline void
427 parse_poke(char **conf, int c)
428 {
429 
430 	**conf = c;
431 }
432 
433 static __inline void
434 parse_advance(char **conf)
435 {
436 
437 	(*conf)++;
438 }
439 
440 static int
441 parse_skipto(char **conf, int mc)
442 {
443 	int c, match;
444 
445 	while (1) {
446 		c = parse_peek(conf);
447 		if (c == 0)
448 			return (PE_EOF);
449 		switch (mc) {
450 		case CC_WHITESPACE:
451 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
452 			break;
453 		case CC_NONWHITESPACE:
454 			if (c == '\n')
455 				return (PE_EOL);
456 			match = (c != ' ' && c != '\t') ? 1 : 0;
457 			break;
458 		default:
459 			match = (c == mc) ? 1 : 0;
460 			break;
461 		}
462 		if (match)
463 			break;
464 		parse_advance(conf);
465 	}
466 	return (0);
467 }
468 
469 static int
470 parse_token(char **conf, char **tok)
471 {
472 	char *p;
473 	size_t len;
474 	int error;
475 
476 	*tok = NULL;
477 	error = parse_skipto(conf, CC_NONWHITESPACE);
478 	if (error)
479 		return (error);
480 	p = *conf;
481 	error = parse_skipto(conf, CC_WHITESPACE);
482 	len = *conf - p;
483 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
484 	bcopy(p, *tok, len);
485 	return (0);
486 }
487 
488 static void
489 parse_dir_ask_printenv(const char *var)
490 {
491 	char *val;
492 
493 	val = kern_getenv(var);
494 	if (val != NULL) {
495 		printf("  %s=%s\n", var, val);
496 		freeenv(val);
497 	}
498 }
499 
500 static int
501 parse_dir_ask(char **conf)
502 {
503 	char name[80];
504 	char *mnt;
505 	int error;
506 
507 	vfs_mountroot_wait();
508 
509 	printf("\nLoader variables:\n");
510 	parse_dir_ask_printenv("vfs.root.mountfrom");
511 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
512 
513 	printf("\nManual root filesystem specification:\n");
514 	printf("  <fstype>:<device> [options]\n");
515 	printf("      Mount <device> using filesystem <fstype>\n");
516 	printf("      and with the specified (optional) option list.\n");
517 	printf("\n");
518 	printf("    eg. ufs:/dev/da0s1a\n");
519 	printf("        zfs:zroot/ROOT/default\n");
520 	printf("        cd9660:/dev/cd0 ro\n");
521 	printf("          (which is equivalent to: ");
522 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
523 	printf("\n");
524 	printf("  ?               List valid disk boot devices\n");
525 	printf("  .               Yield 1 second (for background tasks)\n");
526 	printf("  <empty line>    Abort manual input\n");
527 
528 	do {
529 		error = EINVAL;
530 		printf("\nmountroot> ");
531 		cngets(name, sizeof(name), GETS_ECHO);
532 		if (name[0] == '\0')
533 			break;
534 		if (name[0] == '?' && name[1] == '\0') {
535 			printf("\nList of GEOM managed disk devices:\n  ");
536 			g_dev_print();
537 			continue;
538 		}
539 		if (name[0] == '.' && name[1] == '\0') {
540 			pause("rmask", hz);
541 			continue;
542 		}
543 		mnt = name;
544 		error = parse_mount(&mnt);
545 		if (error == -1)
546 			printf("Invalid file system specification.\n");
547 	} while (error != 0);
548 
549 	return (error);
550 }
551 
552 static int
553 parse_dir_md(char **conf)
554 {
555 	struct stat sb;
556 	struct thread *td;
557 	struct md_ioctl *mdio;
558 	char *path, *tok;
559 	int error, fd, len;
560 
561 	td = curthread;
562 
563 	error = parse_token(conf, &tok);
564 	if (error)
565 		return (error);
566 
567 	len = strlen(tok);
568 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
569 	path = (void *)(mdio + 1);
570 	bcopy(tok, path, len);
571 	free(tok, M_TEMP);
572 
573 	/* Get file status. */
574 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
575 	if (error)
576 		goto out;
577 
578 	/* Open /dev/mdctl so that we can attach/detach. */
579 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
580 	    O_RDWR, 0);
581 	if (error)
582 		goto out;
583 
584 	fd = td->td_retval[0];
585 	mdio->md_version = MDIOVERSION;
586 	mdio->md_type = MD_VNODE;
587 
588 	if (root_mount_mddev != -1) {
589 		mdio->md_unit = root_mount_mddev;
590 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
591 		/* Ignore errors. We don't care. */
592 		root_mount_mddev = -1;
593 	}
594 
595 	mdio->md_file = (void *)(mdio + 1);
596 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
597 	mdio->md_mediasize = sb.st_size;
598 	mdio->md_unit = 0;
599 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
600 	if (error)
601 		goto out;
602 
603 	if (mdio->md_unit > 9) {
604 		printf("rootmount: too many md units\n");
605 		mdio->md_file = NULL;
606 		mdio->md_options = 0;
607 		mdio->md_mediasize = 0;
608 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
609 		/* Ignore errors. We don't care. */
610 		error = ERANGE;
611 		goto out;
612 	}
613 
614 	root_mount_mddev = mdio->md_unit;
615 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
616 
617 	error = kern_close(td, fd);
618 
619  out:
620 	free(mdio, M_TEMP);
621 	return (error);
622 }
623 
624 static int
625 parse_dir_onfail(char **conf)
626 {
627 	char *action;
628 	int error;
629 
630 	error = parse_token(conf, &action);
631 	if (error)
632 		return (error);
633 
634 	if (!strcmp(action, "continue"))
635 		root_mount_onfail = A_CONTINUE;
636 	else if (!strcmp(action, "panic"))
637 		root_mount_onfail = A_PANIC;
638 	else if (!strcmp(action, "reboot"))
639 		root_mount_onfail = A_REBOOT;
640 	else if (!strcmp(action, "retry"))
641 		root_mount_onfail = A_RETRY;
642 	else {
643 		printf("rootmount: %s: unknown action\n", action);
644 		error = EINVAL;
645 	}
646 
647 	free(action, M_TEMP);
648 	return (0);
649 }
650 
651 static int
652 parse_dir_timeout(char **conf)
653 {
654 	char *tok, *endtok;
655 	long secs;
656 	int error;
657 
658 	error = parse_token(conf, &tok);
659 	if (error)
660 		return (error);
661 
662 	secs = strtol(tok, &endtok, 0);
663 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
664 	if (!error)
665 		root_mount_timeout = secs;
666 	free(tok, M_TEMP);
667 	return (error);
668 }
669 
670 static int
671 parse_directive(char **conf)
672 {
673 	char *dir;
674 	int error;
675 
676 	error = parse_token(conf, &dir);
677 	if (error)
678 		return (error);
679 
680 	if (strcmp(dir, ".ask") == 0)
681 		error = parse_dir_ask(conf);
682 	else if (strcmp(dir, ".md") == 0)
683 		error = parse_dir_md(conf);
684 	else if (strcmp(dir, ".onfail") == 0)
685 		error = parse_dir_onfail(conf);
686 	else if (strcmp(dir, ".timeout") == 0)
687 		error = parse_dir_timeout(conf);
688 	else {
689 		printf("mountroot: invalid directive `%s'\n", dir);
690 		/* Ignore the rest of the line. */
691 		(void)parse_skipto(conf, '\n');
692 		error = EINVAL;
693 	}
694 	free(dir, M_TEMP);
695 	return (error);
696 }
697 
698 static int
699 parse_mount_dev_present(const char *dev)
700 {
701 	struct nameidata nd;
702 	int error;
703 
704 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
705 	error = namei(&nd);
706 	if (!error)
707 		vput(nd.ni_vp);
708 	NDFREE(&nd, NDF_ONLY_PNBUF);
709 	return (error != 0) ? 0 : 1;
710 }
711 
712 #define	ERRMSGL	255
713 static int
714 parse_mount(char **conf)
715 {
716 	char *errmsg;
717 	struct mntarg *ma;
718 	char *dev, *fs, *opts, *tok;
719 	int delay, error, timeout;
720 
721 	error = parse_token(conf, &tok);
722 	if (error)
723 		return (error);
724 	fs = tok;
725 	error = parse_skipto(&tok, ':');
726 	if (error) {
727 		free(fs, M_TEMP);
728 		return (error);
729 	}
730 	parse_poke(&tok, '\0');
731 	parse_advance(&tok);
732 	dev = tok;
733 
734 	if (root_mount_mddev != -1) {
735 		/* Handle substitution for the md unit number. */
736 		tok = strstr(dev, "md#");
737 		if (tok != NULL)
738 			tok[2] = '0' + root_mount_mddev;
739 	}
740 
741 	/* Parse options. */
742 	error = parse_token(conf, &tok);
743 	opts = (error == 0) ? tok : NULL;
744 
745 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
746 	    (opts != NULL) ? opts : "");
747 
748 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
749 
750 	if (vfs_byname(fs) == NULL) {
751 		strlcpy(errmsg, "unknown file system", ERRMSGL);
752 		error = ENOENT;
753 		goto out;
754 	}
755 
756 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
757 	if (error != 0)
758 		goto out;
759 
760 	delay = hz / 10;
761 	timeout = root_mount_timeout * hz;
762 
763 	for (;;) {
764 		ma = NULL;
765 		ma = mount_arg(ma, "fstype", fs, -1);
766 		ma = mount_arg(ma, "fspath", "/", -1);
767 		ma = mount_arg(ma, "from", dev, -1);
768 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
769 		ma = mount_arg(ma, "ro", NULL, 0);
770 		ma = parse_mountroot_options(ma, opts);
771 
772 		error = kernel_mount(ma, MNT_ROOTFS);
773 		if (error == 0 || timeout <= 0)
774 			break;
775 
776 		if (root_mount_timeout * hz == timeout ||
777 		    (bootverbose && timeout % hz == 0)) {
778 			printf("Mounting from %s:%s failed with error %d; "
779 			    "retrying for %d more second%s\n", fs, dev, error,
780 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
781 		}
782 		pause("rmretry", delay);
783 		timeout -= delay;
784 	}
785  out:
786 	if (error) {
787 		printf("Mounting from %s:%s failed with error %d",
788 		    fs, dev, error);
789 		if (errmsg[0] != '\0')
790 			printf(": %s", errmsg);
791 		printf(".\n");
792 	}
793 	free(fs, M_TEMP);
794 	free(errmsg, M_TEMP);
795 	if (opts != NULL)
796 		free(opts, M_TEMP);
797 	/* kernel_mount can return -1 on error. */
798 	return ((error < 0) ? EDOOFUS : error);
799 }
800 #undef ERRMSGL
801 
802 static int
803 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
804 {
805 	struct mount *mp;
806 	char *conf;
807 	int error;
808 
809 	root_mount_mddev = -1;
810 
811 retry:
812 	conf = sbuf_data(sb);
813 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
814 	error = (mp == NULL) ? 0 : EDOOFUS;
815 	root_mount_onfail = A_CONTINUE;
816 	while (mp == NULL) {
817 		error = parse_skipto(&conf, CC_NONWHITESPACE);
818 		if (error == PE_EOL) {
819 			parse_advance(&conf);
820 			continue;
821 		}
822 		if (error < 0)
823 			break;
824 		switch (parse_peek(&conf)) {
825 		case '#':
826 			error = parse_skipto(&conf, '\n');
827 			break;
828 		case '.':
829 			error = parse_directive(&conf);
830 			break;
831 		default:
832 			error = parse_mount(&conf);
833 			if (error == -1) {
834 				printf("mountroot: invalid file system "
835 				    "specification.\n");
836 				error = 0;
837 			}
838 			break;
839 		}
840 		if (error < 0)
841 			break;
842 		/* Ignore any trailing garbage on the line. */
843 		if (parse_peek(&conf) != '\n') {
844 			printf("mountroot: advancing to next directive...\n");
845 			(void)parse_skipto(&conf, '\n');
846 		}
847 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
848 	}
849 	if (mp != NULL)
850 		return (0);
851 
852 	/*
853 	 * We failed to mount (a new) root.
854 	 */
855 	switch (root_mount_onfail) {
856 	case A_CONTINUE:
857 		break;
858 	case A_PANIC:
859 		panic("mountroot: unable to (re-)mount root.");
860 		/* NOTREACHED */
861 	case A_RETRY:
862 		goto retry;
863 	case A_REBOOT:
864 		kern_reboot(RB_NOSYNC);
865 		/* NOTREACHED */
866 	}
867 
868 	return (error);
869 }
870 
871 static void
872 vfs_mountroot_conf0(struct sbuf *sb)
873 {
874 	char *s, *tok, *mnt, *opt;
875 	int error;
876 
877 	sbuf_printf(sb, ".onfail panic\n");
878 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
879 	if (boothowto & RB_ASKNAME)
880 		sbuf_printf(sb, ".ask\n");
881 #ifdef ROOTDEVNAME
882 	if (boothowto & RB_DFLTROOT)
883 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
884 #endif
885 	if (boothowto & RB_CDROM) {
886 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
887 		sbuf_printf(sb, ".timeout 0\n");
888 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
889 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
890 	}
891 	s = kern_getenv("vfs.root.mountfrom");
892 	if (s != NULL) {
893 		opt = kern_getenv("vfs.root.mountfrom.options");
894 		tok = s;
895 		error = parse_token(&tok, &mnt);
896 		while (!error) {
897 			sbuf_printf(sb, "%s %s\n", mnt,
898 			    (opt != NULL) ? opt : "");
899 			free(mnt, M_TEMP);
900 			error = parse_token(&tok, &mnt);
901 		}
902 		if (opt != NULL)
903 			freeenv(opt);
904 		freeenv(s);
905 	}
906 	if (rootdevnames[0] != NULL)
907 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
908 	if (rootdevnames[1] != NULL)
909 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
910 #ifdef ROOTDEVNAME
911 	if (!(boothowto & RB_DFLTROOT))
912 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
913 #endif
914 	if (!(boothowto & RB_ASKNAME))
915 		sbuf_printf(sb, ".ask\n");
916 }
917 
918 static int
919 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
920 {
921 	static char buf[128];
922 	struct nameidata nd;
923 	off_t ofs;
924 	ssize_t resid;
925 	int error, flags, len;
926 
927 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
928 	flags = FREAD;
929 	error = vn_open(&nd, &flags, 0, NULL);
930 	if (error)
931 		return (error);
932 
933 	NDFREE(&nd, NDF_ONLY_PNBUF);
934 	ofs = 0;
935 	len = sizeof(buf) - 1;
936 	while (1) {
937 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
938 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
939 		    NOCRED, &resid, td);
940 		if (error)
941 			break;
942 		if (resid == len)
943 			break;
944 		buf[len - resid] = 0;
945 		sbuf_printf(sb, "%s", buf);
946 		ofs += len - resid;
947 	}
948 
949 	VOP_UNLOCK(nd.ni_vp, 0);
950 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
951 	return (error);
952 }
953 
954 static void
955 vfs_mountroot_wait(void)
956 {
957 	struct root_hold_token *h;
958 	struct timeval lastfail;
959 	int curfail;
960 
961 	TSENTER();
962 
963 	curfail = 0;
964 	while (1) {
965 		g_waitidle();
966 		mtx_lock(&root_holds_mtx);
967 		if (LIST_EMPTY(&root_holds)) {
968 			mtx_unlock(&root_holds_mtx);
969 			break;
970 		}
971 		if (ppsratecheck(&lastfail, &curfail, 1)) {
972 			printf("Root mount waiting for:");
973 			LIST_FOREACH(h, &root_holds, list)
974 				printf(" %s", h->who);
975 			printf("\n");
976 		}
977 		TSWAIT("root mount");
978 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
979 		    hz);
980 		TSUNWAIT("root mount");
981 	}
982 
983 	TSEXIT();
984 }
985 
986 static int
987 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
988 {
989 	int delay, timeout;
990 
991 	/*
992 	 * In case of ZFS and NFS we don't have a way to wait for
993 	 * specific device.  Also do the wait if the user forced that
994 	 * behaviour by setting vfs.root_mount_always_wait=1.
995 	 */
996 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
997 	    dev[0] == '\0' || root_mount_always_wait != 0) {
998 		vfs_mountroot_wait();
999 		return (0);
1000 	}
1001 
1002 	/*
1003 	 * Otherwise, no point in waiting if the device is already there.
1004 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1005 	 * eg for geom_part(4) to finish tasting.
1006 	 */
1007 	g_waitidle();
1008 	if (parse_mount_dev_present(dev))
1009 		return (0);
1010 
1011 	/*
1012 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1013 	 * to behave exactly as it used to work before.
1014 	 */
1015 	vfs_mountroot_wait();
1016 	printf("mountroot: waiting for device %s...\n", dev);
1017 	delay = hz / 10;
1018 	timeout = root_mount_timeout * hz;
1019 	do {
1020 		pause("rmdev", delay);
1021 		timeout -= delay;
1022 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1023 
1024 	if (timeout <= 0)
1025 		return (ENODEV);
1026 
1027 	return (0);
1028 }
1029 
1030 void
1031 vfs_mountroot(void)
1032 {
1033 	struct mount *mp;
1034 	struct sbuf *sb;
1035 	struct thread *td;
1036 	time_t timebase;
1037 	int error;
1038 
1039 	mtx_assert(&Giant, MA_NOTOWNED);
1040 
1041 	TSENTER();
1042 
1043 	td = curthread;
1044 
1045 	sb = sbuf_new_auto();
1046 	vfs_mountroot_conf0(sb);
1047 	sbuf_finish(sb);
1048 
1049 	error = vfs_mountroot_devfs(td, &mp);
1050 	while (!error) {
1051 		error = vfs_mountroot_parse(sb, mp);
1052 		if (!error) {
1053 			vfs_mountroot_shuffle(td, mp);
1054 			sbuf_clear(sb);
1055 			error = vfs_mountroot_readconf(td, sb);
1056 			sbuf_finish(sb);
1057 		}
1058 	}
1059 
1060 	sbuf_delete(sb);
1061 
1062 	/*
1063 	 * Iterate over all currently mounted file systems and use
1064 	 * the time stamp found to check and/or initialize the RTC.
1065 	 * Call inittodr() only once and pass it the largest of the
1066 	 * timestamps we encounter.
1067 	 */
1068 	timebase = 0;
1069 	mtx_lock(&mountlist_mtx);
1070 	mp = TAILQ_FIRST(&mountlist);
1071 	while (mp != NULL) {
1072 		if (mp->mnt_time > timebase)
1073 			timebase = mp->mnt_time;
1074 		mp = TAILQ_NEXT(mp, mnt_list);
1075 	}
1076 	mtx_unlock(&mountlist_mtx);
1077 	inittodr(timebase);
1078 
1079 	/* Keep prison0's root in sync with the global rootvnode. */
1080 	mtx_lock(&prison0.pr_mtx);
1081 	prison0.pr_root = rootvnode;
1082 	vref(prison0.pr_root);
1083 	mtx_unlock(&prison0.pr_mtx);
1084 
1085 	mtx_lock(&root_holds_mtx);
1086 	atomic_store_rel_int(&root_mount_complete, 1);
1087 	wakeup(&root_mount_complete);
1088 	mtx_unlock(&root_holds_mtx);
1089 
1090 	EVENTHANDLER_INVOKE(mountroot);
1091 
1092 	TSEXIT();
1093 }
1094 
1095 static struct mntarg *
1096 parse_mountroot_options(struct mntarg *ma, const char *options)
1097 {
1098 	char *p;
1099 	char *name, *name_arg;
1100 	char *val, *val_arg;
1101 	char *opts;
1102 
1103 	if (options == NULL || options[0] == '\0')
1104 		return (ma);
1105 
1106 	p = opts = strdup(options, M_MOUNT);
1107 	if (opts == NULL) {
1108 		return (ma);
1109 	}
1110 
1111 	while((name = strsep(&p, ",")) != NULL) {
1112 		if (name[0] == '\0')
1113 			break;
1114 
1115 		val = strchr(name, '=');
1116 		if (val != NULL) {
1117 			*val = '\0';
1118 			++val;
1119 		}
1120 		if( strcmp(name, "rw") == 0 ||
1121 		    strcmp(name, "noro") == 0) {
1122 			/*
1123 			 * The first time we mount the root file system,
1124 			 * we need to mount 'ro', so We need to ignore
1125 			 * 'rw' and 'noro' mount options.
1126 			 */
1127 			continue;
1128 		}
1129 		name_arg = strdup(name, M_MOUNT);
1130 		val_arg = NULL;
1131 		if (val != NULL)
1132 			val_arg = strdup(val, M_MOUNT);
1133 
1134 		ma = mount_arg(ma, name_arg, val_arg,
1135 		    (val_arg != NULL ? -1 : 0));
1136 	}
1137 	free(opts, M_MOUNT);
1138 	return (ma);
1139 }
1140