xref: /freebsd/sys/kern/vfs_mountroot.c (revision 036d2e814bf0f5d88ffb4b24c159320894541757)
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 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
267 		TAILQ_INIT(opts);
268 		mp->mnt_opt = opts;
269 
270 		mtx_lock(&mountlist_mtx);
271 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
272 		mtx_unlock(&mountlist_mtx);
273 
274 		*mpp = mp;
275 		rootdevmp = mp;
276 		vfs_op_exit(mp);
277 	}
278 
279 	set_rootvnode();
280 
281 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
282 	if (error)
283 		printf("kern_symlink /dev -> / returns %d\n", error);
284 
285 	return (error);
286 }
287 
288 static void
289 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
290 {
291 	struct nameidata nd;
292 	struct mount *mporoot, *mpnroot;
293 	struct vnode *vp, *vporoot, *vpdevfs;
294 	char *fspath;
295 	int error;
296 
297 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
298 
299 	/* Shuffle the mountlist. */
300 	mtx_lock(&mountlist_mtx);
301 	mporoot = TAILQ_FIRST(&mountlist);
302 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
303 	if (mporoot != mpdevfs) {
304 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
305 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
306 	}
307 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
308 	mtx_unlock(&mountlist_mtx);
309 
310 	cache_purgevfs(mporoot, true);
311 	if (mporoot != mpdevfs)
312 		cache_purgevfs(mpdevfs, true);
313 
314 	if (VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot))
315 		panic("vfs_mountroot_shuffle: Cannot find root vnode");
316 
317 	VI_LOCK(vporoot);
318 	vporoot->v_iflag &= ~VI_MOUNT;
319 	VI_UNLOCK(vporoot);
320 	vporoot->v_mountedhere = NULL;
321 	mporoot->mnt_flag &= ~MNT_ROOTFS;
322 	mporoot->mnt_vnodecovered = NULL;
323 	vput(vporoot);
324 
325 	/* Set up the new rootvnode, and purge the cache */
326 	mpnroot->mnt_vnodecovered = NULL;
327 	set_rootvnode();
328 	cache_purgevfs(rootvnode->v_mount, true);
329 
330 	if (mporoot != mpdevfs) {
331 		/* Remount old root under /.mount or /mnt */
332 		fspath = "/.mount";
333 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
334 		    fspath, td);
335 		error = namei(&nd);
336 		if (error) {
337 			NDFREE(&nd, NDF_ONLY_PNBUF);
338 			fspath = "/mnt";
339 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
340 			    fspath, td);
341 			error = namei(&nd);
342 		}
343 		if (!error) {
344 			vp = nd.ni_vp;
345 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
346 			if (!error)
347 				error = vinvalbuf(vp, V_SAVE, 0, 0);
348 			if (!error) {
349 				cache_purge(vp);
350 				mporoot->mnt_vnodecovered = vp;
351 				vp->v_mountedhere = mporoot;
352 				strlcpy(mporoot->mnt_stat.f_mntonname,
353 				    fspath, MNAMELEN);
354 				VOP_UNLOCK(vp, 0);
355 			} else
356 				vput(vp);
357 		}
358 		NDFREE(&nd, NDF_ONLY_PNBUF);
359 
360 		if (error)
361 			printf("mountroot: unable to remount previous root "
362 			    "under /.mount or /mnt (error %d)\n", error);
363 	}
364 
365 	/* Remount devfs under /dev */
366 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
367 	error = namei(&nd);
368 	if (!error) {
369 		vp = nd.ni_vp;
370 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
371 		if (!error)
372 			error = vinvalbuf(vp, V_SAVE, 0, 0);
373 		if (!error) {
374 			vpdevfs = mpdevfs->mnt_vnodecovered;
375 			if (vpdevfs != NULL) {
376 				cache_purge(vpdevfs);
377 				vpdevfs->v_mountedhere = NULL;
378 				vrele(vpdevfs);
379 			}
380 			mpdevfs->mnt_vnodecovered = vp;
381 			vp->v_mountedhere = mpdevfs;
382 			VOP_UNLOCK(vp, 0);
383 		} else
384 			vput(vp);
385 	}
386 	if (error)
387 		printf("mountroot: unable to remount devfs under /dev "
388 		    "(error %d)\n", error);
389 	NDFREE(&nd, NDF_ONLY_PNBUF);
390 
391 	if (mporoot == mpdevfs) {
392 		vfs_unbusy(mpdevfs);
393 		/* Unlink the no longer needed /dev/dev -> / symlink */
394 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
395 		    UIO_SYSSPACE, 0, 0);
396 		if (error)
397 			printf("mountroot: unable to unlink /dev/dev "
398 			    "(error %d)\n", error);
399 	}
400 }
401 
402 /*
403  * Configuration parser.
404  */
405 
406 /* Parser character classes. */
407 #define	CC_WHITESPACE		-1
408 #define	CC_NONWHITESPACE	-2
409 
410 /* Parse errors. */
411 #define	PE_EOF			-1
412 #define	PE_EOL			-2
413 
414 static __inline int
415 parse_peek(char **conf)
416 {
417 
418 	return (**conf);
419 }
420 
421 static __inline void
422 parse_poke(char **conf, int c)
423 {
424 
425 	**conf = c;
426 }
427 
428 static __inline void
429 parse_advance(char **conf)
430 {
431 
432 	(*conf)++;
433 }
434 
435 static int
436 parse_skipto(char **conf, int mc)
437 {
438 	int c, match;
439 
440 	while (1) {
441 		c = parse_peek(conf);
442 		if (c == 0)
443 			return (PE_EOF);
444 		switch (mc) {
445 		case CC_WHITESPACE:
446 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
447 			break;
448 		case CC_NONWHITESPACE:
449 			if (c == '\n')
450 				return (PE_EOL);
451 			match = (c != ' ' && c != '\t') ? 1 : 0;
452 			break;
453 		default:
454 			match = (c == mc) ? 1 : 0;
455 			break;
456 		}
457 		if (match)
458 			break;
459 		parse_advance(conf);
460 	}
461 	return (0);
462 }
463 
464 static int
465 parse_token(char **conf, char **tok)
466 {
467 	char *p;
468 	size_t len;
469 	int error;
470 
471 	*tok = NULL;
472 	error = parse_skipto(conf, CC_NONWHITESPACE);
473 	if (error)
474 		return (error);
475 	p = *conf;
476 	error = parse_skipto(conf, CC_WHITESPACE);
477 	len = *conf - p;
478 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
479 	bcopy(p, *tok, len);
480 	return (0);
481 }
482 
483 static void
484 parse_dir_ask_printenv(const char *var)
485 {
486 	char *val;
487 
488 	val = kern_getenv(var);
489 	if (val != NULL) {
490 		printf("  %s=%s\n", var, val);
491 		freeenv(val);
492 	}
493 }
494 
495 static int
496 parse_dir_ask(char **conf)
497 {
498 	char name[80];
499 	char *mnt;
500 	int error;
501 
502 	vfs_mountroot_wait();
503 
504 	printf("\nLoader variables:\n");
505 	parse_dir_ask_printenv("vfs.root.mountfrom");
506 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
507 
508 	printf("\nManual root filesystem specification:\n");
509 	printf("  <fstype>:<device> [options]\n");
510 	printf("      Mount <device> using filesystem <fstype>\n");
511 	printf("      and with the specified (optional) option list.\n");
512 	printf("\n");
513 	printf("    eg. ufs:/dev/da0s1a\n");
514 	printf("        zfs:zroot/ROOT/default\n");
515 	printf("        cd9660:/dev/cd0 ro\n");
516 	printf("          (which is equivalent to: ");
517 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
518 	printf("\n");
519 	printf("  ?               List valid disk boot devices\n");
520 	printf("  .               Yield 1 second (for background tasks)\n");
521 	printf("  <empty line>    Abort manual input\n");
522 
523 	do {
524 		error = EINVAL;
525 		printf("\nmountroot> ");
526 		cngets(name, sizeof(name), GETS_ECHO);
527 		if (name[0] == '\0')
528 			break;
529 		if (name[0] == '?' && name[1] == '\0') {
530 			printf("\nList of GEOM managed disk devices:\n  ");
531 			g_dev_print();
532 			continue;
533 		}
534 		if (name[0] == '.' && name[1] == '\0') {
535 			pause("rmask", hz);
536 			continue;
537 		}
538 		mnt = name;
539 		error = parse_mount(&mnt);
540 		if (error == -1)
541 			printf("Invalid file system specification.\n");
542 	} while (error != 0);
543 
544 	return (error);
545 }
546 
547 static int
548 parse_dir_md(char **conf)
549 {
550 	struct stat sb;
551 	struct thread *td;
552 	struct md_ioctl *mdio;
553 	char *path, *tok;
554 	int error, fd, len;
555 
556 	td = curthread;
557 
558 	error = parse_token(conf, &tok);
559 	if (error)
560 		return (error);
561 
562 	len = strlen(tok);
563 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
564 	path = (void *)(mdio + 1);
565 	bcopy(tok, path, len);
566 	free(tok, M_TEMP);
567 
568 	/* Get file status. */
569 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
570 	if (error)
571 		goto out;
572 
573 	/* Open /dev/mdctl so that we can attach/detach. */
574 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
575 	    O_RDWR, 0);
576 	if (error)
577 		goto out;
578 
579 	fd = td->td_retval[0];
580 	mdio->md_version = MDIOVERSION;
581 	mdio->md_type = MD_VNODE;
582 
583 	if (root_mount_mddev != -1) {
584 		mdio->md_unit = root_mount_mddev;
585 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
586 		/* Ignore errors. We don't care. */
587 		root_mount_mddev = -1;
588 	}
589 
590 	mdio->md_file = (void *)(mdio + 1);
591 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
592 	mdio->md_mediasize = sb.st_size;
593 	mdio->md_unit = 0;
594 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
595 	if (error)
596 		goto out;
597 
598 	if (mdio->md_unit > 9) {
599 		printf("rootmount: too many md units\n");
600 		mdio->md_file = NULL;
601 		mdio->md_options = 0;
602 		mdio->md_mediasize = 0;
603 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
604 		/* Ignore errors. We don't care. */
605 		error = ERANGE;
606 		goto out;
607 	}
608 
609 	root_mount_mddev = mdio->md_unit;
610 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
611 
612 	error = kern_close(td, fd);
613 
614  out:
615 	free(mdio, M_TEMP);
616 	return (error);
617 }
618 
619 static int
620 parse_dir_onfail(char **conf)
621 {
622 	char *action;
623 	int error;
624 
625 	error = parse_token(conf, &action);
626 	if (error)
627 		return (error);
628 
629 	if (!strcmp(action, "continue"))
630 		root_mount_onfail = A_CONTINUE;
631 	else if (!strcmp(action, "panic"))
632 		root_mount_onfail = A_PANIC;
633 	else if (!strcmp(action, "reboot"))
634 		root_mount_onfail = A_REBOOT;
635 	else if (!strcmp(action, "retry"))
636 		root_mount_onfail = A_RETRY;
637 	else {
638 		printf("rootmount: %s: unknown action\n", action);
639 		error = EINVAL;
640 	}
641 
642 	free(action, M_TEMP);
643 	return (0);
644 }
645 
646 static int
647 parse_dir_timeout(char **conf)
648 {
649 	char *tok, *endtok;
650 	long secs;
651 	int error;
652 
653 	error = parse_token(conf, &tok);
654 	if (error)
655 		return (error);
656 
657 	secs = strtol(tok, &endtok, 0);
658 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
659 	if (!error)
660 		root_mount_timeout = secs;
661 	free(tok, M_TEMP);
662 	return (error);
663 }
664 
665 static int
666 parse_directive(char **conf)
667 {
668 	char *dir;
669 	int error;
670 
671 	error = parse_token(conf, &dir);
672 	if (error)
673 		return (error);
674 
675 	if (strcmp(dir, ".ask") == 0)
676 		error = parse_dir_ask(conf);
677 	else if (strcmp(dir, ".md") == 0)
678 		error = parse_dir_md(conf);
679 	else if (strcmp(dir, ".onfail") == 0)
680 		error = parse_dir_onfail(conf);
681 	else if (strcmp(dir, ".timeout") == 0)
682 		error = parse_dir_timeout(conf);
683 	else {
684 		printf("mountroot: invalid directive `%s'\n", dir);
685 		/* Ignore the rest of the line. */
686 		(void)parse_skipto(conf, '\n');
687 		error = EINVAL;
688 	}
689 	free(dir, M_TEMP);
690 	return (error);
691 }
692 
693 static int
694 parse_mount_dev_present(const char *dev)
695 {
696 	struct nameidata nd;
697 	int error;
698 
699 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
700 	error = namei(&nd);
701 	if (!error)
702 		vput(nd.ni_vp);
703 	NDFREE(&nd, NDF_ONLY_PNBUF);
704 	return (error != 0) ? 0 : 1;
705 }
706 
707 #define	ERRMSGL	255
708 static int
709 parse_mount(char **conf)
710 {
711 	char *errmsg;
712 	struct mntarg *ma;
713 	char *dev, *fs, *opts, *tok;
714 	int delay, error, timeout;
715 
716 	error = parse_token(conf, &tok);
717 	if (error)
718 		return (error);
719 	fs = tok;
720 	error = parse_skipto(&tok, ':');
721 	if (error) {
722 		free(fs, M_TEMP);
723 		return (error);
724 	}
725 	parse_poke(&tok, '\0');
726 	parse_advance(&tok);
727 	dev = tok;
728 
729 	if (root_mount_mddev != -1) {
730 		/* Handle substitution for the md unit number. */
731 		tok = strstr(dev, "md#");
732 		if (tok != NULL)
733 			tok[2] = '0' + root_mount_mddev;
734 	}
735 
736 	/* Parse options. */
737 	error = parse_token(conf, &tok);
738 	opts = (error == 0) ? tok : NULL;
739 
740 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
741 	    (opts != NULL) ? opts : "");
742 
743 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
744 
745 	if (vfs_byname(fs) == NULL) {
746 		strlcpy(errmsg, "unknown file system", ERRMSGL);
747 		error = ENOENT;
748 		goto out;
749 	}
750 
751 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
752 	if (error != 0)
753 		goto out;
754 
755 	delay = hz / 10;
756 	timeout = root_mount_timeout * hz;
757 
758 	for (;;) {
759 		ma = NULL;
760 		ma = mount_arg(ma, "fstype", fs, -1);
761 		ma = mount_arg(ma, "fspath", "/", -1);
762 		ma = mount_arg(ma, "from", dev, -1);
763 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
764 		ma = mount_arg(ma, "ro", NULL, 0);
765 		ma = parse_mountroot_options(ma, opts);
766 
767 		error = kernel_mount(ma, MNT_ROOTFS);
768 		if (error == 0 || timeout <= 0)
769 			break;
770 
771 		if (root_mount_timeout * hz == timeout ||
772 		    (bootverbose && timeout % hz == 0)) {
773 			printf("Mounting from %s:%s failed with error %d; "
774 			    "retrying for %d more second%s\n", fs, dev, error,
775 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
776 		}
777 		pause("rmretry", delay);
778 		timeout -= delay;
779 	}
780  out:
781 	if (error) {
782 		printf("Mounting from %s:%s failed with error %d",
783 		    fs, dev, error);
784 		if (errmsg[0] != '\0')
785 			printf(": %s", errmsg);
786 		printf(".\n");
787 	}
788 	free(fs, M_TEMP);
789 	free(errmsg, M_TEMP);
790 	if (opts != NULL)
791 		free(opts, M_TEMP);
792 	/* kernel_mount can return -1 on error. */
793 	return ((error < 0) ? EDOOFUS : error);
794 }
795 #undef ERRMSGL
796 
797 static int
798 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
799 {
800 	struct mount *mp;
801 	char *conf;
802 	int error;
803 
804 	root_mount_mddev = -1;
805 
806 retry:
807 	conf = sbuf_data(sb);
808 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
809 	error = (mp == NULL) ? 0 : EDOOFUS;
810 	root_mount_onfail = A_CONTINUE;
811 	while (mp == NULL) {
812 		error = parse_skipto(&conf, CC_NONWHITESPACE);
813 		if (error == PE_EOL) {
814 			parse_advance(&conf);
815 			continue;
816 		}
817 		if (error < 0)
818 			break;
819 		switch (parse_peek(&conf)) {
820 		case '#':
821 			error = parse_skipto(&conf, '\n');
822 			break;
823 		case '.':
824 			error = parse_directive(&conf);
825 			break;
826 		default:
827 			error = parse_mount(&conf);
828 			if (error == -1) {
829 				printf("mountroot: invalid file system "
830 				    "specification.\n");
831 				error = 0;
832 			}
833 			break;
834 		}
835 		if (error < 0)
836 			break;
837 		/* Ignore any trailing garbage on the line. */
838 		if (parse_peek(&conf) != '\n') {
839 			printf("mountroot: advancing to next directive...\n");
840 			(void)parse_skipto(&conf, '\n');
841 		}
842 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
843 	}
844 	if (mp != NULL)
845 		return (0);
846 
847 	/*
848 	 * We failed to mount (a new) root.
849 	 */
850 	switch (root_mount_onfail) {
851 	case A_CONTINUE:
852 		break;
853 	case A_PANIC:
854 		panic("mountroot: unable to (re-)mount root.");
855 		/* NOTREACHED */
856 	case A_RETRY:
857 		goto retry;
858 	case A_REBOOT:
859 		kern_reboot(RB_NOSYNC);
860 		/* NOTREACHED */
861 	}
862 
863 	return (error);
864 }
865 
866 static void
867 vfs_mountroot_conf0(struct sbuf *sb)
868 {
869 	char *s, *tok, *mnt, *opt;
870 	int error;
871 
872 	sbuf_printf(sb, ".onfail panic\n");
873 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
874 	if (boothowto & RB_ASKNAME)
875 		sbuf_printf(sb, ".ask\n");
876 #ifdef ROOTDEVNAME
877 	if (boothowto & RB_DFLTROOT)
878 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
879 #endif
880 	if (boothowto & RB_CDROM) {
881 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
882 		sbuf_printf(sb, ".timeout 0\n");
883 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
884 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
885 	}
886 	s = kern_getenv("vfs.root.mountfrom");
887 	if (s != NULL) {
888 		opt = kern_getenv("vfs.root.mountfrom.options");
889 		tok = s;
890 		error = parse_token(&tok, &mnt);
891 		while (!error) {
892 			sbuf_printf(sb, "%s %s\n", mnt,
893 			    (opt != NULL) ? opt : "");
894 			free(mnt, M_TEMP);
895 			error = parse_token(&tok, &mnt);
896 		}
897 		if (opt != NULL)
898 			freeenv(opt);
899 		freeenv(s);
900 	}
901 	if (rootdevnames[0] != NULL)
902 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
903 	if (rootdevnames[1] != NULL)
904 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
905 #ifdef ROOTDEVNAME
906 	if (!(boothowto & RB_DFLTROOT))
907 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
908 #endif
909 	if (!(boothowto & RB_ASKNAME))
910 		sbuf_printf(sb, ".ask\n");
911 }
912 
913 static int
914 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
915 {
916 	static char buf[128];
917 	struct nameidata nd;
918 	off_t ofs;
919 	ssize_t resid;
920 	int error, flags, len;
921 
922 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
923 	flags = FREAD;
924 	error = vn_open(&nd, &flags, 0, NULL);
925 	if (error)
926 		return (error);
927 
928 	NDFREE(&nd, NDF_ONLY_PNBUF);
929 	ofs = 0;
930 	len = sizeof(buf) - 1;
931 	while (1) {
932 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
933 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
934 		    NOCRED, &resid, td);
935 		if (error)
936 			break;
937 		if (resid == len)
938 			break;
939 		buf[len - resid] = 0;
940 		sbuf_printf(sb, "%s", buf);
941 		ofs += len - resid;
942 	}
943 
944 	VOP_UNLOCK(nd.ni_vp, 0);
945 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
946 	return (error);
947 }
948 
949 static void
950 vfs_mountroot_wait(void)
951 {
952 	struct root_hold_token *h;
953 	struct timeval lastfail;
954 	int curfail;
955 
956 	TSENTER();
957 
958 	curfail = 0;
959 	while (1) {
960 		g_waitidle();
961 		mtx_lock(&root_holds_mtx);
962 		if (LIST_EMPTY(&root_holds)) {
963 			mtx_unlock(&root_holds_mtx);
964 			break;
965 		}
966 		if (ppsratecheck(&lastfail, &curfail, 1)) {
967 			printf("Root mount waiting for:");
968 			LIST_FOREACH(h, &root_holds, list)
969 				printf(" %s", h->who);
970 			printf("\n");
971 		}
972 		TSWAIT("root mount");
973 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
974 		    hz);
975 		TSUNWAIT("root mount");
976 	}
977 
978 	TSEXIT();
979 }
980 
981 static int
982 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
983 {
984 	int delay, timeout;
985 
986 	/*
987 	 * In case of ZFS and NFS we don't have a way to wait for
988 	 * specific device.  Also do the wait if the user forced that
989 	 * behaviour by setting vfs.root_mount_always_wait=1.
990 	 */
991 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
992 	    dev[0] == '\0' || root_mount_always_wait != 0) {
993 		vfs_mountroot_wait();
994 		return (0);
995 	}
996 
997 	/*
998 	 * Otherwise, no point in waiting if the device is already there.
999 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1000 	 * eg for geom_part(4) to finish tasting.
1001 	 */
1002 	g_waitidle();
1003 	if (parse_mount_dev_present(dev))
1004 		return (0);
1005 
1006 	/*
1007 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1008 	 * to behave exactly as it used to work before.
1009 	 */
1010 	vfs_mountroot_wait();
1011 	printf("mountroot: waiting for device %s...\n", dev);
1012 	delay = hz / 10;
1013 	timeout = root_mount_timeout * hz;
1014 	do {
1015 		pause("rmdev", delay);
1016 		timeout -= delay;
1017 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1018 
1019 	if (timeout <= 0)
1020 		return (ENODEV);
1021 
1022 	return (0);
1023 }
1024 
1025 void
1026 vfs_mountroot(void)
1027 {
1028 	struct mount *mp;
1029 	struct sbuf *sb;
1030 	struct thread *td;
1031 	time_t timebase;
1032 	int error;
1033 
1034 	mtx_assert(&Giant, MA_NOTOWNED);
1035 
1036 	TSENTER();
1037 
1038 	td = curthread;
1039 
1040 	sb = sbuf_new_auto();
1041 	vfs_mountroot_conf0(sb);
1042 	sbuf_finish(sb);
1043 
1044 	error = vfs_mountroot_devfs(td, &mp);
1045 	while (!error) {
1046 		error = vfs_mountroot_parse(sb, mp);
1047 		if (!error) {
1048 			vfs_mountroot_shuffle(td, mp);
1049 			sbuf_clear(sb);
1050 			error = vfs_mountroot_readconf(td, sb);
1051 			sbuf_finish(sb);
1052 		}
1053 	}
1054 
1055 	sbuf_delete(sb);
1056 
1057 	/*
1058 	 * Iterate over all currently mounted file systems and use
1059 	 * the time stamp found to check and/or initialize the RTC.
1060 	 * Call inittodr() only once and pass it the largest of the
1061 	 * timestamps we encounter.
1062 	 */
1063 	timebase = 0;
1064 	mtx_lock(&mountlist_mtx);
1065 	mp = TAILQ_FIRST(&mountlist);
1066 	while (mp != NULL) {
1067 		if (mp->mnt_time > timebase)
1068 			timebase = mp->mnt_time;
1069 		mp = TAILQ_NEXT(mp, mnt_list);
1070 	}
1071 	mtx_unlock(&mountlist_mtx);
1072 	inittodr(timebase);
1073 
1074 	/* Keep prison0's root in sync with the global rootvnode. */
1075 	mtx_lock(&prison0.pr_mtx);
1076 	prison0.pr_root = rootvnode;
1077 	vref(prison0.pr_root);
1078 	mtx_unlock(&prison0.pr_mtx);
1079 
1080 	mtx_lock(&root_holds_mtx);
1081 	atomic_store_rel_int(&root_mount_complete, 1);
1082 	wakeup(&root_mount_complete);
1083 	mtx_unlock(&root_holds_mtx);
1084 
1085 	EVENTHANDLER_INVOKE(mountroot);
1086 
1087 	TSEXIT();
1088 }
1089 
1090 static struct mntarg *
1091 parse_mountroot_options(struct mntarg *ma, const char *options)
1092 {
1093 	char *p;
1094 	char *name, *name_arg;
1095 	char *val, *val_arg;
1096 	char *opts;
1097 
1098 	if (options == NULL || options[0] == '\0')
1099 		return (ma);
1100 
1101 	p = opts = strdup(options, M_MOUNT);
1102 	if (opts == NULL) {
1103 		return (ma);
1104 	}
1105 
1106 	while((name = strsep(&p, ",")) != NULL) {
1107 		if (name[0] == '\0')
1108 			break;
1109 
1110 		val = strchr(name, '=');
1111 		if (val != NULL) {
1112 			*val = '\0';
1113 			++val;
1114 		}
1115 		if( strcmp(name, "rw") == 0 ||
1116 		    strcmp(name, "noro") == 0) {
1117 			/*
1118 			 * The first time we mount the root file system,
1119 			 * we need to mount 'ro', so We need to ignore
1120 			 * 'rw' and 'noro' mount options.
1121 			 */
1122 			continue;
1123 		}
1124 		name_arg = strdup(name, M_MOUNT);
1125 		val_arg = NULL;
1126 		if (val != NULL)
1127 			val_arg = strdup(val, M_MOUNT);
1128 
1129 		ma = mount_arg(ma, name_arg, val_arg,
1130 		    (val_arg != NULL ? -1 : 0));
1131 	}
1132 	free(opts, M_MOUNT);
1133 	return (ma);
1134 }
1135