xref: /freebsd/sys/kern/vfs_mountroot.c (revision eda14cbc264d6969b02f2b1994cef11148e914f1)
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 static TAILQ_HEAD(, root_hold_token)	root_holds =
115     TAILQ_HEAD_INITIALIZER(root_holds);
116 
117 enum action {
118 	A_CONTINUE,
119 	A_PANIC,
120 	A_REBOOT,
121 	A_RETRY
122 };
123 
124 enum rh_flags {
125 	RH_FREE,
126 	RH_ALLOC,
127 	RH_ARG,
128 };
129 
130 static enum action root_mount_onfail = A_CONTINUE;
131 
132 static int root_mount_mddev;
133 static int root_mount_complete;
134 
135 /* By default wait up to 3 seconds for devices to appear. */
136 static int root_mount_timeout = 3;
137 TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout);
138 
139 static int root_mount_always_wait = 0;
140 SYSCTL_INT(_vfs, OID_AUTO, root_mount_always_wait, CTLFLAG_RDTUN,
141     &root_mount_always_wait, 0,
142     "Wait for root mount holds even if the root device already exists");
143 
144 SYSCTL_PROC(_vfs, OID_AUTO, root_mount_hold,
145     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
146     NULL, 0, sysctl_vfs_root_mount_hold, "A",
147     "List of root mount hold tokens");
148 
149 static int
150 sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS)
151 {
152 	struct sbuf sb;
153 	struct root_hold_token *h;
154 	int error;
155 
156 	sbuf_new(&sb, NULL, 256, SBUF_AUTOEXTEND | SBUF_INCLUDENUL);
157 
158 	mtx_lock(&root_holds_mtx);
159 	TAILQ_FOREACH(h, &root_holds, list) {
160 		if (h != TAILQ_FIRST(&root_holds))
161 			sbuf_putc(&sb, ' ');
162 		sbuf_printf(&sb, "%s", h->who);
163 	}
164 	mtx_unlock(&root_holds_mtx);
165 
166 	error = sbuf_finish(&sb);
167 	if (error == 0)
168 		error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
169 	sbuf_delete(&sb);
170 	return (error);
171 }
172 
173 struct root_hold_token *
174 root_mount_hold(const char *identifier)
175 {
176 	struct root_hold_token *h;
177 
178 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
179 	h->flags = RH_ALLOC;
180 	h->who = identifier;
181 	mtx_lock(&root_holds_mtx);
182 	TSHOLD("root mount");
183 	TAILQ_INSERT_TAIL(&root_holds, h, list);
184 	mtx_unlock(&root_holds_mtx);
185 	return (h);
186 }
187 
188 void
189 root_mount_hold_token(const char *identifier, struct root_hold_token *h)
190 {
191 #ifdef INVARIANTS
192 	struct root_hold_token *t;
193 #endif
194 
195 	h->flags = RH_ARG;
196 	h->who = identifier;
197 	mtx_lock(&root_holds_mtx);
198 #ifdef INVARIANTS
199 	TAILQ_FOREACH(t, &root_holds, list) {
200 		if (t == h) {
201 			panic("Duplicate mount hold by '%s' on %p",
202 			    identifier, h);
203 		}
204 	}
205 #endif
206 	TSHOLD("root mount");
207 	TAILQ_INSERT_TAIL(&root_holds, h, list);
208 	mtx_unlock(&root_holds_mtx);
209 }
210 
211 void
212 root_mount_rel(struct root_hold_token *h)
213 {
214 
215 	if (h == NULL || h->flags == RH_FREE)
216 		return;
217 
218 	mtx_lock(&root_holds_mtx);
219 	TAILQ_REMOVE(&root_holds, h, list);
220 	TSRELEASE("root mount");
221 	wakeup(&root_holds);
222 	mtx_unlock(&root_holds_mtx);
223 	if (h->flags == RH_ALLOC) {
224 		free(h, M_DEVBUF);
225 	} else
226 		h->flags = RH_FREE;
227 }
228 
229 int
230 root_mounted(void)
231 {
232 
233 	/* No mutex is acquired here because int stores are atomic. */
234 	return (root_mount_complete);
235 }
236 
237 static void
238 set_rootvnode(void)
239 {
240 
241 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
242 		panic("set_rootvnode: Cannot find root vnode");
243 
244 	VOP_UNLOCK(rootvnode);
245 
246 	pwd_set_rootvnode();
247 }
248 
249 static int
250 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
251 {
252 	struct vfsoptlist *opts;
253 	struct vfsconf *vfsp;
254 	struct mount *mp;
255 	int error;
256 
257 	*mpp = NULL;
258 
259 	if (rootdevmp != NULL) {
260 		/*
261 		 * Already have /dev; this happens during rerooting.
262 		 */
263 		error = vfs_busy(rootdevmp, 0);
264 		if (error != 0)
265 			return (error);
266 		*mpp = rootdevmp;
267 	} else {
268 		vfsp = vfs_byname("devfs");
269 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
270 		if (vfsp == NULL)
271 			return (ENOENT);
272 
273 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
274 
275 		error = VFS_MOUNT(mp);
276 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
277 		if (error)
278 			return (error);
279 
280 		error = VFS_STATFS(mp, &mp->mnt_stat);
281 		KASSERT(error == 0, ("VFS_STATFS(devfs) failed %d", error));
282 		if (error)
283 			return (error);
284 
285 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
286 		TAILQ_INIT(opts);
287 		mp->mnt_opt = opts;
288 
289 		mtx_lock(&mountlist_mtx);
290 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
291 		mtx_unlock(&mountlist_mtx);
292 
293 		*mpp = mp;
294 		rootdevmp = mp;
295 		vfs_op_exit(mp);
296 	}
297 
298 	set_rootvnode();
299 
300 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
301 	if (error)
302 		printf("kern_symlink /dev -> / returns %d\n", error);
303 
304 	return (error);
305 }
306 
307 static void
308 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
309 {
310 	struct nameidata nd;
311 	struct mount *mporoot, *mpnroot;
312 	struct vnode *vp, *vporoot, *vpdevfs;
313 	char *fspath;
314 	int error;
315 
316 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
317 
318 	/* Shuffle the mountlist. */
319 	mtx_lock(&mountlist_mtx);
320 	mporoot = TAILQ_FIRST(&mountlist);
321 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
322 	if (mporoot != mpdevfs) {
323 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
324 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
325 	}
326 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
327 	mtx_unlock(&mountlist_mtx);
328 
329 	cache_purgevfs(mporoot, true);
330 	if (mporoot != mpdevfs)
331 		cache_purgevfs(mpdevfs, true);
332 
333 	if (VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot))
334 		panic("vfs_mountroot_shuffle: Cannot find root vnode");
335 
336 	VI_LOCK(vporoot);
337 	vporoot->v_iflag &= ~VI_MOUNT;
338 	VI_UNLOCK(vporoot);
339 	vporoot->v_mountedhere = NULL;
340 	mporoot->mnt_flag &= ~MNT_ROOTFS;
341 	mporoot->mnt_vnodecovered = NULL;
342 	vput(vporoot);
343 
344 	/* Set up the new rootvnode, and purge the cache */
345 	mpnroot->mnt_vnodecovered = NULL;
346 	set_rootvnode();
347 	cache_purgevfs(rootvnode->v_mount, true);
348 
349 	if (mporoot != mpdevfs) {
350 		/* Remount old root under /.mount or /mnt */
351 		fspath = "/.mount";
352 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
353 		    fspath, td);
354 		error = namei(&nd);
355 		if (error) {
356 			NDFREE(&nd, NDF_ONLY_PNBUF);
357 			fspath = "/mnt";
358 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
359 			    fspath, td);
360 			error = namei(&nd);
361 		}
362 		if (!error) {
363 			vp = nd.ni_vp;
364 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
365 			if (!error)
366 				error = vinvalbuf(vp, V_SAVE, 0, 0);
367 			if (!error) {
368 				cache_purge(vp);
369 				mporoot->mnt_vnodecovered = vp;
370 				vp->v_mountedhere = mporoot;
371 				strlcpy(mporoot->mnt_stat.f_mntonname,
372 				    fspath, MNAMELEN);
373 				VOP_UNLOCK(vp);
374 			} else
375 				vput(vp);
376 		}
377 		NDFREE(&nd, NDF_ONLY_PNBUF);
378 
379 		if (error)
380 			printf("mountroot: unable to remount previous root "
381 			    "under /.mount or /mnt (error %d)\n", error);
382 	}
383 
384 	/* Remount devfs under /dev */
385 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
386 	error = namei(&nd);
387 	if (!error) {
388 		vp = nd.ni_vp;
389 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
390 		if (!error)
391 			error = vinvalbuf(vp, V_SAVE, 0, 0);
392 		if (!error) {
393 			vpdevfs = mpdevfs->mnt_vnodecovered;
394 			if (vpdevfs != NULL) {
395 				cache_purge(vpdevfs);
396 				vpdevfs->v_mountedhere = NULL;
397 				vrele(vpdevfs);
398 			}
399 			mpdevfs->mnt_vnodecovered = vp;
400 			vp->v_mountedhere = mpdevfs;
401 			VOP_UNLOCK(vp);
402 		} else
403 			vput(vp);
404 	}
405 	if (error)
406 		printf("mountroot: unable to remount devfs under /dev "
407 		    "(error %d)\n", error);
408 	NDFREE(&nd, NDF_ONLY_PNBUF);
409 
410 	if (mporoot == mpdevfs) {
411 		vfs_unbusy(mpdevfs);
412 		/* Unlink the no longer needed /dev/dev -> / symlink */
413 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
414 		    UIO_SYSSPACE, 0, 0);
415 		if (error)
416 			printf("mountroot: unable to unlink /dev/dev "
417 			    "(error %d)\n", error);
418 	}
419 }
420 
421 /*
422  * Configuration parser.
423  */
424 
425 /* Parser character classes. */
426 #define	CC_WHITESPACE		-1
427 #define	CC_NONWHITESPACE	-2
428 
429 /* Parse errors. */
430 #define	PE_EOF			-1
431 #define	PE_EOL			-2
432 
433 static __inline int
434 parse_peek(char **conf)
435 {
436 
437 	return (**conf);
438 }
439 
440 static __inline void
441 parse_poke(char **conf, int c)
442 {
443 
444 	**conf = c;
445 }
446 
447 static __inline void
448 parse_advance(char **conf)
449 {
450 
451 	(*conf)++;
452 }
453 
454 static int
455 parse_skipto(char **conf, int mc)
456 {
457 	int c, match;
458 
459 	while (1) {
460 		c = parse_peek(conf);
461 		if (c == 0)
462 			return (PE_EOF);
463 		switch (mc) {
464 		case CC_WHITESPACE:
465 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
466 			break;
467 		case CC_NONWHITESPACE:
468 			if (c == '\n')
469 				return (PE_EOL);
470 			match = (c != ' ' && c != '\t') ? 1 : 0;
471 			break;
472 		default:
473 			match = (c == mc) ? 1 : 0;
474 			break;
475 		}
476 		if (match)
477 			break;
478 		parse_advance(conf);
479 	}
480 	return (0);
481 }
482 
483 static int
484 parse_token(char **conf, char **tok)
485 {
486 	char *p;
487 	size_t len;
488 	int error;
489 
490 	*tok = NULL;
491 	error = parse_skipto(conf, CC_NONWHITESPACE);
492 	if (error)
493 		return (error);
494 	p = *conf;
495 	error = parse_skipto(conf, CC_WHITESPACE);
496 	len = *conf - p;
497 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
498 	bcopy(p, *tok, len);
499 	return (0);
500 }
501 
502 static void
503 parse_dir_ask_printenv(const char *var)
504 {
505 	char *val;
506 
507 	val = kern_getenv(var);
508 	if (val != NULL) {
509 		printf("  %s=%s\n", var, val);
510 		freeenv(val);
511 	}
512 }
513 
514 static int
515 parse_dir_ask(char **conf)
516 {
517 	char name[80];
518 	char *mnt;
519 	int error;
520 
521 	vfs_mountroot_wait();
522 
523 	printf("\nLoader variables:\n");
524 	parse_dir_ask_printenv("vfs.root.mountfrom");
525 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
526 
527 	printf("\nManual root filesystem specification:\n");
528 	printf("  <fstype>:<device> [options]\n");
529 	printf("      Mount <device> using filesystem <fstype>\n");
530 	printf("      and with the specified (optional) option list.\n");
531 	printf("\n");
532 	printf("    eg. ufs:/dev/da0s1a\n");
533 	printf("        zfs:zroot/ROOT/default\n");
534 	printf("        cd9660:/dev/cd0 ro\n");
535 	printf("          (which is equivalent to: ");
536 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
537 	printf("\n");
538 	printf("  ?               List valid disk boot devices\n");
539 	printf("  .               Yield 1 second (for background tasks)\n");
540 	printf("  <empty line>    Abort manual input\n");
541 
542 	do {
543 		error = EINVAL;
544 		printf("\nmountroot> ");
545 		cngets(name, sizeof(name), GETS_ECHO);
546 		if (name[0] == '\0')
547 			break;
548 		if (name[0] == '?' && name[1] == '\0') {
549 			printf("\nList of GEOM managed disk devices:\n  ");
550 			g_dev_print();
551 			continue;
552 		}
553 		if (name[0] == '.' && name[1] == '\0') {
554 			pause("rmask", hz);
555 			continue;
556 		}
557 		mnt = name;
558 		error = parse_mount(&mnt);
559 		if (error == -1)
560 			printf("Invalid file system specification.\n");
561 	} while (error != 0);
562 
563 	return (error);
564 }
565 
566 static int
567 parse_dir_md(char **conf)
568 {
569 	struct stat sb;
570 	struct thread *td;
571 	struct md_ioctl *mdio;
572 	char *path, *tok;
573 	int error, fd, len;
574 
575 	td = curthread;
576 
577 	error = parse_token(conf, &tok);
578 	if (error)
579 		return (error);
580 
581 	len = strlen(tok);
582 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
583 	path = (void *)(mdio + 1);
584 	bcopy(tok, path, len);
585 	free(tok, M_TEMP);
586 
587 	/* Get file status. */
588 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
589 	if (error)
590 		goto out;
591 
592 	/* Open /dev/mdctl so that we can attach/detach. */
593 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
594 	    O_RDWR, 0);
595 	if (error)
596 		goto out;
597 
598 	fd = td->td_retval[0];
599 	mdio->md_version = MDIOVERSION;
600 	mdio->md_type = MD_VNODE;
601 
602 	if (root_mount_mddev != -1) {
603 		mdio->md_unit = root_mount_mddev;
604 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
605 		/* Ignore errors. We don't care. */
606 		root_mount_mddev = -1;
607 	}
608 
609 	mdio->md_file = (void *)(mdio + 1);
610 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
611 	mdio->md_mediasize = sb.st_size;
612 	mdio->md_unit = 0;
613 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
614 	if (error)
615 		goto out;
616 
617 	if (mdio->md_unit > 9) {
618 		printf("rootmount: too many md units\n");
619 		mdio->md_file = NULL;
620 		mdio->md_options = 0;
621 		mdio->md_mediasize = 0;
622 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
623 		/* Ignore errors. We don't care. */
624 		error = ERANGE;
625 		goto out;
626 	}
627 
628 	root_mount_mddev = mdio->md_unit;
629 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
630 
631 	error = kern_close(td, fd);
632 
633  out:
634 	free(mdio, M_TEMP);
635 	return (error);
636 }
637 
638 static int
639 parse_dir_onfail(char **conf)
640 {
641 	char *action;
642 	int error;
643 
644 	error = parse_token(conf, &action);
645 	if (error)
646 		return (error);
647 
648 	if (!strcmp(action, "continue"))
649 		root_mount_onfail = A_CONTINUE;
650 	else if (!strcmp(action, "panic"))
651 		root_mount_onfail = A_PANIC;
652 	else if (!strcmp(action, "reboot"))
653 		root_mount_onfail = A_REBOOT;
654 	else if (!strcmp(action, "retry"))
655 		root_mount_onfail = A_RETRY;
656 	else {
657 		printf("rootmount: %s: unknown action\n", action);
658 		error = EINVAL;
659 	}
660 
661 	free(action, M_TEMP);
662 	return (0);
663 }
664 
665 static int
666 parse_dir_timeout(char **conf)
667 {
668 	char *tok, *endtok;
669 	long secs;
670 	int error;
671 
672 	error = parse_token(conf, &tok);
673 	if (error)
674 		return (error);
675 
676 	secs = strtol(tok, &endtok, 0);
677 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
678 	if (!error)
679 		root_mount_timeout = secs;
680 	free(tok, M_TEMP);
681 	return (error);
682 }
683 
684 static int
685 parse_directive(char **conf)
686 {
687 	char *dir;
688 	int error;
689 
690 	error = parse_token(conf, &dir);
691 	if (error)
692 		return (error);
693 
694 	if (strcmp(dir, ".ask") == 0)
695 		error = parse_dir_ask(conf);
696 	else if (strcmp(dir, ".md") == 0)
697 		error = parse_dir_md(conf);
698 	else if (strcmp(dir, ".onfail") == 0)
699 		error = parse_dir_onfail(conf);
700 	else if (strcmp(dir, ".timeout") == 0)
701 		error = parse_dir_timeout(conf);
702 	else {
703 		printf("mountroot: invalid directive `%s'\n", dir);
704 		/* Ignore the rest of the line. */
705 		(void)parse_skipto(conf, '\n');
706 		error = EINVAL;
707 	}
708 	free(dir, M_TEMP);
709 	return (error);
710 }
711 
712 static int
713 parse_mount_dev_present(const char *dev)
714 {
715 	struct nameidata nd;
716 	int error;
717 
718 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
719 	error = namei(&nd);
720 	if (!error)
721 		vput(nd.ni_vp);
722 	NDFREE(&nd, NDF_ONLY_PNBUF);
723 	return (error != 0) ? 0 : 1;
724 }
725 
726 #define	ERRMSGL	255
727 static int
728 parse_mount(char **conf)
729 {
730 	char *errmsg;
731 	struct mntarg *ma;
732 	char *dev, *fs, *opts, *tok;
733 	int delay, error, timeout;
734 
735 	error = parse_token(conf, &tok);
736 	if (error)
737 		return (error);
738 	fs = tok;
739 	error = parse_skipto(&tok, ':');
740 	if (error) {
741 		free(fs, M_TEMP);
742 		return (error);
743 	}
744 	parse_poke(&tok, '\0');
745 	parse_advance(&tok);
746 	dev = tok;
747 
748 	if (root_mount_mddev != -1) {
749 		/* Handle substitution for the md unit number. */
750 		tok = strstr(dev, "md#");
751 		if (tok != NULL)
752 			tok[2] = '0' + root_mount_mddev;
753 	}
754 
755 	/* Parse options. */
756 	error = parse_token(conf, &tok);
757 	opts = (error == 0) ? tok : NULL;
758 
759 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
760 	    (opts != NULL) ? opts : "");
761 
762 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
763 
764 	if (vfs_byname(fs) == NULL) {
765 		strlcpy(errmsg, "unknown file system", ERRMSGL);
766 		error = ENOENT;
767 		goto out;
768 	}
769 
770 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
771 	if (error != 0)
772 		goto out;
773 
774 	delay = hz / 10;
775 	timeout = root_mount_timeout * hz;
776 
777 	for (;;) {
778 		ma = NULL;
779 		ma = mount_arg(ma, "fstype", fs, -1);
780 		ma = mount_arg(ma, "fspath", "/", -1);
781 		ma = mount_arg(ma, "from", dev, -1);
782 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
783 		ma = mount_arg(ma, "ro", NULL, 0);
784 		ma = parse_mountroot_options(ma, opts);
785 
786 		error = kernel_mount(ma, MNT_ROOTFS);
787 		if (error == 0 || timeout <= 0)
788 			break;
789 
790 		if (root_mount_timeout * hz == timeout ||
791 		    (bootverbose && timeout % hz == 0)) {
792 			printf("Mounting from %s:%s failed with error %d; "
793 			    "retrying for %d more second%s\n", fs, dev, error,
794 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
795 		}
796 		pause("rmretry", delay);
797 		timeout -= delay;
798 	}
799  out:
800 	if (error) {
801 		printf("Mounting from %s:%s failed with error %d",
802 		    fs, dev, error);
803 		if (errmsg[0] != '\0')
804 			printf(": %s", errmsg);
805 		printf(".\n");
806 	}
807 	free(fs, M_TEMP);
808 	free(errmsg, M_TEMP);
809 	if (opts != NULL)
810 		free(opts, M_TEMP);
811 	/* kernel_mount can return -1 on error. */
812 	return ((error < 0) ? EDOOFUS : error);
813 }
814 #undef ERRMSGL
815 
816 static int
817 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
818 {
819 	struct mount *mp;
820 	char *conf;
821 	int error;
822 
823 	root_mount_mddev = -1;
824 
825 retry:
826 	conf = sbuf_data(sb);
827 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
828 	error = (mp == NULL) ? 0 : EDOOFUS;
829 	root_mount_onfail = A_CONTINUE;
830 	while (mp == NULL) {
831 		error = parse_skipto(&conf, CC_NONWHITESPACE);
832 		if (error == PE_EOL) {
833 			parse_advance(&conf);
834 			continue;
835 		}
836 		if (error < 0)
837 			break;
838 		switch (parse_peek(&conf)) {
839 		case '#':
840 			error = parse_skipto(&conf, '\n');
841 			break;
842 		case '.':
843 			error = parse_directive(&conf);
844 			break;
845 		default:
846 			error = parse_mount(&conf);
847 			if (error == -1) {
848 				printf("mountroot: invalid file system "
849 				    "specification.\n");
850 				error = 0;
851 			}
852 			break;
853 		}
854 		if (error < 0)
855 			break;
856 		/* Ignore any trailing garbage on the line. */
857 		if (parse_peek(&conf) != '\n') {
858 			printf("mountroot: advancing to next directive...\n");
859 			(void)parse_skipto(&conf, '\n');
860 		}
861 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
862 	}
863 	if (mp != NULL)
864 		return (0);
865 
866 	/*
867 	 * We failed to mount (a new) root.
868 	 */
869 	switch (root_mount_onfail) {
870 	case A_CONTINUE:
871 		break;
872 	case A_PANIC:
873 		panic("mountroot: unable to (re-)mount root.");
874 		/* NOTREACHED */
875 	case A_RETRY:
876 		goto retry;
877 	case A_REBOOT:
878 		kern_reboot(RB_NOSYNC);
879 		/* NOTREACHED */
880 	}
881 
882 	return (error);
883 }
884 
885 static void
886 vfs_mountroot_conf0(struct sbuf *sb)
887 {
888 	char *s, *tok, *mnt, *opt;
889 	int error;
890 
891 	sbuf_printf(sb, ".onfail panic\n");
892 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
893 	if (boothowto & RB_ASKNAME)
894 		sbuf_printf(sb, ".ask\n");
895 #ifdef ROOTDEVNAME
896 	if (boothowto & RB_DFLTROOT)
897 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
898 #endif
899 	if (boothowto & RB_CDROM) {
900 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
901 		sbuf_printf(sb, ".timeout 0\n");
902 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
903 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
904 	}
905 	s = kern_getenv("vfs.root.mountfrom");
906 	if (s != NULL) {
907 		opt = kern_getenv("vfs.root.mountfrom.options");
908 		tok = s;
909 		error = parse_token(&tok, &mnt);
910 		while (!error) {
911 			sbuf_printf(sb, "%s %s\n", mnt,
912 			    (opt != NULL) ? opt : "");
913 			free(mnt, M_TEMP);
914 			error = parse_token(&tok, &mnt);
915 		}
916 		if (opt != NULL)
917 			freeenv(opt);
918 		freeenv(s);
919 	}
920 	if (rootdevnames[0] != NULL)
921 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
922 	if (rootdevnames[1] != NULL)
923 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
924 #ifdef ROOTDEVNAME
925 	if (!(boothowto & RB_DFLTROOT))
926 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
927 #endif
928 	if (!(boothowto & RB_ASKNAME))
929 		sbuf_printf(sb, ".ask\n");
930 }
931 
932 static int
933 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
934 {
935 	static char buf[128];
936 	struct nameidata nd;
937 	off_t ofs;
938 	ssize_t resid;
939 	int error, flags, len;
940 
941 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
942 	flags = FREAD;
943 	error = vn_open(&nd, &flags, 0, NULL);
944 	if (error)
945 		return (error);
946 
947 	NDFREE(&nd, NDF_ONLY_PNBUF);
948 	ofs = 0;
949 	len = sizeof(buf) - 1;
950 	while (1) {
951 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
952 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
953 		    NOCRED, &resid, td);
954 		if (error)
955 			break;
956 		if (resid == len)
957 			break;
958 		buf[len - resid] = 0;
959 		sbuf_printf(sb, "%s", buf);
960 		ofs += len - resid;
961 	}
962 
963 	VOP_UNLOCK(nd.ni_vp);
964 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
965 	return (error);
966 }
967 
968 static void
969 vfs_mountroot_wait(void)
970 {
971 	struct root_hold_token *h;
972 	struct timeval lastfail;
973 	int curfail;
974 
975 	TSENTER();
976 
977 	curfail = 0;
978 	while (1) {
979 		g_waitidle();
980 		mtx_lock(&root_holds_mtx);
981 		if (TAILQ_EMPTY(&root_holds)) {
982 			mtx_unlock(&root_holds_mtx);
983 			break;
984 		}
985 		if (ppsratecheck(&lastfail, &curfail, 1)) {
986 			printf("Root mount waiting for:");
987 			TAILQ_FOREACH(h, &root_holds, list)
988 				printf(" %s", h->who);
989 			printf("\n");
990 		}
991 		TSWAIT("root mount");
992 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
993 		    hz);
994 		TSUNWAIT("root mount");
995 	}
996 
997 	TSEXIT();
998 }
999 
1000 static int
1001 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1002 {
1003 	int delay, timeout;
1004 
1005 	/*
1006 	 * In case of ZFS and NFS we don't have a way to wait for
1007 	 * specific device.  Also do the wait if the user forced that
1008 	 * behaviour by setting vfs.root_mount_always_wait=1.
1009 	 */
1010 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1011 	    dev[0] == '\0' || root_mount_always_wait != 0) {
1012 		vfs_mountroot_wait();
1013 		return (0);
1014 	}
1015 
1016 	/*
1017 	 * Otherwise, no point in waiting if the device is already there.
1018 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1019 	 * eg for geom_part(4) to finish tasting.
1020 	 */
1021 	g_waitidle();
1022 	if (parse_mount_dev_present(dev))
1023 		return (0);
1024 
1025 	/*
1026 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1027 	 * to behave exactly as it used to work before.
1028 	 */
1029 	vfs_mountroot_wait();
1030 	printf("mountroot: waiting for device %s...\n", dev);
1031 	delay = hz / 10;
1032 	timeout = root_mount_timeout * hz;
1033 	do {
1034 		pause("rmdev", delay);
1035 		timeout -= delay;
1036 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1037 
1038 	if (timeout <= 0)
1039 		return (ENODEV);
1040 
1041 	return (0);
1042 }
1043 
1044 void
1045 vfs_mountroot(void)
1046 {
1047 	struct mount *mp;
1048 	struct sbuf *sb;
1049 	struct thread *td;
1050 	time_t timebase;
1051 	int error;
1052 
1053 	mtx_assert(&Giant, MA_NOTOWNED);
1054 
1055 	TSENTER();
1056 
1057 	td = curthread;
1058 
1059 	sb = sbuf_new_auto();
1060 	vfs_mountroot_conf0(sb);
1061 	sbuf_finish(sb);
1062 
1063 	error = vfs_mountroot_devfs(td, &mp);
1064 	while (!error) {
1065 		error = vfs_mountroot_parse(sb, mp);
1066 		if (!error) {
1067 			vfs_mountroot_shuffle(td, mp);
1068 			sbuf_clear(sb);
1069 			error = vfs_mountroot_readconf(td, sb);
1070 			sbuf_finish(sb);
1071 		}
1072 	}
1073 
1074 	sbuf_delete(sb);
1075 
1076 	/*
1077 	 * Iterate over all currently mounted file systems and use
1078 	 * the time stamp found to check and/or initialize the RTC.
1079 	 * Call inittodr() only once and pass it the largest of the
1080 	 * timestamps we encounter.
1081 	 */
1082 	timebase = 0;
1083 	mtx_lock(&mountlist_mtx);
1084 	mp = TAILQ_FIRST(&mountlist);
1085 	while (mp != NULL) {
1086 		if (mp->mnt_time > timebase)
1087 			timebase = mp->mnt_time;
1088 		mp = TAILQ_NEXT(mp, mnt_list);
1089 	}
1090 	mtx_unlock(&mountlist_mtx);
1091 	inittodr(timebase);
1092 
1093 	/* Keep prison0's root in sync with the global rootvnode. */
1094 	mtx_lock(&prison0.pr_mtx);
1095 	prison0.pr_root = rootvnode;
1096 	vref(prison0.pr_root);
1097 	mtx_unlock(&prison0.pr_mtx);
1098 
1099 	mtx_lock(&root_holds_mtx);
1100 	atomic_store_rel_int(&root_mount_complete, 1);
1101 	wakeup(&root_mount_complete);
1102 	mtx_unlock(&root_holds_mtx);
1103 
1104 	EVENTHANDLER_INVOKE(mountroot);
1105 
1106 	TSEXIT();
1107 }
1108 
1109 static struct mntarg *
1110 parse_mountroot_options(struct mntarg *ma, const char *options)
1111 {
1112 	char *p;
1113 	char *name, *name_arg;
1114 	char *val, *val_arg;
1115 	char *opts;
1116 
1117 	if (options == NULL || options[0] == '\0')
1118 		return (ma);
1119 
1120 	p = opts = strdup(options, M_MOUNT);
1121 	if (opts == NULL) {
1122 		return (ma);
1123 	}
1124 
1125 	while((name = strsep(&p, ",")) != NULL) {
1126 		if (name[0] == '\0')
1127 			break;
1128 
1129 		val = strchr(name, '=');
1130 		if (val != NULL) {
1131 			*val = '\0';
1132 			++val;
1133 		}
1134 		if( strcmp(name, "rw") == 0 ||
1135 		    strcmp(name, "noro") == 0) {
1136 			/*
1137 			 * The first time we mount the root file system,
1138 			 * we need to mount 'ro', so We need to ignore
1139 			 * 'rw' and 'noro' mount options.
1140 			 */
1141 			continue;
1142 		}
1143 		name_arg = strdup(name, M_MOUNT);
1144 		val_arg = NULL;
1145 		if (val != NULL)
1146 			val_arg = strdup(val, M_MOUNT);
1147 
1148 		ma = mount_arg(ma, name_arg, val_arg,
1149 		    (val_arg != NULL ? -1 : 0));
1150 	}
1151 	free(opts, M_MOUNT);
1152 	return (ma);
1153 }
1154