xref: /freebsd/sys/kern/vfs_mountroot.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
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/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/mdioctl.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/filedesc.h>
59 #include <sys/reboot.h>
60 #include <sys/sbuf.h>
61 #include <sys/stat.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysproto.h>
64 #include <sys/sx.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/systm.h>
68 #include <sys/vnode.h>
69 
70 #include <geom/geom.h>
71 
72 /*
73  * The root filesystem is detailed in the kernel environment variable
74  * vfs.root.mountfrom, which is expected to be in the general format
75  *
76  * <vfsname>:[<path>][	<vfsname>:[<path>] ...]
77  * vfsname   := the name of a VFS known to the kernel and capable
78  *              of being mounted as root
79  * path      := disk device name or other data used by the filesystem
80  *              to locate its physical store
81  *
82  * If the environment variable vfs.root.mountfrom is a space separated list,
83  * each list element is tried in turn and the root filesystem will be mounted
84  * from the first one that succeeds.
85  *
86  * The environment variable vfs.root.mountfrom.options is a comma delimited
87  * set of string mount options.  These mount options must be parseable
88  * by nmount() in the kernel.
89  */
90 
91 static int parse_mount(char **);
92 static struct mntarg *parse_mountroot_options(struct mntarg *, const char *);
93 static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS);
94 static void vfs_mountroot_wait(void);
95 static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev);
96 
97 /*
98  * The vnode of the system's root (/ in the filesystem, without chroot
99  * active.)
100  */
101 struct vnode *rootvnode;
102 
103 /*
104  * Mount of the system's /dev.
105  */
106 struct mount *rootdevmp;
107 
108 char *rootdevnames[2] = {NULL, NULL};
109 
110 struct mtx root_holds_mtx;
111 MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF);
112 
113 struct root_hold_token {
114 	const char			*who;
115 	LIST_ENTRY(root_hold_token)	list;
116 };
117 
118 static LIST_HEAD(, root_hold_token)	root_holds =
119     LIST_HEAD_INITIALIZER(root_holds);
120 
121 enum action {
122 	A_CONTINUE,
123 	A_PANIC,
124 	A_REBOOT,
125 	A_RETRY
126 };
127 
128 static enum action root_mount_onfail = A_CONTINUE;
129 
130 static int root_mount_mddev;
131 static int root_mount_complete;
132 
133 /* By default wait up to 3 seconds for devices to appear. */
134 static int root_mount_timeout = 3;
135 TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout);
136 
137 static int root_mount_always_wait = 0;
138 SYSCTL_INT(_vfs, OID_AUTO, root_mount_always_wait, CTLFLAG_RDTUN,
139     &root_mount_always_wait, 0,
140     "Wait for root mount holds even if the root device already exists");
141 
142 SYSCTL_PROC(_vfs, OID_AUTO, root_mount_hold,
143     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
144     NULL, 0, sysctl_vfs_root_mount_hold, "A",
145     "List of root mount hold tokens");
146 
147 static int
148 sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS)
149 {
150 	struct sbuf sb;
151 	struct root_hold_token *h;
152 	int error;
153 
154 	sbuf_new(&sb, NULL, 256, SBUF_AUTOEXTEND | SBUF_INCLUDENUL);
155 
156 	mtx_lock(&root_holds_mtx);
157 	LIST_FOREACH(h, &root_holds, list) {
158 		if (h != LIST_FIRST(&root_holds))
159 			sbuf_putc(&sb, ' ');
160 		sbuf_printf(&sb, "%s", h->who);
161 	}
162 	mtx_unlock(&root_holds_mtx);
163 
164 	error = sbuf_finish(&sb);
165 	if (error == 0)
166 		error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
167 	sbuf_delete(&sb);
168 	return (error);
169 }
170 
171 struct root_hold_token *
172 root_mount_hold(const char *identifier)
173 {
174 	struct root_hold_token *h;
175 
176 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
177 	h->who = identifier;
178 	mtx_lock(&root_holds_mtx);
179 	TSHOLD("root mount");
180 	LIST_INSERT_HEAD(&root_holds, h, list);
181 	mtx_unlock(&root_holds_mtx);
182 	return (h);
183 }
184 
185 void
186 root_mount_rel(struct root_hold_token *h)
187 {
188 
189 	if (h == NULL)
190 		return;
191 
192 	mtx_lock(&root_holds_mtx);
193 	LIST_REMOVE(h, list);
194 	TSRELEASE("root mount");
195 	wakeup(&root_holds);
196 	mtx_unlock(&root_holds_mtx);
197 	free(h, M_DEVBUF);
198 }
199 
200 int
201 root_mounted(void)
202 {
203 
204 	/* No mutex is acquired here because int stores are atomic. */
205 	return (root_mount_complete);
206 }
207 
208 static void
209 set_rootvnode(void)
210 {
211 	struct proc *p;
212 
213 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
214 		panic("set_rootvnode: Cannot find root vnode");
215 
216 	VOP_UNLOCK(rootvnode, 0);
217 
218 	p = curthread->td_proc;
219 	FILEDESC_XLOCK(p->p_fd);
220 
221 	if (p->p_fd->fd_cdir != NULL)
222 		vrele(p->p_fd->fd_cdir);
223 	p->p_fd->fd_cdir = rootvnode;
224 	VREF(rootvnode);
225 
226 	if (p->p_fd->fd_rdir != NULL)
227 		vrele(p->p_fd->fd_rdir);
228 	p->p_fd->fd_rdir = rootvnode;
229 	VREF(rootvnode);
230 
231 	FILEDESC_XUNLOCK(p->p_fd);
232 }
233 
234 static int
235 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
236 {
237 	struct vfsoptlist *opts;
238 	struct vfsconf *vfsp;
239 	struct mount *mp;
240 	int error;
241 
242 	*mpp = NULL;
243 
244 	if (rootdevmp != NULL) {
245 		/*
246 		 * Already have /dev; this happens during rerooting.
247 		 */
248 		error = vfs_busy(rootdevmp, 0);
249 		if (error != 0)
250 			return (error);
251 		*mpp = rootdevmp;
252 	} else {
253 		vfsp = vfs_byname("devfs");
254 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
255 		if (vfsp == NULL)
256 			return (ENOENT);
257 
258 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
259 
260 		error = VFS_MOUNT(mp);
261 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
262 		if (error)
263 			return (error);
264 
265 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
266 		TAILQ_INIT(opts);
267 		mp->mnt_opt = opts;
268 
269 		mtx_lock(&mountlist_mtx);
270 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
271 		mtx_unlock(&mountlist_mtx);
272 
273 		*mpp = mp;
274 		rootdevmp = mp;
275 	}
276 
277 	set_rootvnode();
278 
279 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
280 	if (error)
281 		printf("kern_symlink /dev -> / returns %d\n", error);
282 
283 	return (error);
284 }
285 
286 static void
287 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
288 {
289 	struct nameidata nd;
290 	struct mount *mporoot, *mpnroot;
291 	struct vnode *vp, *vporoot, *vpdevfs;
292 	char *fspath;
293 	int error;
294 
295 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
296 
297 	/* Shuffle the mountlist. */
298 	mtx_lock(&mountlist_mtx);
299 	mporoot = TAILQ_FIRST(&mountlist);
300 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
301 	if (mporoot != mpdevfs) {
302 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
303 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
304 	}
305 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
306 	mtx_unlock(&mountlist_mtx);
307 
308 	cache_purgevfs(mporoot, true);
309 	if (mporoot != mpdevfs)
310 		cache_purgevfs(mpdevfs, true);
311 
312 	if (VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot))
313 		panic("vfs_mountroot_shuffle: Cannot find root vnode");
314 
315 	VI_LOCK(vporoot);
316 	vporoot->v_iflag &= ~VI_MOUNT;
317 	VI_UNLOCK(vporoot);
318 	vporoot->v_mountedhere = NULL;
319 	mporoot->mnt_flag &= ~MNT_ROOTFS;
320 	mporoot->mnt_vnodecovered = NULL;
321 	vput(vporoot);
322 
323 	/* Set up the new rootvnode, and purge the cache */
324 	mpnroot->mnt_vnodecovered = NULL;
325 	set_rootvnode();
326 	cache_purgevfs(rootvnode->v_mount, true);
327 
328 	if (mporoot != mpdevfs) {
329 		/* Remount old root under /.mount or /mnt */
330 		fspath = "/.mount";
331 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
332 		    fspath, td);
333 		error = namei(&nd);
334 		if (error) {
335 			NDFREE(&nd, NDF_ONLY_PNBUF);
336 			fspath = "/mnt";
337 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
338 			    fspath, td);
339 			error = namei(&nd);
340 		}
341 		if (!error) {
342 			vp = nd.ni_vp;
343 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
344 			if (!error)
345 				error = vinvalbuf(vp, V_SAVE, 0, 0);
346 			if (!error) {
347 				cache_purge(vp);
348 				mporoot->mnt_vnodecovered = vp;
349 				vp->v_mountedhere = mporoot;
350 				strlcpy(mporoot->mnt_stat.f_mntonname,
351 				    fspath, MNAMELEN);
352 				VOP_UNLOCK(vp, 0);
353 			} else
354 				vput(vp);
355 		}
356 		NDFREE(&nd, NDF_ONLY_PNBUF);
357 
358 		if (error)
359 			printf("mountroot: unable to remount previous root "
360 			    "under /.mount or /mnt (error %d)\n", error);
361 	}
362 
363 	/* Remount devfs under /dev */
364 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
365 	error = namei(&nd);
366 	if (!error) {
367 		vp = nd.ni_vp;
368 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
369 		if (!error)
370 			error = vinvalbuf(vp, V_SAVE, 0, 0);
371 		if (!error) {
372 			vpdevfs = mpdevfs->mnt_vnodecovered;
373 			if (vpdevfs != NULL) {
374 				cache_purge(vpdevfs);
375 				vpdevfs->v_mountedhere = NULL;
376 				vrele(vpdevfs);
377 			}
378 			mpdevfs->mnt_vnodecovered = vp;
379 			vp->v_mountedhere = mpdevfs;
380 			VOP_UNLOCK(vp, 0);
381 		} else
382 			vput(vp);
383 	}
384 	if (error)
385 		printf("mountroot: unable to remount devfs under /dev "
386 		    "(error %d)\n", error);
387 	NDFREE(&nd, NDF_ONLY_PNBUF);
388 
389 	if (mporoot == mpdevfs) {
390 		vfs_unbusy(mpdevfs);
391 		/* Unlink the no longer needed /dev/dev -> / symlink */
392 		error = kern_unlinkat(td, AT_FDCWD, "/dev/dev",
393 		    UIO_SYSSPACE, 0, 0);
394 		if (error)
395 			printf("mountroot: unable to unlink /dev/dev "
396 			    "(error %d)\n", error);
397 	}
398 }
399 
400 /*
401  * Configuration parser.
402  */
403 
404 /* Parser character classes. */
405 #define	CC_WHITESPACE		-1
406 #define	CC_NONWHITESPACE	-2
407 
408 /* Parse errors. */
409 #define	PE_EOF			-1
410 #define	PE_EOL			-2
411 
412 static __inline int
413 parse_peek(char **conf)
414 {
415 
416 	return (**conf);
417 }
418 
419 static __inline void
420 parse_poke(char **conf, int c)
421 {
422 
423 	**conf = c;
424 }
425 
426 static __inline void
427 parse_advance(char **conf)
428 {
429 
430 	(*conf)++;
431 }
432 
433 static int
434 parse_skipto(char **conf, int mc)
435 {
436 	int c, match;
437 
438 	while (1) {
439 		c = parse_peek(conf);
440 		if (c == 0)
441 			return (PE_EOF);
442 		switch (mc) {
443 		case CC_WHITESPACE:
444 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
445 			break;
446 		case CC_NONWHITESPACE:
447 			if (c == '\n')
448 				return (PE_EOL);
449 			match = (c != ' ' && c != '\t') ? 1 : 0;
450 			break;
451 		default:
452 			match = (c == mc) ? 1 : 0;
453 			break;
454 		}
455 		if (match)
456 			break;
457 		parse_advance(conf);
458 	}
459 	return (0);
460 }
461 
462 static int
463 parse_token(char **conf, char **tok)
464 {
465 	char *p;
466 	size_t len;
467 	int error;
468 
469 	*tok = NULL;
470 	error = parse_skipto(conf, CC_NONWHITESPACE);
471 	if (error)
472 		return (error);
473 	p = *conf;
474 	error = parse_skipto(conf, CC_WHITESPACE);
475 	len = *conf - p;
476 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
477 	bcopy(p, *tok, len);
478 	return (0);
479 }
480 
481 static void
482 parse_dir_ask_printenv(const char *var)
483 {
484 	char *val;
485 
486 	val = kern_getenv(var);
487 	if (val != NULL) {
488 		printf("  %s=%s\n", var, val);
489 		freeenv(val);
490 	}
491 }
492 
493 static int
494 parse_dir_ask(char **conf)
495 {
496 	char name[80];
497 	char *mnt;
498 	int error;
499 
500 	vfs_mountroot_wait();
501 
502 	printf("\nLoader variables:\n");
503 	parse_dir_ask_printenv("vfs.root.mountfrom");
504 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
505 
506 	printf("\nManual root filesystem specification:\n");
507 	printf("  <fstype>:<device> [options]\n");
508 	printf("      Mount <device> using filesystem <fstype>\n");
509 	printf("      and with the specified (optional) option list.\n");
510 	printf("\n");
511 	printf("    eg. ufs:/dev/da0s1a\n");
512 	printf("        zfs:zroot/ROOT/default\n");
513 	printf("        cd9660:/dev/cd0 ro\n");
514 	printf("          (which is equivalent to: ");
515 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
516 	printf("\n");
517 	printf("  ?               List valid disk boot devices\n");
518 	printf("  .               Yield 1 second (for background tasks)\n");
519 	printf("  <empty line>    Abort manual input\n");
520 
521 	do {
522 		error = EINVAL;
523 		printf("\nmountroot> ");
524 		cngets(name, sizeof(name), GETS_ECHO);
525 		if (name[0] == '\0')
526 			break;
527 		if (name[0] == '?' && name[1] == '\0') {
528 			printf("\nList of GEOM managed disk devices:\n  ");
529 			g_dev_print();
530 			continue;
531 		}
532 		if (name[0] == '.' && name[1] == '\0') {
533 			pause("rmask", hz);
534 			continue;
535 		}
536 		mnt = name;
537 		error = parse_mount(&mnt);
538 		if (error == -1)
539 			printf("Invalid file system specification.\n");
540 	} while (error != 0);
541 
542 	return (error);
543 }
544 
545 static int
546 parse_dir_md(char **conf)
547 {
548 	struct stat sb;
549 	struct thread *td;
550 	struct md_ioctl *mdio;
551 	char *path, *tok;
552 	int error, fd, len;
553 
554 	td = curthread;
555 
556 	error = parse_token(conf, &tok);
557 	if (error)
558 		return (error);
559 
560 	len = strlen(tok);
561 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
562 	path = (void *)(mdio + 1);
563 	bcopy(tok, path, len);
564 	free(tok, M_TEMP);
565 
566 	/* Get file status. */
567 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
568 	if (error)
569 		goto out;
570 
571 	/* Open /dev/mdctl so that we can attach/detach. */
572 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
573 	    O_RDWR, 0);
574 	if (error)
575 		goto out;
576 
577 	fd = td->td_retval[0];
578 	mdio->md_version = MDIOVERSION;
579 	mdio->md_type = MD_VNODE;
580 
581 	if (root_mount_mddev != -1) {
582 		mdio->md_unit = root_mount_mddev;
583 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
584 		/* Ignore errors. We don't care. */
585 		root_mount_mddev = -1;
586 	}
587 
588 	mdio->md_file = (void *)(mdio + 1);
589 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
590 	mdio->md_mediasize = sb.st_size;
591 	mdio->md_unit = 0;
592 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
593 	if (error)
594 		goto out;
595 
596 	if (mdio->md_unit > 9) {
597 		printf("rootmount: too many md units\n");
598 		mdio->md_file = NULL;
599 		mdio->md_options = 0;
600 		mdio->md_mediasize = 0;
601 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
602 		/* Ignore errors. We don't care. */
603 		error = ERANGE;
604 		goto out;
605 	}
606 
607 	root_mount_mddev = mdio->md_unit;
608 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
609 
610 	error = kern_close(td, fd);
611 
612  out:
613 	free(mdio, M_TEMP);
614 	return (error);
615 }
616 
617 static int
618 parse_dir_onfail(char **conf)
619 {
620 	char *action;
621 	int error;
622 
623 	error = parse_token(conf, &action);
624 	if (error)
625 		return (error);
626 
627 	if (!strcmp(action, "continue"))
628 		root_mount_onfail = A_CONTINUE;
629 	else if (!strcmp(action, "panic"))
630 		root_mount_onfail = A_PANIC;
631 	else if (!strcmp(action, "reboot"))
632 		root_mount_onfail = A_REBOOT;
633 	else if (!strcmp(action, "retry"))
634 		root_mount_onfail = A_RETRY;
635 	else {
636 		printf("rootmount: %s: unknown action\n", action);
637 		error = EINVAL;
638 	}
639 
640 	free(action, M_TEMP);
641 	return (0);
642 }
643 
644 static int
645 parse_dir_timeout(char **conf)
646 {
647 	char *tok, *endtok;
648 	long secs;
649 	int error;
650 
651 	error = parse_token(conf, &tok);
652 	if (error)
653 		return (error);
654 
655 	secs = strtol(tok, &endtok, 0);
656 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
657 	if (!error)
658 		root_mount_timeout = secs;
659 	free(tok, M_TEMP);
660 	return (error);
661 }
662 
663 static int
664 parse_directive(char **conf)
665 {
666 	char *dir;
667 	int error;
668 
669 	error = parse_token(conf, &dir);
670 	if (error)
671 		return (error);
672 
673 	if (strcmp(dir, ".ask") == 0)
674 		error = parse_dir_ask(conf);
675 	else if (strcmp(dir, ".md") == 0)
676 		error = parse_dir_md(conf);
677 	else if (strcmp(dir, ".onfail") == 0)
678 		error = parse_dir_onfail(conf);
679 	else if (strcmp(dir, ".timeout") == 0)
680 		error = parse_dir_timeout(conf);
681 	else {
682 		printf("mountroot: invalid directive `%s'\n", dir);
683 		/* Ignore the rest of the line. */
684 		(void)parse_skipto(conf, '\n');
685 		error = EINVAL;
686 	}
687 	free(dir, M_TEMP);
688 	return (error);
689 }
690 
691 static int
692 parse_mount_dev_present(const char *dev)
693 {
694 	struct nameidata nd;
695 	int error;
696 
697 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
698 	error = namei(&nd);
699 	if (!error)
700 		vput(nd.ni_vp);
701 	NDFREE(&nd, NDF_ONLY_PNBUF);
702 	return (error != 0) ? 0 : 1;
703 }
704 
705 #define	ERRMSGL	255
706 static int
707 parse_mount(char **conf)
708 {
709 	char *errmsg;
710 	struct mntarg *ma;
711 	char *dev, *fs, *opts, *tok;
712 	int delay, error, timeout;
713 
714 	error = parse_token(conf, &tok);
715 	if (error)
716 		return (error);
717 	fs = tok;
718 	error = parse_skipto(&tok, ':');
719 	if (error) {
720 		free(fs, M_TEMP);
721 		return (error);
722 	}
723 	parse_poke(&tok, '\0');
724 	parse_advance(&tok);
725 	dev = tok;
726 
727 	if (root_mount_mddev != -1) {
728 		/* Handle substitution for the md unit number. */
729 		tok = strstr(dev, "md#");
730 		if (tok != NULL)
731 			tok[2] = '0' + root_mount_mddev;
732 	}
733 
734 	/* Parse options. */
735 	error = parse_token(conf, &tok);
736 	opts = (error == 0) ? tok : NULL;
737 
738 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
739 	    (opts != NULL) ? opts : "");
740 
741 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
742 
743 	if (vfs_byname(fs) == NULL) {
744 		strlcpy(errmsg, "unknown file system", ERRMSGL);
745 		error = ENOENT;
746 		goto out;
747 	}
748 
749 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
750 	if (error != 0)
751 		goto out;
752 
753 	delay = hz / 10;
754 	timeout = root_mount_timeout * hz;
755 
756 	for (;;) {
757 		ma = NULL;
758 		ma = mount_arg(ma, "fstype", fs, -1);
759 		ma = mount_arg(ma, "fspath", "/", -1);
760 		ma = mount_arg(ma, "from", dev, -1);
761 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
762 		ma = mount_arg(ma, "ro", NULL, 0);
763 		ma = parse_mountroot_options(ma, opts);
764 
765 		error = kernel_mount(ma, MNT_ROOTFS);
766 		if (error == 0 || timeout <= 0)
767 			break;
768 
769 		if (root_mount_timeout * hz == timeout ||
770 		    (bootverbose && timeout % hz == 0)) {
771 			printf("Mounting from %s:%s failed with error %d; "
772 			    "retrying for %d more second%s\n", fs, dev, error,
773 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
774 		}
775 		pause("rmretry", delay);
776 		timeout -= delay;
777 	}
778  out:
779 	if (error) {
780 		printf("Mounting from %s:%s failed with error %d",
781 		    fs, dev, error);
782 		if (errmsg[0] != '\0')
783 			printf(": %s", errmsg);
784 		printf(".\n");
785 	}
786 	free(fs, M_TEMP);
787 	free(errmsg, M_TEMP);
788 	if (opts != NULL)
789 		free(opts, M_TEMP);
790 	/* kernel_mount can return -1 on error. */
791 	return ((error < 0) ? EDOOFUS : error);
792 }
793 #undef ERRMSGL
794 
795 static int
796 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
797 {
798 	struct mount *mp;
799 	char *conf;
800 	int error;
801 
802 	root_mount_mddev = -1;
803 
804 retry:
805 	conf = sbuf_data(sb);
806 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
807 	error = (mp == NULL) ? 0 : EDOOFUS;
808 	root_mount_onfail = A_CONTINUE;
809 	while (mp == NULL) {
810 		error = parse_skipto(&conf, CC_NONWHITESPACE);
811 		if (error == PE_EOL) {
812 			parse_advance(&conf);
813 			continue;
814 		}
815 		if (error < 0)
816 			break;
817 		switch (parse_peek(&conf)) {
818 		case '#':
819 			error = parse_skipto(&conf, '\n');
820 			break;
821 		case '.':
822 			error = parse_directive(&conf);
823 			break;
824 		default:
825 			error = parse_mount(&conf);
826 			if (error == -1) {
827 				printf("mountroot: invalid file system "
828 				    "specification.\n");
829 				error = 0;
830 			}
831 			break;
832 		}
833 		if (error < 0)
834 			break;
835 		/* Ignore any trailing garbage on the line. */
836 		if (parse_peek(&conf) != '\n') {
837 			printf("mountroot: advancing to next directive...\n");
838 			(void)parse_skipto(&conf, '\n');
839 		}
840 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
841 	}
842 	if (mp != NULL)
843 		return (0);
844 
845 	/*
846 	 * We failed to mount (a new) root.
847 	 */
848 	switch (root_mount_onfail) {
849 	case A_CONTINUE:
850 		break;
851 	case A_PANIC:
852 		panic("mountroot: unable to (re-)mount root.");
853 		/* NOTREACHED */
854 	case A_RETRY:
855 		goto retry;
856 	case A_REBOOT:
857 		kern_reboot(RB_NOSYNC);
858 		/* NOTREACHED */
859 	}
860 
861 	return (error);
862 }
863 
864 static void
865 vfs_mountroot_conf0(struct sbuf *sb)
866 {
867 	char *s, *tok, *mnt, *opt;
868 	int error;
869 
870 	sbuf_printf(sb, ".onfail panic\n");
871 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
872 	if (boothowto & RB_ASKNAME)
873 		sbuf_printf(sb, ".ask\n");
874 #ifdef ROOTDEVNAME
875 	if (boothowto & RB_DFLTROOT)
876 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
877 #endif
878 	if (boothowto & RB_CDROM) {
879 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
880 		sbuf_printf(sb, ".timeout 0\n");
881 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
882 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
883 	}
884 	s = kern_getenv("vfs.root.mountfrom");
885 	if (s != NULL) {
886 		opt = kern_getenv("vfs.root.mountfrom.options");
887 		tok = s;
888 		error = parse_token(&tok, &mnt);
889 		while (!error) {
890 			sbuf_printf(sb, "%s %s\n", mnt,
891 			    (opt != NULL) ? opt : "");
892 			free(mnt, M_TEMP);
893 			error = parse_token(&tok, &mnt);
894 		}
895 		if (opt != NULL)
896 			freeenv(opt);
897 		freeenv(s);
898 	}
899 	if (rootdevnames[0] != NULL)
900 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
901 	if (rootdevnames[1] != NULL)
902 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
903 #ifdef ROOTDEVNAME
904 	if (!(boothowto & RB_DFLTROOT))
905 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
906 #endif
907 	if (!(boothowto & RB_ASKNAME))
908 		sbuf_printf(sb, ".ask\n");
909 }
910 
911 static int
912 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
913 {
914 	static char buf[128];
915 	struct nameidata nd;
916 	off_t ofs;
917 	ssize_t resid;
918 	int error, flags, len;
919 
920 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
921 	flags = FREAD;
922 	error = vn_open(&nd, &flags, 0, NULL);
923 	if (error)
924 		return (error);
925 
926 	NDFREE(&nd, NDF_ONLY_PNBUF);
927 	ofs = 0;
928 	len = sizeof(buf) - 1;
929 	while (1) {
930 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
931 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
932 		    NOCRED, &resid, td);
933 		if (error)
934 			break;
935 		if (resid == len)
936 			break;
937 		buf[len - resid] = 0;
938 		sbuf_printf(sb, "%s", buf);
939 		ofs += len - resid;
940 	}
941 
942 	VOP_UNLOCK(nd.ni_vp, 0);
943 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
944 	return (error);
945 }
946 
947 static void
948 vfs_mountroot_wait(void)
949 {
950 	struct root_hold_token *h;
951 	struct timeval lastfail;
952 	int curfail;
953 
954 	TSENTER();
955 
956 	curfail = 0;
957 	while (1) {
958 		g_waitidle();
959 		mtx_lock(&root_holds_mtx);
960 		if (LIST_EMPTY(&root_holds)) {
961 			mtx_unlock(&root_holds_mtx);
962 			break;
963 		}
964 		if (ppsratecheck(&lastfail, &curfail, 1)) {
965 			printf("Root mount waiting for:");
966 			LIST_FOREACH(h, &root_holds, list)
967 				printf(" %s", h->who);
968 			printf("\n");
969 		}
970 		TSWAIT("root mount");
971 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
972 		    hz);
973 		TSUNWAIT("root mount");
974 	}
975 
976 	TSEXIT();
977 }
978 
979 static int
980 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
981 {
982 	int delay, timeout;
983 
984 	/*
985 	 * In case of ZFS and NFS we don't have a way to wait for
986 	 * specific device.  Also do the wait if the user forced that
987 	 * behaviour by setting vfs.root_mount_always_wait=1.
988 	 */
989 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
990 	    dev[0] == '\0' || root_mount_always_wait != 0) {
991 		vfs_mountroot_wait();
992 		return (0);
993 	}
994 
995 	/*
996 	 * Otherwise, no point in waiting if the device is already there.
997 	 * Note that we must wait for GEOM to finish reconfiguring itself,
998 	 * eg for geom_part(4) to finish tasting.
999 	 */
1000 	g_waitidle();
1001 	if (parse_mount_dev_present(dev))
1002 		return (0);
1003 
1004 	/*
1005 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1006 	 * to behave exactly as it used to work before.
1007 	 */
1008 	vfs_mountroot_wait();
1009 	printf("mountroot: waiting for device %s...\n", dev);
1010 	delay = hz / 10;
1011 	timeout = root_mount_timeout * hz;
1012 	do {
1013 		pause("rmdev", delay);
1014 		timeout -= delay;
1015 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1016 
1017 	if (timeout <= 0)
1018 		return (ENODEV);
1019 
1020 	return (0);
1021 }
1022 
1023 void
1024 vfs_mountroot(void)
1025 {
1026 	struct mount *mp;
1027 	struct sbuf *sb;
1028 	struct thread *td;
1029 	time_t timebase;
1030 	int error;
1031 
1032 	mtx_assert(&Giant, MA_NOTOWNED);
1033 
1034 	TSENTER();
1035 
1036 	td = curthread;
1037 
1038 	sb = sbuf_new_auto();
1039 	vfs_mountroot_conf0(sb);
1040 	sbuf_finish(sb);
1041 
1042 	error = vfs_mountroot_devfs(td, &mp);
1043 	while (!error) {
1044 		error = vfs_mountroot_parse(sb, mp);
1045 		if (!error) {
1046 			vfs_mountroot_shuffle(td, mp);
1047 			sbuf_clear(sb);
1048 			error = vfs_mountroot_readconf(td, sb);
1049 			sbuf_finish(sb);
1050 		}
1051 	}
1052 
1053 	sbuf_delete(sb);
1054 
1055 	/*
1056 	 * Iterate over all currently mounted file systems and use
1057 	 * the time stamp found to check and/or initialize the RTC.
1058 	 * Call inittodr() only once and pass it the largest of the
1059 	 * timestamps we encounter.
1060 	 */
1061 	timebase = 0;
1062 	mtx_lock(&mountlist_mtx);
1063 	mp = TAILQ_FIRST(&mountlist);
1064 	while (mp != NULL) {
1065 		if (mp->mnt_time > timebase)
1066 			timebase = mp->mnt_time;
1067 		mp = TAILQ_NEXT(mp, mnt_list);
1068 	}
1069 	mtx_unlock(&mountlist_mtx);
1070 	inittodr(timebase);
1071 
1072 	/* Keep prison0's root in sync with the global rootvnode. */
1073 	mtx_lock(&prison0.pr_mtx);
1074 	prison0.pr_root = rootvnode;
1075 	vref(prison0.pr_root);
1076 	mtx_unlock(&prison0.pr_mtx);
1077 
1078 	mtx_lock(&root_holds_mtx);
1079 	atomic_store_rel_int(&root_mount_complete, 1);
1080 	wakeup(&root_mount_complete);
1081 	mtx_unlock(&root_holds_mtx);
1082 
1083 	EVENTHANDLER_INVOKE(mountroot);
1084 
1085 	TSEXIT();
1086 }
1087 
1088 static struct mntarg *
1089 parse_mountroot_options(struct mntarg *ma, const char *options)
1090 {
1091 	char *p;
1092 	char *name, *name_arg;
1093 	char *val, *val_arg;
1094 	char *opts;
1095 
1096 	if (options == NULL || options[0] == '\0')
1097 		return (ma);
1098 
1099 	p = opts = strdup(options, M_MOUNT);
1100 	if (opts == NULL) {
1101 		return (ma);
1102 	}
1103 
1104 	while((name = strsep(&p, ",")) != NULL) {
1105 		if (name[0] == '\0')
1106 			break;
1107 
1108 		val = strchr(name, '=');
1109 		if (val != NULL) {
1110 			*val = '\0';
1111 			++val;
1112 		}
1113 		if( strcmp(name, "rw") == 0 ||
1114 		    strcmp(name, "noro") == 0) {
1115 			/*
1116 			 * The first time we mount the root file system,
1117 			 * we need to mount 'ro', so We need to ignore
1118 			 * 'rw' and 'noro' mount options.
1119 			 */
1120 			continue;
1121 		}
1122 		name_arg = strdup(name, M_MOUNT);
1123 		val_arg = NULL;
1124 		if (val != NULL)
1125 			val_arg = strdup(val, M_MOUNT);
1126 
1127 		ma = mount_arg(ma, name_arg, val_arg,
1128 		    (val_arg != NULL ? -1 : 0));
1129 	}
1130 	free(opts, M_MOUNT);
1131 	return (ma);
1132 }
1133