xref: /freebsd/sys/kern/vfs_mountroot.c (revision 6683132d54bd6d589889e43dabdc53d35e38a028)
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 	}
277 
278 	set_rootvnode();
279 
280 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
281 	if (error)
282 		printf("kern_symlink /dev -> / returns %d\n", error);
283 
284 	return (error);
285 }
286 
287 static void
288 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
289 {
290 	struct nameidata nd;
291 	struct mount *mporoot, *mpnroot;
292 	struct vnode *vp, *vporoot, *vpdevfs;
293 	char *fspath;
294 	int error;
295 
296 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
297 
298 	/* Shuffle the mountlist. */
299 	mtx_lock(&mountlist_mtx);
300 	mporoot = TAILQ_FIRST(&mountlist);
301 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
302 	if (mporoot != mpdevfs) {
303 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
304 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
305 	}
306 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
307 	mtx_unlock(&mountlist_mtx);
308 
309 	cache_purgevfs(mporoot, true);
310 	if (mporoot != mpdevfs)
311 		cache_purgevfs(mpdevfs, true);
312 
313 	if (VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot))
314 		panic("vfs_mountroot_shuffle: Cannot find root vnode");
315 
316 	VI_LOCK(vporoot);
317 	vporoot->v_iflag &= ~VI_MOUNT;
318 	VI_UNLOCK(vporoot);
319 	vporoot->v_mountedhere = NULL;
320 	mporoot->mnt_flag &= ~MNT_ROOTFS;
321 	mporoot->mnt_vnodecovered = NULL;
322 	vput(vporoot);
323 
324 	/* Set up the new rootvnode, and purge the cache */
325 	mpnroot->mnt_vnodecovered = NULL;
326 	set_rootvnode();
327 	cache_purgevfs(rootvnode->v_mount, true);
328 
329 	if (mporoot != mpdevfs) {
330 		/* Remount old root under /.mount or /mnt */
331 		fspath = "/.mount";
332 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
333 		    fspath, td);
334 		error = namei(&nd);
335 		if (error) {
336 			NDFREE(&nd, NDF_ONLY_PNBUF);
337 			fspath = "/mnt";
338 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
339 			    fspath, td);
340 			error = namei(&nd);
341 		}
342 		if (!error) {
343 			vp = nd.ni_vp;
344 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
345 			if (!error)
346 				error = vinvalbuf(vp, V_SAVE, 0, 0);
347 			if (!error) {
348 				cache_purge(vp);
349 				mporoot->mnt_vnodecovered = vp;
350 				vp->v_mountedhere = mporoot;
351 				strlcpy(mporoot->mnt_stat.f_mntonname,
352 				    fspath, MNAMELEN);
353 				VOP_UNLOCK(vp, 0);
354 			} else
355 				vput(vp);
356 		}
357 		NDFREE(&nd, NDF_ONLY_PNBUF);
358 
359 		if (error)
360 			printf("mountroot: unable to remount previous root "
361 			    "under /.mount or /mnt (error %d)\n", error);
362 	}
363 
364 	/* Remount devfs under /dev */
365 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
366 	error = namei(&nd);
367 	if (!error) {
368 		vp = nd.ni_vp;
369 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
370 		if (!error)
371 			error = vinvalbuf(vp, V_SAVE, 0, 0);
372 		if (!error) {
373 			vpdevfs = mpdevfs->mnt_vnodecovered;
374 			if (vpdevfs != NULL) {
375 				cache_purge(vpdevfs);
376 				vpdevfs->v_mountedhere = NULL;
377 				vrele(vpdevfs);
378 			}
379 			mpdevfs->mnt_vnodecovered = vp;
380 			vp->v_mountedhere = mpdevfs;
381 			VOP_UNLOCK(vp, 0);
382 		} else
383 			vput(vp);
384 	}
385 	if (error)
386 		printf("mountroot: unable to remount devfs under /dev "
387 		    "(error %d)\n", error);
388 	NDFREE(&nd, NDF_ONLY_PNBUF);
389 
390 	if (mporoot == mpdevfs) {
391 		vfs_unbusy(mpdevfs);
392 		/* Unlink the no longer needed /dev/dev -> / symlink */
393 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
394 		    UIO_SYSSPACE, 0, 0);
395 		if (error)
396 			printf("mountroot: unable to unlink /dev/dev "
397 			    "(error %d)\n", error);
398 	}
399 }
400 
401 /*
402  * Configuration parser.
403  */
404 
405 /* Parser character classes. */
406 #define	CC_WHITESPACE		-1
407 #define	CC_NONWHITESPACE	-2
408 
409 /* Parse errors. */
410 #define	PE_EOF			-1
411 #define	PE_EOL			-2
412 
413 static __inline int
414 parse_peek(char **conf)
415 {
416 
417 	return (**conf);
418 }
419 
420 static __inline void
421 parse_poke(char **conf, int c)
422 {
423 
424 	**conf = c;
425 }
426 
427 static __inline void
428 parse_advance(char **conf)
429 {
430 
431 	(*conf)++;
432 }
433 
434 static int
435 parse_skipto(char **conf, int mc)
436 {
437 	int c, match;
438 
439 	while (1) {
440 		c = parse_peek(conf);
441 		if (c == 0)
442 			return (PE_EOF);
443 		switch (mc) {
444 		case CC_WHITESPACE:
445 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
446 			break;
447 		case CC_NONWHITESPACE:
448 			if (c == '\n')
449 				return (PE_EOL);
450 			match = (c != ' ' && c != '\t') ? 1 : 0;
451 			break;
452 		default:
453 			match = (c == mc) ? 1 : 0;
454 			break;
455 		}
456 		if (match)
457 			break;
458 		parse_advance(conf);
459 	}
460 	return (0);
461 }
462 
463 static int
464 parse_token(char **conf, char **tok)
465 {
466 	char *p;
467 	size_t len;
468 	int error;
469 
470 	*tok = NULL;
471 	error = parse_skipto(conf, CC_NONWHITESPACE);
472 	if (error)
473 		return (error);
474 	p = *conf;
475 	error = parse_skipto(conf, CC_WHITESPACE);
476 	len = *conf - p;
477 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
478 	bcopy(p, *tok, len);
479 	return (0);
480 }
481 
482 static void
483 parse_dir_ask_printenv(const char *var)
484 {
485 	char *val;
486 
487 	val = kern_getenv(var);
488 	if (val != NULL) {
489 		printf("  %s=%s\n", var, val);
490 		freeenv(val);
491 	}
492 }
493 
494 static int
495 parse_dir_ask(char **conf)
496 {
497 	char name[80];
498 	char *mnt;
499 	int error;
500 
501 	vfs_mountroot_wait();
502 
503 	printf("\nLoader variables:\n");
504 	parse_dir_ask_printenv("vfs.root.mountfrom");
505 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
506 
507 	printf("\nManual root filesystem specification:\n");
508 	printf("  <fstype>:<device> [options]\n");
509 	printf("      Mount <device> using filesystem <fstype>\n");
510 	printf("      and with the specified (optional) option list.\n");
511 	printf("\n");
512 	printf("    eg. ufs:/dev/da0s1a\n");
513 	printf("        zfs:zroot/ROOT/default\n");
514 	printf("        cd9660:/dev/cd0 ro\n");
515 	printf("          (which is equivalent to: ");
516 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
517 	printf("\n");
518 	printf("  ?               List valid disk boot devices\n");
519 	printf("  .               Yield 1 second (for background tasks)\n");
520 	printf("  <empty line>    Abort manual input\n");
521 
522 	do {
523 		error = EINVAL;
524 		printf("\nmountroot> ");
525 		cngets(name, sizeof(name), GETS_ECHO);
526 		if (name[0] == '\0')
527 			break;
528 		if (name[0] == '?' && name[1] == '\0') {
529 			printf("\nList of GEOM managed disk devices:\n  ");
530 			g_dev_print();
531 			continue;
532 		}
533 		if (name[0] == '.' && name[1] == '\0') {
534 			pause("rmask", hz);
535 			continue;
536 		}
537 		mnt = name;
538 		error = parse_mount(&mnt);
539 		if (error == -1)
540 			printf("Invalid file system specification.\n");
541 	} while (error != 0);
542 
543 	return (error);
544 }
545 
546 static int
547 parse_dir_md(char **conf)
548 {
549 	struct stat sb;
550 	struct thread *td;
551 	struct md_ioctl *mdio;
552 	char *path, *tok;
553 	int error, fd, len;
554 
555 	td = curthread;
556 
557 	error = parse_token(conf, &tok);
558 	if (error)
559 		return (error);
560 
561 	len = strlen(tok);
562 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
563 	path = (void *)(mdio + 1);
564 	bcopy(tok, path, len);
565 	free(tok, M_TEMP);
566 
567 	/* Get file status. */
568 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
569 	if (error)
570 		goto out;
571 
572 	/* Open /dev/mdctl so that we can attach/detach. */
573 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
574 	    O_RDWR, 0);
575 	if (error)
576 		goto out;
577 
578 	fd = td->td_retval[0];
579 	mdio->md_version = MDIOVERSION;
580 	mdio->md_type = MD_VNODE;
581 
582 	if (root_mount_mddev != -1) {
583 		mdio->md_unit = root_mount_mddev;
584 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
585 		/* Ignore errors. We don't care. */
586 		root_mount_mddev = -1;
587 	}
588 
589 	mdio->md_file = (void *)(mdio + 1);
590 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
591 	mdio->md_mediasize = sb.st_size;
592 	mdio->md_unit = 0;
593 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
594 	if (error)
595 		goto out;
596 
597 	if (mdio->md_unit > 9) {
598 		printf("rootmount: too many md units\n");
599 		mdio->md_file = NULL;
600 		mdio->md_options = 0;
601 		mdio->md_mediasize = 0;
602 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
603 		/* Ignore errors. We don't care. */
604 		error = ERANGE;
605 		goto out;
606 	}
607 
608 	root_mount_mddev = mdio->md_unit;
609 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
610 
611 	error = kern_close(td, fd);
612 
613  out:
614 	free(mdio, M_TEMP);
615 	return (error);
616 }
617 
618 static int
619 parse_dir_onfail(char **conf)
620 {
621 	char *action;
622 	int error;
623 
624 	error = parse_token(conf, &action);
625 	if (error)
626 		return (error);
627 
628 	if (!strcmp(action, "continue"))
629 		root_mount_onfail = A_CONTINUE;
630 	else if (!strcmp(action, "panic"))
631 		root_mount_onfail = A_PANIC;
632 	else if (!strcmp(action, "reboot"))
633 		root_mount_onfail = A_REBOOT;
634 	else if (!strcmp(action, "retry"))
635 		root_mount_onfail = A_RETRY;
636 	else {
637 		printf("rootmount: %s: unknown action\n", action);
638 		error = EINVAL;
639 	}
640 
641 	free(action, M_TEMP);
642 	return (0);
643 }
644 
645 static int
646 parse_dir_timeout(char **conf)
647 {
648 	char *tok, *endtok;
649 	long secs;
650 	int error;
651 
652 	error = parse_token(conf, &tok);
653 	if (error)
654 		return (error);
655 
656 	secs = strtol(tok, &endtok, 0);
657 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
658 	if (!error)
659 		root_mount_timeout = secs;
660 	free(tok, M_TEMP);
661 	return (error);
662 }
663 
664 static int
665 parse_directive(char **conf)
666 {
667 	char *dir;
668 	int error;
669 
670 	error = parse_token(conf, &dir);
671 	if (error)
672 		return (error);
673 
674 	if (strcmp(dir, ".ask") == 0)
675 		error = parse_dir_ask(conf);
676 	else if (strcmp(dir, ".md") == 0)
677 		error = parse_dir_md(conf);
678 	else if (strcmp(dir, ".onfail") == 0)
679 		error = parse_dir_onfail(conf);
680 	else if (strcmp(dir, ".timeout") == 0)
681 		error = parse_dir_timeout(conf);
682 	else {
683 		printf("mountroot: invalid directive `%s'\n", dir);
684 		/* Ignore the rest of the line. */
685 		(void)parse_skipto(conf, '\n');
686 		error = EINVAL;
687 	}
688 	free(dir, M_TEMP);
689 	return (error);
690 }
691 
692 static int
693 parse_mount_dev_present(const char *dev)
694 {
695 	struct nameidata nd;
696 	int error;
697 
698 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
699 	error = namei(&nd);
700 	if (!error)
701 		vput(nd.ni_vp);
702 	NDFREE(&nd, NDF_ONLY_PNBUF);
703 	return (error != 0) ? 0 : 1;
704 }
705 
706 #define	ERRMSGL	255
707 static int
708 parse_mount(char **conf)
709 {
710 	char *errmsg;
711 	struct mntarg *ma;
712 	char *dev, *fs, *opts, *tok;
713 	int delay, error, timeout;
714 
715 	error = parse_token(conf, &tok);
716 	if (error)
717 		return (error);
718 	fs = tok;
719 	error = parse_skipto(&tok, ':');
720 	if (error) {
721 		free(fs, M_TEMP);
722 		return (error);
723 	}
724 	parse_poke(&tok, '\0');
725 	parse_advance(&tok);
726 	dev = tok;
727 
728 	if (root_mount_mddev != -1) {
729 		/* Handle substitution for the md unit number. */
730 		tok = strstr(dev, "md#");
731 		if (tok != NULL)
732 			tok[2] = '0' + root_mount_mddev;
733 	}
734 
735 	/* Parse options. */
736 	error = parse_token(conf, &tok);
737 	opts = (error == 0) ? tok : NULL;
738 
739 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
740 	    (opts != NULL) ? opts : "");
741 
742 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
743 
744 	if (vfs_byname(fs) == NULL) {
745 		strlcpy(errmsg, "unknown file system", ERRMSGL);
746 		error = ENOENT;
747 		goto out;
748 	}
749 
750 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
751 	if (error != 0)
752 		goto out;
753 
754 	delay = hz / 10;
755 	timeout = root_mount_timeout * hz;
756 
757 	for (;;) {
758 		ma = NULL;
759 		ma = mount_arg(ma, "fstype", fs, -1);
760 		ma = mount_arg(ma, "fspath", "/", -1);
761 		ma = mount_arg(ma, "from", dev, -1);
762 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
763 		ma = mount_arg(ma, "ro", NULL, 0);
764 		ma = parse_mountroot_options(ma, opts);
765 
766 		error = kernel_mount(ma, MNT_ROOTFS);
767 		if (error == 0 || timeout <= 0)
768 			break;
769 
770 		if (root_mount_timeout * hz == timeout ||
771 		    (bootverbose && timeout % hz == 0)) {
772 			printf("Mounting from %s:%s failed with error %d; "
773 			    "retrying for %d more second%s\n", fs, dev, error,
774 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
775 		}
776 		pause("rmretry", delay);
777 		timeout -= delay;
778 	}
779  out:
780 	if (error) {
781 		printf("Mounting from %s:%s failed with error %d",
782 		    fs, dev, error);
783 		if (errmsg[0] != '\0')
784 			printf(": %s", errmsg);
785 		printf(".\n");
786 	}
787 	free(fs, M_TEMP);
788 	free(errmsg, M_TEMP);
789 	if (opts != NULL)
790 		free(opts, M_TEMP);
791 	/* kernel_mount can return -1 on error. */
792 	return ((error < 0) ? EDOOFUS : error);
793 }
794 #undef ERRMSGL
795 
796 static int
797 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
798 {
799 	struct mount *mp;
800 	char *conf;
801 	int error;
802 
803 	root_mount_mddev = -1;
804 
805 retry:
806 	conf = sbuf_data(sb);
807 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
808 	error = (mp == NULL) ? 0 : EDOOFUS;
809 	root_mount_onfail = A_CONTINUE;
810 	while (mp == NULL) {
811 		error = parse_skipto(&conf, CC_NONWHITESPACE);
812 		if (error == PE_EOL) {
813 			parse_advance(&conf);
814 			continue;
815 		}
816 		if (error < 0)
817 			break;
818 		switch (parse_peek(&conf)) {
819 		case '#':
820 			error = parse_skipto(&conf, '\n');
821 			break;
822 		case '.':
823 			error = parse_directive(&conf);
824 			break;
825 		default:
826 			error = parse_mount(&conf);
827 			if (error == -1) {
828 				printf("mountroot: invalid file system "
829 				    "specification.\n");
830 				error = 0;
831 			}
832 			break;
833 		}
834 		if (error < 0)
835 			break;
836 		/* Ignore any trailing garbage on the line. */
837 		if (parse_peek(&conf) != '\n') {
838 			printf("mountroot: advancing to next directive...\n");
839 			(void)parse_skipto(&conf, '\n');
840 		}
841 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
842 	}
843 	if (mp != NULL)
844 		return (0);
845 
846 	/*
847 	 * We failed to mount (a new) root.
848 	 */
849 	switch (root_mount_onfail) {
850 	case A_CONTINUE:
851 		break;
852 	case A_PANIC:
853 		panic("mountroot: unable to (re-)mount root.");
854 		/* NOTREACHED */
855 	case A_RETRY:
856 		goto retry;
857 	case A_REBOOT:
858 		kern_reboot(RB_NOSYNC);
859 		/* NOTREACHED */
860 	}
861 
862 	return (error);
863 }
864 
865 static void
866 vfs_mountroot_conf0(struct sbuf *sb)
867 {
868 	char *s, *tok, *mnt, *opt;
869 	int error;
870 
871 	sbuf_printf(sb, ".onfail panic\n");
872 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
873 	if (boothowto & RB_ASKNAME)
874 		sbuf_printf(sb, ".ask\n");
875 #ifdef ROOTDEVNAME
876 	if (boothowto & RB_DFLTROOT)
877 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
878 #endif
879 	if (boothowto & RB_CDROM) {
880 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
881 		sbuf_printf(sb, ".timeout 0\n");
882 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
883 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
884 	}
885 	s = kern_getenv("vfs.root.mountfrom");
886 	if (s != NULL) {
887 		opt = kern_getenv("vfs.root.mountfrom.options");
888 		tok = s;
889 		error = parse_token(&tok, &mnt);
890 		while (!error) {
891 			sbuf_printf(sb, "%s %s\n", mnt,
892 			    (opt != NULL) ? opt : "");
893 			free(mnt, M_TEMP);
894 			error = parse_token(&tok, &mnt);
895 		}
896 		if (opt != NULL)
897 			freeenv(opt);
898 		freeenv(s);
899 	}
900 	if (rootdevnames[0] != NULL)
901 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
902 	if (rootdevnames[1] != NULL)
903 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
904 #ifdef ROOTDEVNAME
905 	if (!(boothowto & RB_DFLTROOT))
906 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
907 #endif
908 	if (!(boothowto & RB_ASKNAME))
909 		sbuf_printf(sb, ".ask\n");
910 }
911 
912 static int
913 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
914 {
915 	static char buf[128];
916 	struct nameidata nd;
917 	off_t ofs;
918 	ssize_t resid;
919 	int error, flags, len;
920 
921 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
922 	flags = FREAD;
923 	error = vn_open(&nd, &flags, 0, NULL);
924 	if (error)
925 		return (error);
926 
927 	NDFREE(&nd, NDF_ONLY_PNBUF);
928 	ofs = 0;
929 	len = sizeof(buf) - 1;
930 	while (1) {
931 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
932 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
933 		    NOCRED, &resid, td);
934 		if (error)
935 			break;
936 		if (resid == len)
937 			break;
938 		buf[len - resid] = 0;
939 		sbuf_printf(sb, "%s", buf);
940 		ofs += len - resid;
941 	}
942 
943 	VOP_UNLOCK(nd.ni_vp, 0);
944 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
945 	return (error);
946 }
947 
948 static void
949 vfs_mountroot_wait(void)
950 {
951 	struct root_hold_token *h;
952 	struct timeval lastfail;
953 	int curfail;
954 
955 	TSENTER();
956 
957 	curfail = 0;
958 	while (1) {
959 		g_waitidle();
960 		mtx_lock(&root_holds_mtx);
961 		if (LIST_EMPTY(&root_holds)) {
962 			mtx_unlock(&root_holds_mtx);
963 			break;
964 		}
965 		if (ppsratecheck(&lastfail, &curfail, 1)) {
966 			printf("Root mount waiting for:");
967 			LIST_FOREACH(h, &root_holds, list)
968 				printf(" %s", h->who);
969 			printf("\n");
970 		}
971 		TSWAIT("root mount");
972 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
973 		    hz);
974 		TSUNWAIT("root mount");
975 	}
976 
977 	TSEXIT();
978 }
979 
980 static int
981 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
982 {
983 	int delay, timeout;
984 
985 	/*
986 	 * In case of ZFS and NFS we don't have a way to wait for
987 	 * specific device.  Also do the wait if the user forced that
988 	 * behaviour by setting vfs.root_mount_always_wait=1.
989 	 */
990 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
991 	    dev[0] == '\0' || root_mount_always_wait != 0) {
992 		vfs_mountroot_wait();
993 		return (0);
994 	}
995 
996 	/*
997 	 * Otherwise, no point in waiting if the device is already there.
998 	 * Note that we must wait for GEOM to finish reconfiguring itself,
999 	 * eg for geom_part(4) to finish tasting.
1000 	 */
1001 	g_waitidle();
1002 	if (parse_mount_dev_present(dev))
1003 		return (0);
1004 
1005 	/*
1006 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1007 	 * to behave exactly as it used to work before.
1008 	 */
1009 	vfs_mountroot_wait();
1010 	printf("mountroot: waiting for device %s...\n", dev);
1011 	delay = hz / 10;
1012 	timeout = root_mount_timeout * hz;
1013 	do {
1014 		pause("rmdev", delay);
1015 		timeout -= delay;
1016 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1017 
1018 	if (timeout <= 0)
1019 		return (ENODEV);
1020 
1021 	return (0);
1022 }
1023 
1024 void
1025 vfs_mountroot(void)
1026 {
1027 	struct mount *mp;
1028 	struct sbuf *sb;
1029 	struct thread *td;
1030 	time_t timebase;
1031 	int error;
1032 
1033 	mtx_assert(&Giant, MA_NOTOWNED);
1034 
1035 	TSENTER();
1036 
1037 	td = curthread;
1038 
1039 	sb = sbuf_new_auto();
1040 	vfs_mountroot_conf0(sb);
1041 	sbuf_finish(sb);
1042 
1043 	error = vfs_mountroot_devfs(td, &mp);
1044 	while (!error) {
1045 		error = vfs_mountroot_parse(sb, mp);
1046 		if (!error) {
1047 			vfs_mountroot_shuffle(td, mp);
1048 			sbuf_clear(sb);
1049 			error = vfs_mountroot_readconf(td, sb);
1050 			sbuf_finish(sb);
1051 		}
1052 	}
1053 
1054 	sbuf_delete(sb);
1055 
1056 	/*
1057 	 * Iterate over all currently mounted file systems and use
1058 	 * the time stamp found to check and/or initialize the RTC.
1059 	 * Call inittodr() only once and pass it the largest of the
1060 	 * timestamps we encounter.
1061 	 */
1062 	timebase = 0;
1063 	mtx_lock(&mountlist_mtx);
1064 	mp = TAILQ_FIRST(&mountlist);
1065 	while (mp != NULL) {
1066 		if (mp->mnt_time > timebase)
1067 			timebase = mp->mnt_time;
1068 		mp = TAILQ_NEXT(mp, mnt_list);
1069 	}
1070 	mtx_unlock(&mountlist_mtx);
1071 	inittodr(timebase);
1072 
1073 	/* Keep prison0's root in sync with the global rootvnode. */
1074 	mtx_lock(&prison0.pr_mtx);
1075 	prison0.pr_root = rootvnode;
1076 	vref(prison0.pr_root);
1077 	mtx_unlock(&prison0.pr_mtx);
1078 
1079 	mtx_lock(&root_holds_mtx);
1080 	atomic_store_rel_int(&root_mount_complete, 1);
1081 	wakeup(&root_mount_complete);
1082 	mtx_unlock(&root_holds_mtx);
1083 
1084 	EVENTHANDLER_INVOKE(mountroot);
1085 
1086 	TSEXIT();
1087 }
1088 
1089 static struct mntarg *
1090 parse_mountroot_options(struct mntarg *ma, const char *options)
1091 {
1092 	char *p;
1093 	char *name, *name_arg;
1094 	char *val, *val_arg;
1095 	char *opts;
1096 
1097 	if (options == NULL || options[0] == '\0')
1098 		return (ma);
1099 
1100 	p = opts = strdup(options, M_MOUNT);
1101 	if (opts == NULL) {
1102 		return (ma);
1103 	}
1104 
1105 	while((name = strsep(&p, ",")) != NULL) {
1106 		if (name[0] == '\0')
1107 			break;
1108 
1109 		val = strchr(name, '=');
1110 		if (val != NULL) {
1111 			*val = '\0';
1112 			++val;
1113 		}
1114 		if( strcmp(name, "rw") == 0 ||
1115 		    strcmp(name, "noro") == 0) {
1116 			/*
1117 			 * The first time we mount the root file system,
1118 			 * we need to mount 'ro', so We need to ignore
1119 			 * 'rw' and 'noro' mount options.
1120 			 */
1121 			continue;
1122 		}
1123 		name_arg = strdup(name, M_MOUNT);
1124 		val_arg = NULL;
1125 		if (val != NULL)
1126 			val_arg = strdup(val, M_MOUNT);
1127 
1128 		ma = mount_arg(ma, name_arg, val_arg,
1129 		    (val_arg != NULL ? -1 : 0));
1130 	}
1131 	free(opts, M_MOUNT);
1132 	return (ma);
1133 }
1134