xref: /freebsd/sys/kern/vfs_mountroot.c (revision dd41de95a84d979615a2ef11df6850622bf6184e)
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);
330 	if (mporoot != mpdevfs)
331 		cache_purgevfs(mpdevfs);
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 	vn_irflag_unset_locked(vporoot, VIRF_MOUNTPOINT);
339 	vporoot->v_mountedhere = NULL;
340 	VI_UNLOCK(vporoot);
341 	mporoot->mnt_flag &= ~MNT_ROOTFS;
342 	mporoot->mnt_vnodecovered = NULL;
343 	vput(vporoot);
344 
345 	/* Set up the new rootvnode, and purge the cache */
346 	mpnroot->mnt_vnodecovered = NULL;
347 	set_rootvnode();
348 	cache_purgevfs(rootvnode->v_mount);
349 
350 	if (mporoot != mpdevfs) {
351 		/* Remount old root under /.mount or /mnt */
352 		fspath = "/.mount";
353 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
354 		    fspath, td);
355 		error = namei(&nd);
356 		if (error) {
357 			NDFREE(&nd, NDF_ONLY_PNBUF);
358 			fspath = "/mnt";
359 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
360 			    fspath, td);
361 			error = namei(&nd);
362 		}
363 		if (!error) {
364 			vp = nd.ni_vp;
365 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
366 			if (!error)
367 				error = vinvalbuf(vp, V_SAVE, 0, 0);
368 			if (!error) {
369 				cache_purge(vp);
370 				mporoot->mnt_vnodecovered = vp;
371 				vp->v_mountedhere = mporoot;
372 				strlcpy(mporoot->mnt_stat.f_mntonname,
373 				    fspath, MNAMELEN);
374 				VOP_UNLOCK(vp);
375 			} else
376 				vput(vp);
377 		}
378 		NDFREE(&nd, NDF_ONLY_PNBUF);
379 
380 		if (error)
381 			printf("mountroot: unable to remount previous root "
382 			    "under /.mount or /mnt (error %d)\n", error);
383 	}
384 
385 	/* Remount devfs under /dev */
386 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
387 	error = namei(&nd);
388 	if (!error) {
389 		vp = nd.ni_vp;
390 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
391 		if (!error)
392 			error = vinvalbuf(vp, V_SAVE, 0, 0);
393 		if (!error) {
394 			vpdevfs = mpdevfs->mnt_vnodecovered;
395 			if (vpdevfs != NULL) {
396 				cache_purge(vpdevfs);
397 				VI_LOCK(vpdevfs);
398 				vn_irflag_unset_locked(vpdevfs, VIRF_MOUNTPOINT);
399 				vpdevfs->v_mountedhere = NULL;
400 				VI_UNLOCK(vpdevfs);
401 				vrele(vpdevfs);
402 			}
403 			VI_LOCK(vp);
404 			mpdevfs->mnt_vnodecovered = vp;
405 			vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
406 			vp->v_mountedhere = mpdevfs;
407 			VI_UNLOCK(vp);
408 			VOP_UNLOCK(vp);
409 		} else
410 			vput(vp);
411 	}
412 	if (error)
413 		printf("mountroot: unable to remount devfs under /dev "
414 		    "(error %d)\n", error);
415 	NDFREE(&nd, NDF_ONLY_PNBUF);
416 
417 	if (mporoot == mpdevfs) {
418 		vfs_unbusy(mpdevfs);
419 		/* Unlink the no longer needed /dev/dev -> / symlink */
420 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
421 		    UIO_SYSSPACE, 0, 0);
422 		if (error)
423 			printf("mountroot: unable to unlink /dev/dev "
424 			    "(error %d)\n", error);
425 	}
426 }
427 
428 /*
429  * Configuration parser.
430  */
431 
432 /* Parser character classes. */
433 #define	CC_WHITESPACE		-1
434 #define	CC_NONWHITESPACE	-2
435 
436 /* Parse errors. */
437 #define	PE_EOF			-1
438 #define	PE_EOL			-2
439 
440 static __inline int
441 parse_peek(char **conf)
442 {
443 
444 	return (**conf);
445 }
446 
447 static __inline void
448 parse_poke(char **conf, int c)
449 {
450 
451 	**conf = c;
452 }
453 
454 static __inline void
455 parse_advance(char **conf)
456 {
457 
458 	(*conf)++;
459 }
460 
461 static int
462 parse_skipto(char **conf, int mc)
463 {
464 	int c, match;
465 
466 	while (1) {
467 		c = parse_peek(conf);
468 		if (c == 0)
469 			return (PE_EOF);
470 		switch (mc) {
471 		case CC_WHITESPACE:
472 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
473 			break;
474 		case CC_NONWHITESPACE:
475 			if (c == '\n')
476 				return (PE_EOL);
477 			match = (c != ' ' && c != '\t') ? 1 : 0;
478 			break;
479 		default:
480 			match = (c == mc) ? 1 : 0;
481 			break;
482 		}
483 		if (match)
484 			break;
485 		parse_advance(conf);
486 	}
487 	return (0);
488 }
489 
490 static int
491 parse_token(char **conf, char **tok)
492 {
493 	char *p;
494 	size_t len;
495 	int error;
496 
497 	*tok = NULL;
498 	error = parse_skipto(conf, CC_NONWHITESPACE);
499 	if (error)
500 		return (error);
501 	p = *conf;
502 	error = parse_skipto(conf, CC_WHITESPACE);
503 	len = *conf - p;
504 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
505 	bcopy(p, *tok, len);
506 	return (0);
507 }
508 
509 static void
510 parse_dir_ask_printenv(const char *var)
511 {
512 	char *val;
513 
514 	val = kern_getenv(var);
515 	if (val != NULL) {
516 		printf("  %s=%s\n", var, val);
517 		freeenv(val);
518 	}
519 }
520 
521 static int
522 parse_dir_ask(char **conf)
523 {
524 	char name[80];
525 	char *mnt;
526 	int error;
527 
528 	vfs_mountroot_wait();
529 
530 	printf("\nLoader variables:\n");
531 	parse_dir_ask_printenv("vfs.root.mountfrom");
532 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
533 
534 	printf("\nManual root filesystem specification:\n");
535 	printf("  <fstype>:<device> [options]\n");
536 	printf("      Mount <device> using filesystem <fstype>\n");
537 	printf("      and with the specified (optional) option list.\n");
538 	printf("\n");
539 	printf("    eg. ufs:/dev/da0s1a\n");
540 	printf("        zfs:zroot/ROOT/default\n");
541 	printf("        cd9660:/dev/cd0 ro\n");
542 	printf("          (which is equivalent to: ");
543 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
544 	printf("\n");
545 	printf("  ?               List valid disk boot devices\n");
546 	printf("  .               Yield 1 second (for background tasks)\n");
547 	printf("  <empty line>    Abort manual input\n");
548 
549 	do {
550 		error = EINVAL;
551 		printf("\nmountroot> ");
552 		cngets(name, sizeof(name), GETS_ECHO);
553 		if (name[0] == '\0')
554 			break;
555 		if (name[0] == '?' && name[1] == '\0') {
556 			printf("\nList of GEOM managed disk devices:\n  ");
557 			g_dev_print();
558 			continue;
559 		}
560 		if (name[0] == '.' && name[1] == '\0') {
561 			pause("rmask", hz);
562 			continue;
563 		}
564 		mnt = name;
565 		error = parse_mount(&mnt);
566 		if (error == -1)
567 			printf("Invalid file system specification.\n");
568 	} while (error != 0);
569 
570 	return (error);
571 }
572 
573 static int
574 parse_dir_md(char **conf)
575 {
576 	struct stat sb;
577 	struct thread *td;
578 	struct md_ioctl *mdio;
579 	char *path, *tok;
580 	int error, fd, len;
581 
582 	td = curthread;
583 
584 	error = parse_token(conf, &tok);
585 	if (error)
586 		return (error);
587 
588 	len = strlen(tok);
589 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
590 	path = (void *)(mdio + 1);
591 	bcopy(tok, path, len);
592 	free(tok, M_TEMP);
593 
594 	/* Get file status. */
595 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
596 	if (error)
597 		goto out;
598 
599 	/* Open /dev/mdctl so that we can attach/detach. */
600 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
601 	    O_RDWR, 0);
602 	if (error)
603 		goto out;
604 
605 	fd = td->td_retval[0];
606 	mdio->md_version = MDIOVERSION;
607 	mdio->md_type = MD_VNODE;
608 
609 	if (root_mount_mddev != -1) {
610 		mdio->md_unit = root_mount_mddev;
611 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
612 		/* Ignore errors. We don't care. */
613 		root_mount_mddev = -1;
614 	}
615 
616 	mdio->md_file = (void *)(mdio + 1);
617 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
618 	mdio->md_mediasize = sb.st_size;
619 	mdio->md_unit = 0;
620 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
621 	if (error)
622 		goto out;
623 
624 	if (mdio->md_unit > 9) {
625 		printf("rootmount: too many md units\n");
626 		mdio->md_file = NULL;
627 		mdio->md_options = 0;
628 		mdio->md_mediasize = 0;
629 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
630 		/* Ignore errors. We don't care. */
631 		error = ERANGE;
632 		goto out;
633 	}
634 
635 	root_mount_mddev = mdio->md_unit;
636 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
637 
638 	error = kern_close(td, fd);
639 
640  out:
641 	free(mdio, M_TEMP);
642 	return (error);
643 }
644 
645 static int
646 parse_dir_onfail(char **conf)
647 {
648 	char *action;
649 	int error;
650 
651 	error = parse_token(conf, &action);
652 	if (error)
653 		return (error);
654 
655 	if (!strcmp(action, "continue"))
656 		root_mount_onfail = A_CONTINUE;
657 	else if (!strcmp(action, "panic"))
658 		root_mount_onfail = A_PANIC;
659 	else if (!strcmp(action, "reboot"))
660 		root_mount_onfail = A_REBOOT;
661 	else if (!strcmp(action, "retry"))
662 		root_mount_onfail = A_RETRY;
663 	else {
664 		printf("rootmount: %s: unknown action\n", action);
665 		error = EINVAL;
666 	}
667 
668 	free(action, M_TEMP);
669 	return (0);
670 }
671 
672 static int
673 parse_dir_timeout(char **conf)
674 {
675 	char *tok, *endtok;
676 	long secs;
677 	int error;
678 
679 	error = parse_token(conf, &tok);
680 	if (error)
681 		return (error);
682 
683 	secs = strtol(tok, &endtok, 0);
684 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
685 	if (!error)
686 		root_mount_timeout = secs;
687 	free(tok, M_TEMP);
688 	return (error);
689 }
690 
691 static int
692 parse_directive(char **conf)
693 {
694 	char *dir;
695 	int error;
696 
697 	error = parse_token(conf, &dir);
698 	if (error)
699 		return (error);
700 
701 	if (strcmp(dir, ".ask") == 0)
702 		error = parse_dir_ask(conf);
703 	else if (strcmp(dir, ".md") == 0)
704 		error = parse_dir_md(conf);
705 	else if (strcmp(dir, ".onfail") == 0)
706 		error = parse_dir_onfail(conf);
707 	else if (strcmp(dir, ".timeout") == 0)
708 		error = parse_dir_timeout(conf);
709 	else {
710 		printf("mountroot: invalid directive `%s'\n", dir);
711 		/* Ignore the rest of the line. */
712 		(void)parse_skipto(conf, '\n');
713 		error = EINVAL;
714 	}
715 	free(dir, M_TEMP);
716 	return (error);
717 }
718 
719 static int
720 parse_mount_dev_present(const char *dev)
721 {
722 	struct nameidata nd;
723 	int error;
724 
725 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
726 	error = namei(&nd);
727 	if (!error)
728 		vput(nd.ni_vp);
729 	NDFREE(&nd, NDF_ONLY_PNBUF);
730 	return (error != 0) ? 0 : 1;
731 }
732 
733 #define	ERRMSGL	255
734 static int
735 parse_mount(char **conf)
736 {
737 	char *errmsg;
738 	struct mntarg *ma;
739 	char *dev, *fs, *opts, *tok;
740 	int delay, error, timeout;
741 
742 	error = parse_token(conf, &tok);
743 	if (error)
744 		return (error);
745 	fs = tok;
746 	error = parse_skipto(&tok, ':');
747 	if (error) {
748 		free(fs, M_TEMP);
749 		return (error);
750 	}
751 	parse_poke(&tok, '\0');
752 	parse_advance(&tok);
753 	dev = tok;
754 
755 	if (root_mount_mddev != -1) {
756 		/* Handle substitution for the md unit number. */
757 		tok = strstr(dev, "md#");
758 		if (tok != NULL)
759 			tok[2] = '0' + root_mount_mddev;
760 	}
761 
762 	/* Parse options. */
763 	error = parse_token(conf, &tok);
764 	opts = (error == 0) ? tok : NULL;
765 
766 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
767 	    (opts != NULL) ? opts : "");
768 
769 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
770 
771 	if (vfs_byname(fs) == NULL) {
772 		strlcpy(errmsg, "unknown file system", ERRMSGL);
773 		error = ENOENT;
774 		goto out;
775 	}
776 
777 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
778 	if (error != 0)
779 		goto out;
780 
781 	delay = hz / 10;
782 	timeout = root_mount_timeout * hz;
783 
784 	for (;;) {
785 		ma = NULL;
786 		ma = mount_arg(ma, "fstype", fs, -1);
787 		ma = mount_arg(ma, "fspath", "/", -1);
788 		ma = mount_arg(ma, "from", dev, -1);
789 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
790 		ma = mount_arg(ma, "ro", NULL, 0);
791 		ma = parse_mountroot_options(ma, opts);
792 
793 		error = kernel_mount(ma, MNT_ROOTFS);
794 		if (error == 0 || timeout <= 0)
795 			break;
796 
797 		if (root_mount_timeout * hz == timeout ||
798 		    (bootverbose && timeout % hz == 0)) {
799 			printf("Mounting from %s:%s failed with error %d; "
800 			    "retrying for %d more second%s\n", fs, dev, error,
801 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
802 		}
803 		pause("rmretry", delay);
804 		timeout -= delay;
805 	}
806  out:
807 	if (error) {
808 		printf("Mounting from %s:%s failed with error %d",
809 		    fs, dev, error);
810 		if (errmsg[0] != '\0')
811 			printf(": %s", errmsg);
812 		printf(".\n");
813 	}
814 	free(fs, M_TEMP);
815 	free(errmsg, M_TEMP);
816 	if (opts != NULL)
817 		free(opts, M_TEMP);
818 	/* kernel_mount can return -1 on error. */
819 	return ((error < 0) ? EDOOFUS : error);
820 }
821 #undef ERRMSGL
822 
823 static int
824 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
825 {
826 	struct mount *mp;
827 	char *conf;
828 	int error;
829 
830 	root_mount_mddev = -1;
831 
832 retry:
833 	conf = sbuf_data(sb);
834 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
835 	error = (mp == NULL) ? 0 : EDOOFUS;
836 	root_mount_onfail = A_CONTINUE;
837 	while (mp == NULL) {
838 		error = parse_skipto(&conf, CC_NONWHITESPACE);
839 		if (error == PE_EOL) {
840 			parse_advance(&conf);
841 			continue;
842 		}
843 		if (error < 0)
844 			break;
845 		switch (parse_peek(&conf)) {
846 		case '#':
847 			error = parse_skipto(&conf, '\n');
848 			break;
849 		case '.':
850 			error = parse_directive(&conf);
851 			break;
852 		default:
853 			error = parse_mount(&conf);
854 			if (error == -1) {
855 				printf("mountroot: invalid file system "
856 				    "specification.\n");
857 				error = 0;
858 			}
859 			break;
860 		}
861 		if (error < 0)
862 			break;
863 		/* Ignore any trailing garbage on the line. */
864 		if (parse_peek(&conf) != '\n') {
865 			printf("mountroot: advancing to next directive...\n");
866 			(void)parse_skipto(&conf, '\n');
867 		}
868 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
869 	}
870 	if (mp != NULL)
871 		return (0);
872 
873 	/*
874 	 * We failed to mount (a new) root.
875 	 */
876 	switch (root_mount_onfail) {
877 	case A_CONTINUE:
878 		break;
879 	case A_PANIC:
880 		panic("mountroot: unable to (re-)mount root.");
881 		/* NOTREACHED */
882 	case A_RETRY:
883 		goto retry;
884 	case A_REBOOT:
885 		kern_reboot(RB_NOSYNC);
886 		/* NOTREACHED */
887 	}
888 
889 	return (error);
890 }
891 
892 static void
893 vfs_mountroot_conf0(struct sbuf *sb)
894 {
895 	char *s, *tok, *mnt, *opt;
896 	int error;
897 
898 	sbuf_printf(sb, ".onfail panic\n");
899 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
900 	if (boothowto & RB_ASKNAME)
901 		sbuf_printf(sb, ".ask\n");
902 #ifdef ROOTDEVNAME
903 	if (boothowto & RB_DFLTROOT)
904 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
905 #endif
906 	if (boothowto & RB_CDROM) {
907 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
908 		sbuf_printf(sb, ".timeout 0\n");
909 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
910 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
911 	}
912 	s = kern_getenv("vfs.root.mountfrom");
913 	if (s != NULL) {
914 		opt = kern_getenv("vfs.root.mountfrom.options");
915 		tok = s;
916 		error = parse_token(&tok, &mnt);
917 		while (!error) {
918 			sbuf_printf(sb, "%s %s\n", mnt,
919 			    (opt != NULL) ? opt : "");
920 			free(mnt, M_TEMP);
921 			error = parse_token(&tok, &mnt);
922 		}
923 		if (opt != NULL)
924 			freeenv(opt);
925 		freeenv(s);
926 	}
927 	if (rootdevnames[0] != NULL)
928 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
929 	if (rootdevnames[1] != NULL)
930 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
931 #ifdef ROOTDEVNAME
932 	if (!(boothowto & RB_DFLTROOT))
933 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
934 #endif
935 	if (!(boothowto & RB_ASKNAME))
936 		sbuf_printf(sb, ".ask\n");
937 }
938 
939 static int
940 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
941 {
942 	static char buf[128];
943 	struct nameidata nd;
944 	off_t ofs;
945 	ssize_t resid;
946 	int error, flags, len;
947 
948 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
949 	flags = FREAD;
950 	error = vn_open(&nd, &flags, 0, NULL);
951 	if (error)
952 		return (error);
953 
954 	NDFREE(&nd, NDF_ONLY_PNBUF);
955 	ofs = 0;
956 	len = sizeof(buf) - 1;
957 	while (1) {
958 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
959 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
960 		    NOCRED, &resid, td);
961 		if (error)
962 			break;
963 		if (resid == len)
964 			break;
965 		buf[len - resid] = 0;
966 		sbuf_printf(sb, "%s", buf);
967 		ofs += len - resid;
968 	}
969 
970 	VOP_UNLOCK(nd.ni_vp);
971 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
972 	return (error);
973 }
974 
975 static void
976 vfs_mountroot_wait(void)
977 {
978 	struct root_hold_token *h;
979 	struct timeval lastfail;
980 	int curfail;
981 
982 	TSENTER();
983 
984 	curfail = 0;
985 	while (1) {
986 		g_waitidle();
987 		mtx_lock(&root_holds_mtx);
988 		if (TAILQ_EMPTY(&root_holds)) {
989 			mtx_unlock(&root_holds_mtx);
990 			break;
991 		}
992 		if (ppsratecheck(&lastfail, &curfail, 1)) {
993 			printf("Root mount waiting for:");
994 			TAILQ_FOREACH(h, &root_holds, list)
995 				printf(" %s", h->who);
996 			printf("\n");
997 		}
998 		TSWAIT("root mount");
999 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
1000 		    hz);
1001 		TSUNWAIT("root mount");
1002 	}
1003 
1004 	TSEXIT();
1005 }
1006 
1007 static int
1008 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1009 {
1010 	int delay, timeout;
1011 
1012 	/*
1013 	 * In case of ZFS and NFS we don't have a way to wait for
1014 	 * specific device.  Also do the wait if the user forced that
1015 	 * behaviour by setting vfs.root_mount_always_wait=1.
1016 	 */
1017 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1018 	    dev[0] == '\0' || root_mount_always_wait != 0) {
1019 		vfs_mountroot_wait();
1020 		return (0);
1021 	}
1022 
1023 	/*
1024 	 * Otherwise, no point in waiting if the device is already there.
1025 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1026 	 * eg for geom_part(4) to finish tasting.
1027 	 */
1028 	g_waitidle();
1029 	if (parse_mount_dev_present(dev))
1030 		return (0);
1031 
1032 	/*
1033 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1034 	 * to behave exactly as it used to work before.
1035 	 */
1036 	vfs_mountroot_wait();
1037 	printf("mountroot: waiting for device %s...\n", dev);
1038 	delay = hz / 10;
1039 	timeout = root_mount_timeout * hz;
1040 	do {
1041 		pause("rmdev", delay);
1042 		timeout -= delay;
1043 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1044 
1045 	if (timeout <= 0)
1046 		return (ENODEV);
1047 
1048 	return (0);
1049 }
1050 
1051 void
1052 vfs_mountroot(void)
1053 {
1054 	struct mount *mp;
1055 	struct sbuf *sb;
1056 	struct thread *td;
1057 	time_t timebase;
1058 	int error;
1059 
1060 	mtx_assert(&Giant, MA_NOTOWNED);
1061 
1062 	TSENTER();
1063 
1064 	td = curthread;
1065 
1066 	sb = sbuf_new_auto();
1067 	vfs_mountroot_conf0(sb);
1068 	sbuf_finish(sb);
1069 
1070 	error = vfs_mountroot_devfs(td, &mp);
1071 	while (!error) {
1072 		error = vfs_mountroot_parse(sb, mp);
1073 		if (!error) {
1074 			vfs_mountroot_shuffle(td, mp);
1075 			sbuf_clear(sb);
1076 			error = vfs_mountroot_readconf(td, sb);
1077 			sbuf_finish(sb);
1078 		}
1079 	}
1080 
1081 	sbuf_delete(sb);
1082 
1083 	/*
1084 	 * Iterate over all currently mounted file systems and use
1085 	 * the time stamp found to check and/or initialize the RTC.
1086 	 * Call inittodr() only once and pass it the largest of the
1087 	 * timestamps we encounter.
1088 	 */
1089 	timebase = 0;
1090 	mtx_lock(&mountlist_mtx);
1091 	mp = TAILQ_FIRST(&mountlist);
1092 	while (mp != NULL) {
1093 		if (mp->mnt_time > timebase)
1094 			timebase = mp->mnt_time;
1095 		mp = TAILQ_NEXT(mp, mnt_list);
1096 	}
1097 	mtx_unlock(&mountlist_mtx);
1098 	inittodr(timebase);
1099 
1100 	/* Keep prison0's root in sync with the global rootvnode. */
1101 	mtx_lock(&prison0.pr_mtx);
1102 	prison0.pr_root = rootvnode;
1103 	vref(prison0.pr_root);
1104 	mtx_unlock(&prison0.pr_mtx);
1105 
1106 	mtx_lock(&root_holds_mtx);
1107 	atomic_store_rel_int(&root_mount_complete, 1);
1108 	wakeup(&root_mount_complete);
1109 	mtx_unlock(&root_holds_mtx);
1110 
1111 	EVENTHANDLER_INVOKE(mountroot);
1112 
1113 	TSEXIT();
1114 }
1115 
1116 static struct mntarg *
1117 parse_mountroot_options(struct mntarg *ma, const char *options)
1118 {
1119 	char *p;
1120 	char *name, *name_arg;
1121 	char *val, *val_arg;
1122 	char *opts;
1123 
1124 	if (options == NULL || options[0] == '\0')
1125 		return (ma);
1126 
1127 	p = opts = strdup(options, M_MOUNT);
1128 	if (opts == NULL) {
1129 		return (ma);
1130 	}
1131 
1132 	while((name = strsep(&p, ",")) != NULL) {
1133 		if (name[0] == '\0')
1134 			break;
1135 
1136 		val = strchr(name, '=');
1137 		if (val != NULL) {
1138 			*val = '\0';
1139 			++val;
1140 		}
1141 		if (strcmp(name, "rw") == 0 || strcmp(name, "noro") == 0) {
1142 			/*
1143 			 * The first time we mount the root file system,
1144 			 * we need to mount 'ro', so We need to ignore
1145 			 * 'rw' and 'noro' mount options.
1146 			 */
1147 			continue;
1148 		}
1149 		name_arg = strdup(name, M_MOUNT);
1150 		val_arg = NULL;
1151 		if (val != NULL)
1152 			val_arg = strdup(val, M_MOUNT);
1153 
1154 		ma = mount_arg(ma, name_arg, val_arg,
1155 		    (val_arg != NULL ? -1 : 0));
1156 	}
1157 	free(opts, M_MOUNT);
1158 	return (ma);
1159 }
1160