xref: /freebsd/sys/kern/vfs_mountroot.c (revision fe6060f10f634930ff71b7c50291ddc610da2475)
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 				VI_LOCK(vp);
371 				mporoot->mnt_vnodecovered = vp;
372 				vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
373 				vp->v_mountedhere = mporoot;
374 				strlcpy(mporoot->mnt_stat.f_mntonname,
375 				    fspath, MNAMELEN);
376 				VI_UNLOCK(vp);
377 				VOP_UNLOCK(vp);
378 			} else
379 				vput(vp);
380 		}
381 		NDFREE(&nd, NDF_ONLY_PNBUF);
382 
383 		if (error)
384 			printf("mountroot: unable to remount previous root "
385 			    "under /.mount or /mnt (error %d)\n", error);
386 	}
387 
388 	/* Remount devfs under /dev */
389 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
390 	error = namei(&nd);
391 	if (!error) {
392 		vp = nd.ni_vp;
393 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
394 		if (!error)
395 			error = vinvalbuf(vp, V_SAVE, 0, 0);
396 		if (!error) {
397 			vpdevfs = mpdevfs->mnt_vnodecovered;
398 			if (vpdevfs != NULL) {
399 				cache_purge(vpdevfs);
400 				VI_LOCK(vpdevfs);
401 				vn_irflag_unset_locked(vpdevfs, VIRF_MOUNTPOINT);
402 				vpdevfs->v_mountedhere = NULL;
403 				VI_UNLOCK(vpdevfs);
404 				vrele(vpdevfs);
405 			}
406 			VI_LOCK(vp);
407 			mpdevfs->mnt_vnodecovered = vp;
408 			vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
409 			vp->v_mountedhere = mpdevfs;
410 			VI_UNLOCK(vp);
411 			VOP_UNLOCK(vp);
412 		} else
413 			vput(vp);
414 	}
415 	if (error)
416 		printf("mountroot: unable to remount devfs under /dev "
417 		    "(error %d)\n", error);
418 	NDFREE(&nd, NDF_ONLY_PNBUF);
419 
420 	if (mporoot == mpdevfs) {
421 		vfs_unbusy(mpdevfs);
422 		/* Unlink the no longer needed /dev/dev -> / symlink */
423 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
424 		    UIO_SYSSPACE, 0, 0);
425 		if (error)
426 			printf("mountroot: unable to unlink /dev/dev "
427 			    "(error %d)\n", error);
428 	}
429 }
430 
431 /*
432  * Configuration parser.
433  */
434 
435 /* Parser character classes. */
436 #define	CC_WHITESPACE		-1
437 #define	CC_NONWHITESPACE	-2
438 
439 /* Parse errors. */
440 #define	PE_EOF			-1
441 #define	PE_EOL			-2
442 
443 static __inline int
444 parse_peek(char **conf)
445 {
446 
447 	return (**conf);
448 }
449 
450 static __inline void
451 parse_poke(char **conf, int c)
452 {
453 
454 	**conf = c;
455 }
456 
457 static __inline void
458 parse_advance(char **conf)
459 {
460 
461 	(*conf)++;
462 }
463 
464 static int
465 parse_skipto(char **conf, int mc)
466 {
467 	int c, match;
468 
469 	while (1) {
470 		c = parse_peek(conf);
471 		if (c == 0)
472 			return (PE_EOF);
473 		switch (mc) {
474 		case CC_WHITESPACE:
475 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
476 			break;
477 		case CC_NONWHITESPACE:
478 			if (c == '\n')
479 				return (PE_EOL);
480 			match = (c != ' ' && c != '\t') ? 1 : 0;
481 			break;
482 		default:
483 			match = (c == mc) ? 1 : 0;
484 			break;
485 		}
486 		if (match)
487 			break;
488 		parse_advance(conf);
489 	}
490 	return (0);
491 }
492 
493 static int
494 parse_token(char **conf, char **tok)
495 {
496 	char *p;
497 	size_t len;
498 	int error;
499 
500 	*tok = NULL;
501 	error = parse_skipto(conf, CC_NONWHITESPACE);
502 	if (error)
503 		return (error);
504 	p = *conf;
505 	error = parse_skipto(conf, CC_WHITESPACE);
506 	len = *conf - p;
507 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
508 	bcopy(p, *tok, len);
509 	return (0);
510 }
511 
512 static void
513 parse_dir_ask_printenv(const char *var)
514 {
515 	char *val;
516 
517 	val = kern_getenv(var);
518 	if (val != NULL) {
519 		printf("  %s=%s\n", var, val);
520 		freeenv(val);
521 	}
522 }
523 
524 static int
525 parse_dir_ask(char **conf)
526 {
527 	char name[80];
528 	char *mnt;
529 	int error;
530 
531 	vfs_mountroot_wait();
532 
533 	printf("\nLoader variables:\n");
534 	parse_dir_ask_printenv("vfs.root.mountfrom");
535 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
536 
537 	printf("\nManual root filesystem specification:\n");
538 	printf("  <fstype>:<device> [options]\n");
539 	printf("      Mount <device> using filesystem <fstype>\n");
540 	printf("      and with the specified (optional) option list.\n");
541 	printf("\n");
542 	printf("    eg. ufs:/dev/da0s1a\n");
543 	printf("        zfs:zroot/ROOT/default\n");
544 	printf("        cd9660:/dev/cd0 ro\n");
545 	printf("          (which is equivalent to: ");
546 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
547 	printf("\n");
548 	printf("  ?               List valid disk boot devices\n");
549 	printf("  .               Yield 1 second (for background tasks)\n");
550 	printf("  <empty line>    Abort manual input\n");
551 
552 	do {
553 		error = EINVAL;
554 		printf("\nmountroot> ");
555 		cngets(name, sizeof(name), GETS_ECHO);
556 		if (name[0] == '\0')
557 			break;
558 		if (name[0] == '?' && name[1] == '\0') {
559 			printf("\nList of GEOM managed disk devices:\n  ");
560 			g_dev_print();
561 			continue;
562 		}
563 		if (name[0] == '.' && name[1] == '\0') {
564 			pause("rmask", hz);
565 			continue;
566 		}
567 		mnt = name;
568 		error = parse_mount(&mnt);
569 		if (error == -1)
570 			printf("Invalid file system specification.\n");
571 	} while (error != 0);
572 
573 	return (error);
574 }
575 
576 static int
577 parse_dir_md(char **conf)
578 {
579 	struct stat sb;
580 	struct thread *td;
581 	struct md_ioctl *mdio;
582 	char *path, *tok;
583 	int error, fd, len;
584 
585 	td = curthread;
586 	fd = -1;
587 
588 	error = parse_token(conf, &tok);
589 	if (error)
590 		return (error);
591 
592 	len = strlen(tok);
593 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
594 	path = (void *)(mdio + 1);
595 	bcopy(tok, path, len);
596 	free(tok, M_TEMP);
597 
598 	/* Get file status. */
599 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
600 	if (error)
601 		goto out;
602 
603 	/* Open /dev/mdctl so that we can attach/detach. */
604 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
605 	    O_RDWR, 0);
606 	if (error)
607 		goto out;
608 
609 	fd = td->td_retval[0];
610 	mdio->md_version = MDIOVERSION;
611 	mdio->md_type = MD_VNODE;
612 
613 	if (root_mount_mddev != -1) {
614 		mdio->md_unit = root_mount_mddev;
615 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
616 		/* Ignore errors. We don't care. */
617 		root_mount_mddev = -1;
618 	}
619 
620 	mdio->md_file = (void *)(mdio + 1);
621 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
622 	mdio->md_mediasize = sb.st_size;
623 	mdio->md_unit = 0;
624 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
625 	if (error)
626 		goto out;
627 
628 	if (mdio->md_unit > 9) {
629 		printf("rootmount: too many md units\n");
630 		mdio->md_file = NULL;
631 		mdio->md_options = 0;
632 		mdio->md_mediasize = 0;
633 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
634 		/* Ignore errors. We don't care. */
635 		error = ERANGE;
636 		goto out;
637 	}
638 
639 	root_mount_mddev = mdio->md_unit;
640 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
641 
642  out:
643 	if (fd >= 0)
644 		(void)kern_close(td, fd);
645 	free(mdio, M_TEMP);
646 	return (error);
647 }
648 
649 static int
650 parse_dir_onfail(char **conf)
651 {
652 	char *action;
653 	int error;
654 
655 	error = parse_token(conf, &action);
656 	if (error)
657 		return (error);
658 
659 	if (!strcmp(action, "continue"))
660 		root_mount_onfail = A_CONTINUE;
661 	else if (!strcmp(action, "panic"))
662 		root_mount_onfail = A_PANIC;
663 	else if (!strcmp(action, "reboot"))
664 		root_mount_onfail = A_REBOOT;
665 	else if (!strcmp(action, "retry"))
666 		root_mount_onfail = A_RETRY;
667 	else {
668 		printf("rootmount: %s: unknown action\n", action);
669 		error = EINVAL;
670 	}
671 
672 	free(action, M_TEMP);
673 	return (0);
674 }
675 
676 static int
677 parse_dir_timeout(char **conf)
678 {
679 	char *tok, *endtok;
680 	long secs;
681 	int error;
682 
683 	error = parse_token(conf, &tok);
684 	if (error)
685 		return (error);
686 
687 	secs = strtol(tok, &endtok, 0);
688 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
689 	if (!error)
690 		root_mount_timeout = secs;
691 	free(tok, M_TEMP);
692 	return (error);
693 }
694 
695 static int
696 parse_directive(char **conf)
697 {
698 	char *dir;
699 	int error;
700 
701 	error = parse_token(conf, &dir);
702 	if (error)
703 		return (error);
704 
705 	if (strcmp(dir, ".ask") == 0)
706 		error = parse_dir_ask(conf);
707 	else if (strcmp(dir, ".md") == 0)
708 		error = parse_dir_md(conf);
709 	else if (strcmp(dir, ".onfail") == 0)
710 		error = parse_dir_onfail(conf);
711 	else if (strcmp(dir, ".timeout") == 0)
712 		error = parse_dir_timeout(conf);
713 	else {
714 		printf("mountroot: invalid directive `%s'\n", dir);
715 		/* Ignore the rest of the line. */
716 		(void)parse_skipto(conf, '\n');
717 		error = EINVAL;
718 	}
719 	free(dir, M_TEMP);
720 	return (error);
721 }
722 
723 static int
724 parse_mount_dev_present(const char *dev)
725 {
726 	struct nameidata nd;
727 	int error;
728 
729 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
730 	error = namei(&nd);
731 	if (!error)
732 		vput(nd.ni_vp);
733 	NDFREE(&nd, NDF_ONLY_PNBUF);
734 	return (error != 0) ? 0 : 1;
735 }
736 
737 #define	ERRMSGL	255
738 static int
739 parse_mount(char **conf)
740 {
741 	char *errmsg;
742 	struct mntarg *ma;
743 	char *dev, *fs, *opts, *tok;
744 	int delay, error, timeout;
745 
746 	error = parse_token(conf, &tok);
747 	if (error)
748 		return (error);
749 	fs = tok;
750 	error = parse_skipto(&tok, ':');
751 	if (error) {
752 		free(fs, M_TEMP);
753 		return (error);
754 	}
755 	parse_poke(&tok, '\0');
756 	parse_advance(&tok);
757 	dev = tok;
758 
759 	if (root_mount_mddev != -1) {
760 		/* Handle substitution for the md unit number. */
761 		tok = strstr(dev, "md#");
762 		if (tok != NULL)
763 			tok[2] = '0' + root_mount_mddev;
764 	}
765 
766 	/* Parse options. */
767 	error = parse_token(conf, &tok);
768 	opts = (error == 0) ? tok : NULL;
769 
770 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
771 	    (opts != NULL) ? opts : "");
772 
773 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
774 
775 	if (vfs_byname(fs) == NULL) {
776 		strlcpy(errmsg, "unknown file system", ERRMSGL);
777 		error = ENOENT;
778 		goto out;
779 	}
780 
781 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
782 	if (error != 0)
783 		goto out;
784 
785 	delay = hz / 10;
786 	timeout = root_mount_timeout * hz;
787 
788 	for (;;) {
789 		ma = NULL;
790 		ma = mount_arg(ma, "fstype", fs, -1);
791 		ma = mount_arg(ma, "fspath", "/", -1);
792 		ma = mount_arg(ma, "from", dev, -1);
793 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
794 		ma = mount_arg(ma, "ro", NULL, 0);
795 		ma = parse_mountroot_options(ma, opts);
796 
797 		error = kernel_mount(ma, MNT_ROOTFS);
798 		if (error == 0 || timeout <= 0)
799 			break;
800 
801 		if (root_mount_timeout * hz == timeout ||
802 		    (bootverbose && timeout % hz == 0)) {
803 			printf("Mounting from %s:%s failed with error %d; "
804 			    "retrying for %d more second%s\n", fs, dev, error,
805 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
806 		}
807 		pause("rmretry", delay);
808 		timeout -= delay;
809 	}
810  out:
811 	if (error) {
812 		printf("Mounting from %s:%s failed with error %d",
813 		    fs, dev, error);
814 		if (errmsg[0] != '\0')
815 			printf(": %s", errmsg);
816 		printf(".\n");
817 	}
818 	free(fs, M_TEMP);
819 	free(errmsg, M_TEMP);
820 	if (opts != NULL)
821 		free(opts, M_TEMP);
822 	/* kernel_mount can return -1 on error. */
823 	return ((error < 0) ? EDOOFUS : error);
824 }
825 #undef ERRMSGL
826 
827 static int
828 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
829 {
830 	struct mount *mp;
831 	char *conf;
832 	int error;
833 
834 	root_mount_mddev = -1;
835 
836 retry:
837 	conf = sbuf_data(sb);
838 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
839 	error = (mp == NULL) ? 0 : EDOOFUS;
840 	root_mount_onfail = A_CONTINUE;
841 	while (mp == NULL) {
842 		error = parse_skipto(&conf, CC_NONWHITESPACE);
843 		if (error == PE_EOL) {
844 			parse_advance(&conf);
845 			continue;
846 		}
847 		if (error < 0)
848 			break;
849 		switch (parse_peek(&conf)) {
850 		case '#':
851 			error = parse_skipto(&conf, '\n');
852 			break;
853 		case '.':
854 			error = parse_directive(&conf);
855 			break;
856 		default:
857 			error = parse_mount(&conf);
858 			if (error == -1) {
859 				printf("mountroot: invalid file system "
860 				    "specification.\n");
861 				error = 0;
862 			}
863 			break;
864 		}
865 		if (error < 0)
866 			break;
867 		/* Ignore any trailing garbage on the line. */
868 		if (parse_peek(&conf) != '\n') {
869 			printf("mountroot: advancing to next directive...\n");
870 			(void)parse_skipto(&conf, '\n');
871 		}
872 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
873 	}
874 	if (mp != NULL)
875 		return (0);
876 
877 	/*
878 	 * We failed to mount (a new) root.
879 	 */
880 	switch (root_mount_onfail) {
881 	case A_CONTINUE:
882 		break;
883 	case A_PANIC:
884 		panic("mountroot: unable to (re-)mount root.");
885 		/* NOTREACHED */
886 	case A_RETRY:
887 		goto retry;
888 	case A_REBOOT:
889 		kern_reboot(RB_NOSYNC);
890 		/* NOTREACHED */
891 	}
892 
893 	return (error);
894 }
895 
896 static void
897 vfs_mountroot_conf0(struct sbuf *sb)
898 {
899 	char *s, *tok, *mnt, *opt;
900 	int error;
901 
902 	sbuf_printf(sb, ".onfail panic\n");
903 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
904 	if (boothowto & RB_ASKNAME)
905 		sbuf_printf(sb, ".ask\n");
906 #ifdef ROOTDEVNAME
907 	if (boothowto & RB_DFLTROOT)
908 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
909 #endif
910 	if (boothowto & RB_CDROM) {
911 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
912 		sbuf_printf(sb, ".timeout 0\n");
913 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
914 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
915 	}
916 	s = kern_getenv("vfs.root.mountfrom");
917 	if (s != NULL) {
918 		opt = kern_getenv("vfs.root.mountfrom.options");
919 		tok = s;
920 		error = parse_token(&tok, &mnt);
921 		while (!error) {
922 			sbuf_printf(sb, "%s %s\n", mnt,
923 			    (opt != NULL) ? opt : "");
924 			free(mnt, M_TEMP);
925 			error = parse_token(&tok, &mnt);
926 		}
927 		if (opt != NULL)
928 			freeenv(opt);
929 		freeenv(s);
930 	}
931 	if (rootdevnames[0] != NULL)
932 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
933 	if (rootdevnames[1] != NULL)
934 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
935 #ifdef ROOTDEVNAME
936 	if (!(boothowto & RB_DFLTROOT))
937 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
938 #endif
939 	if (!(boothowto & RB_ASKNAME))
940 		sbuf_printf(sb, ".ask\n");
941 }
942 
943 static int
944 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
945 {
946 	static char buf[128];
947 	struct nameidata nd;
948 	off_t ofs;
949 	ssize_t resid;
950 	int error, flags, len;
951 
952 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
953 	flags = FREAD;
954 	error = vn_open(&nd, &flags, 0, NULL);
955 	if (error)
956 		return (error);
957 
958 	NDFREE(&nd, NDF_ONLY_PNBUF);
959 	ofs = 0;
960 	len = sizeof(buf) - 1;
961 	while (1) {
962 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
963 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
964 		    NOCRED, &resid, td);
965 		if (error)
966 			break;
967 		if (resid == len)
968 			break;
969 		buf[len - resid] = 0;
970 		sbuf_printf(sb, "%s", buf);
971 		ofs += len - resid;
972 	}
973 
974 	VOP_UNLOCK(nd.ni_vp);
975 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
976 	return (error);
977 }
978 
979 static void
980 vfs_mountroot_wait(void)
981 {
982 	struct root_hold_token *h;
983 	struct timeval lastfail;
984 	int curfail;
985 
986 	TSENTER();
987 
988 	curfail = 0;
989 	lastfail.tv_sec = 0;
990 	while (1) {
991 		g_waitidle();
992 		mtx_lock(&root_holds_mtx);
993 		if (TAILQ_EMPTY(&root_holds)) {
994 			mtx_unlock(&root_holds_mtx);
995 			break;
996 		}
997 		if (ppsratecheck(&lastfail, &curfail, 1)) {
998 			printf("Root mount waiting for:");
999 			TAILQ_FOREACH(h, &root_holds, list)
1000 				printf(" %s", h->who);
1001 			printf("\n");
1002 		}
1003 		TSWAIT("root mount");
1004 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
1005 		    hz);
1006 		TSUNWAIT("root mount");
1007 	}
1008 
1009 	TSEXIT();
1010 }
1011 
1012 static int
1013 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1014 {
1015 	int delay, timeout;
1016 
1017 	/*
1018 	 * In case of ZFS and NFS we don't have a way to wait for
1019 	 * specific device.  Also do the wait if the user forced that
1020 	 * behaviour by setting vfs.root_mount_always_wait=1.
1021 	 */
1022 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1023 	    dev[0] == '\0' || root_mount_always_wait != 0) {
1024 		vfs_mountroot_wait();
1025 		return (0);
1026 	}
1027 
1028 	/*
1029 	 * Otherwise, no point in waiting if the device is already there.
1030 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1031 	 * eg for geom_part(4) to finish tasting.
1032 	 */
1033 	g_waitidle();
1034 	if (parse_mount_dev_present(dev))
1035 		return (0);
1036 
1037 	/*
1038 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1039 	 * to behave exactly as it used to work before.
1040 	 */
1041 	vfs_mountroot_wait();
1042 	printf("mountroot: waiting for device %s...\n", dev);
1043 	delay = hz / 10;
1044 	timeout = root_mount_timeout * hz;
1045 	do {
1046 		pause("rmdev", delay);
1047 		timeout -= delay;
1048 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1049 
1050 	if (timeout <= 0)
1051 		return (ENODEV);
1052 
1053 	return (0);
1054 }
1055 
1056 void
1057 vfs_mountroot(void)
1058 {
1059 	struct mount *mp;
1060 	struct sbuf *sb;
1061 	struct thread *td;
1062 	time_t timebase;
1063 	int error;
1064 
1065 	mtx_assert(&Giant, MA_NOTOWNED);
1066 
1067 	TSENTER();
1068 
1069 	td = curthread;
1070 
1071 	sb = sbuf_new_auto();
1072 	vfs_mountroot_conf0(sb);
1073 	sbuf_finish(sb);
1074 
1075 	error = vfs_mountroot_devfs(td, &mp);
1076 	while (!error) {
1077 		error = vfs_mountroot_parse(sb, mp);
1078 		if (!error) {
1079 			vfs_mountroot_shuffle(td, mp);
1080 			sbuf_clear(sb);
1081 			error = vfs_mountroot_readconf(td, sb);
1082 			sbuf_finish(sb);
1083 		}
1084 	}
1085 
1086 	sbuf_delete(sb);
1087 
1088 	/*
1089 	 * Iterate over all currently mounted file systems and use
1090 	 * the time stamp found to check and/or initialize the RTC.
1091 	 * Call inittodr() only once and pass it the largest of the
1092 	 * timestamps we encounter.
1093 	 */
1094 	timebase = 0;
1095 	mtx_lock(&mountlist_mtx);
1096 	mp = TAILQ_FIRST(&mountlist);
1097 	while (mp != NULL) {
1098 		if (mp->mnt_time > timebase)
1099 			timebase = mp->mnt_time;
1100 		mp = TAILQ_NEXT(mp, mnt_list);
1101 	}
1102 	mtx_unlock(&mountlist_mtx);
1103 	inittodr(timebase);
1104 
1105 	/* Keep prison0's root in sync with the global rootvnode. */
1106 	mtx_lock(&prison0.pr_mtx);
1107 	prison0.pr_root = rootvnode;
1108 	vref(prison0.pr_root);
1109 	mtx_unlock(&prison0.pr_mtx);
1110 
1111 	mtx_lock(&root_holds_mtx);
1112 	atomic_store_rel_int(&root_mount_complete, 1);
1113 	wakeup(&root_mount_complete);
1114 	mtx_unlock(&root_holds_mtx);
1115 
1116 	EVENTHANDLER_INVOKE(mountroot);
1117 
1118 	TSEXIT();
1119 }
1120 
1121 static struct mntarg *
1122 parse_mountroot_options(struct mntarg *ma, const char *options)
1123 {
1124 	char *p;
1125 	char *name, *name_arg;
1126 	char *val, *val_arg;
1127 	char *opts;
1128 
1129 	if (options == NULL || options[0] == '\0')
1130 		return (ma);
1131 
1132 	p = opts = strdup(options, M_MOUNT);
1133 	if (opts == NULL) {
1134 		return (ma);
1135 	}
1136 
1137 	while((name = strsep(&p, ",")) != NULL) {
1138 		if (name[0] == '\0')
1139 			break;
1140 
1141 		val = strchr(name, '=');
1142 		if (val != NULL) {
1143 			*val = '\0';
1144 			++val;
1145 		}
1146 		if (strcmp(name, "rw") == 0 || strcmp(name, "noro") == 0) {
1147 			/*
1148 			 * The first time we mount the root file system,
1149 			 * we need to mount 'ro', so We need to ignore
1150 			 * 'rw' and 'noro' mount options.
1151 			 */
1152 			continue;
1153 		}
1154 		name_arg = strdup(name, M_MOUNT);
1155 		val_arg = NULL;
1156 		if (val != NULL)
1157 			val_arg = strdup(val, M_MOUNT);
1158 
1159 		ma = mount_arg(ma, name_arg, val_arg,
1160 		    (val_arg != NULL ? -1 : 0));
1161 	}
1162 	free(opts, M_MOUNT);
1163 	return (ma);
1164 }
1165