xref: /freebsd/sys/kern/vfs_mountroot.c (revision cab6a39d7b343596a5823e65c0f7b426551ec22d)
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 	fd = -1;
584 
585 	error = parse_token(conf, &tok);
586 	if (error)
587 		return (error);
588 
589 	len = strlen(tok);
590 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
591 	path = (void *)(mdio + 1);
592 	bcopy(tok, path, len);
593 	free(tok, M_TEMP);
594 
595 	/* Get file status. */
596 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
597 	if (error)
598 		goto out;
599 
600 	/* Open /dev/mdctl so that we can attach/detach. */
601 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
602 	    O_RDWR, 0);
603 	if (error)
604 		goto out;
605 
606 	fd = td->td_retval[0];
607 	mdio->md_version = MDIOVERSION;
608 	mdio->md_type = MD_VNODE;
609 
610 	if (root_mount_mddev != -1) {
611 		mdio->md_unit = root_mount_mddev;
612 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
613 		/* Ignore errors. We don't care. */
614 		root_mount_mddev = -1;
615 	}
616 
617 	mdio->md_file = (void *)(mdio + 1);
618 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
619 	mdio->md_mediasize = sb.st_size;
620 	mdio->md_unit = 0;
621 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
622 	if (error)
623 		goto out;
624 
625 	if (mdio->md_unit > 9) {
626 		printf("rootmount: too many md units\n");
627 		mdio->md_file = NULL;
628 		mdio->md_options = 0;
629 		mdio->md_mediasize = 0;
630 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
631 		/* Ignore errors. We don't care. */
632 		error = ERANGE;
633 		goto out;
634 	}
635 
636 	root_mount_mddev = mdio->md_unit;
637 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
638 
639  out:
640 	if (fd >= 0)
641 		(void)kern_close(td, fd);
642 	free(mdio, M_TEMP);
643 	return (error);
644 }
645 
646 static int
647 parse_dir_onfail(char **conf)
648 {
649 	char *action;
650 	int error;
651 
652 	error = parse_token(conf, &action);
653 	if (error)
654 		return (error);
655 
656 	if (!strcmp(action, "continue"))
657 		root_mount_onfail = A_CONTINUE;
658 	else if (!strcmp(action, "panic"))
659 		root_mount_onfail = A_PANIC;
660 	else if (!strcmp(action, "reboot"))
661 		root_mount_onfail = A_REBOOT;
662 	else if (!strcmp(action, "retry"))
663 		root_mount_onfail = A_RETRY;
664 	else {
665 		printf("rootmount: %s: unknown action\n", action);
666 		error = EINVAL;
667 	}
668 
669 	free(action, M_TEMP);
670 	return (0);
671 }
672 
673 static int
674 parse_dir_timeout(char **conf)
675 {
676 	char *tok, *endtok;
677 	long secs;
678 	int error;
679 
680 	error = parse_token(conf, &tok);
681 	if (error)
682 		return (error);
683 
684 	secs = strtol(tok, &endtok, 0);
685 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
686 	if (!error)
687 		root_mount_timeout = secs;
688 	free(tok, M_TEMP);
689 	return (error);
690 }
691 
692 static int
693 parse_directive(char **conf)
694 {
695 	char *dir;
696 	int error;
697 
698 	error = parse_token(conf, &dir);
699 	if (error)
700 		return (error);
701 
702 	if (strcmp(dir, ".ask") == 0)
703 		error = parse_dir_ask(conf);
704 	else if (strcmp(dir, ".md") == 0)
705 		error = parse_dir_md(conf);
706 	else if (strcmp(dir, ".onfail") == 0)
707 		error = parse_dir_onfail(conf);
708 	else if (strcmp(dir, ".timeout") == 0)
709 		error = parse_dir_timeout(conf);
710 	else {
711 		printf("mountroot: invalid directive `%s'\n", dir);
712 		/* Ignore the rest of the line. */
713 		(void)parse_skipto(conf, '\n');
714 		error = EINVAL;
715 	}
716 	free(dir, M_TEMP);
717 	return (error);
718 }
719 
720 static int
721 parse_mount_dev_present(const char *dev)
722 {
723 	struct nameidata nd;
724 	int error;
725 
726 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
727 	error = namei(&nd);
728 	if (!error)
729 		vput(nd.ni_vp);
730 	NDFREE(&nd, NDF_ONLY_PNBUF);
731 	return (error != 0) ? 0 : 1;
732 }
733 
734 #define	ERRMSGL	255
735 static int
736 parse_mount(char **conf)
737 {
738 	char *errmsg;
739 	struct mntarg *ma;
740 	char *dev, *fs, *opts, *tok;
741 	int delay, error, timeout;
742 
743 	error = parse_token(conf, &tok);
744 	if (error)
745 		return (error);
746 	fs = tok;
747 	error = parse_skipto(&tok, ':');
748 	if (error) {
749 		free(fs, M_TEMP);
750 		return (error);
751 	}
752 	parse_poke(&tok, '\0');
753 	parse_advance(&tok);
754 	dev = tok;
755 
756 	if (root_mount_mddev != -1) {
757 		/* Handle substitution for the md unit number. */
758 		tok = strstr(dev, "md#");
759 		if (tok != NULL)
760 			tok[2] = '0' + root_mount_mddev;
761 	}
762 
763 	/* Parse options. */
764 	error = parse_token(conf, &tok);
765 	opts = (error == 0) ? tok : NULL;
766 
767 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
768 	    (opts != NULL) ? opts : "");
769 
770 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
771 
772 	if (vfs_byname(fs) == NULL) {
773 		strlcpy(errmsg, "unknown file system", ERRMSGL);
774 		error = ENOENT;
775 		goto out;
776 	}
777 
778 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
779 	if (error != 0)
780 		goto out;
781 
782 	delay = hz / 10;
783 	timeout = root_mount_timeout * hz;
784 
785 	for (;;) {
786 		ma = NULL;
787 		ma = mount_arg(ma, "fstype", fs, -1);
788 		ma = mount_arg(ma, "fspath", "/", -1);
789 		ma = mount_arg(ma, "from", dev, -1);
790 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
791 		ma = mount_arg(ma, "ro", NULL, 0);
792 		ma = parse_mountroot_options(ma, opts);
793 
794 		error = kernel_mount(ma, MNT_ROOTFS);
795 		if (error == 0 || timeout <= 0)
796 			break;
797 
798 		if (root_mount_timeout * hz == timeout ||
799 		    (bootverbose && timeout % hz == 0)) {
800 			printf("Mounting from %s:%s failed with error %d; "
801 			    "retrying for %d more second%s\n", fs, dev, error,
802 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
803 		}
804 		pause("rmretry", delay);
805 		timeout -= delay;
806 	}
807  out:
808 	if (error) {
809 		printf("Mounting from %s:%s failed with error %d",
810 		    fs, dev, error);
811 		if (errmsg[0] != '\0')
812 			printf(": %s", errmsg);
813 		printf(".\n");
814 	}
815 	free(fs, M_TEMP);
816 	free(errmsg, M_TEMP);
817 	if (opts != NULL)
818 		free(opts, M_TEMP);
819 	/* kernel_mount can return -1 on error. */
820 	return ((error < 0) ? EDOOFUS : error);
821 }
822 #undef ERRMSGL
823 
824 static int
825 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
826 {
827 	struct mount *mp;
828 	char *conf;
829 	int error;
830 
831 	root_mount_mddev = -1;
832 
833 retry:
834 	conf = sbuf_data(sb);
835 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
836 	error = (mp == NULL) ? 0 : EDOOFUS;
837 	root_mount_onfail = A_CONTINUE;
838 	while (mp == NULL) {
839 		error = parse_skipto(&conf, CC_NONWHITESPACE);
840 		if (error == PE_EOL) {
841 			parse_advance(&conf);
842 			continue;
843 		}
844 		if (error < 0)
845 			break;
846 		switch (parse_peek(&conf)) {
847 		case '#':
848 			error = parse_skipto(&conf, '\n');
849 			break;
850 		case '.':
851 			error = parse_directive(&conf);
852 			break;
853 		default:
854 			error = parse_mount(&conf);
855 			if (error == -1) {
856 				printf("mountroot: invalid file system "
857 				    "specification.\n");
858 				error = 0;
859 			}
860 			break;
861 		}
862 		if (error < 0)
863 			break;
864 		/* Ignore any trailing garbage on the line. */
865 		if (parse_peek(&conf) != '\n') {
866 			printf("mountroot: advancing to next directive...\n");
867 			(void)parse_skipto(&conf, '\n');
868 		}
869 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
870 	}
871 	if (mp != NULL)
872 		return (0);
873 
874 	/*
875 	 * We failed to mount (a new) root.
876 	 */
877 	switch (root_mount_onfail) {
878 	case A_CONTINUE:
879 		break;
880 	case A_PANIC:
881 		panic("mountroot: unable to (re-)mount root.");
882 		/* NOTREACHED */
883 	case A_RETRY:
884 		goto retry;
885 	case A_REBOOT:
886 		kern_reboot(RB_NOSYNC);
887 		/* NOTREACHED */
888 	}
889 
890 	return (error);
891 }
892 
893 static void
894 vfs_mountroot_conf0(struct sbuf *sb)
895 {
896 	char *s, *tok, *mnt, *opt;
897 	int error;
898 
899 	sbuf_printf(sb, ".onfail panic\n");
900 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
901 	if (boothowto & RB_ASKNAME)
902 		sbuf_printf(sb, ".ask\n");
903 #ifdef ROOTDEVNAME
904 	if (boothowto & RB_DFLTROOT)
905 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
906 #endif
907 	if (boothowto & RB_CDROM) {
908 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
909 		sbuf_printf(sb, ".timeout 0\n");
910 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
911 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
912 	}
913 	s = kern_getenv("vfs.root.mountfrom");
914 	if (s != NULL) {
915 		opt = kern_getenv("vfs.root.mountfrom.options");
916 		tok = s;
917 		error = parse_token(&tok, &mnt);
918 		while (!error) {
919 			sbuf_printf(sb, "%s %s\n", mnt,
920 			    (opt != NULL) ? opt : "");
921 			free(mnt, M_TEMP);
922 			error = parse_token(&tok, &mnt);
923 		}
924 		if (opt != NULL)
925 			freeenv(opt);
926 		freeenv(s);
927 	}
928 	if (rootdevnames[0] != NULL)
929 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
930 	if (rootdevnames[1] != NULL)
931 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
932 #ifdef ROOTDEVNAME
933 	if (!(boothowto & RB_DFLTROOT))
934 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
935 #endif
936 	if (!(boothowto & RB_ASKNAME))
937 		sbuf_printf(sb, ".ask\n");
938 }
939 
940 static int
941 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
942 {
943 	static char buf[128];
944 	struct nameidata nd;
945 	off_t ofs;
946 	ssize_t resid;
947 	int error, flags, len;
948 
949 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
950 	flags = FREAD;
951 	error = vn_open(&nd, &flags, 0, NULL);
952 	if (error)
953 		return (error);
954 
955 	NDFREE(&nd, NDF_ONLY_PNBUF);
956 	ofs = 0;
957 	len = sizeof(buf) - 1;
958 	while (1) {
959 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
960 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
961 		    NOCRED, &resid, td);
962 		if (error)
963 			break;
964 		if (resid == len)
965 			break;
966 		buf[len - resid] = 0;
967 		sbuf_printf(sb, "%s", buf);
968 		ofs += len - resid;
969 	}
970 
971 	VOP_UNLOCK(nd.ni_vp);
972 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
973 	return (error);
974 }
975 
976 static void
977 vfs_mountroot_wait(void)
978 {
979 	struct root_hold_token *h;
980 	struct timeval lastfail;
981 	int curfail;
982 
983 	TSENTER();
984 
985 	curfail = 0;
986 	lastfail.tv_sec = 0;
987 	while (1) {
988 		g_waitidle();
989 		mtx_lock(&root_holds_mtx);
990 		if (TAILQ_EMPTY(&root_holds)) {
991 			mtx_unlock(&root_holds_mtx);
992 			break;
993 		}
994 		if (ppsratecheck(&lastfail, &curfail, 1)) {
995 			printf("Root mount waiting for:");
996 			TAILQ_FOREACH(h, &root_holds, list)
997 				printf(" %s", h->who);
998 			printf("\n");
999 		}
1000 		TSWAIT("root mount");
1001 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
1002 		    hz);
1003 		TSUNWAIT("root mount");
1004 	}
1005 
1006 	TSEXIT();
1007 }
1008 
1009 static int
1010 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1011 {
1012 	int delay, timeout;
1013 
1014 	/*
1015 	 * In case of ZFS and NFS we don't have a way to wait for
1016 	 * specific device.  Also do the wait if the user forced that
1017 	 * behaviour by setting vfs.root_mount_always_wait=1.
1018 	 */
1019 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1020 	    dev[0] == '\0' || root_mount_always_wait != 0) {
1021 		vfs_mountroot_wait();
1022 		return (0);
1023 	}
1024 
1025 	/*
1026 	 * Otherwise, no point in waiting if the device is already there.
1027 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1028 	 * eg for geom_part(4) to finish tasting.
1029 	 */
1030 	g_waitidle();
1031 	if (parse_mount_dev_present(dev))
1032 		return (0);
1033 
1034 	/*
1035 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1036 	 * to behave exactly as it used to work before.
1037 	 */
1038 	vfs_mountroot_wait();
1039 	printf("mountroot: waiting for device %s...\n", dev);
1040 	delay = hz / 10;
1041 	timeout = root_mount_timeout * hz;
1042 	do {
1043 		pause("rmdev", delay);
1044 		timeout -= delay;
1045 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1046 
1047 	if (timeout <= 0)
1048 		return (ENODEV);
1049 
1050 	return (0);
1051 }
1052 
1053 void
1054 vfs_mountroot(void)
1055 {
1056 	struct mount *mp;
1057 	struct sbuf *sb;
1058 	struct thread *td;
1059 	time_t timebase;
1060 	int error;
1061 
1062 	mtx_assert(&Giant, MA_NOTOWNED);
1063 
1064 	TSENTER();
1065 
1066 	td = curthread;
1067 
1068 	sb = sbuf_new_auto();
1069 	vfs_mountroot_conf0(sb);
1070 	sbuf_finish(sb);
1071 
1072 	error = vfs_mountroot_devfs(td, &mp);
1073 	while (!error) {
1074 		error = vfs_mountroot_parse(sb, mp);
1075 		if (!error) {
1076 			vfs_mountroot_shuffle(td, mp);
1077 			sbuf_clear(sb);
1078 			error = vfs_mountroot_readconf(td, sb);
1079 			sbuf_finish(sb);
1080 		}
1081 	}
1082 
1083 	sbuf_delete(sb);
1084 
1085 	/*
1086 	 * Iterate over all currently mounted file systems and use
1087 	 * the time stamp found to check and/or initialize the RTC.
1088 	 * Call inittodr() only once and pass it the largest of the
1089 	 * timestamps we encounter.
1090 	 */
1091 	timebase = 0;
1092 	mtx_lock(&mountlist_mtx);
1093 	mp = TAILQ_FIRST(&mountlist);
1094 	while (mp != NULL) {
1095 		if (mp->mnt_time > timebase)
1096 			timebase = mp->mnt_time;
1097 		mp = TAILQ_NEXT(mp, mnt_list);
1098 	}
1099 	mtx_unlock(&mountlist_mtx);
1100 	inittodr(timebase);
1101 
1102 	/* Keep prison0's root in sync with the global rootvnode. */
1103 	mtx_lock(&prison0.pr_mtx);
1104 	prison0.pr_root = rootvnode;
1105 	vref(prison0.pr_root);
1106 	mtx_unlock(&prison0.pr_mtx);
1107 
1108 	mtx_lock(&root_holds_mtx);
1109 	atomic_store_rel_int(&root_mount_complete, 1);
1110 	wakeup(&root_mount_complete);
1111 	mtx_unlock(&root_holds_mtx);
1112 
1113 	EVENTHANDLER_INVOKE(mountroot);
1114 
1115 	TSEXIT();
1116 }
1117 
1118 static struct mntarg *
1119 parse_mountroot_options(struct mntarg *ma, const char *options)
1120 {
1121 	char *p;
1122 	char *name, *name_arg;
1123 	char *val, *val_arg;
1124 	char *opts;
1125 
1126 	if (options == NULL || options[0] == '\0')
1127 		return (ma);
1128 
1129 	p = opts = strdup(options, M_MOUNT);
1130 	if (opts == NULL) {
1131 		return (ma);
1132 	}
1133 
1134 	while((name = strsep(&p, ",")) != NULL) {
1135 		if (name[0] == '\0')
1136 			break;
1137 
1138 		val = strchr(name, '=');
1139 		if (val != NULL) {
1140 			*val = '\0';
1141 			++val;
1142 		}
1143 		if (strcmp(name, "rw") == 0 || strcmp(name, "noro") == 0) {
1144 			/*
1145 			 * The first time we mount the root file system,
1146 			 * we need to mount 'ro', so We need to ignore
1147 			 * 'rw' and 'noro' mount options.
1148 			 */
1149 			continue;
1150 		}
1151 		name_arg = strdup(name, M_MOUNT);
1152 		val_arg = NULL;
1153 		if (val != NULL)
1154 			val_arg = strdup(val, M_MOUNT);
1155 
1156 		ma = mount_arg(ma, name_arg, val_arg,
1157 		    (val_arg != NULL ? -1 : 0));
1158 	}
1159 	free(opts, M_MOUNT);
1160 	return (ma);
1161 }
1162