xref: /freebsd/sys/kern/vfs_mountroot.c (revision 2e4311906d8c8dc7a7c726345268253bca6d4acc)
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, fspath);
354 		error = namei(&nd);
355 		if (error) {
356 			NDFREE(&nd, NDF_ONLY_PNBUF);
357 			fspath = "/mnt";
358 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
359 			    fspath);
360 			error = namei(&nd);
361 		}
362 		if (!error) {
363 			vp = nd.ni_vp;
364 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
365 			if (!error)
366 				error = vinvalbuf(vp, V_SAVE, 0, 0);
367 			if (!error) {
368 				cache_purge(vp);
369 				VI_LOCK(vp);
370 				mporoot->mnt_vnodecovered = vp;
371 				vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
372 				vp->v_mountedhere = mporoot;
373 				strlcpy(mporoot->mnt_stat.f_mntonname,
374 				    fspath, MNAMELEN);
375 				VI_UNLOCK(vp);
376 				VOP_UNLOCK(vp);
377 			} else
378 				vput(vp);
379 		}
380 		NDFREE(&nd, NDF_ONLY_PNBUF);
381 
382 		if (error)
383 			printf("mountroot: unable to remount previous root "
384 			    "under /.mount or /mnt (error %d)\n", error);
385 	}
386 
387 	/* Remount devfs under /dev */
388 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev");
389 	error = namei(&nd);
390 	if (!error) {
391 		vp = nd.ni_vp;
392 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
393 		if (!error)
394 			error = vinvalbuf(vp, V_SAVE, 0, 0);
395 		if (!error) {
396 			vpdevfs = mpdevfs->mnt_vnodecovered;
397 			if (vpdevfs != NULL) {
398 				cache_purge(vpdevfs);
399 				VI_LOCK(vpdevfs);
400 				vn_irflag_unset_locked(vpdevfs, VIRF_MOUNTPOINT);
401 				vpdevfs->v_mountedhere = NULL;
402 				VI_UNLOCK(vpdevfs);
403 				vrele(vpdevfs);
404 			}
405 			VI_LOCK(vp);
406 			mpdevfs->mnt_vnodecovered = vp;
407 			vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
408 			vp->v_mountedhere = mpdevfs;
409 			VI_UNLOCK(vp);
410 			VOP_UNLOCK(vp);
411 		} else
412 			vput(vp);
413 	}
414 	if (error)
415 		printf("mountroot: unable to remount devfs under /dev "
416 		    "(error %d)\n", error);
417 	NDFREE(&nd, NDF_ONLY_PNBUF);
418 
419 	if (mporoot == mpdevfs) {
420 		vfs_unbusy(mpdevfs);
421 		/* Unlink the no longer needed /dev/dev -> / symlink */
422 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
423 		    UIO_SYSSPACE, 0, 0);
424 		if (error)
425 			printf("mountroot: unable to unlink /dev/dev "
426 			    "(error %d)\n", error);
427 	}
428 }
429 
430 /*
431  * Configuration parser.
432  */
433 
434 /* Parser character classes. */
435 #define	CC_WHITESPACE		-1
436 #define	CC_NONWHITESPACE	-2
437 
438 /* Parse errors. */
439 #define	PE_EOF			-1
440 #define	PE_EOL			-2
441 
442 static __inline int
443 parse_peek(char **conf)
444 {
445 
446 	return (**conf);
447 }
448 
449 static __inline void
450 parse_poke(char **conf, int c)
451 {
452 
453 	**conf = c;
454 }
455 
456 static __inline void
457 parse_advance(char **conf)
458 {
459 
460 	(*conf)++;
461 }
462 
463 static int
464 parse_skipto(char **conf, int mc)
465 {
466 	int c, match;
467 
468 	while (1) {
469 		c = parse_peek(conf);
470 		if (c == 0)
471 			return (PE_EOF);
472 		switch (mc) {
473 		case CC_WHITESPACE:
474 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
475 			break;
476 		case CC_NONWHITESPACE:
477 			if (c == '\n')
478 				return (PE_EOL);
479 			match = (c != ' ' && c != '\t') ? 1 : 0;
480 			break;
481 		default:
482 			match = (c == mc) ? 1 : 0;
483 			break;
484 		}
485 		if (match)
486 			break;
487 		parse_advance(conf);
488 	}
489 	return (0);
490 }
491 
492 static int
493 parse_token(char **conf, char **tok)
494 {
495 	char *p;
496 	size_t len;
497 	int error;
498 
499 	*tok = NULL;
500 	error = parse_skipto(conf, CC_NONWHITESPACE);
501 	if (error)
502 		return (error);
503 	p = *conf;
504 	error = parse_skipto(conf, CC_WHITESPACE);
505 	len = *conf - p;
506 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
507 	bcopy(p, *tok, len);
508 	return (0);
509 }
510 
511 static void
512 parse_dir_ask_printenv(const char *var)
513 {
514 	char *val;
515 
516 	val = kern_getenv(var);
517 	if (val != NULL) {
518 		printf("  %s=%s\n", var, val);
519 		freeenv(val);
520 	}
521 }
522 
523 static int
524 parse_dir_ask(char **conf)
525 {
526 	char name[80];
527 	char *mnt;
528 	int error;
529 
530 	vfs_mountroot_wait();
531 
532 	printf("\nLoader variables:\n");
533 	parse_dir_ask_printenv("vfs.root.mountfrom");
534 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
535 
536 	printf("\nManual root filesystem specification:\n");
537 	printf("  <fstype>:<device> [options]\n");
538 	printf("      Mount <device> using filesystem <fstype>\n");
539 	printf("      and with the specified (optional) option list.\n");
540 	printf("\n");
541 	printf("    eg. ufs:/dev/da0s1a\n");
542 	printf("        zfs:zroot/ROOT/default\n");
543 	printf("        cd9660:/dev/cd0 ro\n");
544 	printf("          (which is equivalent to: ");
545 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
546 	printf("\n");
547 	printf("  ?               List valid disk boot devices\n");
548 	printf("  .               Yield 1 second (for background tasks)\n");
549 	printf("  <empty line>    Abort manual input\n");
550 
551 	do {
552 		error = EINVAL;
553 		printf("\nmountroot> ");
554 		cngets(name, sizeof(name), GETS_ECHO);
555 		if (name[0] == '\0')
556 			break;
557 		if (name[0] == '?' && name[1] == '\0') {
558 			printf("\nList of GEOM managed disk devices:\n  ");
559 			g_dev_print();
560 			continue;
561 		}
562 		if (name[0] == '.' && name[1] == '\0') {
563 			pause("rmask", hz);
564 			continue;
565 		}
566 		mnt = name;
567 		error = parse_mount(&mnt);
568 		if (error == -1)
569 			printf("Invalid file system specification.\n");
570 	} while (error != 0);
571 
572 	return (error);
573 }
574 
575 static int
576 parse_dir_md(char **conf)
577 {
578 	struct stat sb;
579 	struct thread *td;
580 	struct md_ioctl *mdio;
581 	char *path, *tok;
582 	int error, fd, len;
583 
584 	td = curthread;
585 	fd = -1;
586 
587 	error = parse_token(conf, &tok);
588 	if (error)
589 		return (error);
590 
591 	len = strlen(tok);
592 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
593 	path = (void *)(mdio + 1);
594 	bcopy(tok, path, len);
595 	free(tok, M_TEMP);
596 
597 	/* Get file status. */
598 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
599 	if (error)
600 		goto out;
601 
602 	/* Open /dev/mdctl so that we can attach/detach. */
603 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
604 	    O_RDWR, 0);
605 	if (error)
606 		goto out;
607 
608 	fd = td->td_retval[0];
609 	mdio->md_version = MDIOVERSION;
610 	mdio->md_type = MD_VNODE;
611 
612 	if (root_mount_mddev != -1) {
613 		mdio->md_unit = root_mount_mddev;
614 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
615 		/* Ignore errors. We don't care. */
616 		root_mount_mddev = -1;
617 	}
618 
619 	mdio->md_file = (void *)(mdio + 1);
620 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
621 	mdio->md_mediasize = sb.st_size;
622 	mdio->md_unit = 0;
623 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
624 	if (error)
625 		goto out;
626 
627 	if (mdio->md_unit > 9) {
628 		printf("rootmount: too many md units\n");
629 		mdio->md_file = NULL;
630 		mdio->md_options = 0;
631 		mdio->md_mediasize = 0;
632 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
633 		/* Ignore errors. We don't care. */
634 		error = ERANGE;
635 		goto out;
636 	}
637 
638 	root_mount_mddev = mdio->md_unit;
639 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
640 
641  out:
642 	if (fd >= 0)
643 		(void)kern_close(td, fd);
644 	free(mdio, M_TEMP);
645 	return (error);
646 }
647 
648 static int
649 parse_dir_onfail(char **conf)
650 {
651 	char *action;
652 	int error;
653 
654 	error = parse_token(conf, &action);
655 	if (error)
656 		return (error);
657 
658 	if (!strcmp(action, "continue"))
659 		root_mount_onfail = A_CONTINUE;
660 	else if (!strcmp(action, "panic"))
661 		root_mount_onfail = A_PANIC;
662 	else if (!strcmp(action, "reboot"))
663 		root_mount_onfail = A_REBOOT;
664 	else if (!strcmp(action, "retry"))
665 		root_mount_onfail = A_RETRY;
666 	else {
667 		printf("rootmount: %s: unknown action\n", action);
668 		error = EINVAL;
669 	}
670 
671 	free(action, M_TEMP);
672 	return (0);
673 }
674 
675 static int
676 parse_dir_timeout(char **conf)
677 {
678 	char *tok, *endtok;
679 	long secs;
680 	int error;
681 
682 	error = parse_token(conf, &tok);
683 	if (error)
684 		return (error);
685 
686 	secs = strtol(tok, &endtok, 0);
687 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
688 	if (!error)
689 		root_mount_timeout = secs;
690 	free(tok, M_TEMP);
691 	return (error);
692 }
693 
694 static int
695 parse_directive(char **conf)
696 {
697 	char *dir;
698 	int error;
699 
700 	error = parse_token(conf, &dir);
701 	if (error)
702 		return (error);
703 
704 	if (strcmp(dir, ".ask") == 0)
705 		error = parse_dir_ask(conf);
706 	else if (strcmp(dir, ".md") == 0)
707 		error = parse_dir_md(conf);
708 	else if (strcmp(dir, ".onfail") == 0)
709 		error = parse_dir_onfail(conf);
710 	else if (strcmp(dir, ".timeout") == 0)
711 		error = parse_dir_timeout(conf);
712 	else {
713 		printf("mountroot: invalid directive `%s'\n", dir);
714 		/* Ignore the rest of the line. */
715 		(void)parse_skipto(conf, '\n');
716 		error = EINVAL;
717 	}
718 	free(dir, M_TEMP);
719 	return (error);
720 }
721 
722 static int
723 parse_mount_dev_present(const char *dev)
724 {
725 	struct nameidata nd;
726 	int error;
727 
728 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev);
729 	error = namei(&nd);
730 	if (!error)
731 		vput(nd.ni_vp);
732 	NDFREE(&nd, NDF_ONLY_PNBUF);
733 	return (error != 0) ? 0 : 1;
734 }
735 
736 #define	ERRMSGL	255
737 static int
738 parse_mount(char **conf)
739 {
740 	char *errmsg;
741 	struct mntarg *ma;
742 	char *dev, *fs, *opts, *tok;
743 	int delay, error, timeout;
744 
745 	error = parse_token(conf, &tok);
746 	if (error)
747 		return (error);
748 	fs = tok;
749 	error = parse_skipto(&tok, ':');
750 	if (error) {
751 		free(fs, M_TEMP);
752 		return (error);
753 	}
754 	parse_poke(&tok, '\0');
755 	parse_advance(&tok);
756 	dev = tok;
757 
758 	if (root_mount_mddev != -1) {
759 		/* Handle substitution for the md unit number. */
760 		tok = strstr(dev, "md#");
761 		if (tok != NULL)
762 			tok[2] = '0' + root_mount_mddev;
763 	}
764 
765 	/* Parse options. */
766 	error = parse_token(conf, &tok);
767 	opts = (error == 0) ? tok : NULL;
768 
769 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
770 	    (opts != NULL) ? opts : "");
771 
772 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
773 
774 	if (vfs_byname(fs) == NULL) {
775 		strlcpy(errmsg, "unknown file system", ERRMSGL);
776 		error = ENOENT;
777 		goto out;
778 	}
779 
780 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
781 	if (error != 0)
782 		goto out;
783 
784 	delay = hz / 10;
785 	timeout = root_mount_timeout * hz;
786 
787 	for (;;) {
788 		ma = NULL;
789 		ma = mount_arg(ma, "fstype", fs, -1);
790 		ma = mount_arg(ma, "fspath", "/", -1);
791 		ma = mount_arg(ma, "from", dev, -1);
792 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
793 		ma = mount_arg(ma, "ro", NULL, 0);
794 		ma = parse_mountroot_options(ma, opts);
795 
796 		error = kernel_mount(ma, MNT_ROOTFS);
797 		if (error == 0 || timeout <= 0)
798 			break;
799 
800 		if (root_mount_timeout * hz == timeout ||
801 		    (bootverbose && timeout % hz == 0)) {
802 			printf("Mounting from %s:%s failed with error %d; "
803 			    "retrying for %d more second%s\n", fs, dev, error,
804 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
805 		}
806 		pause("rmretry", delay);
807 		timeout -= delay;
808 	}
809  out:
810 	if (error) {
811 		printf("Mounting from %s:%s failed with error %d",
812 		    fs, dev, error);
813 		if (errmsg[0] != '\0')
814 			printf(": %s", errmsg);
815 		printf(".\n");
816 	}
817 	free(fs, M_TEMP);
818 	free(errmsg, M_TEMP);
819 	if (opts != NULL)
820 		free(opts, M_TEMP);
821 	/* kernel_mount can return -1 on error. */
822 	return ((error < 0) ? EDOOFUS : error);
823 }
824 #undef ERRMSGL
825 
826 static int
827 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
828 {
829 	struct mount *mp;
830 	char *conf;
831 	int error;
832 
833 	root_mount_mddev = -1;
834 
835 retry:
836 	conf = sbuf_data(sb);
837 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
838 	error = (mp == NULL) ? 0 : EDOOFUS;
839 	root_mount_onfail = A_CONTINUE;
840 	while (mp == NULL) {
841 		error = parse_skipto(&conf, CC_NONWHITESPACE);
842 		if (error == PE_EOL) {
843 			parse_advance(&conf);
844 			continue;
845 		}
846 		if (error < 0)
847 			break;
848 		switch (parse_peek(&conf)) {
849 		case '#':
850 			error = parse_skipto(&conf, '\n');
851 			break;
852 		case '.':
853 			error = parse_directive(&conf);
854 			break;
855 		default:
856 			error = parse_mount(&conf);
857 			if (error == -1) {
858 				printf("mountroot: invalid file system "
859 				    "specification.\n");
860 				error = 0;
861 			}
862 			break;
863 		}
864 		if (error < 0)
865 			break;
866 		/* Ignore any trailing garbage on the line. */
867 		if (parse_peek(&conf) != '\n') {
868 			printf("mountroot: advancing to next directive...\n");
869 			(void)parse_skipto(&conf, '\n');
870 		}
871 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
872 	}
873 	if (mp != NULL)
874 		return (0);
875 
876 	/*
877 	 * We failed to mount (a new) root.
878 	 */
879 	switch (root_mount_onfail) {
880 	case A_CONTINUE:
881 		break;
882 	case A_PANIC:
883 		panic("mountroot: unable to (re-)mount root.");
884 		/* NOTREACHED */
885 	case A_RETRY:
886 		goto retry;
887 	case A_REBOOT:
888 		kern_reboot(RB_NOSYNC);
889 		/* NOTREACHED */
890 	}
891 
892 	return (error);
893 }
894 
895 static void
896 vfs_mountroot_conf0(struct sbuf *sb)
897 {
898 	char *s, *tok, *mnt, *opt;
899 	int error;
900 
901 	sbuf_printf(sb, ".onfail panic\n");
902 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
903 	if (boothowto & RB_ASKNAME)
904 		sbuf_printf(sb, ".ask\n");
905 #ifdef ROOTDEVNAME
906 	if (boothowto & RB_DFLTROOT)
907 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
908 #endif
909 	if (boothowto & RB_CDROM) {
910 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
911 		sbuf_printf(sb, ".timeout 0\n");
912 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
913 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
914 	}
915 	s = kern_getenv("vfs.root.mountfrom");
916 	if (s != NULL) {
917 		opt = kern_getenv("vfs.root.mountfrom.options");
918 		tok = s;
919 		error = parse_token(&tok, &mnt);
920 		while (!error) {
921 			sbuf_printf(sb, "%s %s\n", mnt,
922 			    (opt != NULL) ? opt : "");
923 			free(mnt, M_TEMP);
924 			error = parse_token(&tok, &mnt);
925 		}
926 		if (opt != NULL)
927 			freeenv(opt);
928 		freeenv(s);
929 	}
930 	if (rootdevnames[0] != NULL)
931 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
932 	if (rootdevnames[1] != NULL)
933 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
934 #ifdef ROOTDEVNAME
935 	if (!(boothowto & RB_DFLTROOT))
936 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
937 #endif
938 	if (!(boothowto & RB_ASKNAME))
939 		sbuf_printf(sb, ".ask\n");
940 }
941 
942 static int
943 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
944 {
945 	static char buf[128];
946 	struct nameidata nd;
947 	off_t ofs;
948 	ssize_t resid;
949 	int error, flags, len;
950 
951 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf");
952 	flags = FREAD;
953 	error = vn_open(&nd, &flags, 0, NULL);
954 	if (error)
955 		return (error);
956 
957 	NDFREE(&nd, NDF_ONLY_PNBUF);
958 	ofs = 0;
959 	len = sizeof(buf) - 1;
960 	while (1) {
961 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
962 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
963 		    NOCRED, &resid, td);
964 		if (error)
965 			break;
966 		if (resid == len)
967 			break;
968 		buf[len - resid] = 0;
969 		sbuf_printf(sb, "%s", buf);
970 		ofs += len - resid;
971 	}
972 
973 	VOP_UNLOCK(nd.ni_vp);
974 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
975 	return (error);
976 }
977 
978 static void
979 vfs_mountroot_wait(void)
980 {
981 	struct root_hold_token *h;
982 	struct timeval lastfail;
983 	int curfail;
984 
985 	TSENTER();
986 
987 	curfail = 0;
988 	lastfail.tv_sec = 0;
989 	ppsratecheck(&lastfail, &curfail, 1);
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 	g_waitidle();
1009 
1010 	TSEXIT();
1011 }
1012 
1013 static int
1014 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1015 {
1016 	int delay, timeout;
1017 
1018 	/*
1019 	 * In case of ZFS and NFS we don't have a way to wait for
1020 	 * specific device.  Also do the wait if the user forced that
1021 	 * behaviour by setting vfs.root_mount_always_wait=1.
1022 	 */
1023 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1024 	    dev[0] == '\0' || root_mount_always_wait != 0) {
1025 		vfs_mountroot_wait();
1026 		return (0);
1027 	}
1028 
1029 	/*
1030 	 * Otherwise, no point in waiting if the device is already there.
1031 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1032 	 * eg for geom_part(4) to finish tasting.
1033 	 */
1034 	g_waitidle();
1035 	if (parse_mount_dev_present(dev))
1036 		return (0);
1037 
1038 	/*
1039 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1040 	 * to behave exactly as it used to work before.
1041 	 */
1042 	vfs_mountroot_wait();
1043 	if (parse_mount_dev_present(dev))
1044 		return (0);
1045 	printf("mountroot: waiting for device %s...\n", dev);
1046 	delay = hz / 10;
1047 	timeout = root_mount_timeout * hz;
1048 	do {
1049 		pause("rmdev", delay);
1050 		timeout -= delay;
1051 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1052 
1053 	if (timeout <= 0)
1054 		return (ENODEV);
1055 
1056 	return (0);
1057 }
1058 
1059 void
1060 vfs_mountroot(void)
1061 {
1062 	struct mount *mp;
1063 	struct sbuf *sb;
1064 	struct thread *td;
1065 	time_t timebase;
1066 	int error;
1067 
1068 	mtx_assert(&Giant, MA_NOTOWNED);
1069 
1070 	TSENTER();
1071 
1072 	td = curthread;
1073 
1074 	sb = sbuf_new_auto();
1075 	vfs_mountroot_conf0(sb);
1076 	sbuf_finish(sb);
1077 
1078 	error = vfs_mountroot_devfs(td, &mp);
1079 	while (!error) {
1080 		error = vfs_mountroot_parse(sb, mp);
1081 		if (!error) {
1082 			vfs_mountroot_shuffle(td, mp);
1083 			sbuf_clear(sb);
1084 			error = vfs_mountroot_readconf(td, sb);
1085 			sbuf_finish(sb);
1086 		}
1087 	}
1088 
1089 	sbuf_delete(sb);
1090 
1091 	/*
1092 	 * Iterate over all currently mounted file systems and use
1093 	 * the time stamp found to check and/or initialize the RTC.
1094 	 * Call inittodr() only once and pass it the largest of the
1095 	 * timestamps we encounter.
1096 	 */
1097 	timebase = 0;
1098 	mtx_lock(&mountlist_mtx);
1099 	mp = TAILQ_FIRST(&mountlist);
1100 	while (mp != NULL) {
1101 		if (mp->mnt_time > timebase)
1102 			timebase = mp->mnt_time;
1103 		mp = TAILQ_NEXT(mp, mnt_list);
1104 	}
1105 	mtx_unlock(&mountlist_mtx);
1106 	inittodr(timebase);
1107 
1108 	/* Keep prison0's root in sync with the global rootvnode. */
1109 	mtx_lock(&prison0.pr_mtx);
1110 	prison0.pr_root = rootvnode;
1111 	vref(prison0.pr_root);
1112 	mtx_unlock(&prison0.pr_mtx);
1113 
1114 	mtx_lock(&root_holds_mtx);
1115 	atomic_store_rel_int(&root_mount_complete, 1);
1116 	wakeup(&root_mount_complete);
1117 	mtx_unlock(&root_holds_mtx);
1118 
1119 	EVENTHANDLER_INVOKE(mountroot);
1120 
1121 	TSEXIT();
1122 }
1123 
1124 static struct mntarg *
1125 parse_mountroot_options(struct mntarg *ma, const char *options)
1126 {
1127 	char *p;
1128 	char *name, *name_arg;
1129 	char *val, *val_arg;
1130 	char *opts;
1131 
1132 	if (options == NULL || options[0] == '\0')
1133 		return (ma);
1134 
1135 	p = opts = strdup(options, M_MOUNT);
1136 	if (opts == NULL) {
1137 		return (ma);
1138 	}
1139 
1140 	while((name = strsep(&p, ",")) != NULL) {
1141 		if (name[0] == '\0')
1142 			break;
1143 
1144 		val = strchr(name, '=');
1145 		if (val != NULL) {
1146 			*val = '\0';
1147 			++val;
1148 		}
1149 		if (strcmp(name, "rw") == 0 || strcmp(name, "noro") == 0) {
1150 			/*
1151 			 * The first time we mount the root file system,
1152 			 * we need to mount 'ro', so We need to ignore
1153 			 * 'rw' and 'noro' mount options.
1154 			 */
1155 			continue;
1156 		}
1157 		name_arg = strdup(name, M_MOUNT);
1158 		val_arg = NULL;
1159 		if (val != NULL)
1160 			val_arg = strdup(val, M_MOUNT);
1161 
1162 		ma = mount_arg(ma, name_arg, val_arg,
1163 		    (val_arg != NULL ? -1 : 0));
1164 	}
1165 	free(opts, M_MOUNT);
1166 	return (ma);
1167 }
1168