xref: /freebsd/sys/kern/vfs_mountroot.c (revision e4c66ddabdb470bab319705c1834a4867c508a43)
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/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/mdioctl.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/filedesc.h>
59 #include <sys/reboot.h>
60 #include <sys/sbuf.h>
61 #include <sys/stat.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysproto.h>
64 #include <sys/sx.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/systm.h>
68 #include <sys/vnode.h>
69 
70 #include <geom/geom.h>
71 
72 /*
73  * The root filesystem is detailed in the kernel environment variable
74  * vfs.root.mountfrom, which is expected to be in the general format
75  *
76  * <vfsname>:[<path>][	<vfsname>:[<path>] ...]
77  * vfsname   := the name of a VFS known to the kernel and capable
78  *              of being mounted as root
79  * path      := disk device name or other data used by the filesystem
80  *              to locate its physical store
81  *
82  * If the environment variable vfs.root.mountfrom is a space separated list,
83  * each list element is tried in turn and the root filesystem will be mounted
84  * from the first one that succeeds.
85  *
86  * The environment variable vfs.root.mountfrom.options is a comma delimited
87  * set of string mount options.  These mount options must be parseable
88  * by nmount() in the kernel.
89  */
90 
91 static int parse_mount(char **);
92 static struct mntarg *parse_mountroot_options(struct mntarg *, const char *);
93 static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS);
94 static void vfs_mountroot_wait(void);
95 static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev);
96 
97 /*
98  * The vnode of the system's root (/ in the filesystem, without chroot
99  * active.)
100  */
101 struct vnode *rootvnode;
102 
103 /*
104  * Mount of the system's /dev.
105  */
106 struct mount *rootdevmp;
107 
108 char *rootdevnames[2] = {NULL, NULL};
109 
110 struct mtx root_holds_mtx;
111 MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF);
112 
113 struct root_hold_token {
114 	const char			*who;
115 	LIST_ENTRY(root_hold_token)	list;
116 };
117 
118 static LIST_HEAD(, root_hold_token)	root_holds =
119     LIST_HEAD_INITIALIZER(root_holds);
120 
121 enum action {
122 	A_CONTINUE,
123 	A_PANIC,
124 	A_REBOOT,
125 	A_RETRY
126 };
127 
128 static enum action root_mount_onfail = A_CONTINUE;
129 
130 static int root_mount_mddev;
131 static int root_mount_complete;
132 
133 /* By default wait up to 3 seconds for devices to appear. */
134 static int root_mount_timeout = 3;
135 TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout);
136 
137 static int root_mount_always_wait = 0;
138 SYSCTL_INT(_vfs, OID_AUTO, root_mount_always_wait, CTLFLAG_RDTUN,
139     &root_mount_always_wait, 0,
140     "Wait for root mount holds even if the root device already exists");
141 
142 SYSCTL_PROC(_vfs, OID_AUTO, root_mount_hold,
143     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
144     NULL, 0, sysctl_vfs_root_mount_hold, "A",
145     "List of root mount hold tokens");
146 
147 static int
148 sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS)
149 {
150 	struct sbuf sb;
151 	struct root_hold_token *h;
152 	int error;
153 
154 	sbuf_new(&sb, NULL, 256, SBUF_AUTOEXTEND | SBUF_INCLUDENUL);
155 
156 	mtx_lock(&root_holds_mtx);
157 	LIST_FOREACH(h, &root_holds, list) {
158 		if (h != LIST_FIRST(&root_holds))
159 			sbuf_putc(&sb, ' ');
160 		sbuf_printf(&sb, "%s", h->who);
161 	}
162 	mtx_unlock(&root_holds_mtx);
163 
164 	error = sbuf_finish(&sb);
165 	if (error == 0)
166 		error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
167 	sbuf_delete(&sb);
168 	return (error);
169 }
170 
171 struct root_hold_token *
172 root_mount_hold(const char *identifier)
173 {
174 	struct root_hold_token *h;
175 
176 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
177 	h->who = identifier;
178 	mtx_lock(&root_holds_mtx);
179 	TSHOLD("root mount");
180 	LIST_INSERT_HEAD(&root_holds, h, list);
181 	mtx_unlock(&root_holds_mtx);
182 	return (h);
183 }
184 
185 void
186 root_mount_rel(struct root_hold_token *h)
187 {
188 
189 	if (h == NULL)
190 		return;
191 
192 	mtx_lock(&root_holds_mtx);
193 	LIST_REMOVE(h, list);
194 	TSRELEASE("root mount");
195 	wakeup(&root_holds);
196 	mtx_unlock(&root_holds_mtx);
197 	free(h, M_DEVBUF);
198 }
199 
200 int
201 root_mounted(void)
202 {
203 
204 	/* No mutex is acquired here because int stores are atomic. */
205 	return (root_mount_complete);
206 }
207 
208 static void
209 set_rootvnode(void)
210 {
211 	struct proc *p;
212 
213 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
214 		panic("Cannot find root vnode");
215 
216 	VOP_UNLOCK(rootvnode, 0);
217 
218 	p = curthread->td_proc;
219 	FILEDESC_XLOCK(p->p_fd);
220 
221 	if (p->p_fd->fd_cdir != NULL)
222 		vrele(p->p_fd->fd_cdir);
223 	p->p_fd->fd_cdir = rootvnode;
224 	VREF(rootvnode);
225 
226 	if (p->p_fd->fd_rdir != NULL)
227 		vrele(p->p_fd->fd_rdir);
228 	p->p_fd->fd_rdir = rootvnode;
229 	VREF(rootvnode);
230 
231 	FILEDESC_XUNLOCK(p->p_fd);
232 }
233 
234 static int
235 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
236 {
237 	struct vfsoptlist *opts;
238 	struct vfsconf *vfsp;
239 	struct mount *mp;
240 	int error;
241 
242 	*mpp = NULL;
243 
244 	if (rootdevmp != NULL) {
245 		/*
246 		 * Already have /dev; this happens during rerooting.
247 		 */
248 		error = vfs_busy(rootdevmp, 0);
249 		if (error != 0)
250 			return (error);
251 		*mpp = rootdevmp;
252 	} else {
253 		vfsp = vfs_byname("devfs");
254 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
255 		if (vfsp == NULL)
256 			return (ENOENT);
257 
258 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
259 
260 		error = VFS_MOUNT(mp);
261 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
262 		if (error)
263 			return (error);
264 
265 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
266 		TAILQ_INIT(opts);
267 		mp->mnt_opt = opts;
268 
269 		mtx_lock(&mountlist_mtx);
270 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
271 		mtx_unlock(&mountlist_mtx);
272 
273 		*mpp = mp;
274 		rootdevmp = mp;
275 	}
276 
277 	set_rootvnode();
278 
279 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
280 	if (error)
281 		printf("kern_symlink /dev -> / returns %d\n", error);
282 
283 	return (error);
284 }
285 
286 static void
287 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
288 {
289 	struct nameidata nd;
290 	struct mount *mporoot, *mpnroot;
291 	struct vnode *vp, *vporoot, *vpdevfs;
292 	char *fspath;
293 	int error;
294 
295 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
296 
297 	/* Shuffle the mountlist. */
298 	mtx_lock(&mountlist_mtx);
299 	mporoot = TAILQ_FIRST(&mountlist);
300 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
301 	if (mporoot != mpdevfs) {
302 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
303 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
304 	}
305 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
306 	mtx_unlock(&mountlist_mtx);
307 
308 	cache_purgevfs(mporoot, true);
309 	if (mporoot != mpdevfs)
310 		cache_purgevfs(mpdevfs, true);
311 
312 	VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot);
313 
314 	VI_LOCK(vporoot);
315 	vporoot->v_iflag &= ~VI_MOUNT;
316 	VI_UNLOCK(vporoot);
317 	vporoot->v_mountedhere = NULL;
318 	mporoot->mnt_flag &= ~MNT_ROOTFS;
319 	mporoot->mnt_vnodecovered = NULL;
320 	vput(vporoot);
321 
322 	/* Set up the new rootvnode, and purge the cache */
323 	mpnroot->mnt_vnodecovered = NULL;
324 	set_rootvnode();
325 	cache_purgevfs(rootvnode->v_mount, true);
326 
327 	if (mporoot != mpdevfs) {
328 		/* Remount old root under /.mount or /mnt */
329 		fspath = "/.mount";
330 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
331 		    fspath, td);
332 		error = namei(&nd);
333 		if (error) {
334 			NDFREE(&nd, NDF_ONLY_PNBUF);
335 			fspath = "/mnt";
336 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
337 			    fspath, td);
338 			error = namei(&nd);
339 		}
340 		if (!error) {
341 			vp = nd.ni_vp;
342 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
343 			if (!error)
344 				error = vinvalbuf(vp, V_SAVE, 0, 0);
345 			if (!error) {
346 				cache_purge(vp);
347 				mporoot->mnt_vnodecovered = vp;
348 				vp->v_mountedhere = mporoot;
349 				strlcpy(mporoot->mnt_stat.f_mntonname,
350 				    fspath, MNAMELEN);
351 				VOP_UNLOCK(vp, 0);
352 			} else
353 				vput(vp);
354 		}
355 		NDFREE(&nd, NDF_ONLY_PNBUF);
356 
357 		if (error)
358 			printf("mountroot: unable to remount previous root "
359 			    "under /.mount or /mnt (error %d)\n", error);
360 	}
361 
362 	/* Remount devfs under /dev */
363 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
364 	error = namei(&nd);
365 	if (!error) {
366 		vp = nd.ni_vp;
367 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
368 		if (!error)
369 			error = vinvalbuf(vp, V_SAVE, 0, 0);
370 		if (!error) {
371 			vpdevfs = mpdevfs->mnt_vnodecovered;
372 			if (vpdevfs != NULL) {
373 				cache_purge(vpdevfs);
374 				vpdevfs->v_mountedhere = NULL;
375 				vrele(vpdevfs);
376 			}
377 			mpdevfs->mnt_vnodecovered = vp;
378 			vp->v_mountedhere = mpdevfs;
379 			VOP_UNLOCK(vp, 0);
380 		} else
381 			vput(vp);
382 	}
383 	if (error)
384 		printf("mountroot: unable to remount devfs under /dev "
385 		    "(error %d)\n", error);
386 	NDFREE(&nd, NDF_ONLY_PNBUF);
387 
388 	if (mporoot == mpdevfs) {
389 		vfs_unbusy(mpdevfs);
390 		/* Unlink the no longer needed /dev/dev -> / symlink */
391 		error = kern_unlinkat(td, AT_FDCWD, "/dev/dev",
392 		    UIO_SYSSPACE, 0);
393 		if (error)
394 			printf("mountroot: unable to unlink /dev/dev "
395 			    "(error %d)\n", error);
396 	}
397 }
398 
399 /*
400  * Configuration parser.
401  */
402 
403 /* Parser character classes. */
404 #define	CC_WHITESPACE		-1
405 #define	CC_NONWHITESPACE	-2
406 
407 /* Parse errors. */
408 #define	PE_EOF			-1
409 #define	PE_EOL			-2
410 
411 static __inline int
412 parse_peek(char **conf)
413 {
414 
415 	return (**conf);
416 }
417 
418 static __inline void
419 parse_poke(char **conf, int c)
420 {
421 
422 	**conf = c;
423 }
424 
425 static __inline void
426 parse_advance(char **conf)
427 {
428 
429 	(*conf)++;
430 }
431 
432 static int
433 parse_skipto(char **conf, int mc)
434 {
435 	int c, match;
436 
437 	while (1) {
438 		c = parse_peek(conf);
439 		if (c == 0)
440 			return (PE_EOF);
441 		switch (mc) {
442 		case CC_WHITESPACE:
443 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
444 			break;
445 		case CC_NONWHITESPACE:
446 			if (c == '\n')
447 				return (PE_EOL);
448 			match = (c != ' ' && c != '\t') ? 1 : 0;
449 			break;
450 		default:
451 			match = (c == mc) ? 1 : 0;
452 			break;
453 		}
454 		if (match)
455 			break;
456 		parse_advance(conf);
457 	}
458 	return (0);
459 }
460 
461 static int
462 parse_token(char **conf, char **tok)
463 {
464 	char *p;
465 	size_t len;
466 	int error;
467 
468 	*tok = NULL;
469 	error = parse_skipto(conf, CC_NONWHITESPACE);
470 	if (error)
471 		return (error);
472 	p = *conf;
473 	error = parse_skipto(conf, CC_WHITESPACE);
474 	len = *conf - p;
475 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
476 	bcopy(p, *tok, len);
477 	return (0);
478 }
479 
480 static void
481 parse_dir_ask_printenv(const char *var)
482 {
483 	char *val;
484 
485 	val = kern_getenv(var);
486 	if (val != NULL) {
487 		printf("  %s=%s\n", var, val);
488 		freeenv(val);
489 	}
490 }
491 
492 static int
493 parse_dir_ask(char **conf)
494 {
495 	char name[80];
496 	char *mnt;
497 	int error;
498 
499 	vfs_mountroot_wait();
500 
501 	printf("\nLoader variables:\n");
502 	parse_dir_ask_printenv("vfs.root.mountfrom");
503 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
504 
505 	printf("\nManual root filesystem specification:\n");
506 	printf("  <fstype>:<device> [options]\n");
507 	printf("      Mount <device> using filesystem <fstype>\n");
508 	printf("      and with the specified (optional) option list.\n");
509 	printf("\n");
510 	printf("    eg. ufs:/dev/da0s1a\n");
511 	printf("        zfs:tank\n");
512 	printf("        cd9660:/dev/cd0 ro\n");
513 	printf("          (which is equivalent to: ");
514 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
515 	printf("\n");
516 	printf("  ?               List valid disk boot devices\n");
517 	printf("  .               Yield 1 second (for background tasks)\n");
518 	printf("  <empty line>    Abort manual input\n");
519 
520 	do {
521 		error = EINVAL;
522 		printf("\nmountroot> ");
523 		cngets(name, sizeof(name), GETS_ECHO);
524 		if (name[0] == '\0')
525 			break;
526 		if (name[0] == '?' && name[1] == '\0') {
527 			printf("\nList of GEOM managed disk devices:\n  ");
528 			g_dev_print();
529 			continue;
530 		}
531 		if (name[0] == '.' && name[1] == '\0') {
532 			pause("rmask", hz);
533 			continue;
534 		}
535 		mnt = name;
536 		error = parse_mount(&mnt);
537 		if (error == -1)
538 			printf("Invalid file system specification.\n");
539 	} while (error != 0);
540 
541 	return (error);
542 }
543 
544 static int
545 parse_dir_md(char **conf)
546 {
547 	struct stat sb;
548 	struct thread *td;
549 	struct md_ioctl *mdio;
550 	char *path, *tok;
551 	int error, fd, len;
552 
553 	td = curthread;
554 
555 	error = parse_token(conf, &tok);
556 	if (error)
557 		return (error);
558 
559 	len = strlen(tok);
560 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
561 	path = (void *)(mdio + 1);
562 	bcopy(tok, path, len);
563 	free(tok, M_TEMP);
564 
565 	/* Get file status. */
566 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
567 	if (error)
568 		goto out;
569 
570 	/* Open /dev/mdctl so that we can attach/detach. */
571 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
572 	    O_RDWR, 0);
573 	if (error)
574 		goto out;
575 
576 	fd = td->td_retval[0];
577 	mdio->md_version = MDIOVERSION;
578 	mdio->md_type = MD_VNODE;
579 
580 	if (root_mount_mddev != -1) {
581 		mdio->md_unit = root_mount_mddev;
582 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
583 		/* Ignore errors. We don't care. */
584 		root_mount_mddev = -1;
585 	}
586 
587 	mdio->md_file = (void *)(mdio + 1);
588 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
589 	mdio->md_mediasize = sb.st_size;
590 	mdio->md_unit = 0;
591 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
592 	if (error)
593 		goto out;
594 
595 	if (mdio->md_unit > 9) {
596 		printf("rootmount: too many md units\n");
597 		mdio->md_file = NULL;
598 		mdio->md_options = 0;
599 		mdio->md_mediasize = 0;
600 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
601 		/* Ignore errors. We don't care. */
602 		error = ERANGE;
603 		goto out;
604 	}
605 
606 	root_mount_mddev = mdio->md_unit;
607 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
608 
609 	error = kern_close(td, fd);
610 
611  out:
612 	free(mdio, M_TEMP);
613 	return (error);
614 }
615 
616 static int
617 parse_dir_onfail(char **conf)
618 {
619 	char *action;
620 	int error;
621 
622 	error = parse_token(conf, &action);
623 	if (error)
624 		return (error);
625 
626 	if (!strcmp(action, "continue"))
627 		root_mount_onfail = A_CONTINUE;
628 	else if (!strcmp(action, "panic"))
629 		root_mount_onfail = A_PANIC;
630 	else if (!strcmp(action, "reboot"))
631 		root_mount_onfail = A_REBOOT;
632 	else if (!strcmp(action, "retry"))
633 		root_mount_onfail = A_RETRY;
634 	else {
635 		printf("rootmount: %s: unknown action\n", action);
636 		error = EINVAL;
637 	}
638 
639 	free(action, M_TEMP);
640 	return (0);
641 }
642 
643 static int
644 parse_dir_timeout(char **conf)
645 {
646 	char *tok, *endtok;
647 	long secs;
648 	int error;
649 
650 	error = parse_token(conf, &tok);
651 	if (error)
652 		return (error);
653 
654 	secs = strtol(tok, &endtok, 0);
655 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
656 	if (!error)
657 		root_mount_timeout = secs;
658 	free(tok, M_TEMP);
659 	return (error);
660 }
661 
662 static int
663 parse_directive(char **conf)
664 {
665 	char *dir;
666 	int error;
667 
668 	error = parse_token(conf, &dir);
669 	if (error)
670 		return (error);
671 
672 	if (strcmp(dir, ".ask") == 0)
673 		error = parse_dir_ask(conf);
674 	else if (strcmp(dir, ".md") == 0)
675 		error = parse_dir_md(conf);
676 	else if (strcmp(dir, ".onfail") == 0)
677 		error = parse_dir_onfail(conf);
678 	else if (strcmp(dir, ".timeout") == 0)
679 		error = parse_dir_timeout(conf);
680 	else {
681 		printf("mountroot: invalid directive `%s'\n", dir);
682 		/* Ignore the rest of the line. */
683 		(void)parse_skipto(conf, '\n');
684 		error = EINVAL;
685 	}
686 	free(dir, M_TEMP);
687 	return (error);
688 }
689 
690 static int
691 parse_mount_dev_present(const char *dev)
692 {
693 	struct nameidata nd;
694 	int error;
695 
696 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
697 	error = namei(&nd);
698 	if (!error)
699 		vput(nd.ni_vp);
700 	NDFREE(&nd, NDF_ONLY_PNBUF);
701 	return (error != 0) ? 0 : 1;
702 }
703 
704 #define	ERRMSGL	255
705 static int
706 parse_mount(char **conf)
707 {
708 	char *errmsg;
709 	struct mntarg *ma;
710 	char *dev, *fs, *opts, *tok;
711 	int delay, error, timeout;
712 
713 	error = parse_token(conf, &tok);
714 	if (error)
715 		return (error);
716 	fs = tok;
717 	error = parse_skipto(&tok, ':');
718 	if (error) {
719 		free(fs, M_TEMP);
720 		return (error);
721 	}
722 	parse_poke(&tok, '\0');
723 	parse_advance(&tok);
724 	dev = tok;
725 
726 	if (root_mount_mddev != -1) {
727 		/* Handle substitution for the md unit number. */
728 		tok = strstr(dev, "md#");
729 		if (tok != NULL)
730 			tok[2] = '0' + root_mount_mddev;
731 	}
732 
733 	/* Parse options. */
734 	error = parse_token(conf, &tok);
735 	opts = (error == 0) ? tok : NULL;
736 
737 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
738 	    (opts != NULL) ? opts : "");
739 
740 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
741 
742 	if (vfs_byname(fs) == NULL) {
743 		strlcpy(errmsg, "unknown file system", ERRMSGL);
744 		error = ENOENT;
745 		goto out;
746 	}
747 
748 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
749 	if (error != 0)
750 		goto out;
751 
752 	delay = hz / 10;
753 	timeout = root_mount_timeout * hz;
754 
755 	for (;;) {
756 		ma = NULL;
757 		ma = mount_arg(ma, "fstype", fs, -1);
758 		ma = mount_arg(ma, "fspath", "/", -1);
759 		ma = mount_arg(ma, "from", dev, -1);
760 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
761 		ma = mount_arg(ma, "ro", NULL, 0);
762 		ma = parse_mountroot_options(ma, opts);
763 
764 		error = kernel_mount(ma, MNT_ROOTFS);
765 		if (error == 0 || timeout <= 0)
766 			break;
767 
768 		if (root_mount_timeout * hz == timeout ||
769 		    (bootverbose && timeout % hz == 0)) {
770 			printf("Mounting from %s:%s failed with error %d; "
771 			    "retrying for %d more second%s\n", fs, dev, error,
772 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
773 		}
774 		pause("rmretry", delay);
775 		timeout -= delay;
776 	}
777  out:
778 	if (error) {
779 		printf("Mounting from %s:%s failed with error %d",
780 		    fs, dev, error);
781 		if (errmsg[0] != '\0')
782 			printf(": %s", errmsg);
783 		printf(".\n");
784 	}
785 	free(fs, M_TEMP);
786 	free(errmsg, M_TEMP);
787 	if (opts != NULL)
788 		free(opts, M_TEMP);
789 	/* kernel_mount can return -1 on error. */
790 	return ((error < 0) ? EDOOFUS : error);
791 }
792 #undef ERRMSGL
793 
794 static int
795 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
796 {
797 	struct mount *mp;
798 	char *conf;
799 	int error;
800 
801 	root_mount_mddev = -1;
802 
803 retry:
804 	conf = sbuf_data(sb);
805 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
806 	error = (mp == NULL) ? 0 : EDOOFUS;
807 	root_mount_onfail = A_CONTINUE;
808 	while (mp == NULL) {
809 		error = parse_skipto(&conf, CC_NONWHITESPACE);
810 		if (error == PE_EOL) {
811 			parse_advance(&conf);
812 			continue;
813 		}
814 		if (error < 0)
815 			break;
816 		switch (parse_peek(&conf)) {
817 		case '#':
818 			error = parse_skipto(&conf, '\n');
819 			break;
820 		case '.':
821 			error = parse_directive(&conf);
822 			break;
823 		default:
824 			error = parse_mount(&conf);
825 			if (error == -1) {
826 				printf("mountroot: invalid file system "
827 				    "specification.\n");
828 				error = 0;
829 			}
830 			break;
831 		}
832 		if (error < 0)
833 			break;
834 		/* Ignore any trailing garbage on the line. */
835 		if (parse_peek(&conf) != '\n') {
836 			printf("mountroot: advancing to next directive...\n");
837 			(void)parse_skipto(&conf, '\n');
838 		}
839 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
840 	}
841 	if (mp != NULL)
842 		return (0);
843 
844 	/*
845 	 * We failed to mount (a new) root.
846 	 */
847 	switch (root_mount_onfail) {
848 	case A_CONTINUE:
849 		break;
850 	case A_PANIC:
851 		panic("mountroot: unable to (re-)mount root.");
852 		/* NOTREACHED */
853 	case A_RETRY:
854 		goto retry;
855 	case A_REBOOT:
856 		kern_reboot(RB_NOSYNC);
857 		/* NOTREACHED */
858 	}
859 
860 	return (error);
861 }
862 
863 static void
864 vfs_mountroot_conf0(struct sbuf *sb)
865 {
866 	char *s, *tok, *mnt, *opt;
867 	int error;
868 
869 	sbuf_printf(sb, ".onfail panic\n");
870 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
871 	if (boothowto & RB_ASKNAME)
872 		sbuf_printf(sb, ".ask\n");
873 #ifdef ROOTDEVNAME
874 	if (boothowto & RB_DFLTROOT)
875 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
876 #endif
877 	if (boothowto & RB_CDROM) {
878 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
879 		sbuf_printf(sb, ".timeout 0\n");
880 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
881 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
882 	}
883 	s = kern_getenv("vfs.root.mountfrom");
884 	if (s != NULL) {
885 		opt = kern_getenv("vfs.root.mountfrom.options");
886 		tok = s;
887 		error = parse_token(&tok, &mnt);
888 		while (!error) {
889 			sbuf_printf(sb, "%s %s\n", mnt,
890 			    (opt != NULL) ? opt : "");
891 			free(mnt, M_TEMP);
892 			error = parse_token(&tok, &mnt);
893 		}
894 		if (opt != NULL)
895 			freeenv(opt);
896 		freeenv(s);
897 	}
898 	if (rootdevnames[0] != NULL)
899 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
900 	if (rootdevnames[1] != NULL)
901 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
902 #ifdef ROOTDEVNAME
903 	if (!(boothowto & RB_DFLTROOT))
904 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
905 #endif
906 	if (!(boothowto & RB_ASKNAME))
907 		sbuf_printf(sb, ".ask\n");
908 }
909 
910 static int
911 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
912 {
913 	static char buf[128];
914 	struct nameidata nd;
915 	off_t ofs;
916 	ssize_t resid;
917 	int error, flags, len;
918 
919 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
920 	flags = FREAD;
921 	error = vn_open(&nd, &flags, 0, NULL);
922 	if (error)
923 		return (error);
924 
925 	NDFREE(&nd, NDF_ONLY_PNBUF);
926 	ofs = 0;
927 	len = sizeof(buf) - 1;
928 	while (1) {
929 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
930 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
931 		    NOCRED, &resid, td);
932 		if (error)
933 			break;
934 		if (resid == len)
935 			break;
936 		buf[len - resid] = 0;
937 		sbuf_printf(sb, "%s", buf);
938 		ofs += len - resid;
939 	}
940 
941 	VOP_UNLOCK(nd.ni_vp, 0);
942 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
943 	return (error);
944 }
945 
946 static void
947 vfs_mountroot_wait(void)
948 {
949 	struct root_hold_token *h;
950 	struct timeval lastfail;
951 	int curfail;
952 
953 	TSENTER();
954 
955 	curfail = 0;
956 	while (1) {
957 		g_waitidle();
958 		mtx_lock(&root_holds_mtx);
959 		if (LIST_EMPTY(&root_holds)) {
960 			mtx_unlock(&root_holds_mtx);
961 			break;
962 		}
963 		if (ppsratecheck(&lastfail, &curfail, 1)) {
964 			printf("Root mount waiting for:");
965 			LIST_FOREACH(h, &root_holds, list)
966 				printf(" %s", h->who);
967 			printf("\n");
968 		}
969 		TSWAIT("root mount");
970 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
971 		    hz);
972 		TSUNWAIT("root mount");
973 	}
974 
975 	TSEXIT();
976 }
977 
978 static int
979 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
980 {
981 	int delay, timeout;
982 
983 	/*
984 	 * In case of ZFS and NFS we don't have a way to wait for
985 	 * specific device.  Also do the wait if the user forced that
986 	 * behaviour by setting vfs.root_mount_always_wait=1.
987 	 */
988 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
989 	    dev[0] == '\0' || root_mount_always_wait != 0) {
990 		vfs_mountroot_wait();
991 		return (0);
992 	}
993 
994 	/*
995 	 * Otherwise, no point in waiting if the device is already there.
996 	 * Note that we must wait for GEOM to finish reconfiguring itself,
997 	 * eg for geom_part(4) to finish tasting.
998 	 */
999 	g_waitidle();
1000 	if (parse_mount_dev_present(dev))
1001 		return (0);
1002 
1003 	/*
1004 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1005 	 * to behave exactly as it used to work before.
1006 	 */
1007 	vfs_mountroot_wait();
1008 	printf("mountroot: waiting for device %s...\n", dev);
1009 	delay = hz / 10;
1010 	timeout = root_mount_timeout * hz;
1011 	do {
1012 		pause("rmdev", delay);
1013 		timeout -= delay;
1014 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1015 
1016 	if (timeout <= 0)
1017 		return (ENODEV);
1018 
1019 	return (0);
1020 }
1021 
1022 void
1023 vfs_mountroot(void)
1024 {
1025 	struct mount *mp;
1026 	struct sbuf *sb;
1027 	struct thread *td;
1028 	time_t timebase;
1029 	int error;
1030 
1031 	mtx_assert(&Giant, MA_NOTOWNED);
1032 
1033 	TSENTER();
1034 
1035 	td = curthread;
1036 
1037 	sb = sbuf_new_auto();
1038 	vfs_mountroot_conf0(sb);
1039 	sbuf_finish(sb);
1040 
1041 	error = vfs_mountroot_devfs(td, &mp);
1042 	while (!error) {
1043 		error = vfs_mountroot_parse(sb, mp);
1044 		if (!error) {
1045 			vfs_mountroot_shuffle(td, mp);
1046 			sbuf_clear(sb);
1047 			error = vfs_mountroot_readconf(td, sb);
1048 			sbuf_finish(sb);
1049 		}
1050 	}
1051 
1052 	sbuf_delete(sb);
1053 
1054 	/*
1055 	 * Iterate over all currently mounted file systems and use
1056 	 * the time stamp found to check and/or initialize the RTC.
1057 	 * Call inittodr() only once and pass it the largest of the
1058 	 * timestamps we encounter.
1059 	 */
1060 	timebase = 0;
1061 	mtx_lock(&mountlist_mtx);
1062 	mp = TAILQ_FIRST(&mountlist);
1063 	while (mp != NULL) {
1064 		if (mp->mnt_time > timebase)
1065 			timebase = mp->mnt_time;
1066 		mp = TAILQ_NEXT(mp, mnt_list);
1067 	}
1068 	mtx_unlock(&mountlist_mtx);
1069 	inittodr(timebase);
1070 
1071 	/* Keep prison0's root in sync with the global rootvnode. */
1072 	mtx_lock(&prison0.pr_mtx);
1073 	prison0.pr_root = rootvnode;
1074 	vref(prison0.pr_root);
1075 	mtx_unlock(&prison0.pr_mtx);
1076 
1077 	mtx_lock(&root_holds_mtx);
1078 	atomic_store_rel_int(&root_mount_complete, 1);
1079 	wakeup(&root_mount_complete);
1080 	mtx_unlock(&root_holds_mtx);
1081 
1082 	EVENTHANDLER_INVOKE(mountroot);
1083 
1084 	TSEXIT();
1085 }
1086 
1087 static struct mntarg *
1088 parse_mountroot_options(struct mntarg *ma, const char *options)
1089 {
1090 	char *p;
1091 	char *name, *name_arg;
1092 	char *val, *val_arg;
1093 	char *opts;
1094 
1095 	if (options == NULL || options[0] == '\0')
1096 		return (ma);
1097 
1098 	p = opts = strdup(options, M_MOUNT);
1099 	if (opts == NULL) {
1100 		return (ma);
1101 	}
1102 
1103 	while((name = strsep(&p, ",")) != NULL) {
1104 		if (name[0] == '\0')
1105 			break;
1106 
1107 		val = strchr(name, '=');
1108 		if (val != NULL) {
1109 			*val = '\0';
1110 			++val;
1111 		}
1112 		if( strcmp(name, "rw") == 0 ||
1113 		    strcmp(name, "noro") == 0) {
1114 			/*
1115 			 * The first time we mount the root file system,
1116 			 * we need to mount 'ro', so We need to ignore
1117 			 * 'rw' and 'noro' mount options.
1118 			 */
1119 			continue;
1120 		}
1121 		name_arg = strdup(name, M_MOUNT);
1122 		val_arg = NULL;
1123 		if (val != NULL)
1124 			val_arg = strdup(val, M_MOUNT);
1125 
1126 		ma = mount_arg(ma, name_arg, val_arg,
1127 		    (val_arg != NULL ? -1 : 0));
1128 	}
1129 	free(opts, M_MOUNT);
1130 	return (ma);
1131 }
1132