xref: /freebsd/sys/kern/vfs_mountroot.c (revision 193d9e768ba63fcfb187cfd17f461f7d41345048)
1 /*-
2  * Copyright (c) 2010 Marcel Moolenaar
3  * Copyright (c) 1999-2004 Poul-Henning Kamp
4  * Copyright (c) 1999 Michael Smith
5  * Copyright (c) 1989, 1993
6  *      The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 #include "opt_rootdevname.h"
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include <sys/param.h>
44 #include <sys/conf.h>
45 #include <sys/cons.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mdioctl.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/filedesc.h>
57 #include <sys/reboot.h>
58 #include <sys/sbuf.h>
59 #include <sys/stat.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/systm.h>
66 #include <sys/vnode.h>
67 
68 #include <geom/geom.h>
69 
70 /*
71  * The root filesystem is detailed in the kernel environment variable
72  * vfs.root.mountfrom, which is expected to be in the general format
73  *
74  * <vfsname>:[<path>][	<vfsname>:[<path>] ...]
75  * vfsname   := the name of a VFS known to the kernel and capable
76  *              of being mounted as root
77  * path      := disk device name or other data used by the filesystem
78  *              to locate its physical store
79  *
80  * If the environment variable vfs.root.mountfrom is a space separated list,
81  * each list element is tried in turn and the root filesystem will be mounted
82  * from the first one that succeeds.
83  *
84  * The environment variable vfs.root.mountfrom.options is a comma delimited
85  * set of string mount options.  These mount options must be parseable
86  * by nmount() in the kernel.
87  */
88 
89 static int parse_mount(char **);
90 static struct mntarg *parse_mountroot_options(struct mntarg *, const char *);
91 static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS);
92 static void vfs_mountroot_wait(void);
93 static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev);
94 
95 /*
96  * The vnode of the system's root (/ in the filesystem, without chroot
97  * active.)
98  */
99 struct vnode *rootvnode;
100 
101 /*
102  * Mount of the system's /dev.
103  */
104 struct mount *rootdevmp;
105 
106 char *rootdevnames[2] = {NULL, NULL};
107 
108 struct mtx root_holds_mtx;
109 MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF);
110 
111 struct root_hold_token {
112 	const char			*who;
113 	LIST_ENTRY(root_hold_token)	list;
114 };
115 
116 static LIST_HEAD(, root_hold_token)	root_holds =
117     LIST_HEAD_INITIALIZER(root_holds);
118 
119 enum action {
120 	A_CONTINUE,
121 	A_PANIC,
122 	A_REBOOT,
123 	A_RETRY
124 };
125 
126 static enum action root_mount_onfail = A_CONTINUE;
127 
128 static int root_mount_mddev;
129 static int root_mount_complete;
130 
131 /* By default wait up to 3 seconds for devices to appear. */
132 static int root_mount_timeout = 3;
133 TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout);
134 
135 static int root_mount_always_wait = 0;
136 SYSCTL_INT(_vfs, OID_AUTO, root_mount_always_wait, CTLFLAG_RDTUN,
137     &root_mount_always_wait, 0,
138     "Wait for root mount holds even if the root device already exists");
139 
140 SYSCTL_PROC(_vfs, OID_AUTO, root_mount_hold,
141     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
142     NULL, 0, sysctl_vfs_root_mount_hold, "A",
143     "List of root mount hold tokens");
144 
145 static int
146 sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS)
147 {
148 	struct sbuf sb;
149 	struct root_hold_token *h;
150 	int error;
151 
152 	sbuf_new(&sb, NULL, 256, SBUF_AUTOEXTEND | SBUF_INCLUDENUL);
153 
154 	mtx_lock(&root_holds_mtx);
155 	LIST_FOREACH(h, &root_holds, list) {
156 		if (h != LIST_FIRST(&root_holds))
157 			sbuf_putc(&sb, ' ');
158 		sbuf_printf(&sb, "%s", h->who);
159 	}
160 	mtx_unlock(&root_holds_mtx);
161 
162 	error = sbuf_finish(&sb);
163 	if (error == 0)
164 		error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
165 	sbuf_delete(&sb);
166 	return (error);
167 }
168 
169 struct root_hold_token *
170 root_mount_hold(const char *identifier)
171 {
172 	struct root_hold_token *h;
173 
174 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
175 	h->who = identifier;
176 	mtx_lock(&root_holds_mtx);
177 	LIST_INSERT_HEAD(&root_holds, h, list);
178 	mtx_unlock(&root_holds_mtx);
179 	return (h);
180 }
181 
182 void
183 root_mount_rel(struct root_hold_token *h)
184 {
185 
186 	KASSERT(h != NULL, ("%s: NULL token", __func__));
187 
188 	mtx_lock(&root_holds_mtx);
189 	LIST_REMOVE(h, list);
190 	wakeup(&root_holds);
191 	mtx_unlock(&root_holds_mtx);
192 	free(h, M_DEVBUF);
193 }
194 
195 int
196 root_mounted(void)
197 {
198 
199 	/* No mutex is acquired here because int stores are atomic. */
200 	return (root_mount_complete);
201 }
202 
203 static void
204 set_rootvnode(void)
205 {
206 	struct proc *p;
207 
208 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
209 		panic("Cannot find root vnode");
210 
211 	VOP_UNLOCK(rootvnode, 0);
212 
213 	p = curthread->td_proc;
214 	FILEDESC_XLOCK(p->p_fd);
215 
216 	if (p->p_fd->fd_cdir != NULL)
217 		vrele(p->p_fd->fd_cdir);
218 	p->p_fd->fd_cdir = rootvnode;
219 	VREF(rootvnode);
220 
221 	if (p->p_fd->fd_rdir != NULL)
222 		vrele(p->p_fd->fd_rdir);
223 	p->p_fd->fd_rdir = rootvnode;
224 	VREF(rootvnode);
225 
226 	FILEDESC_XUNLOCK(p->p_fd);
227 }
228 
229 static int
230 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
231 {
232 	struct vfsoptlist *opts;
233 	struct vfsconf *vfsp;
234 	struct mount *mp;
235 	int error;
236 
237 	*mpp = NULL;
238 
239 	if (rootdevmp != NULL) {
240 		/*
241 		 * Already have /dev; this happens during rerooting.
242 		 */
243 		error = vfs_busy(rootdevmp, 0);
244 		if (error != 0)
245 			return (error);
246 		*mpp = rootdevmp;
247 	} else {
248 		vfsp = vfs_byname("devfs");
249 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
250 		if (vfsp == NULL)
251 			return (ENOENT);
252 
253 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
254 
255 		error = VFS_MOUNT(mp);
256 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
257 		if (error)
258 			return (error);
259 
260 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
261 		TAILQ_INIT(opts);
262 		mp->mnt_opt = opts;
263 
264 		mtx_lock(&mountlist_mtx);
265 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
266 		mtx_unlock(&mountlist_mtx);
267 
268 		*mpp = mp;
269 		rootdevmp = mp;
270 	}
271 
272 	set_rootvnode();
273 
274 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
275 	if (error)
276 		printf("kern_symlink /dev -> / returns %d\n", error);
277 
278 	return (error);
279 }
280 
281 static void
282 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
283 {
284 	struct nameidata nd;
285 	struct mount *mporoot, *mpnroot;
286 	struct vnode *vp, *vporoot, *vpdevfs;
287 	char *fspath;
288 	int error;
289 
290 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
291 
292 	/* Shuffle the mountlist. */
293 	mtx_lock(&mountlist_mtx);
294 	mporoot = TAILQ_FIRST(&mountlist);
295 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
296 	if (mporoot != mpdevfs) {
297 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
298 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
299 	}
300 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
301 	mtx_unlock(&mountlist_mtx);
302 
303 	cache_purgevfs(mporoot, true);
304 	if (mporoot != mpdevfs)
305 		cache_purgevfs(mpdevfs, true);
306 
307 	VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot);
308 
309 	VI_LOCK(vporoot);
310 	vporoot->v_iflag &= ~VI_MOUNT;
311 	VI_UNLOCK(vporoot);
312 	vporoot->v_mountedhere = NULL;
313 	mporoot->mnt_flag &= ~MNT_ROOTFS;
314 	mporoot->mnt_vnodecovered = NULL;
315 	vput(vporoot);
316 
317 	/* Set up the new rootvnode, and purge the cache */
318 	mpnroot->mnt_vnodecovered = NULL;
319 	set_rootvnode();
320 	cache_purgevfs(rootvnode->v_mount, true);
321 
322 	if (mporoot != mpdevfs) {
323 		/* Remount old root under /.mount or /mnt */
324 		fspath = "/.mount";
325 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
326 		    fspath, td);
327 		error = namei(&nd);
328 		if (error) {
329 			NDFREE(&nd, NDF_ONLY_PNBUF);
330 			fspath = "/mnt";
331 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
332 			    fspath, td);
333 			error = namei(&nd);
334 		}
335 		if (!error) {
336 			vp = nd.ni_vp;
337 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
338 			if (!error)
339 				error = vinvalbuf(vp, V_SAVE, 0, 0);
340 			if (!error) {
341 				cache_purge(vp);
342 				mporoot->mnt_vnodecovered = vp;
343 				vp->v_mountedhere = mporoot;
344 				strlcpy(mporoot->mnt_stat.f_mntonname,
345 				    fspath, MNAMELEN);
346 				VOP_UNLOCK(vp, 0);
347 			} else
348 				vput(vp);
349 		}
350 		NDFREE(&nd, NDF_ONLY_PNBUF);
351 
352 		if (error)
353 			printf("mountroot: unable to remount previous root "
354 			    "under /.mount or /mnt (error %d)\n", error);
355 	}
356 
357 	/* Remount devfs under /dev */
358 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
359 	error = namei(&nd);
360 	if (!error) {
361 		vp = nd.ni_vp;
362 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
363 		if (!error)
364 			error = vinvalbuf(vp, V_SAVE, 0, 0);
365 		if (!error) {
366 			vpdevfs = mpdevfs->mnt_vnodecovered;
367 			if (vpdevfs != NULL) {
368 				cache_purge(vpdevfs);
369 				vpdevfs->v_mountedhere = NULL;
370 				vrele(vpdevfs);
371 			}
372 			mpdevfs->mnt_vnodecovered = vp;
373 			vp->v_mountedhere = mpdevfs;
374 			VOP_UNLOCK(vp, 0);
375 		} else
376 			vput(vp);
377 	}
378 	if (error)
379 		printf("mountroot: unable to remount devfs under /dev "
380 		    "(error %d)\n", error);
381 	NDFREE(&nd, NDF_ONLY_PNBUF);
382 
383 	if (mporoot == mpdevfs) {
384 		vfs_unbusy(mpdevfs);
385 		/* Unlink the no longer needed /dev/dev -> / symlink */
386 		error = kern_unlinkat(td, AT_FDCWD, "/dev/dev",
387 		    UIO_SYSSPACE, 0);
388 		if (error)
389 			printf("mountroot: unable to unlink /dev/dev "
390 			    "(error %d)\n", error);
391 	}
392 }
393 
394 /*
395  * Configuration parser.
396  */
397 
398 /* Parser character classes. */
399 #define	CC_WHITESPACE		-1
400 #define	CC_NONWHITESPACE	-2
401 
402 /* Parse errors. */
403 #define	PE_EOF			-1
404 #define	PE_EOL			-2
405 
406 static __inline int
407 parse_peek(char **conf)
408 {
409 
410 	return (**conf);
411 }
412 
413 static __inline void
414 parse_poke(char **conf, int c)
415 {
416 
417 	**conf = c;
418 }
419 
420 static __inline void
421 parse_advance(char **conf)
422 {
423 
424 	(*conf)++;
425 }
426 
427 static int
428 parse_skipto(char **conf, int mc)
429 {
430 	int c, match;
431 
432 	while (1) {
433 		c = parse_peek(conf);
434 		if (c == 0)
435 			return (PE_EOF);
436 		switch (mc) {
437 		case CC_WHITESPACE:
438 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
439 			break;
440 		case CC_NONWHITESPACE:
441 			if (c == '\n')
442 				return (PE_EOL);
443 			match = (c != ' ' && c != '\t') ? 1 : 0;
444 			break;
445 		default:
446 			match = (c == mc) ? 1 : 0;
447 			break;
448 		}
449 		if (match)
450 			break;
451 		parse_advance(conf);
452 	}
453 	return (0);
454 }
455 
456 static int
457 parse_token(char **conf, char **tok)
458 {
459 	char *p;
460 	size_t len;
461 	int error;
462 
463 	*tok = NULL;
464 	error = parse_skipto(conf, CC_NONWHITESPACE);
465 	if (error)
466 		return (error);
467 	p = *conf;
468 	error = parse_skipto(conf, CC_WHITESPACE);
469 	len = *conf - p;
470 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
471 	bcopy(p, *tok, len);
472 	return (0);
473 }
474 
475 static void
476 parse_dir_ask_printenv(const char *var)
477 {
478 	char *val;
479 
480 	val = kern_getenv(var);
481 	if (val != NULL) {
482 		printf("  %s=%s\n", var, val);
483 		freeenv(val);
484 	}
485 }
486 
487 static int
488 parse_dir_ask(char **conf)
489 {
490 	char name[80];
491 	char *mnt;
492 	int error;
493 
494 	vfs_mountroot_wait();
495 
496 	printf("\nLoader variables:\n");
497 	parse_dir_ask_printenv("vfs.root.mountfrom");
498 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
499 
500 	printf("\nManual root filesystem specification:\n");
501 	printf("  <fstype>:<device> [options]\n");
502 	printf("      Mount <device> using filesystem <fstype>\n");
503 	printf("      and with the specified (optional) option list.\n");
504 	printf("\n");
505 	printf("    eg. ufs:/dev/da0s1a\n");
506 	printf("        zfs:tank\n");
507 	printf("        cd9660:/dev/cd0 ro\n");
508 	printf("          (which is equivalent to: ");
509 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
510 	printf("\n");
511 	printf("  ?               List valid disk boot devices\n");
512 	printf("  .               Yield 1 second (for background tasks)\n");
513 	printf("  <empty line>    Abort manual input\n");
514 
515 	do {
516 		error = EINVAL;
517 		printf("\nmountroot> ");
518 		cngets(name, sizeof(name), GETS_ECHO);
519 		if (name[0] == '\0')
520 			break;
521 		if (name[0] == '?' && name[1] == '\0') {
522 			printf("\nList of GEOM managed disk devices:\n  ");
523 			g_dev_print();
524 			continue;
525 		}
526 		if (name[0] == '.' && name[1] == '\0') {
527 			pause("rmask", hz);
528 			continue;
529 		}
530 		mnt = name;
531 		error = parse_mount(&mnt);
532 		if (error == -1)
533 			printf("Invalid file system specification.\n");
534 	} while (error != 0);
535 
536 	return (error);
537 }
538 
539 static int
540 parse_dir_md(char **conf)
541 {
542 	struct stat sb;
543 	struct thread *td;
544 	struct md_ioctl *mdio;
545 	char *path, *tok;
546 	int error, fd, len;
547 
548 	td = curthread;
549 
550 	error = parse_token(conf, &tok);
551 	if (error)
552 		return (error);
553 
554 	len = strlen(tok);
555 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
556 	path = (void *)(mdio + 1);
557 	bcopy(tok, path, len);
558 	free(tok, M_TEMP);
559 
560 	/* Get file status. */
561 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
562 	if (error)
563 		goto out;
564 
565 	/* Open /dev/mdctl so that we can attach/detach. */
566 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
567 	    O_RDWR, 0);
568 	if (error)
569 		goto out;
570 
571 	fd = td->td_retval[0];
572 	mdio->md_version = MDIOVERSION;
573 	mdio->md_type = MD_VNODE;
574 
575 	if (root_mount_mddev != -1) {
576 		mdio->md_unit = root_mount_mddev;
577 		DROP_GIANT();
578 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
579 		PICKUP_GIANT();
580 		/* Ignore errors. We don't care. */
581 		root_mount_mddev = -1;
582 	}
583 
584 	mdio->md_file = (void *)(mdio + 1);
585 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
586 	mdio->md_mediasize = sb.st_size;
587 	mdio->md_unit = 0;
588 	DROP_GIANT();
589 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
590 	PICKUP_GIANT();
591 	if (error)
592 		goto out;
593 
594 	if (mdio->md_unit > 9) {
595 		printf("rootmount: too many md units\n");
596 		mdio->md_file = NULL;
597 		mdio->md_options = 0;
598 		mdio->md_mediasize = 0;
599 		DROP_GIANT();
600 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
601 		PICKUP_GIANT();
602 		/* Ignore errors. We don't care. */
603 		error = ERANGE;
604 		goto out;
605 	}
606 
607 	root_mount_mddev = mdio->md_unit;
608 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
609 
610 	error = kern_close(td, fd);
611 
612  out:
613 	free(mdio, M_TEMP);
614 	return (error);
615 }
616 
617 static int
618 parse_dir_onfail(char **conf)
619 {
620 	char *action;
621 	int error;
622 
623 	error = parse_token(conf, &action);
624 	if (error)
625 		return (error);
626 
627 	if (!strcmp(action, "continue"))
628 		root_mount_onfail = A_CONTINUE;
629 	else if (!strcmp(action, "panic"))
630 		root_mount_onfail = A_PANIC;
631 	else if (!strcmp(action, "reboot"))
632 		root_mount_onfail = A_REBOOT;
633 	else if (!strcmp(action, "retry"))
634 		root_mount_onfail = A_RETRY;
635 	else {
636 		printf("rootmount: %s: unknown action\n", action);
637 		error = EINVAL;
638 	}
639 
640 	free(action, M_TEMP);
641 	return (0);
642 }
643 
644 static int
645 parse_dir_timeout(char **conf)
646 {
647 	char *tok, *endtok;
648 	long secs;
649 	int error;
650 
651 	error = parse_token(conf, &tok);
652 	if (error)
653 		return (error);
654 
655 	secs = strtol(tok, &endtok, 0);
656 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
657 	if (!error)
658 		root_mount_timeout = secs;
659 	free(tok, M_TEMP);
660 	return (error);
661 }
662 
663 static int
664 parse_directive(char **conf)
665 {
666 	char *dir;
667 	int error;
668 
669 	error = parse_token(conf, &dir);
670 	if (error)
671 		return (error);
672 
673 	if (strcmp(dir, ".ask") == 0)
674 		error = parse_dir_ask(conf);
675 	else if (strcmp(dir, ".md") == 0)
676 		error = parse_dir_md(conf);
677 	else if (strcmp(dir, ".onfail") == 0)
678 		error = parse_dir_onfail(conf);
679 	else if (strcmp(dir, ".timeout") == 0)
680 		error = parse_dir_timeout(conf);
681 	else {
682 		printf("mountroot: invalid directive `%s'\n", dir);
683 		/* Ignore the rest of the line. */
684 		(void)parse_skipto(conf, '\n');
685 		error = EINVAL;
686 	}
687 	free(dir, M_TEMP);
688 	return (error);
689 }
690 
691 static int
692 parse_mount_dev_present(const char *dev)
693 {
694 	struct nameidata nd;
695 	int error;
696 
697 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
698 	error = namei(&nd);
699 	if (!error)
700 		vput(nd.ni_vp);
701 	NDFREE(&nd, NDF_ONLY_PNBUF);
702 	return (error != 0) ? 0 : 1;
703 }
704 
705 #define	ERRMSGL	255
706 static int
707 parse_mount(char **conf)
708 {
709 	char *errmsg;
710 	struct mntarg *ma;
711 	char *dev, *fs, *opts, *tok;
712 	int error;
713 
714 	error = parse_token(conf, &tok);
715 	if (error)
716 		return (error);
717 	fs = tok;
718 	error = parse_skipto(&tok, ':');
719 	if (error) {
720 		free(fs, M_TEMP);
721 		return (error);
722 	}
723 	parse_poke(&tok, '\0');
724 	parse_advance(&tok);
725 	dev = tok;
726 
727 	if (root_mount_mddev != -1) {
728 		/* Handle substitution for the md unit number. */
729 		tok = strstr(dev, "md#");
730 		if (tok != NULL)
731 			tok[2] = '0' + root_mount_mddev;
732 	}
733 
734 	/* Parse options. */
735 	error = parse_token(conf, &tok);
736 	opts = (error == 0) ? tok : NULL;
737 
738 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
739 	    (opts != NULL) ? opts : "");
740 
741 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
742 
743 	if (vfs_byname(fs) == NULL) {
744 		strlcpy(errmsg, "unknown file system", ERRMSGL);
745 		error = ENOENT;
746 		goto out;
747 	}
748 
749 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
750 	if (error != 0)
751 		goto out;
752 
753 	ma = NULL;
754 	ma = mount_arg(ma, "fstype", fs, -1);
755 	ma = mount_arg(ma, "fspath", "/", -1);
756 	ma = mount_arg(ma, "from", dev, -1);
757 	ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
758 	ma = mount_arg(ma, "ro", NULL, 0);
759 	ma = parse_mountroot_options(ma, opts);
760 	error = kernel_mount(ma, MNT_ROOTFS);
761 
762  out:
763 	if (error) {
764 		printf("Mounting from %s:%s failed with error %d",
765 		    fs, dev, error);
766 		if (errmsg[0] != '\0')
767 			printf(": %s", errmsg);
768 		printf(".\n");
769 	}
770 	free(fs, M_TEMP);
771 	free(errmsg, M_TEMP);
772 	if (opts != NULL)
773 		free(opts, M_TEMP);
774 	/* kernel_mount can return -1 on error. */
775 	return ((error < 0) ? EDOOFUS : error);
776 }
777 #undef ERRMSGL
778 
779 static int
780 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
781 {
782 	struct mount *mp;
783 	char *conf;
784 	int error;
785 
786 	root_mount_mddev = -1;
787 
788 retry:
789 	conf = sbuf_data(sb);
790 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
791 	error = (mp == NULL) ? 0 : EDOOFUS;
792 	root_mount_onfail = A_CONTINUE;
793 	while (mp == NULL) {
794 		error = parse_skipto(&conf, CC_NONWHITESPACE);
795 		if (error == PE_EOL) {
796 			parse_advance(&conf);
797 			continue;
798 		}
799 		if (error < 0)
800 			break;
801 		switch (parse_peek(&conf)) {
802 		case '#':
803 			error = parse_skipto(&conf, '\n');
804 			break;
805 		case '.':
806 			error = parse_directive(&conf);
807 			break;
808 		default:
809 			error = parse_mount(&conf);
810 			if (error == -1) {
811 				printf("mountroot: invalid file system "
812 				    "specification.\n");
813 				error = 0;
814 			}
815 			break;
816 		}
817 		if (error < 0)
818 			break;
819 		/* Ignore any trailing garbage on the line. */
820 		if (parse_peek(&conf) != '\n') {
821 			printf("mountroot: advancing to next directive...\n");
822 			(void)parse_skipto(&conf, '\n');
823 		}
824 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
825 	}
826 	if (mp != NULL)
827 		return (0);
828 
829 	/*
830 	 * We failed to mount (a new) root.
831 	 */
832 	switch (root_mount_onfail) {
833 	case A_CONTINUE:
834 		break;
835 	case A_PANIC:
836 		panic("mountroot: unable to (re-)mount root.");
837 		/* NOTREACHED */
838 	case A_RETRY:
839 		goto retry;
840 	case A_REBOOT:
841 		kern_reboot(RB_NOSYNC);
842 		/* NOTREACHED */
843 	}
844 
845 	return (error);
846 }
847 
848 static void
849 vfs_mountroot_conf0(struct sbuf *sb)
850 {
851 	char *s, *tok, *mnt, *opt;
852 	int error;
853 
854 	sbuf_printf(sb, ".onfail panic\n");
855 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
856 	if (boothowto & RB_ASKNAME)
857 		sbuf_printf(sb, ".ask\n");
858 #ifdef ROOTDEVNAME
859 	if (boothowto & RB_DFLTROOT)
860 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
861 #endif
862 	if (boothowto & RB_CDROM) {
863 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
864 		sbuf_printf(sb, ".timeout 0\n");
865 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
866 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
867 	}
868 	s = kern_getenv("vfs.root.mountfrom");
869 	if (s != NULL) {
870 		opt = kern_getenv("vfs.root.mountfrom.options");
871 		tok = s;
872 		error = parse_token(&tok, &mnt);
873 		while (!error) {
874 			sbuf_printf(sb, "%s %s\n", mnt,
875 			    (opt != NULL) ? opt : "");
876 			free(mnt, M_TEMP);
877 			error = parse_token(&tok, &mnt);
878 		}
879 		if (opt != NULL)
880 			freeenv(opt);
881 		freeenv(s);
882 	}
883 	if (rootdevnames[0] != NULL)
884 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
885 	if (rootdevnames[1] != NULL)
886 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
887 #ifdef ROOTDEVNAME
888 	if (!(boothowto & RB_DFLTROOT))
889 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
890 #endif
891 	if (!(boothowto & RB_ASKNAME))
892 		sbuf_printf(sb, ".ask\n");
893 }
894 
895 static int
896 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
897 {
898 	static char buf[128];
899 	struct nameidata nd;
900 	off_t ofs;
901 	ssize_t resid;
902 	int error, flags, len;
903 
904 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
905 	flags = FREAD;
906 	error = vn_open(&nd, &flags, 0, NULL);
907 	if (error)
908 		return (error);
909 
910 	NDFREE(&nd, NDF_ONLY_PNBUF);
911 	ofs = 0;
912 	len = sizeof(buf) - 1;
913 	while (1) {
914 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
915 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
916 		    NOCRED, &resid, td);
917 		if (error)
918 			break;
919 		if (resid == len)
920 			break;
921 		buf[len - resid] = 0;
922 		sbuf_printf(sb, "%s", buf);
923 		ofs += len - resid;
924 	}
925 
926 	VOP_UNLOCK(nd.ni_vp, 0);
927 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
928 	return (error);
929 }
930 
931 static void
932 vfs_mountroot_wait(void)
933 {
934 	struct root_hold_token *h;
935 	struct timeval lastfail;
936 	int curfail;
937 
938 	curfail = 0;
939 	while (1) {
940 		DROP_GIANT();
941 		g_waitidle();
942 		PICKUP_GIANT();
943 		mtx_lock(&root_holds_mtx);
944 		if (LIST_EMPTY(&root_holds)) {
945 			mtx_unlock(&root_holds_mtx);
946 			break;
947 		}
948 		if (ppsratecheck(&lastfail, &curfail, 1)) {
949 			printf("Root mount waiting for:");
950 			LIST_FOREACH(h, &root_holds, list)
951 				printf(" %s", h->who);
952 			printf("\n");
953 		}
954 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
955 		    hz);
956 	}
957 }
958 
959 static int
960 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
961 {
962 	int delay, timeout;
963 
964 	/*
965 	 * In case of ZFS and NFS we don't have a way to wait for
966 	 * specific device.  Also do the wait if the user forced that
967 	 * behaviour by setting vfs.root_mount_always_wait=1.
968 	 */
969 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
970 	    dev[0] == '\0' || root_mount_always_wait != 0) {
971 		vfs_mountroot_wait();
972 		return (0);
973 	}
974 
975 	/*
976 	 * Otherwise, no point in waiting if the device is already there.
977 	 * Note that we must wait for GEOM to finish reconfiguring itself,
978 	 * eg for geom_part(4) to finish tasting.
979 	 */
980 	DROP_GIANT();
981 	g_waitidle();
982 	PICKUP_GIANT();
983 	if (parse_mount_dev_present(dev))
984 		return (0);
985 
986 	/*
987 	 * No luck.  Let's wait.  This code looks weird, but it's that way
988 	 * to behave exactly as it used to work before.
989 	 */
990 	vfs_mountroot_wait();
991 	printf("mountroot: waiting for device %s...\n", dev);
992 	delay = hz / 10;
993 	timeout = root_mount_timeout * hz;
994 	do {
995 		pause("rmdev", delay);
996 		timeout -= delay;
997 	} while (timeout > 0 && !parse_mount_dev_present(dev));
998 
999 	if (timeout <= 0)
1000 		return (ENODEV);
1001 
1002 	return (0);
1003 }
1004 
1005 void
1006 vfs_mountroot(void)
1007 {
1008 	struct mount *mp;
1009 	struct sbuf *sb;
1010 	struct thread *td;
1011 	time_t timebase;
1012 	int error;
1013 
1014 	td = curthread;
1015 
1016 	sb = sbuf_new_auto();
1017 	vfs_mountroot_conf0(sb);
1018 	sbuf_finish(sb);
1019 
1020 	error = vfs_mountroot_devfs(td, &mp);
1021 	while (!error) {
1022 		error = vfs_mountroot_parse(sb, mp);
1023 		if (!error) {
1024 			vfs_mountroot_shuffle(td, mp);
1025 			sbuf_clear(sb);
1026 			error = vfs_mountroot_readconf(td, sb);
1027 			sbuf_finish(sb);
1028 		}
1029 	}
1030 
1031 	sbuf_delete(sb);
1032 
1033 	/*
1034 	 * Iterate over all currently mounted file systems and use
1035 	 * the time stamp found to check and/or initialize the RTC.
1036 	 * Call inittodr() only once and pass it the largest of the
1037 	 * timestamps we encounter.
1038 	 */
1039 	timebase = 0;
1040 	mtx_lock(&mountlist_mtx);
1041 	mp = TAILQ_FIRST(&mountlist);
1042 	while (mp != NULL) {
1043 		if (mp->mnt_time > timebase)
1044 			timebase = mp->mnt_time;
1045 		mp = TAILQ_NEXT(mp, mnt_list);
1046 	}
1047 	mtx_unlock(&mountlist_mtx);
1048 	inittodr(timebase);
1049 
1050 	/* Keep prison0's root in sync with the global rootvnode. */
1051 	mtx_lock(&prison0.pr_mtx);
1052 	prison0.pr_root = rootvnode;
1053 	vref(prison0.pr_root);
1054 	mtx_unlock(&prison0.pr_mtx);
1055 
1056 	mtx_lock(&root_holds_mtx);
1057 	atomic_store_rel_int(&root_mount_complete, 1);
1058 	wakeup(&root_mount_complete);
1059 	mtx_unlock(&root_holds_mtx);
1060 
1061 	EVENTHANDLER_INVOKE(mountroot);
1062 }
1063 
1064 static struct mntarg *
1065 parse_mountroot_options(struct mntarg *ma, const char *options)
1066 {
1067 	char *p;
1068 	char *name, *name_arg;
1069 	char *val, *val_arg;
1070 	char *opts;
1071 
1072 	if (options == NULL || options[0] == '\0')
1073 		return (ma);
1074 
1075 	p = opts = strdup(options, M_MOUNT);
1076 	if (opts == NULL) {
1077 		return (ma);
1078 	}
1079 
1080 	while((name = strsep(&p, ",")) != NULL) {
1081 		if (name[0] == '\0')
1082 			break;
1083 
1084 		val = strchr(name, '=');
1085 		if (val != NULL) {
1086 			*val = '\0';
1087 			++val;
1088 		}
1089 		if( strcmp(name, "rw") == 0 ||
1090 		    strcmp(name, "noro") == 0) {
1091 			/*
1092 			 * The first time we mount the root file system,
1093 			 * we need to mount 'ro', so We need to ignore
1094 			 * 'rw' and 'noro' mount options.
1095 			 */
1096 			continue;
1097 		}
1098 		name_arg = strdup(name, M_MOUNT);
1099 		val_arg = NULL;
1100 		if (val != NULL)
1101 			val_arg = strdup(val, M_MOUNT);
1102 
1103 		ma = mount_arg(ma, name_arg, val_arg,
1104 		    (val_arg != NULL ? -1 : 0));
1105 	}
1106 	free(opts, M_MOUNT);
1107 	return (ma);
1108 }
1109