xref: /freebsd/sys/kern/vfs_mountroot.c (revision 783d3ff6d7fae619db8a7990b8a6387de0c677b5)
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/param.h>
43 #include <sys/conf.h>
44 #include <sys/cons.h>
45 #include <sys/eventhandler.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/systm.h>
65 #include <sys/vnode.h>
66 
67 #include <geom/geom.h>
68 
69 /*
70  * The root filesystem is detailed in the kernel environment variable
71  * vfs.root.mountfrom, which is expected to be in the general format
72  *
73  * <vfsname>:[<path>][	<vfsname>:[<path>] ...]
74  * vfsname   := the name of a VFS known to the kernel and capable
75  *              of being mounted as root
76  * path      := disk device name or other data used by the filesystem
77  *              to locate its physical store
78  *
79  * If the environment variable vfs.root.mountfrom is a space separated list,
80  * each list element is tried in turn and the root filesystem will be mounted
81  * from the first one that succeeds.
82  *
83  * The environment variable vfs.root.mountfrom.options is a comma delimited
84  * set of string mount options.  These mount options must be parseable
85  * by nmount() in the kernel.
86  */
87 
88 static int parse_mount(char **);
89 static struct mntarg *parse_mountroot_options(struct mntarg *, const char *);
90 static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS);
91 static void vfs_mountroot_wait(void);
92 static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev);
93 
94 /*
95  * The vnode of the system's root (/ in the filesystem, without chroot
96  * active.)
97  */
98 struct vnode *rootvnode;
99 
100 /*
101  * Mount of the system's /dev.
102  */
103 struct mount *rootdevmp;
104 
105 char *rootdevnames[2] = {NULL, NULL};
106 
107 struct mtx root_holds_mtx;
108 MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF);
109 
110 static TAILQ_HEAD(, root_hold_token)	root_holds =
111     TAILQ_HEAD_INITIALIZER(root_holds);
112 
113 enum action {
114 	A_CONTINUE,
115 	A_PANIC,
116 	A_REBOOT,
117 	A_RETRY
118 };
119 
120 enum rh_flags {
121 	RH_FREE,
122 	RH_ALLOC,
123 	RH_ARG,
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 	TAILQ_FOREACH(h, &root_holds, list) {
156 		if (h != TAILQ_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->flags = RH_ALLOC;
176 	h->who = identifier;
177 	mtx_lock(&root_holds_mtx);
178 	TSHOLD("root mount");
179 	TAILQ_INSERT_TAIL(&root_holds, h, list);
180 	mtx_unlock(&root_holds_mtx);
181 	return (h);
182 }
183 
184 void
185 root_mount_hold_token(const char *identifier, struct root_hold_token *h)
186 {
187 #ifdef INVARIANTS
188 	struct root_hold_token *t;
189 #endif
190 
191 	h->flags = RH_ARG;
192 	h->who = identifier;
193 	mtx_lock(&root_holds_mtx);
194 #ifdef INVARIANTS
195 	TAILQ_FOREACH(t, &root_holds, list) {
196 		if (t == h) {
197 			panic("Duplicate mount hold by '%s' on %p",
198 			    identifier, h);
199 		}
200 	}
201 #endif
202 	TSHOLD("root mount");
203 	TAILQ_INSERT_TAIL(&root_holds, h, list);
204 	mtx_unlock(&root_holds_mtx);
205 }
206 
207 void
208 root_mount_rel(struct root_hold_token *h)
209 {
210 
211 	if (h == NULL || h->flags == RH_FREE)
212 		return;
213 
214 	mtx_lock(&root_holds_mtx);
215 	TAILQ_REMOVE(&root_holds, h, list);
216 	TSRELEASE("root mount");
217 	wakeup(&root_holds);
218 	mtx_unlock(&root_holds_mtx);
219 	if (h->flags == RH_ALLOC) {
220 		free(h, M_DEVBUF);
221 	} else
222 		h->flags = RH_FREE;
223 }
224 
225 int
226 root_mounted(void)
227 {
228 
229 	/* No mutex is acquired here because int stores are atomic. */
230 	return (root_mount_complete);
231 }
232 
233 static void
234 set_rootvnode(void)
235 {
236 
237 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
238 		panic("set_rootvnode: Cannot find root vnode");
239 
240 	VOP_UNLOCK(rootvnode);
241 
242 	pwd_set_rootvnode();
243 }
244 
245 static int
246 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
247 {
248 	struct vfsoptlist *opts;
249 	struct vfsconf *vfsp;
250 	struct mount *mp;
251 	int error;
252 
253 	*mpp = NULL;
254 
255 	if (rootdevmp != NULL) {
256 		/*
257 		 * Already have /dev; this happens during rerooting.
258 		 */
259 		error = vfs_busy(rootdevmp, 0);
260 		if (error != 0)
261 			return (error);
262 		*mpp = rootdevmp;
263 	} else {
264 		vfsp = vfs_byname("devfs");
265 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
266 		if (vfsp == NULL)
267 			return (ENOENT);
268 
269 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
270 
271 		error = VFS_MOUNT(mp);
272 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
273 		if (error)
274 			return (error);
275 
276 		error = VFS_STATFS(mp, &mp->mnt_stat);
277 		KASSERT(error == 0, ("VFS_STATFS(devfs) failed %d", error));
278 		if (error)
279 			return (error);
280 
281 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
282 		TAILQ_INIT(opts);
283 		mp->mnt_opt = opts;
284 
285 		mtx_lock(&mountlist_mtx);
286 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
287 		mtx_unlock(&mountlist_mtx);
288 
289 		*mpp = mp;
290 		rootdevmp = mp;
291 		vfs_op_exit(mp);
292 	}
293 
294 	set_rootvnode();
295 
296 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
297 	if (error)
298 		printf("kern_symlink /dev -> / returns %d\n", error);
299 
300 	return (error);
301 }
302 
303 static void
304 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
305 {
306 	struct nameidata nd;
307 	struct mount *mporoot, *mpnroot;
308 	struct vnode *vp, *vporoot, *vpdevfs;
309 	char *fspath;
310 	int error;
311 
312 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
313 
314 	/* Shuffle the mountlist. */
315 	mtx_lock(&mountlist_mtx);
316 	mporoot = TAILQ_FIRST(&mountlist);
317 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
318 	if (mporoot != mpdevfs) {
319 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
320 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
321 	}
322 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
323 	mtx_unlock(&mountlist_mtx);
324 
325 	cache_purgevfs(mporoot);
326 	if (mporoot != mpdevfs)
327 		cache_purgevfs(mpdevfs);
328 
329 	if (VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot))
330 		panic("vfs_mountroot_shuffle: Cannot find root vnode");
331 
332 	VI_LOCK(vporoot);
333 	vporoot->v_iflag &= ~VI_MOUNT;
334 	vn_irflag_unset_locked(vporoot, VIRF_MOUNTPOINT);
335 	vporoot->v_mountedhere = NULL;
336 	VI_UNLOCK(vporoot);
337 	mporoot->mnt_flag &= ~MNT_ROOTFS;
338 	mporoot->mnt_vnodecovered = NULL;
339 	vput(vporoot);
340 
341 	/* Set up the new rootvnode, and purge the cache */
342 	mpnroot->mnt_vnodecovered = NULL;
343 	set_rootvnode();
344 	cache_purgevfs(rootvnode->v_mount);
345 
346 	if (mporoot != mpdevfs) {
347 		/* Remount old root under /.mount or /mnt */
348 		fspath = "/.mount";
349 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath);
350 		error = namei(&nd);
351 		if (error) {
352 			fspath = "/mnt";
353 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
354 			    fspath);
355 			error = namei(&nd);
356 		}
357 		if (!error) {
358 			NDFREE_PNBUF(&nd);
359 			vp = nd.ni_vp;
360 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
361 			if (!error)
362 				error = vinvalbuf(vp, V_SAVE, 0, 0);
363 			if (!error) {
364 				cache_purge(vp);
365 				VI_LOCK(vp);
366 				mporoot->mnt_vnodecovered = vp;
367 				vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
368 				vp->v_mountedhere = mporoot;
369 				strlcpy(mporoot->mnt_stat.f_mntonname,
370 				    fspath, MNAMELEN);
371 				VI_UNLOCK(vp);
372 				VOP_UNLOCK(vp);
373 			} else
374 				vput(vp);
375 		}
376 
377 		if (error)
378 			printf("mountroot: unable to remount previous root "
379 			    "under /.mount or /mnt (error %d)\n", error);
380 	}
381 
382 	/* Remount devfs under /dev */
383 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev");
384 	error = namei(&nd);
385 	if (!error) {
386 		NDFREE_PNBUF(&nd);
387 		vp = nd.ni_vp;
388 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
389 		if (!error)
390 			error = vinvalbuf(vp, V_SAVE, 0, 0);
391 		if (!error) {
392 			vpdevfs = mpdevfs->mnt_vnodecovered;
393 			if (vpdevfs != NULL) {
394 				cache_purge(vpdevfs);
395 				VI_LOCK(vpdevfs);
396 				vn_irflag_unset_locked(vpdevfs, VIRF_MOUNTPOINT);
397 				vpdevfs->v_mountedhere = NULL;
398 				VI_UNLOCK(vpdevfs);
399 				vrele(vpdevfs);
400 			}
401 			VI_LOCK(vp);
402 			mpdevfs->mnt_vnodecovered = vp;
403 			vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
404 			vp->v_mountedhere = mpdevfs;
405 			VI_UNLOCK(vp);
406 			VOP_UNLOCK(vp);
407 		} else
408 			vput(vp);
409 	}
410 	if (error)
411 		printf("mountroot: unable to remount devfs under /dev "
412 		    "(error %d)\n", error);
413 
414 	if (mporoot == mpdevfs) {
415 		vfs_unbusy(mpdevfs);
416 		/* Unlink the no longer needed /dev/dev -> / symlink */
417 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
418 		    UIO_SYSSPACE, 0, 0);
419 		if (error)
420 			printf("mountroot: unable to unlink /dev/dev "
421 			    "(error %d)\n", error);
422 	}
423 }
424 
425 /*
426  * Configuration parser.
427  */
428 
429 /* Parser character classes. */
430 #define	CC_WHITESPACE		-1
431 #define	CC_NONWHITESPACE	-2
432 
433 /* Parse errors. */
434 #define	PE_EOF			-1
435 #define	PE_EOL			-2
436 
437 static __inline int
438 parse_peek(char **conf)
439 {
440 
441 	return (**conf);
442 }
443 
444 static __inline void
445 parse_poke(char **conf, int c)
446 {
447 
448 	**conf = c;
449 }
450 
451 static __inline void
452 parse_advance(char **conf)
453 {
454 
455 	(*conf)++;
456 }
457 
458 static int
459 parse_skipto(char **conf, int mc)
460 {
461 	int c, match;
462 
463 	while (1) {
464 		c = parse_peek(conf);
465 		if (c == 0)
466 			return (PE_EOF);
467 		switch (mc) {
468 		case CC_WHITESPACE:
469 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
470 			break;
471 		case CC_NONWHITESPACE:
472 			if (c == '\n')
473 				return (PE_EOL);
474 			match = (c != ' ' && c != '\t') ? 1 : 0;
475 			break;
476 		default:
477 			match = (c == mc) ? 1 : 0;
478 			break;
479 		}
480 		if (match)
481 			break;
482 		parse_advance(conf);
483 	}
484 	return (0);
485 }
486 
487 static int
488 parse_token(char **conf, char **tok)
489 {
490 	char *p;
491 	size_t len;
492 	int error;
493 
494 	*tok = NULL;
495 	error = parse_skipto(conf, CC_NONWHITESPACE);
496 	if (error)
497 		return (error);
498 	p = *conf;
499 	error = parse_skipto(conf, CC_WHITESPACE);
500 	len = *conf - p;
501 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
502 	bcopy(p, *tok, len);
503 	return (0);
504 }
505 
506 static void
507 parse_dir_ask_printenv(const char *var)
508 {
509 	char *val;
510 
511 	val = kern_getenv(var);
512 	if (val != NULL) {
513 		printf("  %s=%s\n", var, val);
514 		freeenv(val);
515 	}
516 }
517 
518 static int
519 parse_dir_ask(char **conf)
520 {
521 	char name[80];
522 	char *mnt;
523 	int error;
524 
525 	vfs_mountroot_wait();
526 
527 	printf("\nLoader variables:\n");
528 	parse_dir_ask_printenv("vfs.root.mountfrom");
529 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
530 
531 	printf("\nManual root filesystem specification:\n");
532 	printf("  <fstype>:<device> [options]\n");
533 	printf("      Mount <device> using filesystem <fstype>\n");
534 	printf("      and with the specified (optional) option list.\n");
535 	printf("\n");
536 	printf("    eg. ufs:/dev/da0s1a\n");
537 	printf("        zfs:zroot/ROOT/default\n");
538 	printf("        cd9660:/dev/cd0 ro\n");
539 	printf("          (which is equivalent to: ");
540 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
541 	printf("\n");
542 	printf("  ?               List valid disk boot devices\n");
543 	printf("  .               Yield 1 second (for background tasks)\n");
544 	printf("  <empty line>    Abort manual input\n");
545 
546 	do {
547 		error = EINVAL;
548 		printf("\nmountroot> ");
549 		cngets(name, sizeof(name), GETS_ECHO);
550 		if (name[0] == '\0')
551 			break;
552 		if (name[0] == '?' && name[1] == '\0') {
553 			printf("\nList of GEOM managed disk devices:\n  ");
554 			g_dev_print();
555 			continue;
556 		}
557 		if (name[0] == '.' && name[1] == '\0') {
558 			pause("rmask", hz);
559 			continue;
560 		}
561 		mnt = name;
562 		error = parse_mount(&mnt);
563 		if (error == -1)
564 			printf("Invalid file system specification.\n");
565 	} while (error != 0);
566 
567 	return (error);
568 }
569 
570 static int
571 parse_dir_md(char **conf)
572 {
573 	struct stat sb;
574 	struct thread *td;
575 	struct md_ioctl *mdio;
576 	char *path, *tok;
577 	int error, fd, len;
578 
579 	td = curthread;
580 	fd = -1;
581 
582 	error = parse_token(conf, &tok);
583 	if (error)
584 		return (error);
585 
586 	len = strlen(tok);
587 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
588 	path = (void *)(mdio + 1);
589 	bcopy(tok, path, len);
590 	free(tok, M_TEMP);
591 
592 	/* Get file status. */
593 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb);
594 	if (error)
595 		goto out;
596 
597 	/* Open /dev/mdctl so that we can attach/detach. */
598 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
599 	    O_RDWR, 0);
600 	if (error)
601 		goto out;
602 
603 	fd = td->td_retval[0];
604 	mdio->md_version = MDIOVERSION;
605 	mdio->md_type = MD_VNODE;
606 
607 	if (root_mount_mddev != -1) {
608 		mdio->md_unit = root_mount_mddev;
609 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
610 		/* Ignore errors. We don't care. */
611 		root_mount_mddev = -1;
612 	}
613 
614 	mdio->md_file = (void *)(mdio + 1);
615 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
616 	mdio->md_mediasize = sb.st_size;
617 	mdio->md_unit = 0;
618 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
619 	if (error)
620 		goto out;
621 
622 	if (mdio->md_unit > 9) {
623 		printf("rootmount: too many md units\n");
624 		mdio->md_file = NULL;
625 		mdio->md_options = 0;
626 		mdio->md_mediasize = 0;
627 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
628 		/* Ignore errors. We don't care. */
629 		error = ERANGE;
630 		goto out;
631 	}
632 
633 	root_mount_mddev = mdio->md_unit;
634 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
635 
636  out:
637 	if (fd >= 0)
638 		(void)kern_close(td, fd);
639 	free(mdio, M_TEMP);
640 	return (error);
641 }
642 
643 static int
644 parse_dir_onfail(char **conf)
645 {
646 	char *action;
647 	int error;
648 
649 	error = parse_token(conf, &action);
650 	if (error)
651 		return (error);
652 
653 	if (!strcmp(action, "continue"))
654 		root_mount_onfail = A_CONTINUE;
655 	else if (!strcmp(action, "panic"))
656 		root_mount_onfail = A_PANIC;
657 	else if (!strcmp(action, "reboot"))
658 		root_mount_onfail = A_REBOOT;
659 	else if (!strcmp(action, "retry"))
660 		root_mount_onfail = A_RETRY;
661 	else {
662 		printf("rootmount: %s: unknown action\n", action);
663 		error = EINVAL;
664 	}
665 
666 	free(action, M_TEMP);
667 	return (0);
668 }
669 
670 static int
671 parse_dir_timeout(char **conf)
672 {
673 	char *tok, *endtok;
674 	long secs;
675 	int error;
676 
677 	error = parse_token(conf, &tok);
678 	if (error)
679 		return (error);
680 
681 	secs = strtol(tok, &endtok, 0);
682 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
683 	if (!error)
684 		root_mount_timeout = secs;
685 	free(tok, M_TEMP);
686 	return (error);
687 }
688 
689 static int
690 parse_directive(char **conf)
691 {
692 	char *dir;
693 	int error;
694 
695 	error = parse_token(conf, &dir);
696 	if (error)
697 		return (error);
698 
699 	if (strcmp(dir, ".ask") == 0)
700 		error = parse_dir_ask(conf);
701 	else if (strcmp(dir, ".md") == 0)
702 		error = parse_dir_md(conf);
703 	else if (strcmp(dir, ".onfail") == 0)
704 		error = parse_dir_onfail(conf);
705 	else if (strcmp(dir, ".timeout") == 0)
706 		error = parse_dir_timeout(conf);
707 	else {
708 		printf("mountroot: invalid directive `%s'\n", dir);
709 		/* Ignore the rest of the line. */
710 		(void)parse_skipto(conf, '\n');
711 		error = EINVAL;
712 	}
713 	free(dir, M_TEMP);
714 	return (error);
715 }
716 
717 static bool
718 parse_mount_dev_present(const char *dev)
719 {
720 	struct nameidata nd;
721 	int error;
722 
723 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, dev);
724 	error = namei(&nd);
725 	if (error != 0)
726 		return (false);
727 	vrele(nd.ni_vp);
728 	NDFREE_PNBUF(&nd);
729 	return (true);
730 }
731 
732 #define	ERRMSGL	255
733 static int
734 parse_mount(char **conf)
735 {
736 	char *errmsg;
737 	struct mntarg *ma;
738 	char *dev, *fs, *opts, *tok;
739 	int delay, error, timeout;
740 
741 	error = parse_token(conf, &tok);
742 	if (error)
743 		return (error);
744 	fs = tok;
745 	error = parse_skipto(&tok, ':');
746 	if (error) {
747 		free(fs, M_TEMP);
748 		return (error);
749 	}
750 	parse_poke(&tok, '\0');
751 	parse_advance(&tok);
752 	dev = tok;
753 
754 	if (root_mount_mddev != -1) {
755 		/* Handle substitution for the md unit number. */
756 		tok = strstr(dev, "md#");
757 		if (tok != NULL)
758 			tok[2] = '0' + root_mount_mddev;
759 	}
760 
761 	/* Parse options. */
762 	error = parse_token(conf, &tok);
763 	opts = (error == 0) ? tok : NULL;
764 
765 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
766 	    (opts != NULL) ? opts : "");
767 
768 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
769 
770 	if (vfs_byname(fs) == NULL) {
771 		strlcpy(errmsg, "unknown file system", ERRMSGL);
772 		error = ENOENT;
773 		goto out;
774 	}
775 
776 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
777 	if (error != 0)
778 		goto out;
779 
780 	delay = hz / 10;
781 	timeout = root_mount_timeout * hz;
782 
783 	for (;;) {
784 		ma = NULL;
785 		ma = mount_arg(ma, "fstype", fs, -1);
786 		ma = mount_arg(ma, "fspath", "/", -1);
787 		ma = mount_arg(ma, "from", dev, -1);
788 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
789 		ma = mount_arg(ma, "ro", NULL, 0);
790 		ma = parse_mountroot_options(ma, opts);
791 
792 		error = kernel_mount(ma, MNT_ROOTFS);
793 		if (error == 0 || error == EILSEQ || timeout <= 0)
794 			break;
795 
796 		if (root_mount_timeout * hz == timeout ||
797 		    (bootverbose && timeout % hz == 0)) {
798 			printf("Mounting from %s:%s failed with error %d; "
799 			    "retrying for %d more second%s\n", fs, dev, error,
800 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
801 		}
802 		pause("rmretry", delay);
803 		timeout -= delay;
804 	}
805  out:
806 	if (error) {
807 		printf("Mounting from %s:%s failed with error %d",
808 		    fs, dev, error);
809 		if (errmsg[0] != '\0')
810 			printf(": %s", errmsg);
811 		printf(".\n");
812 	}
813 	free(fs, M_TEMP);
814 	free(errmsg, M_TEMP);
815 	if (opts != NULL)
816 		free(opts, M_TEMP);
817 	/* kernel_mount can return -1 on error. */
818 	return ((error < 0) ? EDOOFUS : error);
819 }
820 #undef ERRMSGL
821 
822 static int
823 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
824 {
825 	struct mount *mp;
826 	char *conf;
827 	int error;
828 
829 	root_mount_mddev = -1;
830 
831 retry:
832 	conf = sbuf_data(sb);
833 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
834 	error = (mp == NULL) ? 0 : EDOOFUS;
835 	root_mount_onfail = A_CONTINUE;
836 	while (mp == NULL) {
837 		error = parse_skipto(&conf, CC_NONWHITESPACE);
838 		if (error == PE_EOL) {
839 			parse_advance(&conf);
840 			continue;
841 		}
842 		if (error < 0)
843 			break;
844 		switch (parse_peek(&conf)) {
845 		case '#':
846 			error = parse_skipto(&conf, '\n');
847 			break;
848 		case '.':
849 			error = parse_directive(&conf);
850 			break;
851 		default:
852 			error = parse_mount(&conf);
853 			if (error == -1) {
854 				printf("mountroot: invalid file system "
855 				    "specification.\n");
856 				error = 0;
857 			}
858 			break;
859 		}
860 		if (error < 0)
861 			break;
862 		/* Ignore any trailing garbage on the line. */
863 		if (parse_peek(&conf) != '\n') {
864 			printf("mountroot: advancing to next directive...\n");
865 			(void)parse_skipto(&conf, '\n');
866 		}
867 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
868 	}
869 	if (mp != NULL)
870 		return (0);
871 
872 	/*
873 	 * We failed to mount (a new) root.
874 	 */
875 	switch (root_mount_onfail) {
876 	case A_CONTINUE:
877 		break;
878 	case A_PANIC:
879 		panic("mountroot: unable to (re-)mount root.");
880 		/* NOTREACHED */
881 	case A_RETRY:
882 		goto retry;
883 	case A_REBOOT:
884 		kern_reboot(RB_NOSYNC);
885 		/* NOTREACHED */
886 	}
887 
888 	return (error);
889 }
890 
891 static void
892 vfs_mountroot_conf0(struct sbuf *sb)
893 {
894 	char *s, *tok, *mnt, *opt;
895 	int error;
896 
897 	sbuf_cat(sb, ".onfail panic\n");
898 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
899 	if (boothowto & RB_ASKNAME)
900 		sbuf_cat(sb, ".ask\n");
901 #ifdef ROOTDEVNAME
902 	if (boothowto & RB_DFLTROOT)
903 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
904 #endif
905 	if (boothowto & RB_CDROM) {
906 		sbuf_cat(sb, "cd9660:/dev/cd0 ro\n");
907 		sbuf_cat(sb, ".timeout 0\n");
908 		sbuf_cat(sb, "cd9660:/dev/cd1 ro\n");
909 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
910 	}
911 	s = kern_getenv("vfs.root.mountfrom");
912 	if (s != NULL) {
913 		opt = kern_getenv("vfs.root.mountfrom.options");
914 		tok = s;
915 		error = parse_token(&tok, &mnt);
916 		while (!error) {
917 			sbuf_printf(sb, "%s %s\n", mnt,
918 			    (opt != NULL) ? opt : "");
919 			free(mnt, M_TEMP);
920 			error = parse_token(&tok, &mnt);
921 		}
922 		if (opt != NULL)
923 			freeenv(opt);
924 		freeenv(s);
925 	}
926 	if (rootdevnames[0] != NULL)
927 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
928 	if (rootdevnames[1] != NULL)
929 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
930 #ifdef ROOTDEVNAME
931 	if (!(boothowto & RB_DFLTROOT))
932 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
933 #endif
934 	if (!(boothowto & RB_ASKNAME))
935 		sbuf_cat(sb, ".ask\n");
936 }
937 
938 static int
939 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
940 {
941 	static char buf[128];
942 	struct nameidata nd;
943 	off_t ofs;
944 	ssize_t resid;
945 	int error, flags, len;
946 
947 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf");
948 	flags = FREAD;
949 	error = vn_open(&nd, &flags, 0, NULL);
950 	if (error)
951 		return (error);
952 
953 	NDFREE_PNBUF(&nd);
954 	ofs = 0;
955 	len = sizeof(buf) - 1;
956 	while (1) {
957 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
958 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
959 		    NOCRED, &resid, td);
960 		if (error)
961 			break;
962 		if (resid == len)
963 			break;
964 		buf[len - resid] = 0;
965 		sbuf_printf(sb, "%s", buf);
966 		ofs += len - resid;
967 	}
968 
969 	VOP_UNLOCK(nd.ni_vp);
970 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
971 	return (error);
972 }
973 
974 static void
975 vfs_mountroot_wait(void)
976 {
977 	struct root_hold_token *h;
978 	struct thread *td;
979 	struct timeval lastfail;
980 	int curfail;
981 
982 	TSENTER();
983 
984 	curfail = 0;
985 	lastfail.tv_sec = 0;
986 	eventratecheck(&lastfail, &curfail, 1);
987 	td = curthread;
988 	while (1) {
989 		g_waitidle(td);
990 		mtx_lock(&root_holds_mtx);
991 		if (TAILQ_EMPTY(&root_holds)) {
992 			mtx_unlock(&root_holds_mtx);
993 			break;
994 		}
995 		if (eventratecheck(&lastfail, &curfail, 1)) {
996 			printf("Root mount waiting for:");
997 			TAILQ_FOREACH(h, &root_holds, list)
998 				printf(" %s", h->who);
999 			printf("\n");
1000 		}
1001 		TSWAIT("root mount");
1002 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
1003 		    hz);
1004 		TSUNWAIT("root mount");
1005 	}
1006 	g_waitidle(td);
1007 
1008 	TSEXIT();
1009 }
1010 
1011 static int
1012 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1013 {
1014 	int delay, timeout;
1015 
1016 	/*
1017 	 * In case of ZFS and NFS we don't have a way to wait for
1018 	 * specific device.  Also do the wait if the user forced that
1019 	 * behaviour by setting vfs.root_mount_always_wait=1.
1020 	 */
1021 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1022 	    dev[0] == '\0' || root_mount_always_wait != 0) {
1023 		vfs_mountroot_wait();
1024 		return (0);
1025 	}
1026 
1027 	/*
1028 	 * Otherwise, no point in waiting if the device is already there.
1029 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1030 	 * eg for geom_part(4) to finish tasting.
1031 	 */
1032 	g_waitidle(curthread);
1033 	if (parse_mount_dev_present(dev))
1034 		return (0);
1035 
1036 	/*
1037 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1038 	 * to behave exactly as it used to work before.
1039 	 */
1040 	vfs_mountroot_wait();
1041 	if (parse_mount_dev_present(dev))
1042 		return (0);
1043 	printf("mountroot: waiting for device %s...\n", dev);
1044 	delay = hz / 10;
1045 	timeout = root_mount_timeout * hz;
1046 	do {
1047 		pause("rmdev", delay);
1048 		timeout -= delay;
1049 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1050 
1051 	if (timeout <= 0)
1052 		return (ENODEV);
1053 
1054 	return (0);
1055 }
1056 
1057 void
1058 vfs_mountroot(void)
1059 {
1060 	struct mount *mp;
1061 	struct sbuf *sb;
1062 	struct thread *td;
1063 	time_t timebase;
1064 	int error;
1065 
1066 	mtx_assert(&Giant, MA_NOTOWNED);
1067 
1068 	TSENTER();
1069 
1070 	td = curthread;
1071 
1072 	sb = sbuf_new_auto();
1073 	vfs_mountroot_conf0(sb);
1074 	sbuf_finish(sb);
1075 
1076 	error = vfs_mountroot_devfs(td, &mp);
1077 	while (!error) {
1078 		error = vfs_mountroot_parse(sb, mp);
1079 		if (!error) {
1080 			vfs_mountroot_shuffle(td, mp);
1081 			sbuf_clear(sb);
1082 			error = vfs_mountroot_readconf(td, sb);
1083 			sbuf_finish(sb);
1084 		}
1085 	}
1086 
1087 	sbuf_delete(sb);
1088 
1089 	/*
1090 	 * Iterate over all currently mounted file systems and use
1091 	 * the time stamp found to check and/or initialize the RTC.
1092 	 * Call inittodr() only once and pass it the largest of the
1093 	 * timestamps we encounter.
1094 	 */
1095 	timebase = 0;
1096 	mtx_lock(&mountlist_mtx);
1097 	mp = TAILQ_FIRST(&mountlist);
1098 	while (mp != NULL) {
1099 		if (mp->mnt_time > timebase)
1100 			timebase = mp->mnt_time;
1101 		mp = TAILQ_NEXT(mp, mnt_list);
1102 	}
1103 	mtx_unlock(&mountlist_mtx);
1104 	inittodr(timebase);
1105 
1106 	/* Keep prison0's root in sync with the global rootvnode. */
1107 	mtx_lock(&prison0.pr_mtx);
1108 	prison0.pr_root = rootvnode;
1109 	vref(prison0.pr_root);
1110 	mtx_unlock(&prison0.pr_mtx);
1111 
1112 	mtx_lock(&root_holds_mtx);
1113 	atomic_store_rel_int(&root_mount_complete, 1);
1114 	wakeup(&root_mount_complete);
1115 	mtx_unlock(&root_holds_mtx);
1116 
1117 	EVENTHANDLER_INVOKE(mountroot);
1118 
1119 	TSEXIT();
1120 }
1121 
1122 static struct mntarg *
1123 parse_mountroot_options(struct mntarg *ma, const char *options)
1124 {
1125 	char *p;
1126 	char *name, *name_arg;
1127 	char *val, *val_arg;
1128 	char *opts;
1129 
1130 	if (options == NULL || options[0] == '\0')
1131 		return (ma);
1132 
1133 	p = opts = strdup(options, M_MOUNT);
1134 	if (opts == NULL) {
1135 		return (ma);
1136 	}
1137 
1138 	while((name = strsep(&p, ",")) != NULL) {
1139 		if (name[0] == '\0')
1140 			break;
1141 
1142 		val = strchr(name, '=');
1143 		if (val != NULL) {
1144 			*val = '\0';
1145 			++val;
1146 		}
1147 		if (strcmp(name, "rw") == 0 || strcmp(name, "noro") == 0) {
1148 			/*
1149 			 * The first time we mount the root file system,
1150 			 * we need to mount 'ro', so We need to ignore
1151 			 * 'rw' and 'noro' mount options.
1152 			 */
1153 			continue;
1154 		}
1155 		name_arg = strdup(name, M_MOUNT);
1156 		val_arg = NULL;
1157 		if (val != NULL)
1158 			val_arg = strdup(val, M_MOUNT);
1159 
1160 		ma = mount_arg(ma, name_arg, val_arg,
1161 		    (val_arg != NULL ? -1 : 0));
1162 	}
1163 	free(opts, M_MOUNT);
1164 	return (ma);
1165 }
1166