xref: /freebsd/sys/kern/vfs_mountroot.c (revision af23369a6deaaeb612ab266eb88b8bb8d560c322)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2010 Marcel Moolenaar
5  * Copyright (c) 1999-2004 Poul-Henning Kamp
6  * Copyright (c) 1999 Michael Smith
7  * Copyright (c) 1989, 1993
8  *      The Regents of the University of California.  All rights reserved.
9  * (c) UNIX System Laboratories, Inc.
10  * All or some portions of this file are derived from material licensed
11  * to the University of California by American Telephone and Telegraph
12  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
13  * the permission of UNIX System Laboratories, Inc.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  */
39 
40 #include "opt_rootdevname.h"
41 
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
44 
45 #include <sys/param.h>
46 #include <sys/conf.h>
47 #include <sys/cons.h>
48 #include <sys/eventhandler.h>
49 #include <sys/fcntl.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/malloc.h>
53 #include <sys/mdioctl.h>
54 #include <sys/mount.h>
55 #include <sys/mutex.h>
56 #include <sys/namei.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/filedesc.h>
60 #include <sys/reboot.h>
61 #include <sys/sbuf.h>
62 #include <sys/stat.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysproto.h>
65 #include <sys/sx.h>
66 #include <sys/sysctl.h>
67 #include <sys/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 static TAILQ_HEAD(, root_hold_token)	root_holds =
114     TAILQ_HEAD_INITIALIZER(root_holds);
115 
116 enum action {
117 	A_CONTINUE,
118 	A_PANIC,
119 	A_REBOOT,
120 	A_RETRY
121 };
122 
123 enum rh_flags {
124 	RH_FREE,
125 	RH_ALLOC,
126 	RH_ARG,
127 };
128 
129 static enum action root_mount_onfail = A_CONTINUE;
130 
131 static int root_mount_mddev;
132 static int root_mount_complete;
133 
134 /* By default wait up to 3 seconds for devices to appear. */
135 static int root_mount_timeout = 3;
136 TUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout);
137 
138 static int root_mount_always_wait = 0;
139 SYSCTL_INT(_vfs, OID_AUTO, root_mount_always_wait, CTLFLAG_RDTUN,
140     &root_mount_always_wait, 0,
141     "Wait for root mount holds even if the root device already exists");
142 
143 SYSCTL_PROC(_vfs, OID_AUTO, root_mount_hold,
144     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE,
145     NULL, 0, sysctl_vfs_root_mount_hold, "A",
146     "List of root mount hold tokens");
147 
148 static int
149 sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS)
150 {
151 	struct sbuf sb;
152 	struct root_hold_token *h;
153 	int error;
154 
155 	sbuf_new(&sb, NULL, 256, SBUF_AUTOEXTEND | SBUF_INCLUDENUL);
156 
157 	mtx_lock(&root_holds_mtx);
158 	TAILQ_FOREACH(h, &root_holds, list) {
159 		if (h != TAILQ_FIRST(&root_holds))
160 			sbuf_putc(&sb, ' ');
161 		sbuf_printf(&sb, "%s", h->who);
162 	}
163 	mtx_unlock(&root_holds_mtx);
164 
165 	error = sbuf_finish(&sb);
166 	if (error == 0)
167 		error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb));
168 	sbuf_delete(&sb);
169 	return (error);
170 }
171 
172 struct root_hold_token *
173 root_mount_hold(const char *identifier)
174 {
175 	struct root_hold_token *h;
176 
177 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
178 	h->flags = RH_ALLOC;
179 	h->who = identifier;
180 	mtx_lock(&root_holds_mtx);
181 	TSHOLD("root mount");
182 	TAILQ_INSERT_TAIL(&root_holds, h, list);
183 	mtx_unlock(&root_holds_mtx);
184 	return (h);
185 }
186 
187 void
188 root_mount_hold_token(const char *identifier, struct root_hold_token *h)
189 {
190 #ifdef INVARIANTS
191 	struct root_hold_token *t;
192 #endif
193 
194 	h->flags = RH_ARG;
195 	h->who = identifier;
196 	mtx_lock(&root_holds_mtx);
197 #ifdef INVARIANTS
198 	TAILQ_FOREACH(t, &root_holds, list) {
199 		if (t == h) {
200 			panic("Duplicate mount hold by '%s' on %p",
201 			    identifier, h);
202 		}
203 	}
204 #endif
205 	TSHOLD("root mount");
206 	TAILQ_INSERT_TAIL(&root_holds, h, list);
207 	mtx_unlock(&root_holds_mtx);
208 }
209 
210 void
211 root_mount_rel(struct root_hold_token *h)
212 {
213 
214 	if (h == NULL || h->flags == RH_FREE)
215 		return;
216 
217 	mtx_lock(&root_holds_mtx);
218 	TAILQ_REMOVE(&root_holds, h, list);
219 	TSRELEASE("root mount");
220 	wakeup(&root_holds);
221 	mtx_unlock(&root_holds_mtx);
222 	if (h->flags == RH_ALLOC) {
223 		free(h, M_DEVBUF);
224 	} else
225 		h->flags = RH_FREE;
226 }
227 
228 int
229 root_mounted(void)
230 {
231 
232 	/* No mutex is acquired here because int stores are atomic. */
233 	return (root_mount_complete);
234 }
235 
236 static void
237 set_rootvnode(void)
238 {
239 
240 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
241 		panic("set_rootvnode: Cannot find root vnode");
242 
243 	VOP_UNLOCK(rootvnode);
244 
245 	pwd_set_rootvnode();
246 }
247 
248 static int
249 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
250 {
251 	struct vfsoptlist *opts;
252 	struct vfsconf *vfsp;
253 	struct mount *mp;
254 	int error;
255 
256 	*mpp = NULL;
257 
258 	if (rootdevmp != NULL) {
259 		/*
260 		 * Already have /dev; this happens during rerooting.
261 		 */
262 		error = vfs_busy(rootdevmp, 0);
263 		if (error != 0)
264 			return (error);
265 		*mpp = rootdevmp;
266 	} else {
267 		vfsp = vfs_byname("devfs");
268 		KASSERT(vfsp != NULL, ("Could not find devfs by name"));
269 		if (vfsp == NULL)
270 			return (ENOENT);
271 
272 		mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
273 
274 		error = VFS_MOUNT(mp);
275 		KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
276 		if (error)
277 			return (error);
278 
279 		error = VFS_STATFS(mp, &mp->mnt_stat);
280 		KASSERT(error == 0, ("VFS_STATFS(devfs) failed %d", error));
281 		if (error)
282 			return (error);
283 
284 		opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
285 		TAILQ_INIT(opts);
286 		mp->mnt_opt = opts;
287 
288 		mtx_lock(&mountlist_mtx);
289 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
290 		mtx_unlock(&mountlist_mtx);
291 
292 		*mpp = mp;
293 		rootdevmp = mp;
294 		vfs_op_exit(mp);
295 	}
296 
297 	set_rootvnode();
298 
299 	error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
300 	if (error)
301 		printf("kern_symlink /dev -> / returns %d\n", error);
302 
303 	return (error);
304 }
305 
306 static void
307 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
308 {
309 	struct nameidata nd;
310 	struct mount *mporoot, *mpnroot;
311 	struct vnode *vp, *vporoot, *vpdevfs;
312 	char *fspath;
313 	int error;
314 
315 	mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
316 
317 	/* Shuffle the mountlist. */
318 	mtx_lock(&mountlist_mtx);
319 	mporoot = TAILQ_FIRST(&mountlist);
320 	TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
321 	if (mporoot != mpdevfs) {
322 		TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
323 		TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
324 	}
325 	TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
326 	mtx_unlock(&mountlist_mtx);
327 
328 	cache_purgevfs(mporoot);
329 	if (mporoot != mpdevfs)
330 		cache_purgevfs(mpdevfs);
331 
332 	if (VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot))
333 		panic("vfs_mountroot_shuffle: Cannot find root vnode");
334 
335 	VI_LOCK(vporoot);
336 	vporoot->v_iflag &= ~VI_MOUNT;
337 	vn_irflag_unset_locked(vporoot, VIRF_MOUNTPOINT);
338 	vporoot->v_mountedhere = NULL;
339 	VI_UNLOCK(vporoot);
340 	mporoot->mnt_flag &= ~MNT_ROOTFS;
341 	mporoot->mnt_vnodecovered = NULL;
342 	vput(vporoot);
343 
344 	/* Set up the new rootvnode, and purge the cache */
345 	mpnroot->mnt_vnodecovered = NULL;
346 	set_rootvnode();
347 	cache_purgevfs(rootvnode->v_mount);
348 
349 	if (mporoot != mpdevfs) {
350 		/* Remount old root under /.mount or /mnt */
351 		fspath = "/.mount";
352 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath);
353 		error = namei(&nd);
354 		if (error) {
355 			NDFREE_PNBUF(&nd);
356 			fspath = "/mnt";
357 			NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
358 			    fspath);
359 			error = namei(&nd);
360 		}
361 		if (!error) {
362 			vp = nd.ni_vp;
363 			error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
364 			if (!error)
365 				error = vinvalbuf(vp, V_SAVE, 0, 0);
366 			if (!error) {
367 				cache_purge(vp);
368 				VI_LOCK(vp);
369 				mporoot->mnt_vnodecovered = vp;
370 				vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
371 				vp->v_mountedhere = mporoot;
372 				strlcpy(mporoot->mnt_stat.f_mntonname,
373 				    fspath, MNAMELEN);
374 				VI_UNLOCK(vp);
375 				VOP_UNLOCK(vp);
376 			} else
377 				vput(vp);
378 		}
379 		NDFREE_PNBUF(&nd);
380 
381 		if (error)
382 			printf("mountroot: unable to remount previous root "
383 			    "under /.mount or /mnt (error %d)\n", error);
384 	}
385 
386 	/* Remount devfs under /dev */
387 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev");
388 	error = namei(&nd);
389 	if (!error) {
390 		vp = nd.ni_vp;
391 		error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
392 		if (!error)
393 			error = vinvalbuf(vp, V_SAVE, 0, 0);
394 		if (!error) {
395 			vpdevfs = mpdevfs->mnt_vnodecovered;
396 			if (vpdevfs != NULL) {
397 				cache_purge(vpdevfs);
398 				VI_LOCK(vpdevfs);
399 				vn_irflag_unset_locked(vpdevfs, VIRF_MOUNTPOINT);
400 				vpdevfs->v_mountedhere = NULL;
401 				VI_UNLOCK(vpdevfs);
402 				vrele(vpdevfs);
403 			}
404 			VI_LOCK(vp);
405 			mpdevfs->mnt_vnodecovered = vp;
406 			vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
407 			vp->v_mountedhere = mpdevfs;
408 			VI_UNLOCK(vp);
409 			VOP_UNLOCK(vp);
410 		} else
411 			vput(vp);
412 	}
413 	if (error)
414 		printf("mountroot: unable to remount devfs under /dev "
415 		    "(error %d)\n", error);
416 	NDFREE_PNBUF(&nd);
417 
418 	if (mporoot == mpdevfs) {
419 		vfs_unbusy(mpdevfs);
420 		/* Unlink the no longer needed /dev/dev -> / symlink */
421 		error = kern_funlinkat(td, AT_FDCWD, "/dev/dev", FD_NONE,
422 		    UIO_SYSSPACE, 0, 0);
423 		if (error)
424 			printf("mountroot: unable to unlink /dev/dev "
425 			    "(error %d)\n", error);
426 	}
427 }
428 
429 /*
430  * Configuration parser.
431  */
432 
433 /* Parser character classes. */
434 #define	CC_WHITESPACE		-1
435 #define	CC_NONWHITESPACE	-2
436 
437 /* Parse errors. */
438 #define	PE_EOF			-1
439 #define	PE_EOL			-2
440 
441 static __inline int
442 parse_peek(char **conf)
443 {
444 
445 	return (**conf);
446 }
447 
448 static __inline void
449 parse_poke(char **conf, int c)
450 {
451 
452 	**conf = c;
453 }
454 
455 static __inline void
456 parse_advance(char **conf)
457 {
458 
459 	(*conf)++;
460 }
461 
462 static int
463 parse_skipto(char **conf, int mc)
464 {
465 	int c, match;
466 
467 	while (1) {
468 		c = parse_peek(conf);
469 		if (c == 0)
470 			return (PE_EOF);
471 		switch (mc) {
472 		case CC_WHITESPACE:
473 			match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
474 			break;
475 		case CC_NONWHITESPACE:
476 			if (c == '\n')
477 				return (PE_EOL);
478 			match = (c != ' ' && c != '\t') ? 1 : 0;
479 			break;
480 		default:
481 			match = (c == mc) ? 1 : 0;
482 			break;
483 		}
484 		if (match)
485 			break;
486 		parse_advance(conf);
487 	}
488 	return (0);
489 }
490 
491 static int
492 parse_token(char **conf, char **tok)
493 {
494 	char *p;
495 	size_t len;
496 	int error;
497 
498 	*tok = NULL;
499 	error = parse_skipto(conf, CC_NONWHITESPACE);
500 	if (error)
501 		return (error);
502 	p = *conf;
503 	error = parse_skipto(conf, CC_WHITESPACE);
504 	len = *conf - p;
505 	*tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
506 	bcopy(p, *tok, len);
507 	return (0);
508 }
509 
510 static void
511 parse_dir_ask_printenv(const char *var)
512 {
513 	char *val;
514 
515 	val = kern_getenv(var);
516 	if (val != NULL) {
517 		printf("  %s=%s\n", var, val);
518 		freeenv(val);
519 	}
520 }
521 
522 static int
523 parse_dir_ask(char **conf)
524 {
525 	char name[80];
526 	char *mnt;
527 	int error;
528 
529 	vfs_mountroot_wait();
530 
531 	printf("\nLoader variables:\n");
532 	parse_dir_ask_printenv("vfs.root.mountfrom");
533 	parse_dir_ask_printenv("vfs.root.mountfrom.options");
534 
535 	printf("\nManual root filesystem specification:\n");
536 	printf("  <fstype>:<device> [options]\n");
537 	printf("      Mount <device> using filesystem <fstype>\n");
538 	printf("      and with the specified (optional) option list.\n");
539 	printf("\n");
540 	printf("    eg. ufs:/dev/da0s1a\n");
541 	printf("        zfs:zroot/ROOT/default\n");
542 	printf("        cd9660:/dev/cd0 ro\n");
543 	printf("          (which is equivalent to: ");
544 	printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
545 	printf("\n");
546 	printf("  ?               List valid disk boot devices\n");
547 	printf("  .               Yield 1 second (for background tasks)\n");
548 	printf("  <empty line>    Abort manual input\n");
549 
550 	do {
551 		error = EINVAL;
552 		printf("\nmountroot> ");
553 		cngets(name, sizeof(name), GETS_ECHO);
554 		if (name[0] == '\0')
555 			break;
556 		if (name[0] == '?' && name[1] == '\0') {
557 			printf("\nList of GEOM managed disk devices:\n  ");
558 			g_dev_print();
559 			continue;
560 		}
561 		if (name[0] == '.' && name[1] == '\0') {
562 			pause("rmask", hz);
563 			continue;
564 		}
565 		mnt = name;
566 		error = parse_mount(&mnt);
567 		if (error == -1)
568 			printf("Invalid file system specification.\n");
569 	} while (error != 0);
570 
571 	return (error);
572 }
573 
574 static int
575 parse_dir_md(char **conf)
576 {
577 	struct stat sb;
578 	struct thread *td;
579 	struct md_ioctl *mdio;
580 	char *path, *tok;
581 	int error, fd, len;
582 
583 	td = curthread;
584 	fd = -1;
585 
586 	error = parse_token(conf, &tok);
587 	if (error)
588 		return (error);
589 
590 	len = strlen(tok);
591 	mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
592 	path = (void *)(mdio + 1);
593 	bcopy(tok, path, len);
594 	free(tok, M_TEMP);
595 
596 	/* Get file status. */
597 	error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
598 	if (error)
599 		goto out;
600 
601 	/* Open /dev/mdctl so that we can attach/detach. */
602 	error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
603 	    O_RDWR, 0);
604 	if (error)
605 		goto out;
606 
607 	fd = td->td_retval[0];
608 	mdio->md_version = MDIOVERSION;
609 	mdio->md_type = MD_VNODE;
610 
611 	if (root_mount_mddev != -1) {
612 		mdio->md_unit = root_mount_mddev;
613 		(void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
614 		/* Ignore errors. We don't care. */
615 		root_mount_mddev = -1;
616 	}
617 
618 	mdio->md_file = (void *)(mdio + 1);
619 	mdio->md_options = MD_AUTOUNIT | MD_READONLY;
620 	mdio->md_mediasize = sb.st_size;
621 	mdio->md_unit = 0;
622 	error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
623 	if (error)
624 		goto out;
625 
626 	if (mdio->md_unit > 9) {
627 		printf("rootmount: too many md units\n");
628 		mdio->md_file = NULL;
629 		mdio->md_options = 0;
630 		mdio->md_mediasize = 0;
631 		error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
632 		/* Ignore errors. We don't care. */
633 		error = ERANGE;
634 		goto out;
635 	}
636 
637 	root_mount_mddev = mdio->md_unit;
638 	printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
639 
640  out:
641 	if (fd >= 0)
642 		(void)kern_close(td, fd);
643 	free(mdio, M_TEMP);
644 	return (error);
645 }
646 
647 static int
648 parse_dir_onfail(char **conf)
649 {
650 	char *action;
651 	int error;
652 
653 	error = parse_token(conf, &action);
654 	if (error)
655 		return (error);
656 
657 	if (!strcmp(action, "continue"))
658 		root_mount_onfail = A_CONTINUE;
659 	else if (!strcmp(action, "panic"))
660 		root_mount_onfail = A_PANIC;
661 	else if (!strcmp(action, "reboot"))
662 		root_mount_onfail = A_REBOOT;
663 	else if (!strcmp(action, "retry"))
664 		root_mount_onfail = A_RETRY;
665 	else {
666 		printf("rootmount: %s: unknown action\n", action);
667 		error = EINVAL;
668 	}
669 
670 	free(action, M_TEMP);
671 	return (0);
672 }
673 
674 static int
675 parse_dir_timeout(char **conf)
676 {
677 	char *tok, *endtok;
678 	long secs;
679 	int error;
680 
681 	error = parse_token(conf, &tok);
682 	if (error)
683 		return (error);
684 
685 	secs = strtol(tok, &endtok, 0);
686 	error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
687 	if (!error)
688 		root_mount_timeout = secs;
689 	free(tok, M_TEMP);
690 	return (error);
691 }
692 
693 static int
694 parse_directive(char **conf)
695 {
696 	char *dir;
697 	int error;
698 
699 	error = parse_token(conf, &dir);
700 	if (error)
701 		return (error);
702 
703 	if (strcmp(dir, ".ask") == 0)
704 		error = parse_dir_ask(conf);
705 	else if (strcmp(dir, ".md") == 0)
706 		error = parse_dir_md(conf);
707 	else if (strcmp(dir, ".onfail") == 0)
708 		error = parse_dir_onfail(conf);
709 	else if (strcmp(dir, ".timeout") == 0)
710 		error = parse_dir_timeout(conf);
711 	else {
712 		printf("mountroot: invalid directive `%s'\n", dir);
713 		/* Ignore the rest of the line. */
714 		(void)parse_skipto(conf, '\n');
715 		error = EINVAL;
716 	}
717 	free(dir, M_TEMP);
718 	return (error);
719 }
720 
721 static bool
722 parse_mount_dev_present(const char *dev)
723 {
724 	struct nameidata nd;
725 	int error;
726 
727 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, dev);
728 	error = namei(&nd);
729 	if (error != 0)
730 		return (false);
731 	vrele(nd.ni_vp);
732 	NDFREE_PNBUF(&nd);
733 	return (true);
734 }
735 
736 #define	ERRMSGL	255
737 static int
738 parse_mount(char **conf)
739 {
740 	char *errmsg;
741 	struct mntarg *ma;
742 	char *dev, *fs, *opts, *tok;
743 	int delay, error, timeout;
744 
745 	error = parse_token(conf, &tok);
746 	if (error)
747 		return (error);
748 	fs = tok;
749 	error = parse_skipto(&tok, ':');
750 	if (error) {
751 		free(fs, M_TEMP);
752 		return (error);
753 	}
754 	parse_poke(&tok, '\0');
755 	parse_advance(&tok);
756 	dev = tok;
757 
758 	if (root_mount_mddev != -1) {
759 		/* Handle substitution for the md unit number. */
760 		tok = strstr(dev, "md#");
761 		if (tok != NULL)
762 			tok[2] = '0' + root_mount_mddev;
763 	}
764 
765 	/* Parse options. */
766 	error = parse_token(conf, &tok);
767 	opts = (error == 0) ? tok : NULL;
768 
769 	printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
770 	    (opts != NULL) ? opts : "");
771 
772 	errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
773 
774 	if (vfs_byname(fs) == NULL) {
775 		strlcpy(errmsg, "unknown file system", ERRMSGL);
776 		error = ENOENT;
777 		goto out;
778 	}
779 
780 	error = vfs_mountroot_wait_if_neccessary(fs, dev);
781 	if (error != 0)
782 		goto out;
783 
784 	delay = hz / 10;
785 	timeout = root_mount_timeout * hz;
786 
787 	for (;;) {
788 		ma = NULL;
789 		ma = mount_arg(ma, "fstype", fs, -1);
790 		ma = mount_arg(ma, "fspath", "/", -1);
791 		ma = mount_arg(ma, "from", dev, -1);
792 		ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
793 		ma = mount_arg(ma, "ro", NULL, 0);
794 		ma = parse_mountroot_options(ma, opts);
795 
796 		error = kernel_mount(ma, MNT_ROOTFS);
797 		if (error == 0 || error == EILSEQ || timeout <= 0)
798 			break;
799 
800 		if (root_mount_timeout * hz == timeout ||
801 		    (bootverbose && timeout % hz == 0)) {
802 			printf("Mounting from %s:%s failed with error %d; "
803 			    "retrying for %d more second%s\n", fs, dev, error,
804 			    timeout / hz, (timeout / hz > 1) ? "s" : "");
805 		}
806 		pause("rmretry", delay);
807 		timeout -= delay;
808 	}
809  out:
810 	if (error) {
811 		printf("Mounting from %s:%s failed with error %d",
812 		    fs, dev, error);
813 		if (errmsg[0] != '\0')
814 			printf(": %s", errmsg);
815 		printf(".\n");
816 	}
817 	free(fs, M_TEMP);
818 	free(errmsg, M_TEMP);
819 	if (opts != NULL)
820 		free(opts, M_TEMP);
821 	/* kernel_mount can return -1 on error. */
822 	return ((error < 0) ? EDOOFUS : error);
823 }
824 #undef ERRMSGL
825 
826 static int
827 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
828 {
829 	struct mount *mp;
830 	char *conf;
831 	int error;
832 
833 	root_mount_mddev = -1;
834 
835 retry:
836 	conf = sbuf_data(sb);
837 	mp = TAILQ_NEXT(mpdevfs, mnt_list);
838 	error = (mp == NULL) ? 0 : EDOOFUS;
839 	root_mount_onfail = A_CONTINUE;
840 	while (mp == NULL) {
841 		error = parse_skipto(&conf, CC_NONWHITESPACE);
842 		if (error == PE_EOL) {
843 			parse_advance(&conf);
844 			continue;
845 		}
846 		if (error < 0)
847 			break;
848 		switch (parse_peek(&conf)) {
849 		case '#':
850 			error = parse_skipto(&conf, '\n');
851 			break;
852 		case '.':
853 			error = parse_directive(&conf);
854 			break;
855 		default:
856 			error = parse_mount(&conf);
857 			if (error == -1) {
858 				printf("mountroot: invalid file system "
859 				    "specification.\n");
860 				error = 0;
861 			}
862 			break;
863 		}
864 		if (error < 0)
865 			break;
866 		/* Ignore any trailing garbage on the line. */
867 		if (parse_peek(&conf) != '\n') {
868 			printf("mountroot: advancing to next directive...\n");
869 			(void)parse_skipto(&conf, '\n');
870 		}
871 		mp = TAILQ_NEXT(mpdevfs, mnt_list);
872 	}
873 	if (mp != NULL)
874 		return (0);
875 
876 	/*
877 	 * We failed to mount (a new) root.
878 	 */
879 	switch (root_mount_onfail) {
880 	case A_CONTINUE:
881 		break;
882 	case A_PANIC:
883 		panic("mountroot: unable to (re-)mount root.");
884 		/* NOTREACHED */
885 	case A_RETRY:
886 		goto retry;
887 	case A_REBOOT:
888 		kern_reboot(RB_NOSYNC);
889 		/* NOTREACHED */
890 	}
891 
892 	return (error);
893 }
894 
895 static void
896 vfs_mountroot_conf0(struct sbuf *sb)
897 {
898 	char *s, *tok, *mnt, *opt;
899 	int error;
900 
901 	sbuf_printf(sb, ".onfail panic\n");
902 	sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
903 	if (boothowto & RB_ASKNAME)
904 		sbuf_printf(sb, ".ask\n");
905 #ifdef ROOTDEVNAME
906 	if (boothowto & RB_DFLTROOT)
907 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
908 #endif
909 	if (boothowto & RB_CDROM) {
910 		sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
911 		sbuf_printf(sb, ".timeout 0\n");
912 		sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
913 		sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
914 	}
915 	s = kern_getenv("vfs.root.mountfrom");
916 	if (s != NULL) {
917 		opt = kern_getenv("vfs.root.mountfrom.options");
918 		tok = s;
919 		error = parse_token(&tok, &mnt);
920 		while (!error) {
921 			sbuf_printf(sb, "%s %s\n", mnt,
922 			    (opt != NULL) ? opt : "");
923 			free(mnt, M_TEMP);
924 			error = parse_token(&tok, &mnt);
925 		}
926 		if (opt != NULL)
927 			freeenv(opt);
928 		freeenv(s);
929 	}
930 	if (rootdevnames[0] != NULL)
931 		sbuf_printf(sb, "%s\n", rootdevnames[0]);
932 	if (rootdevnames[1] != NULL)
933 		sbuf_printf(sb, "%s\n", rootdevnames[1]);
934 #ifdef ROOTDEVNAME
935 	if (!(boothowto & RB_DFLTROOT))
936 		sbuf_printf(sb, "%s\n", ROOTDEVNAME);
937 #endif
938 	if (!(boothowto & RB_ASKNAME))
939 		sbuf_printf(sb, ".ask\n");
940 }
941 
942 static int
943 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
944 {
945 	static char buf[128];
946 	struct nameidata nd;
947 	off_t ofs;
948 	ssize_t resid;
949 	int error, flags, len;
950 
951 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf");
952 	flags = FREAD;
953 	error = vn_open(&nd, &flags, 0, NULL);
954 	if (error)
955 		return (error);
956 
957 	NDFREE_PNBUF(&nd);
958 	ofs = 0;
959 	len = sizeof(buf) - 1;
960 	while (1) {
961 		error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
962 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
963 		    NOCRED, &resid, td);
964 		if (error)
965 			break;
966 		if (resid == len)
967 			break;
968 		buf[len - resid] = 0;
969 		sbuf_printf(sb, "%s", buf);
970 		ofs += len - resid;
971 	}
972 
973 	VOP_UNLOCK(nd.ni_vp);
974 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
975 	return (error);
976 }
977 
978 static void
979 vfs_mountroot_wait(void)
980 {
981 	struct root_hold_token *h;
982 	struct thread *td;
983 	struct timeval lastfail;
984 	int curfail;
985 
986 	TSENTER();
987 
988 	curfail = 0;
989 	lastfail.tv_sec = 0;
990 	eventratecheck(&lastfail, &curfail, 1);
991 	td = curthread;
992 	while (1) {
993 		g_waitidle(td);
994 		mtx_lock(&root_holds_mtx);
995 		if (TAILQ_EMPTY(&root_holds)) {
996 			mtx_unlock(&root_holds_mtx);
997 			break;
998 		}
999 		if (eventratecheck(&lastfail, &curfail, 1)) {
1000 			printf("Root mount waiting for:");
1001 			TAILQ_FOREACH(h, &root_holds, list)
1002 				printf(" %s", h->who);
1003 			printf("\n");
1004 		}
1005 		TSWAIT("root mount");
1006 		msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
1007 		    hz);
1008 		TSUNWAIT("root mount");
1009 	}
1010 	g_waitidle(td);
1011 
1012 	TSEXIT();
1013 }
1014 
1015 static int
1016 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1017 {
1018 	int delay, timeout;
1019 
1020 	/*
1021 	 * In case of ZFS and NFS we don't have a way to wait for
1022 	 * specific device.  Also do the wait if the user forced that
1023 	 * behaviour by setting vfs.root_mount_always_wait=1.
1024 	 */
1025 	if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1026 	    dev[0] == '\0' || root_mount_always_wait != 0) {
1027 		vfs_mountroot_wait();
1028 		return (0);
1029 	}
1030 
1031 	/*
1032 	 * Otherwise, no point in waiting if the device is already there.
1033 	 * Note that we must wait for GEOM to finish reconfiguring itself,
1034 	 * eg for geom_part(4) to finish tasting.
1035 	 */
1036 	g_waitidle(curthread);
1037 	if (parse_mount_dev_present(dev))
1038 		return (0);
1039 
1040 	/*
1041 	 * No luck.  Let's wait.  This code looks weird, but it's that way
1042 	 * to behave exactly as it used to work before.
1043 	 */
1044 	vfs_mountroot_wait();
1045 	if (parse_mount_dev_present(dev))
1046 		return (0);
1047 	printf("mountroot: waiting for device %s...\n", dev);
1048 	delay = hz / 10;
1049 	timeout = root_mount_timeout * hz;
1050 	do {
1051 		pause("rmdev", delay);
1052 		timeout -= delay;
1053 	} while (timeout > 0 && !parse_mount_dev_present(dev));
1054 
1055 	if (timeout <= 0)
1056 		return (ENODEV);
1057 
1058 	return (0);
1059 }
1060 
1061 void
1062 vfs_mountroot(void)
1063 {
1064 	struct mount *mp;
1065 	struct sbuf *sb;
1066 	struct thread *td;
1067 	time_t timebase;
1068 	int error;
1069 
1070 	mtx_assert(&Giant, MA_NOTOWNED);
1071 
1072 	TSENTER();
1073 
1074 	td = curthread;
1075 
1076 	sb = sbuf_new_auto();
1077 	vfs_mountroot_conf0(sb);
1078 	sbuf_finish(sb);
1079 
1080 	error = vfs_mountroot_devfs(td, &mp);
1081 	while (!error) {
1082 		error = vfs_mountroot_parse(sb, mp);
1083 		if (!error) {
1084 			vfs_mountroot_shuffle(td, mp);
1085 			sbuf_clear(sb);
1086 			error = vfs_mountroot_readconf(td, sb);
1087 			sbuf_finish(sb);
1088 		}
1089 	}
1090 
1091 	sbuf_delete(sb);
1092 
1093 	/*
1094 	 * Iterate over all currently mounted file systems and use
1095 	 * the time stamp found to check and/or initialize the RTC.
1096 	 * Call inittodr() only once and pass it the largest of the
1097 	 * timestamps we encounter.
1098 	 */
1099 	timebase = 0;
1100 	mtx_lock(&mountlist_mtx);
1101 	mp = TAILQ_FIRST(&mountlist);
1102 	while (mp != NULL) {
1103 		if (mp->mnt_time > timebase)
1104 			timebase = mp->mnt_time;
1105 		mp = TAILQ_NEXT(mp, mnt_list);
1106 	}
1107 	mtx_unlock(&mountlist_mtx);
1108 	inittodr(timebase);
1109 
1110 	/* Keep prison0's root in sync with the global rootvnode. */
1111 	mtx_lock(&prison0.pr_mtx);
1112 	prison0.pr_root = rootvnode;
1113 	vref(prison0.pr_root);
1114 	mtx_unlock(&prison0.pr_mtx);
1115 
1116 	mtx_lock(&root_holds_mtx);
1117 	atomic_store_rel_int(&root_mount_complete, 1);
1118 	wakeup(&root_mount_complete);
1119 	mtx_unlock(&root_holds_mtx);
1120 
1121 	EVENTHANDLER_INVOKE(mountroot);
1122 
1123 	TSEXIT();
1124 }
1125 
1126 static struct mntarg *
1127 parse_mountroot_options(struct mntarg *ma, const char *options)
1128 {
1129 	char *p;
1130 	char *name, *name_arg;
1131 	char *val, *val_arg;
1132 	char *opts;
1133 
1134 	if (options == NULL || options[0] == '\0')
1135 		return (ma);
1136 
1137 	p = opts = strdup(options, M_MOUNT);
1138 	if (opts == NULL) {
1139 		return (ma);
1140 	}
1141 
1142 	while((name = strsep(&p, ",")) != NULL) {
1143 		if (name[0] == '\0')
1144 			break;
1145 
1146 		val = strchr(name, '=');
1147 		if (val != NULL) {
1148 			*val = '\0';
1149 			++val;
1150 		}
1151 		if (strcmp(name, "rw") == 0 || strcmp(name, "noro") == 0) {
1152 			/*
1153 			 * The first time we mount the root file system,
1154 			 * we need to mount 'ro', so We need to ignore
1155 			 * 'rw' and 'noro' mount options.
1156 			 */
1157 			continue;
1158 		}
1159 		name_arg = strdup(name, M_MOUNT);
1160 		val_arg = NULL;
1161 		if (val != NULL)
1162 			val_arg = strdup(val, M_MOUNT);
1163 
1164 		ma = mount_arg(ma, name_arg, val_arg,
1165 		    (val_arg != NULL ? -1 : 0));
1166 	}
1167 	free(opts, M_MOUNT);
1168 	return (ma);
1169 }
1170