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: stable/12/sys/kern/vfs_mountroot.c 370440 2021-08-29 17:01:04Z kevans $");
44
45 #include <sys/param.h>
46 #include <sys/conf.h>
47 #include <sys/cons.h>
48 #include <sys/fcntl.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/mdioctl.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/filedesc.h>
59 #include <sys/reboot.h>
60 #include <sys/sbuf.h>
61 #include <sys/stat.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysproto.h>
64 #include <sys/sx.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/systm.h>
68 #include <sys/vnode.h>
69
70 #include <geom/geom.h>
71
72 /*
73 * The root filesystem is detailed in the kernel environment variable
74 * vfs.root.mountfrom, which is expected to be in the general format
75 *
76 * <vfsname>:[<path>][ <vfsname>:[<path>] ...]
77 * vfsname := the name of a VFS known to the kernel and capable
78 * of being mounted as root
79 * path := disk device name or other data used by the filesystem
80 * to locate its physical store
81 *
82 * If the environment variable vfs.root.mountfrom is a space separated list,
83 * each list element is tried in turn and the root filesystem will be mounted
84 * from the first one that succeeds.
85 *
86 * The environment variable vfs.root.mountfrom.options is a comma delimited
87 * set of string mount options. These mount options must be parseable
88 * by nmount() in the kernel.
89 */
90
91 static int parse_mount(char **);
92 static struct mntarg *parse_mountroot_options(struct mntarg *, const char *);
93 static int sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS);
94 static void vfs_mountroot_wait(void);
95 static int vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev);
96
97 /*
98 * The vnode of the system's root (/ in the filesystem, without chroot
99 * active.)
100 */
101 struct vnode *rootvnode;
102
103 /*
104 * Mount of the system's /dev.
105 */
106 struct mount *rootdevmp;
107
108 char *rootdevnames[2] = {NULL, NULL};
109
110 struct mtx root_holds_mtx;
111 MTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF);
112
113 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
sysctl_vfs_root_mount_hold(SYSCTL_HANDLER_ARGS)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 *
root_mount_hold(const char * identifier)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
root_mount_hold_token(const char * identifier,struct root_hold_token * h)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
root_mount_rel(struct root_hold_token * h)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
root_mounted(void)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
set_rootvnode(void)237 set_rootvnode(void)
238 {
239 struct proc *p;
240
241 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode))
242 panic("Cannot find root vnode");
243
244 VOP_UNLOCK(rootvnode, 0);
245
246 p = curthread->td_proc;
247 FILEDESC_XLOCK(p->p_fd);
248
249 if (p->p_fd->fd_cdir != NULL)
250 vrele(p->p_fd->fd_cdir);
251 p->p_fd->fd_cdir = rootvnode;
252 VREF(rootvnode);
253
254 if (p->p_fd->fd_rdir != NULL)
255 vrele(p->p_fd->fd_rdir);
256 p->p_fd->fd_rdir = rootvnode;
257 VREF(rootvnode);
258
259 FILEDESC_XUNLOCK(p->p_fd);
260 }
261
262 static int
vfs_mountroot_devfs(struct thread * td,struct mount ** mpp)263 vfs_mountroot_devfs(struct thread *td, struct mount **mpp)
264 {
265 struct vfsoptlist *opts;
266 struct vfsconf *vfsp;
267 struct mount *mp;
268 int error;
269
270 *mpp = NULL;
271
272 if (rootdevmp != NULL) {
273 /*
274 * Already have /dev; this happens during rerooting.
275 */
276 error = vfs_busy(rootdevmp, 0);
277 if (error != 0)
278 return (error);
279 *mpp = rootdevmp;
280 } else {
281 vfsp = vfs_byname("devfs");
282 KASSERT(vfsp != NULL, ("Could not find devfs by name"));
283 if (vfsp == NULL)
284 return (ENOENT);
285
286 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
287
288 error = VFS_MOUNT(mp);
289 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
290 if (error)
291 return (error);
292
293 error = VFS_STATFS(mp, &mp->mnt_stat);
294 KASSERT(error == 0, ("VFS_STATFS(devfs) failed %d", error));
295 if (error)
296 return (error);
297
298 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
299 TAILQ_INIT(opts);
300 mp->mnt_opt = opts;
301
302 mtx_lock(&mountlist_mtx);
303 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
304 mtx_unlock(&mountlist_mtx);
305
306 *mpp = mp;
307 rootdevmp = mp;
308 }
309
310 set_rootvnode();
311
312 error = kern_symlinkat(td, "/", AT_FDCWD, "dev", UIO_SYSSPACE);
313 if (error)
314 printf("kern_symlink /dev -> / returns %d\n", error);
315
316 return (error);
317 }
318
319 static void
vfs_mountroot_shuffle(struct thread * td,struct mount * mpdevfs)320 vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs)
321 {
322 struct nameidata nd;
323 struct mount *mporoot, *mpnroot;
324 struct vnode *vp, *vporoot, *vpdevfs;
325 char *fspath;
326 int error;
327
328 mpnroot = TAILQ_NEXT(mpdevfs, mnt_list);
329
330 /* Shuffle the mountlist. */
331 mtx_lock(&mountlist_mtx);
332 mporoot = TAILQ_FIRST(&mountlist);
333 TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list);
334 if (mporoot != mpdevfs) {
335 TAILQ_REMOVE(&mountlist, mpnroot, mnt_list);
336 TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list);
337 }
338 TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list);
339 mtx_unlock(&mountlist_mtx);
340
341 cache_purgevfs(mporoot, true);
342 if (mporoot != mpdevfs)
343 cache_purgevfs(mpdevfs, true);
344
345 VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot);
346
347 VI_LOCK(vporoot);
348 vporoot->v_iflag &= ~VI_MOUNT;
349 VI_UNLOCK(vporoot);
350 vporoot->v_mountedhere = NULL;
351 mporoot->mnt_flag &= ~MNT_ROOTFS;
352 mporoot->mnt_vnodecovered = NULL;
353 vput(vporoot);
354
355 /* Set up the new rootvnode, and purge the cache */
356 mpnroot->mnt_vnodecovered = NULL;
357 set_rootvnode();
358 cache_purgevfs(rootvnode->v_mount, true);
359
360 if (mporoot != mpdevfs) {
361 /* Remount old root under /.mount or /mnt */
362 fspath = "/.mount";
363 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
364 fspath, td);
365 error = namei(&nd);
366 if (error) {
367 NDFREE(&nd, NDF_ONLY_PNBUF);
368 fspath = "/mnt";
369 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
370 fspath, td);
371 error = namei(&nd);
372 }
373 if (!error) {
374 vp = nd.ni_vp;
375 error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
376 if (!error)
377 error = vinvalbuf(vp, V_SAVE, 0, 0);
378 if (!error) {
379 cache_purge(vp);
380 mporoot->mnt_vnodecovered = vp;
381 vp->v_mountedhere = mporoot;
382 strlcpy(mporoot->mnt_stat.f_mntonname,
383 fspath, MNAMELEN);
384 VOP_UNLOCK(vp, 0);
385 } else
386 vput(vp);
387 }
388 NDFREE(&nd, NDF_ONLY_PNBUF);
389
390 if (error)
391 printf("mountroot: unable to remount previous root "
392 "under /.mount or /mnt (error %d)\n", error);
393 }
394
395 /* Remount devfs under /dev */
396 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
397 error = namei(&nd);
398 if (!error) {
399 vp = nd.ni_vp;
400 error = (vp->v_type == VDIR) ? 0 : ENOTDIR;
401 if (!error)
402 error = vinvalbuf(vp, V_SAVE, 0, 0);
403 if (!error) {
404 vpdevfs = mpdevfs->mnt_vnodecovered;
405 if (vpdevfs != NULL) {
406 cache_purge(vpdevfs);
407 vpdevfs->v_mountedhere = NULL;
408 vrele(vpdevfs);
409 }
410 mpdevfs->mnt_vnodecovered = vp;
411 vp->v_mountedhere = mpdevfs;
412 VOP_UNLOCK(vp, 0);
413 } else
414 vput(vp);
415 }
416 if (error)
417 printf("mountroot: unable to remount devfs under /dev "
418 "(error %d)\n", error);
419 NDFREE(&nd, NDF_ONLY_PNBUF);
420
421 if (mporoot == mpdevfs) {
422 vfs_unbusy(mpdevfs);
423 /* Unlink the no longer needed /dev/dev -> / symlink */
424 error = kern_unlinkat(td, AT_FDCWD, "/dev/dev",
425 UIO_SYSSPACE, 0, 0);
426 if (error)
427 printf("mountroot: unable to unlink /dev/dev "
428 "(error %d)\n", error);
429 }
430 }
431
432 /*
433 * Configuration parser.
434 */
435
436 /* Parser character classes. */
437 #define CC_WHITESPACE -1
438 #define CC_NONWHITESPACE -2
439
440 /* Parse errors. */
441 #define PE_EOF -1
442 #define PE_EOL -2
443
444 static __inline int
parse_peek(char ** conf)445 parse_peek(char **conf)
446 {
447
448 return (**conf);
449 }
450
451 static __inline void
parse_poke(char ** conf,int c)452 parse_poke(char **conf, int c)
453 {
454
455 **conf = c;
456 }
457
458 static __inline void
parse_advance(char ** conf)459 parse_advance(char **conf)
460 {
461
462 (*conf)++;
463 }
464
465 static int
parse_skipto(char ** conf,int mc)466 parse_skipto(char **conf, int mc)
467 {
468 int c, match;
469
470 while (1) {
471 c = parse_peek(conf);
472 if (c == 0)
473 return (PE_EOF);
474 switch (mc) {
475 case CC_WHITESPACE:
476 match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0;
477 break;
478 case CC_NONWHITESPACE:
479 if (c == '\n')
480 return (PE_EOL);
481 match = (c != ' ' && c != '\t') ? 1 : 0;
482 break;
483 default:
484 match = (c == mc) ? 1 : 0;
485 break;
486 }
487 if (match)
488 break;
489 parse_advance(conf);
490 }
491 return (0);
492 }
493
494 static int
parse_token(char ** conf,char ** tok)495 parse_token(char **conf, char **tok)
496 {
497 char *p;
498 size_t len;
499 int error;
500
501 *tok = NULL;
502 error = parse_skipto(conf, CC_NONWHITESPACE);
503 if (error)
504 return (error);
505 p = *conf;
506 error = parse_skipto(conf, CC_WHITESPACE);
507 len = *conf - p;
508 *tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO);
509 bcopy(p, *tok, len);
510 return (0);
511 }
512
513 static void
parse_dir_ask_printenv(const char * var)514 parse_dir_ask_printenv(const char *var)
515 {
516 char *val;
517
518 val = kern_getenv(var);
519 if (val != NULL) {
520 printf(" %s=%s\n", var, val);
521 freeenv(val);
522 }
523 }
524
525 static int
parse_dir_ask(char ** conf)526 parse_dir_ask(char **conf)
527 {
528 char name[80];
529 char *mnt;
530 int error;
531
532 vfs_mountroot_wait();
533
534 printf("\nLoader variables:\n");
535 parse_dir_ask_printenv("vfs.root.mountfrom");
536 parse_dir_ask_printenv("vfs.root.mountfrom.options");
537
538 printf("\nManual root filesystem specification:\n");
539 printf(" <fstype>:<device> [options]\n");
540 printf(" Mount <device> using filesystem <fstype>\n");
541 printf(" and with the specified (optional) option list.\n");
542 printf("\n");
543 printf(" eg. ufs:/dev/da0s1a\n");
544 printf(" zfs:zroot/ROOT/default\n");
545 printf(" cd9660:/dev/cd0 ro\n");
546 printf(" (which is equivalent to: ");
547 printf("mount -t cd9660 -o ro /dev/cd0 /)\n");
548 printf("\n");
549 printf(" ? List valid disk boot devices\n");
550 printf(" . Yield 1 second (for background tasks)\n");
551 printf(" <empty line> Abort manual input\n");
552
553 do {
554 error = EINVAL;
555 printf("\nmountroot> ");
556 cngets(name, sizeof(name), GETS_ECHO);
557 if (name[0] == '\0')
558 break;
559 if (name[0] == '?' && name[1] == '\0') {
560 printf("\nList of GEOM managed disk devices:\n ");
561 g_dev_print();
562 continue;
563 }
564 if (name[0] == '.' && name[1] == '\0') {
565 pause("rmask", hz);
566 continue;
567 }
568 mnt = name;
569 error = parse_mount(&mnt);
570 if (error == -1)
571 printf("Invalid file system specification.\n");
572 } while (error != 0);
573
574 return (error);
575 }
576
577 static int
parse_dir_md(char ** conf)578 parse_dir_md(char **conf)
579 {
580 struct stat sb;
581 struct thread *td;
582 struct md_ioctl *mdio;
583 char *path, *tok;
584 int error, fd, len;
585
586 td = curthread;
587 fd = -1;
588
589 error = parse_token(conf, &tok);
590 if (error)
591 return (error);
592
593 len = strlen(tok);
594 mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO);
595 path = (void *)(mdio + 1);
596 bcopy(tok, path, len);
597 free(tok, M_TEMP);
598
599 /* Get file status. */
600 error = kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &sb, NULL);
601 if (error)
602 goto out;
603
604 /* Open /dev/mdctl so that we can attach/detach. */
605 error = kern_openat(td, AT_FDCWD, "/dev/" MDCTL_NAME, UIO_SYSSPACE,
606 O_RDWR, 0);
607 if (error)
608 goto out;
609
610 fd = td->td_retval[0];
611 mdio->md_version = MDIOVERSION;
612 mdio->md_type = MD_VNODE;
613
614 if (root_mount_mddev != -1) {
615 mdio->md_unit = root_mount_mddev;
616 (void)kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
617 /* Ignore errors. We don't care. */
618 root_mount_mddev = -1;
619 }
620
621 mdio->md_file = (void *)(mdio + 1);
622 mdio->md_options = MD_AUTOUNIT | MD_READONLY;
623 mdio->md_mediasize = sb.st_size;
624 mdio->md_unit = 0;
625 error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio);
626 if (error)
627 goto out;
628
629 if (mdio->md_unit > 9) {
630 printf("rootmount: too many md units\n");
631 mdio->md_file = NULL;
632 mdio->md_options = 0;
633 mdio->md_mediasize = 0;
634 error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio);
635 /* Ignore errors. We don't care. */
636 error = ERANGE;
637 goto out;
638 }
639
640 root_mount_mddev = mdio->md_unit;
641 printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file);
642
643 out:
644 if (fd >= 0)
645 (void)kern_close(td, fd);
646 free(mdio, M_TEMP);
647 return (error);
648 }
649
650 static int
parse_dir_onfail(char ** conf)651 parse_dir_onfail(char **conf)
652 {
653 char *action;
654 int error;
655
656 error = parse_token(conf, &action);
657 if (error)
658 return (error);
659
660 if (!strcmp(action, "continue"))
661 root_mount_onfail = A_CONTINUE;
662 else if (!strcmp(action, "panic"))
663 root_mount_onfail = A_PANIC;
664 else if (!strcmp(action, "reboot"))
665 root_mount_onfail = A_REBOOT;
666 else if (!strcmp(action, "retry"))
667 root_mount_onfail = A_RETRY;
668 else {
669 printf("rootmount: %s: unknown action\n", action);
670 error = EINVAL;
671 }
672
673 free(action, M_TEMP);
674 return (0);
675 }
676
677 static int
parse_dir_timeout(char ** conf)678 parse_dir_timeout(char **conf)
679 {
680 char *tok, *endtok;
681 long secs;
682 int error;
683
684 error = parse_token(conf, &tok);
685 if (error)
686 return (error);
687
688 secs = strtol(tok, &endtok, 0);
689 error = (secs < 0 || *endtok != '\0') ? EINVAL : 0;
690 if (!error)
691 root_mount_timeout = secs;
692 free(tok, M_TEMP);
693 return (error);
694 }
695
696 static int
parse_directive(char ** conf)697 parse_directive(char **conf)
698 {
699 char *dir;
700 int error;
701
702 error = parse_token(conf, &dir);
703 if (error)
704 return (error);
705
706 if (strcmp(dir, ".ask") == 0)
707 error = parse_dir_ask(conf);
708 else if (strcmp(dir, ".md") == 0)
709 error = parse_dir_md(conf);
710 else if (strcmp(dir, ".onfail") == 0)
711 error = parse_dir_onfail(conf);
712 else if (strcmp(dir, ".timeout") == 0)
713 error = parse_dir_timeout(conf);
714 else {
715 printf("mountroot: invalid directive `%s'\n", dir);
716 /* Ignore the rest of the line. */
717 (void)parse_skipto(conf, '\n');
718 error = EINVAL;
719 }
720 free(dir, M_TEMP);
721 return (error);
722 }
723
724 static int
parse_mount_dev_present(const char * dev)725 parse_mount_dev_present(const char *dev)
726 {
727 struct nameidata nd;
728 int error;
729
730 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread);
731 error = namei(&nd);
732 if (!error)
733 vput(nd.ni_vp);
734 NDFREE(&nd, NDF_ONLY_PNBUF);
735 return (error != 0) ? 0 : 1;
736 }
737
738 #define ERRMSGL 255
739 static int
parse_mount(char ** conf)740 parse_mount(char **conf)
741 {
742 char *errmsg;
743 struct mntarg *ma;
744 char *dev, *fs, *opts, *tok;
745 int delay, error, timeout;
746
747 error = parse_token(conf, &tok);
748 if (error)
749 return (error);
750 fs = tok;
751 error = parse_skipto(&tok, ':');
752 if (error) {
753 free(fs, M_TEMP);
754 return (error);
755 }
756 parse_poke(&tok, '\0');
757 parse_advance(&tok);
758 dev = tok;
759
760 if (root_mount_mddev != -1) {
761 /* Handle substitution for the md unit number. */
762 tok = strstr(dev, "md#");
763 if (tok != NULL)
764 tok[2] = '0' + root_mount_mddev;
765 }
766
767 /* Parse options. */
768 error = parse_token(conf, &tok);
769 opts = (error == 0) ? tok : NULL;
770
771 printf("Trying to mount root from %s:%s [%s]...\n", fs, dev,
772 (opts != NULL) ? opts : "");
773
774 errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO);
775
776 if (vfs_byname(fs) == NULL) {
777 strlcpy(errmsg, "unknown file system", ERRMSGL);
778 error = ENOENT;
779 goto out;
780 }
781
782 error = vfs_mountroot_wait_if_neccessary(fs, dev);
783 if (error != 0)
784 goto out;
785
786 delay = hz / 10;
787 timeout = root_mount_timeout * hz;
788
789 for (;;) {
790 ma = NULL;
791 ma = mount_arg(ma, "fstype", fs, -1);
792 ma = mount_arg(ma, "fspath", "/", -1);
793 ma = mount_arg(ma, "from", dev, -1);
794 ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL);
795 ma = mount_arg(ma, "ro", NULL, 0);
796 ma = parse_mountroot_options(ma, opts);
797
798 error = kernel_mount(ma, MNT_ROOTFS);
799 if (error == 0 || timeout <= 0)
800 break;
801
802 if (root_mount_timeout * hz == timeout ||
803 (bootverbose && timeout % hz == 0)) {
804 printf("Mounting from %s:%s failed with error %d; "
805 "retrying for %d more second%s\n", fs, dev, error,
806 timeout / hz, (timeout / hz > 1) ? "s" : "");
807 }
808 pause("rmretry", delay);
809 timeout -= delay;
810 }
811 out:
812 if (error) {
813 printf("Mounting from %s:%s failed with error %d",
814 fs, dev, error);
815 if (errmsg[0] != '\0')
816 printf(": %s", errmsg);
817 printf(".\n");
818 }
819 free(fs, M_TEMP);
820 free(errmsg, M_TEMP);
821 if (opts != NULL)
822 free(opts, M_TEMP);
823 /* kernel_mount can return -1 on error. */
824 return ((error < 0) ? EDOOFUS : error);
825 }
826 #undef ERRMSGL
827
828 static int
vfs_mountroot_parse(struct sbuf * sb,struct mount * mpdevfs)829 vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs)
830 {
831 struct mount *mp;
832 char *conf;
833 int error;
834
835 root_mount_mddev = -1;
836
837 retry:
838 conf = sbuf_data(sb);
839 mp = TAILQ_NEXT(mpdevfs, mnt_list);
840 error = (mp == NULL) ? 0 : EDOOFUS;
841 root_mount_onfail = A_CONTINUE;
842 while (mp == NULL) {
843 error = parse_skipto(&conf, CC_NONWHITESPACE);
844 if (error == PE_EOL) {
845 parse_advance(&conf);
846 continue;
847 }
848 if (error < 0)
849 break;
850 switch (parse_peek(&conf)) {
851 case '#':
852 error = parse_skipto(&conf, '\n');
853 break;
854 case '.':
855 error = parse_directive(&conf);
856 break;
857 default:
858 error = parse_mount(&conf);
859 if (error == -1) {
860 printf("mountroot: invalid file system "
861 "specification.\n");
862 error = 0;
863 }
864 break;
865 }
866 if (error < 0)
867 break;
868 /* Ignore any trailing garbage on the line. */
869 if (parse_peek(&conf) != '\n') {
870 printf("mountroot: advancing to next directive...\n");
871 (void)parse_skipto(&conf, '\n');
872 }
873 mp = TAILQ_NEXT(mpdevfs, mnt_list);
874 }
875 if (mp != NULL)
876 return (0);
877
878 /*
879 * We failed to mount (a new) root.
880 */
881 switch (root_mount_onfail) {
882 case A_CONTINUE:
883 break;
884 case A_PANIC:
885 panic("mountroot: unable to (re-)mount root.");
886 /* NOTREACHED */
887 case A_RETRY:
888 goto retry;
889 case A_REBOOT:
890 kern_reboot(RB_NOSYNC);
891 /* NOTREACHED */
892 }
893
894 return (error);
895 }
896
897 static void
vfs_mountroot_conf0(struct sbuf * sb)898 vfs_mountroot_conf0(struct sbuf *sb)
899 {
900 char *s, *tok, *mnt, *opt;
901 int error;
902
903 sbuf_printf(sb, ".onfail panic\n");
904 sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
905 if (boothowto & RB_ASKNAME)
906 sbuf_printf(sb, ".ask\n");
907 #ifdef ROOTDEVNAME
908 if (boothowto & RB_DFLTROOT)
909 sbuf_printf(sb, "%s\n", ROOTDEVNAME);
910 #endif
911 if (boothowto & RB_CDROM) {
912 sbuf_printf(sb, "cd9660:/dev/cd0 ro\n");
913 sbuf_printf(sb, ".timeout 0\n");
914 sbuf_printf(sb, "cd9660:/dev/cd1 ro\n");
915 sbuf_printf(sb, ".timeout %d\n", root_mount_timeout);
916 }
917 s = kern_getenv("vfs.root.mountfrom");
918 if (s != NULL) {
919 opt = kern_getenv("vfs.root.mountfrom.options");
920 tok = s;
921 error = parse_token(&tok, &mnt);
922 while (!error) {
923 sbuf_printf(sb, "%s %s\n", mnt,
924 (opt != NULL) ? opt : "");
925 free(mnt, M_TEMP);
926 error = parse_token(&tok, &mnt);
927 }
928 if (opt != NULL)
929 freeenv(opt);
930 freeenv(s);
931 }
932 if (rootdevnames[0] != NULL)
933 sbuf_printf(sb, "%s\n", rootdevnames[0]);
934 if (rootdevnames[1] != NULL)
935 sbuf_printf(sb, "%s\n", rootdevnames[1]);
936 #ifdef ROOTDEVNAME
937 if (!(boothowto & RB_DFLTROOT))
938 sbuf_printf(sb, "%s\n", ROOTDEVNAME);
939 #endif
940 if (!(boothowto & RB_ASKNAME))
941 sbuf_printf(sb, ".ask\n");
942 }
943
944 static int
vfs_mountroot_readconf(struct thread * td,struct sbuf * sb)945 vfs_mountroot_readconf(struct thread *td, struct sbuf *sb)
946 {
947 static char buf[128];
948 struct nameidata nd;
949 off_t ofs;
950 ssize_t resid;
951 int error, flags, len;
952
953 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td);
954 flags = FREAD;
955 error = vn_open(&nd, &flags, 0, NULL);
956 if (error)
957 return (error);
958
959 NDFREE(&nd, NDF_ONLY_PNBUF);
960 ofs = 0;
961 len = sizeof(buf) - 1;
962 while (1) {
963 error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs,
964 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
965 NOCRED, &resid, td);
966 if (error)
967 break;
968 if (resid == len)
969 break;
970 buf[len - resid] = 0;
971 sbuf_printf(sb, "%s", buf);
972 ofs += len - resid;
973 }
974
975 VOP_UNLOCK(nd.ni_vp, 0);
976 vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
977 return (error);
978 }
979
980 static void
vfs_mountroot_wait(void)981 vfs_mountroot_wait(void)
982 {
983 struct root_hold_token *h;
984 struct timeval lastfail;
985 int curfail;
986
987 TSENTER();
988
989 curfail = 0;
990 while (1) {
991 g_waitidle();
992 mtx_lock(&root_holds_mtx);
993 if (TAILQ_EMPTY(&root_holds)) {
994 mtx_unlock(&root_holds_mtx);
995 break;
996 }
997 if (ppsratecheck(&lastfail, &curfail, 1)) {
998 printf("Root mount waiting for:");
999 TAILQ_FOREACH(h, &root_holds, list)
1000 printf(" %s", h->who);
1001 printf("\n");
1002 }
1003 TSWAIT("root mount");
1004 msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold",
1005 hz);
1006 TSUNWAIT("root mount");
1007 }
1008
1009 TSEXIT();
1010 }
1011
1012 static int
vfs_mountroot_wait_if_neccessary(const char * fs,const char * dev)1013 vfs_mountroot_wait_if_neccessary(const char *fs, const char *dev)
1014 {
1015 int delay, timeout;
1016
1017 /*
1018 * In case of ZFS and NFS we don't have a way to wait for
1019 * specific device. Also do the wait if the user forced that
1020 * behaviour by setting vfs.root_mount_always_wait=1.
1021 */
1022 if (strcmp(fs, "zfs") == 0 || strstr(fs, "nfs") != NULL ||
1023 dev[0] == '\0' || root_mount_always_wait != 0) {
1024 vfs_mountroot_wait();
1025 return (0);
1026 }
1027
1028 /*
1029 * Otherwise, no point in waiting if the device is already there.
1030 * Note that we must wait for GEOM to finish reconfiguring itself,
1031 * eg for geom_part(4) to finish tasting.
1032 */
1033 g_waitidle();
1034 if (parse_mount_dev_present(dev))
1035 return (0);
1036
1037 /*
1038 * No luck. Let's wait. This code looks weird, but it's that way
1039 * to behave exactly as it used to work before.
1040 */
1041 vfs_mountroot_wait();
1042 printf("mountroot: waiting for device %s...\n", dev);
1043 delay = hz / 10;
1044 timeout = root_mount_timeout * hz;
1045 do {
1046 pause("rmdev", delay);
1047 timeout -= delay;
1048 } while (timeout > 0 && !parse_mount_dev_present(dev));
1049
1050 if (timeout <= 0)
1051 return (ENODEV);
1052
1053 return (0);
1054 }
1055
1056 void
vfs_mountroot(void)1057 vfs_mountroot(void)
1058 {
1059 struct mount *mp;
1060 struct sbuf *sb;
1061 struct thread *td;
1062 time_t timebase;
1063 int error;
1064
1065 mtx_assert(&Giant, MA_NOTOWNED);
1066
1067 TSENTER();
1068
1069 td = curthread;
1070
1071 sb = sbuf_new_auto();
1072 vfs_mountroot_conf0(sb);
1073 sbuf_finish(sb);
1074
1075 error = vfs_mountroot_devfs(td, &mp);
1076 while (!error) {
1077 error = vfs_mountroot_parse(sb, mp);
1078 if (!error) {
1079 vfs_mountroot_shuffle(td, mp);
1080 sbuf_clear(sb);
1081 error = vfs_mountroot_readconf(td, sb);
1082 sbuf_finish(sb);
1083 }
1084 }
1085
1086 sbuf_delete(sb);
1087
1088 /*
1089 * Iterate over all currently mounted file systems and use
1090 * the time stamp found to check and/or initialize the RTC.
1091 * Call inittodr() only once and pass it the largest of the
1092 * timestamps we encounter.
1093 */
1094 timebase = 0;
1095 mtx_lock(&mountlist_mtx);
1096 mp = TAILQ_FIRST(&mountlist);
1097 while (mp != NULL) {
1098 if (mp->mnt_time > timebase)
1099 timebase = mp->mnt_time;
1100 mp = TAILQ_NEXT(mp, mnt_list);
1101 }
1102 mtx_unlock(&mountlist_mtx);
1103 inittodr(timebase);
1104
1105 /* Keep prison0's root in sync with the global rootvnode. */
1106 mtx_lock(&prison0.pr_mtx);
1107 prison0.pr_root = rootvnode;
1108 vref(prison0.pr_root);
1109 mtx_unlock(&prison0.pr_mtx);
1110
1111 mtx_lock(&root_holds_mtx);
1112 atomic_store_rel_int(&root_mount_complete, 1);
1113 wakeup(&root_mount_complete);
1114 mtx_unlock(&root_holds_mtx);
1115
1116 EVENTHANDLER_INVOKE(mountroot);
1117
1118 TSEXIT();
1119 }
1120
1121 static struct mntarg *
parse_mountroot_options(struct mntarg * ma,const char * options)1122 parse_mountroot_options(struct mntarg *ma, const char *options)
1123 {
1124 char *p;
1125 char *name, *name_arg;
1126 char *val, *val_arg;
1127 char *opts;
1128
1129 if (options == NULL || options[0] == '\0')
1130 return (ma);
1131
1132 p = opts = strdup(options, M_MOUNT);
1133 if (opts == NULL) {
1134 return (ma);
1135 }
1136
1137 while((name = strsep(&p, ",")) != NULL) {
1138 if (name[0] == '\0')
1139 break;
1140
1141 val = strchr(name, '=');
1142 if (val != NULL) {
1143 *val = '\0';
1144 ++val;
1145 }
1146 if( strcmp(name, "rw") == 0 ||
1147 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