xref: /freebsd/sys/kern/vfs_mount.c (revision c0b9f4fe659b6839541970eb5675e57f4d814969)
1 /*-
2  * Copyright (c) 1999-2004 Poul-Henning Kamp
3  * Copyright (c) 1999 Michael Smith
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/conf.h>
42 #include <sys/jail.h>
43 #include <sys/kernel.h>
44 #include <sys/libkern.h>
45 #include <sys/mac.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/proc.h>
51 #include <sys/filedesc.h>
52 #include <sys/reboot.h>
53 #include <sys/syscallsubr.h>
54 #include <sys/sysproto.h>
55 #include <sys/sx.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysent.h>
58 #include <sys/systm.h>
59 #include <sys/vnode.h>
60 
61 #include <geom/geom.h>
62 
63 #include <machine/stdarg.h>
64 
65 #include "opt_rootdevname.h"
66 #include "opt_ddb.h"
67 #include "opt_mac.h"
68 
69 #ifdef DDB
70 #include <ddb/ddb.h>
71 #endif
72 
73 #define	ROOTNAME		"root_device"
74 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
75 
76 static int	vfs_domount(struct thread *td, const char *fstype,
77 		    char *fspath, int fsflags, void *fsdata);
78 static struct mount *vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp,
79 		    const char *fspath, struct thread *td);
80 static int	vfs_mountroot_ask(void);
81 static int	vfs_mountroot_try(const char *mountfrom);
82 static int	vfs_donmount(struct thread *td, int fsflags,
83 		    struct uio *fsoptions);
84 static void	free_mntarg(struct mntarg *ma);
85 static void	vfs_mount_destroy(struct mount *, struct thread *);
86 static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
87 
88 static int	usermount = 0;
89 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
90     "Unprivileged users may mount and unmount file systems");
91 
92 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
93 
94 /* List of mounted filesystems. */
95 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
96 
97 /* For any iteration/modification of mountlist */
98 struct mtx mountlist_mtx;
99 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
100 
101 TAILQ_HEAD(vfsoptlist, vfsopt);
102 struct vfsopt {
103 	TAILQ_ENTRY(vfsopt) link;
104 	char	*name;
105 	void	*value;
106 	int	len;
107 };
108 
109 /*
110  * The vnode of the system's root (/ in the filesystem, without chroot
111  * active.)
112  */
113 struct vnode	*rootvnode;
114 
115 /*
116  * The root filesystem is detailed in the kernel environment variable
117  * vfs.root.mountfrom, which is expected to be in the general format
118  *
119  * <vfsname>:[<path>]
120  * vfsname   := the name of a VFS known to the kernel and capable
121  *              of being mounted as root
122  * path      := disk device name or other data used by the filesystem
123  *              to locate its physical store
124  */
125 
126 /*
127  * Global opts, taken by all filesystems
128  */
129 static const char *global_opts[] = {
130 	"errmsg",
131 	"fstype",
132 	"fspath",
133 	"rdonly",
134 	"ro",
135 	"rw",
136 	"suid",
137 	"exec",
138 	NULL
139 };
140 
141 /*
142  * The root specifiers we will try if RB_CDROM is specified.
143  */
144 static char *cdrom_rootdevnames[] = {
145 	"cd9660:cd0",
146 	"cd9660:acd0",
147 	NULL
148 };
149 
150 /* legacy find-root code */
151 char		*rootdevnames[2] = {NULL, NULL};
152 #ifndef ROOTDEVNAME
153 #  define ROOTDEVNAME NULL
154 #endif
155 static const char	*ctrootdevname = ROOTDEVNAME;
156 
157 /*
158  * ---------------------------------------------------------------------
159  * Functions for building and sanitizing the mount options
160  */
161 
162 /* Remove one mount option. */
163 static void
164 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
165 {
166 
167 	TAILQ_REMOVE(opts, opt, link);
168 	free(opt->name, M_MOUNT);
169 	if (opt->value != NULL)
170 		free(opt->value, M_MOUNT);
171 #ifdef INVARIANTS
172 	else if (opt->len != 0)
173 		panic("%s: mount option with NULL value but length != 0",
174 		    __func__);
175 #endif
176 	free(opt, M_MOUNT);
177 }
178 
179 /* Release all resources related to the mount options. */
180 static void
181 vfs_freeopts(struct vfsoptlist *opts)
182 {
183 	struct vfsopt *opt;
184 
185 	while (!TAILQ_EMPTY(opts)) {
186 		opt = TAILQ_FIRST(opts);
187 		vfs_freeopt(opts, opt);
188 	}
189 	free(opts, M_MOUNT);
190 }
191 
192 /*
193  * Check if options are equal (with or without the "no" prefix).
194  */
195 static int
196 vfs_equalopts(const char *opt1, const char *opt2)
197 {
198 
199 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
200 	if (strcmp(opt1, opt2) == 0)
201 		return (1);
202 	/* "noopt" vs. "opt" */
203 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
204 		return (1);
205 	/* "opt" vs. "noopt" */
206 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
207 		return (1);
208 	return (0);
209 }
210 
211 /*
212  * If a mount option is specified several times,
213  * (with or without the "no" prefix) only keep
214  * the last occurence of it.
215  */
216 static void
217 vfs_sanitizeopts(struct vfsoptlist *opts)
218 {
219 	struct vfsopt *opt, *opt2, *tmp;
220 
221 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
222 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
223 		while (opt2 != NULL) {
224 			if (vfs_equalopts(opt->name, opt2->name)) {
225 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
226 				vfs_freeopt(opts, opt2);
227 				opt2 = tmp;
228 			} else {
229 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
230 			}
231 		}
232 	}
233 }
234 
235 /*
236  * Build a linked list of mount options from a struct uio.
237  */
238 static int
239 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
240 {
241 	struct vfsoptlist *opts;
242 	struct vfsopt *opt;
243 	size_t memused;
244 	unsigned int i, iovcnt;
245 	int error, namelen, optlen;
246 
247 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
248 	TAILQ_INIT(opts);
249 	memused = 0;
250 	iovcnt = auio->uio_iovcnt;
251 	for (i = 0; i < iovcnt; i += 2) {
252 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
253 		namelen = auio->uio_iov[i].iov_len;
254 		optlen = auio->uio_iov[i + 1].iov_len;
255 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
256 		opt->value = NULL;
257 		opt->len = 0;
258 
259 		/*
260 		 * Do this early, so jumps to "bad" will free the current
261 		 * option.
262 		 */
263 		TAILQ_INSERT_TAIL(opts, opt, link);
264 		memused += sizeof(struct vfsopt) + optlen + namelen;
265 
266 		/*
267 		 * Avoid consuming too much memory, and attempts to overflow
268 		 * memused.
269 		 */
270 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
271 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
272 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
273 			error = EINVAL;
274 			goto bad;
275 		}
276 
277 		if (auio->uio_segflg == UIO_SYSSPACE) {
278 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
279 		} else {
280 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
281 			    namelen);
282 			if (error)
283 				goto bad;
284 		}
285 		/* Ensure names are null-terminated strings. */
286 		if (opt->name[namelen - 1] != '\0') {
287 			error = EINVAL;
288 			goto bad;
289 		}
290 		if (optlen != 0) {
291 			opt->len = optlen;
292 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
293 			if (auio->uio_segflg == UIO_SYSSPACE) {
294 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
295 				    optlen);
296 			} else {
297 				error = copyin(auio->uio_iov[i + 1].iov_base,
298 				    opt->value, optlen);
299 				if (error)
300 					goto bad;
301 			}
302 		}
303 	}
304 	vfs_sanitizeopts(opts);
305 	*options = opts;
306 	return (0);
307 bad:
308 	vfs_freeopts(opts);
309 	return (error);
310 }
311 
312 /*
313  * Merge the old mount options with the new ones passed
314  * in the MNT_UPDATE case.
315  */
316 static void
317 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
318 {
319 	struct vfsopt *opt, *opt2, *new;
320 
321 	TAILQ_FOREACH(opt, opts, link) {
322 		/*
323 		 * Check that this option hasn't been redefined
324 		 * nor cancelled with a "no" mount option.
325 		 */
326 		opt2 = TAILQ_FIRST(toopts);
327 		while (opt2 != NULL) {
328 			if (strcmp(opt2->name, opt->name) == 0)
329 				goto next;
330 			if (strncmp(opt2->name, "no", 2) == 0 &&
331 			    strcmp(opt2->name + 2, opt->name) == 0) {
332 				vfs_freeopt(toopts, opt2);
333 				goto next;
334 			}
335 			opt2 = TAILQ_NEXT(opt2, link);
336 		}
337 		/* We want this option, duplicate it. */
338 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
339 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
340 		strcpy(new->name, opt->name);
341 		if (opt->len != 0) {
342 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
343 			bcopy(opt->value, new->value, opt->len);
344 		} else {
345 			new->value = NULL;
346 		}
347 		new->len = opt->len;
348 		TAILQ_INSERT_TAIL(toopts, new, link);
349 next:
350 		continue;
351 	}
352 }
353 
354 /*
355  * ---------------------------------------------------------------------
356  * Mount a filesystem
357  */
358 int
359 nmount(td, uap)
360 	struct thread *td;
361 	struct nmount_args /* {
362 		struct iovec *iovp;
363 		unsigned int iovcnt;
364 		int flags;
365 	} */ *uap;
366 {
367 	struct uio *auio;
368 	struct iovec *iov;
369 	unsigned int i;
370 	int error;
371 	u_int iovcnt;
372 
373 	/* Kick out MNT_ROOTFS early as it is legal internally */
374 	if (uap->flags & MNT_ROOTFS)
375 		return (EINVAL);
376 
377 	iovcnt = uap->iovcnt;
378 	/*
379 	 * Check that we have an even number of iovec's
380 	 * and that we have at least two options.
381 	 */
382 	if ((iovcnt & 1) || (iovcnt < 4))
383 		return (EINVAL);
384 
385 	error = copyinuio(uap->iovp, iovcnt, &auio);
386 	if (error)
387 		return (error);
388 	iov = auio->uio_iov;
389 	for (i = 0; i < iovcnt; i++) {
390 		if (iov->iov_len > MMAXOPTIONLEN) {
391 			free(auio, M_IOV);
392 			return (EINVAL);
393 		}
394 		iov++;
395 	}
396 	error = vfs_donmount(td, uap->flags, auio);
397 
398 	free(auio, M_IOV);
399 	return (error);
400 }
401 
402 /*
403  * ---------------------------------------------------------------------
404  * Various utility functions
405  */
406 
407 /*
408  * Allocate and initialize the mount point struct.
409  */
410 static struct mount *
411 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
412     const char *fspath, struct thread *td)
413 {
414 	struct mount *mp;
415 
416 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
417 	TAILQ_INIT(&mp->mnt_nvnodelist);
418 	mp->mnt_nvnodelistsize = 0;
419 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
420 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
421 	vfs_busy(mp, LK_NOWAIT, 0, td);
422 	mp->mnt_op = vfsp->vfc_vfsops;
423 	mp->mnt_vfc = vfsp;
424 	vfsp->vfc_refcount++;
425 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
426 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
427 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
428 	mp->mnt_vnodecovered = vp;
429 	mp->mnt_cred = crdup(td->td_ucred);
430 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
431 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
432 	mp->mnt_iosize_max = DFLTPHYS;
433 #ifdef MAC
434 	mac_init_mount(mp);
435 	mac_create_mount(td->td_ucred, mp);
436 #endif
437 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
438 	return (mp);
439 }
440 
441 /*
442  * Destroy the mount struct previously allocated by vfs_mount_alloc().
443  */
444 static void
445 vfs_mount_destroy(struct mount *mp, struct thread *td)
446 {
447 
448 	mp->mnt_vfc->vfc_refcount--;
449 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
450 		panic("unmount: dangling vnode");
451 	vfs_unbusy(mp,td);
452 	lockdestroy(&mp->mnt_lock);
453 	MNT_ILOCK(mp);
454 	if (mp->mnt_kern_flag & MNTK_MWAIT)
455 		wakeup(mp);
456 	MNT_IUNLOCK(mp);
457 	mtx_destroy(&mp->mnt_mtx);
458 #ifdef MAC
459 	mac_destroy_mount(mp);
460 #endif
461 	if (mp->mnt_opt != NULL)
462 		vfs_freeopts(mp->mnt_opt);
463 	crfree(mp->mnt_cred);
464 	free(mp, M_MOUNT);
465 }
466 
467 static int
468 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
469 {
470 	struct vfsoptlist *optlist;
471 	char *fstype, *fspath, *errmsg;
472 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
473 
474 	errmsg_len = 0;
475 	errmsg_pos = -1;
476 
477 	error = vfs_buildopts(fsoptions, &optlist);
478 	if (error)
479 		return (error);
480 
481 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
482 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
483 	else
484 		errmsg_len = 0;
485 
486 	/*
487 	 * We need these two options before the others,
488 	 * and they are mandatory for any filesystem.
489 	 * Ensure they are NUL terminated as well.
490 	 */
491 	fstypelen = 0;
492 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
493 	if (error || fstype[fstypelen - 1] != '\0') {
494 		error = EINVAL;
495 		if (errmsg != NULL)
496 			strncpy(errmsg, "Invalid fstype", errmsg_len);
497 		goto bail;
498 	}
499 	fspathlen = 0;
500 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
501 	if (error || fspath[fspathlen - 1] != '\0') {
502 		error = EINVAL;
503 		if (errmsg != NULL)
504 			strncpy(errmsg, "Invalid fspath", errmsg_len);
505 		goto bail;
506 	}
507 
508 	/*
509 	 * We need to see if we have the "update" option
510 	 * before we call vfs_domount(), since vfs_domount() has special
511 	 * logic based on MNT_UPDATE.  This is very important
512 	 * when we want to update the root filesystem.
513 	 */
514 	if (vfs_getopt(optlist, "update", NULL, NULL) == 0)
515 		fsflags |= MNT_UPDATE;
516 
517 	if (vfs_getopt(optlist, "async", NULL, NULL) == 0)
518 		fsflags |= MNT_ASYNC;
519 
520 	if (vfs_getopt(optlist, "force", NULL, NULL) == 0)
521 		fsflags |= MNT_FORCE;
522 
523 	if (vfs_getopt(optlist, "multilabel", NULL, NULL) == 0)
524 		fsflags |= MNT_MULTILABEL;
525 
526 	if (vfs_getopt(optlist, "noasync", NULL, NULL) == 0)
527 		fsflags &= ~MNT_ASYNC;
528 
529 	if (vfs_getopt(optlist, "noatime", NULL, NULL) == 0)
530 		fsflags |= MNT_NOATIME;
531 
532 	if (vfs_getopt(optlist, "noclusterr", NULL, NULL) == 0)
533 		fsflags |= MNT_NOCLUSTERR;
534 
535 	if (vfs_getopt(optlist, "noclusterw", NULL, NULL) == 0)
536 		fsflags |= MNT_NOCLUSTERW;
537 
538 	if (vfs_getopt(optlist, "noexec", NULL, NULL) == 0)
539 		fsflags |= MNT_NOEXEC;
540 
541 	if (vfs_getopt(optlist, "nosuid", NULL, NULL) == 0)
542 		fsflags |= MNT_NOSUID;
543 
544 	if (vfs_getopt(optlist, "nosymfollow", NULL, NULL) == 0)
545 		fsflags |= MNT_NOSYMFOLLOW;
546 
547 	if (vfs_getopt(optlist, "noro", NULL, NULL) == 0)
548 		fsflags &= ~MNT_RDONLY;
549 
550 	if (vfs_getopt(optlist, "ro", NULL, NULL) == 0)
551 		fsflags |= MNT_RDONLY;
552 
553 	if (vfs_getopt(optlist, "rdonly", NULL, NULL) == 0)
554 		fsflags |= MNT_RDONLY;
555 
556 	if (vfs_getopt(optlist, "rw", NULL, NULL) == 0)
557 		fsflags &= ~MNT_RDONLY;
558 
559 	if (vfs_getopt(optlist, "snapshot", NULL, NULL) == 0)
560 		fsflags |= MNT_SNAPSHOT;
561 
562 	if (vfs_getopt(optlist, "suiddir", NULL, NULL) == 0)
563 		fsflags |= MNT_SUIDDIR;
564 
565 	if (vfs_getopt(optlist, "sync", NULL, NULL) == 0)
566 		fsflags |= MNT_SYNCHRONOUS;
567 
568 	if (vfs_getopt(optlist, "union", NULL, NULL) == 0)
569 		fsflags |= MNT_UNION;
570 
571 	/*
572 	 * Be ultra-paranoid about making sure the type and fspath
573 	 * variables will fit in our mp buffers, including the
574 	 * terminating NUL.
575 	 */
576 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
577 		error = ENAMETOOLONG;
578 		goto bail;
579 	}
580 
581 	mtx_lock(&Giant);
582 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
583 	mtx_unlock(&Giant);
584 bail:
585 	/* copyout the errmsg */
586 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
587 	    && errmsg_len > 0 && errmsg != NULL) {
588 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
589 			strncpy(fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
590 			    errmsg,
591 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
592 		} else {
593 			copystr(errmsg,
594 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
595 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len,
596 			    NULL);
597 		}
598 	}
599 
600 	if (error != 0)
601 		vfs_freeopts(optlist);
602 	return (error);
603 }
604 
605 /*
606  * ---------------------------------------------------------------------
607  * Old mount API.
608  */
609 #ifndef _SYS_SYSPROTO_H_
610 struct mount_args {
611 	char	*type;
612 	char	*path;
613 	int	flags;
614 	caddr_t	data;
615 };
616 #endif
617 /* ARGSUSED */
618 int
619 mount(td, uap)
620 	struct thread *td;
621 	struct mount_args /* {
622 		char *type;
623 		char *path;
624 		int flags;
625 		caddr_t data;
626 	} */ *uap;
627 {
628 	char *fstype;
629 	struct vfsconf *vfsp = NULL;
630 	struct mntarg *ma = NULL;
631 	int error;
632 
633 	/* Kick out MNT_ROOTFS early as it is legal internally */
634 	uap->flags &= ~MNT_ROOTFS;
635 
636 	if (uap->data == NULL)
637 		return (EINVAL);
638 
639 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
640 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
641 	if (!error) {
642 		mtx_lock(&Giant);	/* XXX ? */
643 		vfsp = vfs_byname_kld(fstype, td, &error);
644 		mtx_unlock(&Giant);
645 	}
646 	free(fstype, M_TEMP);
647 	if (error)
648 		return (error);
649 	if (vfsp == NULL)
650 		return (ENOENT);
651 	if (vfsp->vfc_vfsops->vfs_cmount == NULL)
652 		return (EOPNOTSUPP);
653 
654 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
655 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
656 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
657 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
658 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
659 
660 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
661 	return (error);
662 }
663 
664 
665 /*
666  * vfs_domount(): actually attempt a filesystem mount.
667  */
668 static int
669 vfs_domount(
670 	struct thread *td,	/* Flags common to all filesystems. */
671 	const char *fstype,	/* Filesystem type. */
672 	char *fspath,		/* Mount path. */
673 	int fsflags,		/* Flags common to all filesystems. */
674 	void *fsdata		/* Options local to the filesystem. */
675 	)
676 {
677 	struct vnode *vp;
678 	struct mount *mp;
679 	struct vfsconf *vfsp;
680 	int error, flag = 0, kern_flag = 0;
681 	struct vattr va;
682 	struct nameidata nd;
683 
684 	mtx_assert(&Giant, MA_OWNED);
685 
686 	/*
687 	 * Be ultra-paranoid about making sure the type and fspath
688 	 * variables will fit in our mp buffers, including the
689 	 * terminating NUL.
690 	 */
691 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
692 		return (ENAMETOOLONG);
693 
694 	if (jailed(td->td_ucred))
695 		return (EPERM);
696 	if (usermount == 0) {
697 		if ((error = suser(td)) != 0)
698 			return (error);
699 	}
700 
701 	/*
702 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
703 	 */
704 	if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) {
705 		if ((error = suser(td)) != 0)
706 			return (error);
707 	}
708 	/*
709 	 * Silently enforce MNT_NOSUID and MNT_USER for
710 	 * unprivileged users.
711 	 */
712 	if (suser(td) != 0)
713 		fsflags |= MNT_NOSUID | MNT_USER;
714 	/*
715 	 * Get vnode to be covered
716 	 */
717 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
718 	if ((error = namei(&nd)) != 0)
719 		return (error);
720 	NDFREE(&nd, NDF_ONLY_PNBUF);
721 	vp = nd.ni_vp;
722 	if (fsflags & MNT_UPDATE) {
723 		if ((vp->v_vflag & VV_ROOT) == 0) {
724 			vput(vp);
725 			return (EINVAL);
726 		}
727 		mp = vp->v_mount;
728 		flag = mp->mnt_flag;
729 		kern_flag = mp->mnt_kern_flag;
730 		/*
731 		 * We only allow the filesystem to be reloaded if it
732 		 * is currently mounted read-only.
733 		 */
734 		if ((fsflags & MNT_RELOAD) &&
735 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
736 			vput(vp);
737 			return (EOPNOTSUPP);	/* Needs translation */
738 		}
739 		/*
740 		 * Only privileged root, or (if MNT_USER is set) the user that
741 		 * did the original mount is permitted to update it.
742 		 */
743 		error = vfs_suser(mp, td);
744 		if (error) {
745 			vput(vp);
746 			return (error);
747 		}
748 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
749 			vput(vp);
750 			return (EBUSY);
751 		}
752 		VI_LOCK(vp);
753 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
754 		    vp->v_mountedhere != NULL) {
755 			VI_UNLOCK(vp);
756 			vfs_unbusy(mp, td);
757 			vput(vp);
758 			return (EBUSY);
759 		}
760 		vp->v_iflag |= VI_MOUNT;
761 		VI_UNLOCK(vp);
762 		mp->mnt_flag |= fsflags &
763 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
764 		VOP_UNLOCK(vp, 0, td);
765 		mp->mnt_optnew = fsdata;
766 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
767 	} else {
768 		/*
769 		 * If the user is not root, ensure that they own the directory
770 		 * onto which we are attempting to mount.
771 		 */
772 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
773 		if (error) {
774 			vput(vp);
775 			return (error);
776 		}
777 		if (va.va_uid != td->td_ucred->cr_uid) {
778 			if ((error = suser(td)) != 0) {
779 				vput(vp);
780 				return (error);
781 			}
782 		}
783 		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
784 		if (error != 0) {
785 			vput(vp);
786 			return (error);
787 		}
788 		if (vp->v_type != VDIR) {
789 			vput(vp);
790 			return (ENOTDIR);
791 		}
792 		vfsp = vfs_byname_kld(fstype, td, &error);
793 		if (vfsp == NULL) {
794 			vput(vp);
795 			return (error);
796 		}
797 		VI_LOCK(vp);
798 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
799 		    vp->v_mountedhere != NULL) {
800 			VI_UNLOCK(vp);
801 			vput(vp);
802 			return (EBUSY);
803 		}
804 		vp->v_iflag |= VI_MOUNT;
805 		VI_UNLOCK(vp);
806 
807 		/*
808 		 * Allocate and initialize the filesystem.
809 		 */
810 		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
811 		VOP_UNLOCK(vp, 0, td);
812 
813 		/* XXXMAC: pass to vfs_mount_alloc? */
814 		mp->mnt_optnew = fsdata;
815 	}
816 
817 	/*
818 	 * Set the mount level flags.
819 	 */
820 	if (fsflags & MNT_RDONLY)
821 		mp->mnt_flag |= MNT_RDONLY;
822 	mp->mnt_flag &=~ MNT_UPDATEMASK;
823 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS);
824 	/*
825 	 * Mount the filesystem.
826 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
827 	 * get.  No freeing of cn_pnbuf.
828 	 */
829         error = VFS_MOUNT(mp, td);
830 	if (!error) {
831 		if (mp->mnt_opt != NULL)
832 			vfs_freeopts(mp->mnt_opt);
833 		mp->mnt_opt = mp->mnt_optnew;
834 		VFS_STATFS(mp, &mp->mnt_stat, td);
835 	}
836 	/*
837 	 * Prevent external consumers of mount options from reading
838 	 * mnt_optnew.
839 	*/
840 	mp->mnt_optnew = NULL;
841 	if (mp->mnt_flag & MNT_UPDATE) {
842 		mp->mnt_flag &=
843 		    ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
844 		if (error) {
845 			mp->mnt_flag = flag;
846 			mp->mnt_kern_flag = kern_flag;
847 		}
848 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
849 			if (mp->mnt_syncer == NULL)
850 				error = vfs_allocate_syncvnode(mp);
851 		} else {
852 			if (mp->mnt_syncer != NULL)
853 				vrele(mp->mnt_syncer);
854 			mp->mnt_syncer = NULL;
855 		}
856 		vfs_unbusy(mp, td);
857 		VI_LOCK(vp);
858 		vp->v_iflag &= ~VI_MOUNT;
859 		VI_UNLOCK(vp);
860 		vrele(vp);
861 		return (error);
862 	}
863 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
864 	/*
865 	 * Put the new filesystem on the mount list after root.
866 	 */
867 	cache_purge(vp);
868 	if (!error) {
869 		struct vnode *newdp;
870 
871 		VI_LOCK(vp);
872 		vp->v_iflag &= ~VI_MOUNT;
873 		VI_UNLOCK(vp);
874 		vp->v_mountedhere = mp;
875 		mtx_lock(&mountlist_mtx);
876 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
877 		mtx_unlock(&mountlist_mtx);
878 		vfs_event_signal(NULL, VQ_MOUNT, 0);
879 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
880 			panic("mount: lost mount");
881 		mountcheckdirs(vp, newdp);
882 		vput(newdp);
883 		VOP_UNLOCK(vp, 0, td);
884 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
885 			error = vfs_allocate_syncvnode(mp);
886 		vfs_unbusy(mp, td);
887 		if (error)
888 			vrele(vp);
889 	} else {
890 		VI_LOCK(vp);
891 		vp->v_iflag &= ~VI_MOUNT;
892 		VI_UNLOCK(vp);
893 		vfs_mount_destroy(mp, td);
894 		vput(vp);
895 	}
896 	return (error);
897 }
898 
899 /*
900  * ---------------------------------------------------------------------
901  * Unmount a filesystem.
902  *
903  * Note: unmount takes a path to the vnode mounted on as argument,
904  * not special file (as before).
905  */
906 #ifndef _SYS_SYSPROTO_H_
907 struct unmount_args {
908 	char	*path;
909 	int	flags;
910 };
911 #endif
912 /* ARGSUSED */
913 int
914 unmount(td, uap)
915 	struct thread *td;
916 	register struct unmount_args /* {
917 		char *path;
918 		int flags;
919 	} */ *uap;
920 {
921 	struct mount *mp;
922 	char *pathbuf;
923 	int error, id0, id1;
924 
925 	if (jailed(td->td_ucred))
926 		return (EPERM);
927 	if (usermount == 0) {
928 		if ((error = suser(td)) != 0)
929 			return (error);
930 	}
931 
932 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
933 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
934 	if (error) {
935 		free(pathbuf, M_TEMP);
936 		return (error);
937 	}
938 	if (uap->flags & MNT_BYFSID) {
939 		/* Decode the filesystem ID. */
940 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
941 			free(pathbuf, M_TEMP);
942 			return (EINVAL);
943 		}
944 
945 		mtx_lock(&mountlist_mtx);
946 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
947 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
948 			    mp->mnt_stat.f_fsid.val[1] == id1)
949 				break;
950 		}
951 		mtx_unlock(&mountlist_mtx);
952 	} else {
953 		mtx_lock(&mountlist_mtx);
954 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
955 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
956 				break;
957 		}
958 		mtx_unlock(&mountlist_mtx);
959 	}
960 	free(pathbuf, M_TEMP);
961 	if (mp == NULL) {
962 		/*
963 		 * Previously we returned ENOENT for a nonexistent path and
964 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
965 		 * now, so in the !MNT_BYFSID case return the more likely
966 		 * EINVAL for compatibility.
967 		 */
968 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
969 	}
970 
971 	/*
972 	 * Only privileged root, or (if MNT_USER is set) the user that did the
973 	 * original mount is permitted to unmount this filesystem.
974 	 */
975 	error = vfs_suser(mp, td);
976 	if (error)
977 		return (error);
978 
979 	/*
980 	 * Don't allow unmounting the root filesystem.
981 	 */
982 	if (mp->mnt_flag & MNT_ROOTFS)
983 		return (EINVAL);
984 	mtx_lock(&Giant);
985 	error = dounmount(mp, uap->flags, td);
986 	mtx_unlock(&Giant);
987 	return (error);
988 }
989 
990 /*
991  * Do the actual filesystem unmount.
992  */
993 int
994 dounmount(mp, flags, td)
995 	struct mount *mp;
996 	int flags;
997 	struct thread *td;
998 {
999 	struct vnode *coveredvp, *fsrootvp;
1000 	int error;
1001 	int async_flag;
1002 
1003 	mtx_assert(&Giant, MA_OWNED);
1004 
1005 	MNT_ILOCK(mp);
1006 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1007 		MNT_IUNLOCK(mp);
1008 		return (EBUSY);
1009 	}
1010 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1011 	/* Allow filesystems to detect that a forced unmount is in progress. */
1012 	if (flags & MNT_FORCE)
1013 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1014 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1015 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1016 	if (error) {
1017 		MNT_ILOCK(mp);
1018 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1019 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1020 			wakeup(mp);
1021 		MNT_IUNLOCK(mp);
1022 		return (error);
1023 	}
1024 	vn_start_write(NULL, &mp, V_WAIT);
1025 
1026 	if (mp->mnt_flag & MNT_EXPUBLIC)
1027 		vfs_setpublicfs(NULL, NULL, NULL);
1028 
1029 	vfs_msync(mp, MNT_WAIT);
1030 	async_flag = mp->mnt_flag & MNT_ASYNC;
1031 	mp->mnt_flag &= ~MNT_ASYNC;
1032 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1033 	if (mp->mnt_syncer != NULL)
1034 		vrele(mp->mnt_syncer);
1035 	/*
1036 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1037 	 * vnode to the covered vnode.  For non-forced unmounts we want
1038 	 * such references to cause an EBUSY error.
1039 	 */
1040 	if ((flags & MNT_FORCE) &&
1041 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1042 		if (mp->mnt_vnodecovered != NULL)
1043 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1044 		if (fsrootvp == rootvnode) {
1045 			vrele(rootvnode);
1046 			rootvnode = NULL;
1047 		}
1048 		vput(fsrootvp);
1049 	}
1050 	if (((mp->mnt_flag & MNT_RDONLY) ||
1051 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1052 	    (flags & MNT_FORCE)) {
1053 		error = VFS_UNMOUNT(mp, flags, td);
1054 	}
1055 	vn_finished_write(mp);
1056 	if (error) {
1057 		/* Undo cdir/rdir and rootvnode changes made above. */
1058 		if ((flags & MNT_FORCE) &&
1059 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1060 			if (mp->mnt_vnodecovered != NULL)
1061 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1062 			if (rootvnode == NULL) {
1063 				rootvnode = fsrootvp;
1064 				vref(rootvnode);
1065 			}
1066 			vput(fsrootvp);
1067 		}
1068 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1069 			(void) vfs_allocate_syncvnode(mp);
1070 		MNT_ILOCK(mp);
1071 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1072 		mp->mnt_flag |= async_flag;
1073 		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1074 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1075 			wakeup(mp);
1076 		MNT_IUNLOCK(mp);
1077 		return (error);
1078 	}
1079 	mtx_lock(&mountlist_mtx);
1080 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1081 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1082 		coveredvp->v_mountedhere = NULL;
1083 	mtx_unlock(&mountlist_mtx);
1084 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1085 	vfs_mount_destroy(mp, td);
1086 	if (coveredvp != NULL)
1087 		vrele(coveredvp);
1088 	return (0);
1089 }
1090 
1091 /*
1092  * ---------------------------------------------------------------------
1093  * Mounting of root filesystem
1094  *
1095  */
1096 
1097 struct root_hold_token {
1098 	const char 			*who;
1099 	LIST_ENTRY(root_hold_token)	list;
1100 };
1101 
1102 static LIST_HEAD(, root_hold_token)	root_holds =
1103     LIST_HEAD_INITIALIZER(&root_holds);
1104 
1105 struct root_hold_token *
1106 root_mount_hold(const char *identifier)
1107 {
1108 	struct root_hold_token *h;
1109 
1110 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1111 	h->who = identifier;
1112 	mtx_lock(&mountlist_mtx);
1113 	LIST_INSERT_HEAD(&root_holds, h, list);
1114 	mtx_unlock(&mountlist_mtx);
1115 	return (h);
1116 }
1117 
1118 void
1119 root_mount_rel(struct root_hold_token *h)
1120 {
1121 
1122 	mtx_lock(&mountlist_mtx);
1123 	LIST_REMOVE(h, list);
1124 	wakeup(&root_holds);
1125 	mtx_unlock(&mountlist_mtx);
1126 	free(h, M_DEVBUF);
1127 }
1128 
1129 static void
1130 root_mount_wait(void)
1131 {
1132 	struct root_hold_token *h;
1133 
1134 	for (;;) {
1135 		DROP_GIANT();
1136 		g_waitidle();
1137 		PICKUP_GIANT();
1138 		mtx_lock(&mountlist_mtx);
1139 		if (LIST_EMPTY(&root_holds)) {
1140 			mtx_unlock(&mountlist_mtx);
1141 			break;
1142 		}
1143 		printf("Root mount waiting for:");
1144 		LIST_FOREACH(h, &root_holds, list)
1145 			printf(" %s", h->who);
1146 		printf("\n");
1147 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1148 		    hz);
1149 	}
1150 }
1151 
1152 static void
1153 set_rootvnode(struct thread *td)
1154 {
1155 	struct proc *p;
1156 
1157 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1158 		panic("Cannot find root vnode");
1159 
1160 	p = td->td_proc;
1161 	FILEDESC_LOCK(p->p_fd);
1162 
1163 	if (p->p_fd->fd_cdir != NULL)
1164 		vrele(p->p_fd->fd_cdir);
1165 	p->p_fd->fd_cdir = rootvnode;
1166 	VREF(rootvnode);
1167 
1168 	if (p->p_fd->fd_rdir != NULL)
1169 		vrele(p->p_fd->fd_rdir);
1170 	p->p_fd->fd_rdir = rootvnode;
1171 	VREF(rootvnode);
1172 
1173 	FILEDESC_UNLOCK(p->p_fd);
1174 
1175 	VOP_UNLOCK(rootvnode, 0, td);
1176 }
1177 
1178 /*
1179  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1180  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1181  */
1182 
1183 static void
1184 devfs_first(void)
1185 {
1186 	struct thread *td = curthread;
1187 	struct vfsoptlist *opts;
1188 	struct vfsconf *vfsp;
1189 	struct mount *mp = NULL;
1190 	int error;
1191 
1192 	vfsp = vfs_byname("devfs");
1193 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1194 	if (vfsp == NULL)
1195 		return;
1196 
1197 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1198 
1199 	error = VFS_MOUNT(mp, td);
1200 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1201 	if (error)
1202 		return;
1203 
1204 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1205 	TAILQ_INIT(opts);
1206 	mp->mnt_opt = opts;
1207 
1208 	mtx_lock(&mountlist_mtx);
1209 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1210 	mtx_unlock(&mountlist_mtx);
1211 
1212 	set_rootvnode(td);
1213 
1214 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1215 	if (error)
1216 		printf("kern_symlink /dev -> / returns %d\n", error);
1217 }
1218 
1219 /*
1220  * Surgically move our devfs to be mounted on /dev.
1221  */
1222 
1223 static void
1224 devfs_fixup(struct thread *td)
1225 {
1226 	struct nameidata nd;
1227 	int error;
1228 	struct vnode *vp, *dvp;
1229 	struct mount *mp;
1230 
1231 	/* Remove our devfs mount from the mountlist and purge the cache */
1232 	mtx_lock(&mountlist_mtx);
1233 	mp = TAILQ_FIRST(&mountlist);
1234 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1235 	mtx_unlock(&mountlist_mtx);
1236 	cache_purgevfs(mp);
1237 
1238 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1239 	VI_LOCK(dvp);
1240 	dvp->v_iflag &= ~VI_MOUNT;
1241 	dvp->v_mountedhere = NULL;
1242 	VI_UNLOCK(dvp);
1243 
1244 	/* Set up the real rootvnode, and purge the cache */
1245 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1246 	set_rootvnode(td);
1247 	cache_purgevfs(rootvnode->v_mount);
1248 
1249 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1250 	error = namei(&nd);
1251 	if (error) {
1252 		printf("Lookup of /dev for devfs, error: %d\n", error);
1253 		return;
1254 	}
1255 	NDFREE(&nd, NDF_ONLY_PNBUF);
1256 	vp = nd.ni_vp;
1257 	if (vp->v_type != VDIR) {
1258 		vput(vp);
1259 	}
1260 	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1261 	if (error) {
1262 		vput(vp);
1263 	}
1264 	cache_purge(vp);
1265 	mp->mnt_vnodecovered = vp;
1266 	vp->v_mountedhere = mp;
1267 	mtx_lock(&mountlist_mtx);
1268 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1269 	mtx_unlock(&mountlist_mtx);
1270 	VOP_UNLOCK(vp, 0, td);
1271 	vfs_unbusy(mp, td);
1272 	vput(dvp);
1273 
1274 	/* Unlink the no longer needed /dev/dev -> / symlink */
1275 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1276 }
1277 
1278 /*
1279  * Report errors during filesystem mounting.
1280  */
1281 void
1282 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1283 {
1284 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1285 	va_list ap;
1286 	int error, len;
1287 	char *errmsg;
1288 
1289 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1290 	if (error || errmsg == NULL || len <= 0)
1291 		return;
1292 
1293 	va_start(ap, fmt);
1294 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1295 	va_end(ap);
1296 }
1297 
1298 /*
1299  * Find and mount the root filesystem
1300  */
1301 void
1302 vfs_mountroot(void)
1303 {
1304 	char *cp;
1305 	int error, i, asked = 0;
1306 
1307 	root_mount_wait();
1308 
1309 	devfs_first();
1310 
1311 	/*
1312 	 * We are booted with instructions to prompt for the root filesystem.
1313 	 */
1314 	if (boothowto & RB_ASKNAME) {
1315 		if (!vfs_mountroot_ask())
1316 			return;
1317 		asked = 1;
1318 	}
1319 
1320 	/*
1321 	 * The root filesystem information is compiled in, and we are
1322 	 * booted with instructions to use it.
1323 	 */
1324 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1325 		if (!vfs_mountroot_try(ctrootdevname))
1326 			return;
1327 		ctrootdevname = NULL;
1328 	}
1329 
1330 	/*
1331 	 * We've been given the generic "use CDROM as root" flag.  This is
1332 	 * necessary because one media may be used in many different
1333 	 * devices, so we need to search for them.
1334 	 */
1335 	if (boothowto & RB_CDROM) {
1336 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1337 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1338 				return;
1339 		}
1340 	}
1341 
1342 	/*
1343 	 * Try to use the value read by the loader from /etc/fstab, or
1344 	 * supplied via some other means.  This is the preferred
1345 	 * mechanism.
1346 	 */
1347 	cp = getenv("vfs.root.mountfrom");
1348 	if (cp != NULL) {
1349 		error = vfs_mountroot_try(cp);
1350 		freeenv(cp);
1351 		if (!error)
1352 			return;
1353 	}
1354 
1355 	/*
1356 	 * Try values that may have been computed by code during boot
1357 	 */
1358 	if (!vfs_mountroot_try(rootdevnames[0]))
1359 		return;
1360 	if (!vfs_mountroot_try(rootdevnames[1]))
1361 		return;
1362 
1363 	/*
1364 	 * If we (still) have a compiled-in default, try it.
1365 	 */
1366 	if (ctrootdevname != NULL)
1367 		if (!vfs_mountroot_try(ctrootdevname))
1368 			return;
1369 	/*
1370 	 * Everything so far has failed, prompt on the console if we haven't
1371 	 * already tried that.
1372 	 */
1373 	if (!asked)
1374 		if (!vfs_mountroot_ask())
1375 			return;
1376 
1377 	panic("Root mount failed, startup aborted.");
1378 }
1379 
1380 /*
1381  * Mount (mountfrom) as the root filesystem.
1382  */
1383 static int
1384 vfs_mountroot_try(const char *mountfrom)
1385 {
1386 	struct mount	*mp;
1387 	char		*vfsname, *path;
1388 	time_t		timebase;
1389 	int		error;
1390 	char		patt[32];
1391 
1392 	vfsname = NULL;
1393 	path    = NULL;
1394 	mp      = NULL;
1395 	error   = EINVAL;
1396 
1397 	if (mountfrom == NULL)
1398 		return (error);		/* don't complain */
1399 	printf("Trying to mount root from %s\n", mountfrom);
1400 
1401 	/* parse vfs name and path */
1402 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1403 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1404 	vfsname[0] = path[0] = 0;
1405 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1406 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1407 		goto out;
1408 
1409 	if (path[0] == '\0')
1410 		strcpy(path, ROOTNAME);
1411 
1412 	error = kernel_vmount(
1413 	    MNT_RDONLY | MNT_ROOTFS,
1414 	    "fstype", vfsname,
1415 	    "fspath", "/",
1416 	    "from", path,
1417 	    NULL);
1418 	if (error == 0) {
1419 		/*
1420 		 * We mount devfs prior to mounting the / FS, so the first
1421 		 * entry will typically be devfs.
1422 		 */
1423 		mp = TAILQ_FIRST(&mountlist);
1424 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1425 
1426 		/*
1427 		 * Iterate over all currently mounted file systems and use
1428 		 * the time stamp found to check and/or initialize the RTC.
1429 		 * Typically devfs has no time stamp and the only other FS
1430 		 * is the actual / FS.
1431 		 * Call inittodr() only once and pass it the largest of the
1432 		 * timestamps we encounter.
1433 		 */
1434 		timebase = 0;
1435 		do {
1436 			if (mp->mnt_time > timebase)
1437 				timebase = mp->mnt_time;
1438 			mp = TAILQ_NEXT(mp, mnt_list);
1439 		} while (mp != NULL);
1440 		inittodr(timebase);
1441 
1442 		devfs_fixup(curthread);
1443 	}
1444 out:
1445 	free(path, M_MOUNT);
1446 	free(vfsname, M_MOUNT);
1447 	return (error);
1448 }
1449 
1450 /*
1451  * ---------------------------------------------------------------------
1452  * Interactive root filesystem selection code.
1453  */
1454 
1455 static int
1456 vfs_mountroot_ask(void)
1457 {
1458 	char name[128];
1459 
1460 	for(;;) {
1461 		printf("\nManual root filesystem specification:\n");
1462 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1463 #if defined(__i386__) || defined(__ia64__)
1464 		printf("                       eg. ufs:da0s1a\n");
1465 #else
1466 		printf("                       eg. ufs:/dev/da0a\n");
1467 #endif
1468 		printf("  ?                  List valid disk boot devices\n");
1469 		printf("  <empty line>       Abort manual input\n");
1470 		printf("\nmountroot> ");
1471 		gets(name, sizeof(name), 1);
1472 		if (name[0] == '\0')
1473 			return (1);
1474 		if (name[0] == '?') {
1475 			printf("\nList of GEOM managed disk devices:\n  ");
1476 			g_dev_print();
1477 			continue;
1478 		}
1479 		if (!vfs_mountroot_try(name))
1480 			return (0);
1481 	}
1482 }
1483 
1484 /*
1485  * ---------------------------------------------------------------------
1486  * Functions for querying mount options/arguments from filesystems.
1487  */
1488 
1489 /*
1490  * Check that no unknown options are given
1491  */
1492 int
1493 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1494 {
1495 	struct vfsopt *opt;
1496 	const char **t, *p;
1497 
1498 
1499 	TAILQ_FOREACH(opt, opts, link) {
1500 		p = opt->name;
1501 		if (p[0] == 'n' && p[1] == 'o')
1502 			p += 2;
1503 		for(t = global_opts; *t != NULL; t++)
1504 			if (!strcmp(*t, p))
1505 				break;
1506 		if (*t != NULL)
1507 			continue;
1508 		for(t = legal; *t != NULL; t++)
1509 			if (!strcmp(*t, p))
1510 				break;
1511 		if (*t != NULL)
1512 			continue;
1513 		printf("mount option <%s> is unknown\n", p);
1514 		return (EINVAL);
1515 	}
1516 	return (0);
1517 }
1518 
1519 /*
1520  * Get a mount option by its name.
1521  *
1522  * Return 0 if the option was found, ENOENT otherwise.
1523  * If len is non-NULL it will be filled with the length
1524  * of the option. If buf is non-NULL, it will be filled
1525  * with the address of the option.
1526  */
1527 int
1528 vfs_getopt(opts, name, buf, len)
1529 	struct vfsoptlist *opts;
1530 	const char *name;
1531 	void **buf;
1532 	int *len;
1533 {
1534 	struct vfsopt *opt;
1535 
1536 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1537 
1538 	TAILQ_FOREACH(opt, opts, link) {
1539 		if (strcmp(name, opt->name) == 0) {
1540 			if (len != NULL)
1541 				*len = opt->len;
1542 			if (buf != NULL)
1543 				*buf = opt->value;
1544 			return (0);
1545 		}
1546 	}
1547 	return (ENOENT);
1548 }
1549 
1550 static int
1551 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1552 {
1553 	struct vfsopt *opt;
1554 	int i;
1555 
1556 	if (opts == NULL)
1557 		return (-1);
1558 
1559 	i = 0;
1560 	TAILQ_FOREACH(opt, opts, link) {
1561 		if (strcmp(name, opt->name) == 0)
1562 			return (i);
1563 		++i;
1564 	}
1565 	return (-1);
1566 }
1567 
1568 char *
1569 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1570 {
1571 	struct vfsopt *opt;
1572 
1573 	*error = 0;
1574 	TAILQ_FOREACH(opt, opts, link) {
1575 		if (strcmp(name, opt->name) != 0)
1576 			continue;
1577 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1578 			*error = EINVAL;
1579 			return (NULL);
1580 		}
1581 		return (opt->value);
1582 	}
1583 	return (NULL);
1584 }
1585 
1586 int
1587 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1588 {
1589 	struct vfsopt *opt;
1590 
1591 	TAILQ_FOREACH(opt, opts, link) {
1592 		if (strcmp(name, opt->name) == 0) {
1593 			if (w != NULL)
1594 				*w |= val;
1595 			return (1);
1596 		}
1597 	}
1598 	if (w != NULL)
1599 		*w &= ~val;
1600 	return (0);
1601 }
1602 
1603 int
1604 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1605 {
1606 	va_list ap;
1607 	struct vfsopt *opt;
1608 	int ret;
1609 
1610 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1611 
1612 	TAILQ_FOREACH(opt, opts, link) {
1613 		if (strcmp(name, opt->name) != 0)
1614 			continue;
1615 		if (((char *)opt->value)[opt->len - 1] != '\0')
1616 			return (0);
1617 		va_start(ap, fmt);
1618 		ret = vsscanf(opt->value, fmt, ap);
1619 		va_end(ap);
1620 		return (ret);
1621 	}
1622 	return (0);
1623 }
1624 
1625 /*
1626  * Find and copy a mount option.
1627  *
1628  * The size of the buffer has to be specified
1629  * in len, if it is not the same length as the
1630  * mount option, EINVAL is returned.
1631  * Returns ENOENT if the option is not found.
1632  */
1633 int
1634 vfs_copyopt(opts, name, dest, len)
1635 	struct vfsoptlist *opts;
1636 	const char *name;
1637 	void *dest;
1638 	int len;
1639 {
1640 	struct vfsopt *opt;
1641 
1642 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1643 
1644 	TAILQ_FOREACH(opt, opts, link) {
1645 		if (strcmp(name, opt->name) == 0) {
1646 			if (len != opt->len)
1647 				return (EINVAL);
1648 			bcopy(opt->value, dest, opt->len);
1649 			return (0);
1650 		}
1651 	}
1652 	return (ENOENT);
1653 }
1654 
1655 /*
1656  * This is a helper function for filesystems to traverse their
1657  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1658  */
1659 
1660 struct vnode *
1661 __mnt_vnode_next(struct vnode **nvp, struct mount *mp)
1662 {
1663 	struct vnode *vp;
1664 
1665 	mtx_assert(&mp->mnt_mtx, MA_OWNED);
1666 
1667 	vp = *nvp;
1668 	/* Check if we are done */
1669 	if (vp == NULL)
1670 		return (NULL);
1671 	/* If our next vnode is no longer ours, start over */
1672 	if (vp->v_mount != mp)
1673 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1674 	/* Save pointer to next vnode in list */
1675 	if (vp != NULL)
1676 		*nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1677 	else
1678 		*nvp = NULL;
1679 	return (vp);
1680 }
1681 
1682 int
1683 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
1684 {
1685 	int error;
1686 
1687 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
1688 	if (sbp != &mp->mnt_stat)
1689 		*sbp = mp->mnt_stat;
1690 	return (error);
1691 }
1692 
1693 void
1694 vfs_mountedfrom(struct mount *mp, const char *from)
1695 {
1696 
1697 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1698 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1699 	    sizeof mp->mnt_stat.f_mntfromname);
1700 }
1701 
1702 /*
1703  * ---------------------------------------------------------------------
1704  * This is the api for building mount args and mounting filesystems from
1705  * inside the kernel.
1706  *
1707  * The API works by accumulation of individual args.  First error is
1708  * latched.
1709  *
1710  * XXX: should be documented in new manpage kernel_mount(9)
1711  */
1712 
1713 /* A memory allocation which must be freed when we are done */
1714 struct mntaarg {
1715 	SLIST_ENTRY(mntaarg)	next;
1716 };
1717 
1718 /* The header for the mount arguments */
1719 struct mntarg {
1720 	struct iovec *v;
1721 	int len;
1722 	int error;
1723 	SLIST_HEAD(, mntaarg)	list;
1724 };
1725 
1726 /*
1727  * Add a boolean argument.
1728  *
1729  * flag is the boolean value.
1730  * name must start with "no".
1731  */
1732 struct mntarg *
1733 mount_argb(struct mntarg *ma, int flag, const char *name)
1734 {
1735 
1736 	KASSERT(name[0] == 'n' && name[1] == 'o',
1737 	    ("mount_argb(...,%s): name must start with 'no'", name));
1738 
1739 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1740 }
1741 
1742 /*
1743  * Add an argument printf style
1744  */
1745 struct mntarg *
1746 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1747 {
1748 	va_list ap;
1749 	struct mntaarg *maa;
1750 	struct sbuf *sb;
1751 	int len;
1752 
1753 	if (ma == NULL) {
1754 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1755 		SLIST_INIT(&ma->list);
1756 	}
1757 	if (ma->error)
1758 		return (ma);
1759 
1760 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1761 	    M_MOUNT, M_WAITOK);
1762 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1763 	ma->v[ma->len].iov_len = strlen(name) + 1;
1764 	ma->len++;
1765 
1766 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
1767 	va_start(ap, fmt);
1768 	sbuf_vprintf(sb, fmt, ap);
1769 	va_end(ap);
1770 	sbuf_finish(sb);
1771 	len = sbuf_len(sb) + 1;
1772 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1773 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1774 	bcopy(sbuf_data(sb), maa + 1, len);
1775 	sbuf_delete(sb);
1776 
1777 	ma->v[ma->len].iov_base = maa + 1;
1778 	ma->v[ma->len].iov_len = len;
1779 	ma->len++;
1780 
1781 	return (ma);
1782 }
1783 
1784 /*
1785  * Add an argument which is a userland string.
1786  */
1787 struct mntarg *
1788 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1789 {
1790 	struct mntaarg *maa;
1791 	char *tbuf;
1792 
1793 	if (val == NULL)
1794 		return (ma);
1795 	if (ma == NULL) {
1796 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1797 		SLIST_INIT(&ma->list);
1798 	}
1799 	if (ma->error)
1800 		return (ma);
1801 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1802 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1803 	tbuf = (void *)(maa + 1);
1804 	ma->error = copyinstr(val, tbuf, len, NULL);
1805 	return (mount_arg(ma, name, tbuf, -1));
1806 }
1807 
1808 /*
1809  * Plain argument.
1810  *
1811  * If length is -1, use printf.
1812  */
1813 struct mntarg *
1814 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
1815 {
1816 
1817 	if (ma == NULL) {
1818 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1819 		SLIST_INIT(&ma->list);
1820 	}
1821 	if (ma->error)
1822 		return (ma);
1823 
1824 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1825 	    M_MOUNT, M_WAITOK);
1826 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1827 	ma->v[ma->len].iov_len = strlen(name) + 1;
1828 	ma->len++;
1829 
1830 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
1831 	if (len < 0)
1832 		ma->v[ma->len].iov_len = strlen(val) + 1;
1833 	else
1834 		ma->v[ma->len].iov_len = len;
1835 	ma->len++;
1836 	return (ma);
1837 }
1838 
1839 /*
1840  * Free a mntarg structure
1841  */
1842 static void
1843 free_mntarg(struct mntarg *ma)
1844 {
1845 	struct mntaarg *maa;
1846 
1847 	while (!SLIST_EMPTY(&ma->list)) {
1848 		maa = SLIST_FIRST(&ma->list);
1849 		SLIST_REMOVE_HEAD(&ma->list, next);
1850 		free(maa, M_MOUNT);
1851 	}
1852 	free(ma->v, M_MOUNT);
1853 	free(ma, M_MOUNT);
1854 }
1855 
1856 /*
1857  * Mount a filesystem
1858  */
1859 int
1860 kernel_mount(struct mntarg *ma, int flags)
1861 {
1862 	struct uio auio;
1863 	int error;
1864 
1865 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
1866 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
1867 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
1868 
1869 	auio.uio_iov = ma->v;
1870 	auio.uio_iovcnt = ma->len;
1871 	auio.uio_segflg = UIO_SYSSPACE;
1872 
1873 	error = ma->error;
1874 	if (!error)
1875 		error = vfs_donmount(curthread, flags, &auio);
1876 	free_mntarg(ma);
1877 	return (error);
1878 }
1879 
1880 /*
1881  * A printflike function to mount a filesystem.
1882  */
1883 int
1884 kernel_vmount(int flags, ...)
1885 {
1886 	struct mntarg *ma = NULL;
1887 	va_list ap;
1888 	const char *cp;
1889 	const void *vp;
1890 	int error;
1891 
1892 	va_start(ap, flags);
1893 	for (;;) {
1894 		cp = va_arg(ap, const char *);
1895 		if (cp == NULL)
1896 			break;
1897 		vp = va_arg(ap, const void *);
1898 		ma = mount_arg(ma, cp, vp, -1);
1899 	}
1900 	va_end(ap);
1901 
1902 	error = kernel_mount(ma, flags);
1903 	return (error);
1904 }
1905