xref: /freebsd/sys/kern/vfs_mount.c (revision ae83180158c4c937f170e31eff311b18c0286a93)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  * Copyright (c) 1999 Michael Smith
39  * All rights reserved.
40  * Copyright (c) 1999 Poul-Henning Kamp
41  * All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  *
52  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
53  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
56  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62  * SUCH DAMAGE.
63  *
64  * $FreeBSD$
65  */
66 
67 #include <sys/param.h>
68 #include <sys/conf.h>
69 #include <sys/cons.h>
70 #include <sys/kernel.h>
71 #include <sys/linker.h>
72 #include <sys/mac.h>
73 #include <sys/malloc.h>
74 #include <sys/mount.h>
75 #include <sys/mutex.h>
76 #include <sys/namei.h>
77 #include <sys/proc.h>
78 #include <sys/reboot.h>
79 #include <sys/sysproto.h>
80 #include <sys/sx.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysent.h>
83 #include <sys/systm.h>
84 #include <sys/vnode.h>
85 
86 #include <machine/stdarg.h>
87 
88 #include "opt_rootdevname.h"
89 #include "opt_ddb.h"
90 #include "opt_mac.h"
91 
92 #ifdef DDB
93 #include <ddb/ddb.h>
94 #endif
95 
96 #define ROOTNAME	"root_device"
97 
98 static void	checkdirs(struct vnode *olddp, struct vnode *newdp);
99 static int	vfs_nmount(struct thread *td, int, struct uio *);
100 static int	vfs_mountroot_try(char *mountfrom);
101 static int	vfs_mountroot_ask(void);
102 static void	gets(char *cp);
103 
104 static int	usermount = 0;	/* if 1, non-root can mount fs. */
105 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
106 
107 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
108 
109 /* List of mounted filesystems. */
110 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
111 
112 /* For any iteration/modification of mountlist */
113 struct mtx mountlist_mtx;
114 
115 /* For any iteration/modification of mnt_vnodelist */
116 struct mtx mntvnode_mtx;
117 
118 /*
119  * The vnode of the system's root (/ in the filesystem, without chroot
120  * active.)
121  */
122 struct vnode	*rootvnode;
123 
124 /*
125  * The root filesystem is detailed in the kernel environment variable
126  * vfs.root.mountfrom, which is expected to be in the general format
127  *
128  * <vfsname>:[<path>]
129  * vfsname   := the name of a VFS known to the kernel and capable
130  *              of being mounted as root
131  * path      := disk device name or other data used by the filesystem
132  *              to locate its physical store
133  */
134 
135 /*
136  * The root specifiers we will try if RB_CDROM is specified.
137  */
138 static char *cdrom_rootdevnames[] = {
139 	"cd9660:cd0a",
140 	"cd9660:acd0a",
141 	"cd9660:wcd0a",
142 	NULL
143 };
144 
145 /* legacy find-root code */
146 char		*rootdevnames[2] = {NULL, NULL};
147 static int	setrootbyname(char *name);
148 dev_t		rootdev = NODEV;
149 
150 /* Remove one mount option. */
151 static void
152 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
153 {
154 
155 	TAILQ_REMOVE(opts, opt, link);
156 	free(opt->name, M_MOUNT);
157 	if (opt->value != NULL)
158 		free(opt->value, M_MOUNT);
159 #ifdef INVARIANTS
160 	else if (opt->len != 0)
161 		panic("%s: mount option with NULL value but length != 0",
162 		    __func__);
163 #endif
164 	free(opt, M_MOUNT);
165 }
166 
167 /* Release all resources related to the mount options. */
168 static void
169 vfs_freeopts(struct vfsoptlist *opts)
170 {
171 	struct vfsopt *opt;
172 
173 	while (!TAILQ_EMPTY(opts)) {
174 		opt = TAILQ_FIRST(opts);
175 		vfs_freeopt(opts, opt);
176 	}
177 	free(opts, M_MOUNT);
178 }
179 
180 /*
181  * If a mount option is specified several times,
182  * (with or without the "no" prefix) only keep
183  * the last occurence of it.
184  */
185 static void
186 vfs_sanitizeopts(struct vfsoptlist *opts)
187 {
188 	struct vfsopt *opt, *opt2, *tmp;
189 	int noopt;
190 
191 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
192 		if (strncmp(opt->name, "no", 2) == 0)
193 			noopt = 1;
194 		else
195 			noopt = 0;
196 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
197 		while (opt2 != NULL) {
198 			if (strcmp(opt2->name, opt->name) == 0 ||
199 			    (noopt && strcmp(opt->name + 2, opt2->name) == 0) ||
200 			    (!noopt && strncmp(opt2->name, "no", 2) == 0 &&
201 			    strcmp(opt2->name + 2, opt->name) == 0)) {
202 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
203 				vfs_freeopt(opts, opt2);
204 				opt2 = tmp;
205 			} else {
206 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
207 			}
208 		}
209 	}
210 }
211 
212 /*
213  * Build a linked list of mount options from a struct uio.
214  */
215 static int
216 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
217 {
218 	struct vfsoptlist *opts;
219 	struct vfsopt *opt;
220 	unsigned int i, iovcnt;
221 	int error, namelen, optlen;
222 
223 	iovcnt = auio->uio_iovcnt;
224 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
225 	TAILQ_INIT(opts);
226 	for (i = 0; i < iovcnt; i += 2) {
227 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
228 		namelen = auio->uio_iov[i].iov_len;
229 		optlen = auio->uio_iov[i + 1].iov_len;
230 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
231 		opt->value = NULL;
232 		if (auio->uio_segflg == UIO_SYSSPACE) {
233 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
234 		} else {
235 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
236 			    namelen);
237 			if (error)
238 				goto bad;
239 		}
240 		opt->len = optlen;
241 		if (optlen != 0) {
242 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
243 			if (auio->uio_segflg == UIO_SYSSPACE) {
244 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
245 				    optlen);
246 			} else {
247 				error = copyin(auio->uio_iov[i + 1].iov_base,
248 				    opt->value, optlen);
249 				if (error)
250 					goto bad;
251 			}
252 		}
253 		TAILQ_INSERT_TAIL(opts, opt, link);
254 	}
255 	vfs_sanitizeopts(opts);
256 	*options = opts;
257 	return (0);
258 bad:
259 	vfs_freeopts(opts);
260 	return (error);
261 }
262 
263 /*
264  * Merge the old mount options with the new ones passed
265  * in the MNT_UPDATE case.
266  */
267 static void
268 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
269 {
270 	struct vfsopt *opt, *opt2, *new;
271 
272 	TAILQ_FOREACH(opt, opts, link) {
273 		/*
274 		 * Check that this option hasn't been redefined
275 		 * nor cancelled with a "no" mount option.
276 		 */
277 		opt2 = TAILQ_FIRST(toopts);
278 		while (opt2 != NULL) {
279 			if (strcmp(opt2->name, opt->name) == 0)
280 				goto next;
281 			if (strncmp(opt2->name, "no", 2) == 0 &&
282 			    strcmp(opt2->name + 2, opt->name) == 0) {
283 				vfs_freeopt(toopts, opt2);
284 				goto next;
285 			}
286 			opt2 = TAILQ_NEXT(opt2, link);
287 		}
288 		/* We want this option, duplicate it. */
289 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
290 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
291 		strcpy(new->name, opt->name);
292 		if (opt->len != 0) {
293 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
294 			bcopy(opt->value, new->value, opt->len);
295 		} else {
296 			new->value = NULL;
297 		}
298 		new->len = opt->len;
299 		TAILQ_INSERT_TAIL(toopts, new, link);
300 next:
301 		continue;
302 	}
303 }
304 
305 /*
306  * New mount API.
307  */
308 int
309 nmount(td, uap)
310 	struct thread *td;
311 	struct nmount_args /* {
312 		syscallarg(struct iovec *) iovp;
313 		syscallarg(unsigned int) iovcnt;
314 		syscallarg(int) flags;
315 	} */ *uap;
316 {
317 	struct uio auio;
318 	struct iovec *iov, *needfree;
319 	struct iovec aiov[UIO_SMALLIOV];
320 	unsigned int i;
321 	int error;
322 	u_int iovlen, iovcnt;
323 
324 	iovcnt = SCARG(uap, iovcnt);
325 	iovlen = iovcnt * sizeof (struct iovec);
326 	/*
327 	 * Check that we have an even number of iovec's
328 	 * and that we have at least two options.
329 	 */
330 	if ((iovcnt & 1) || (iovcnt < 4) || (iovcnt > UIO_MAXIOV))
331 		return (EINVAL);
332 
333 	if (iovcnt > UIO_SMALLIOV) {
334 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
335 		needfree = iov;
336 	} else {
337 		iov = aiov;
338 		needfree = NULL;
339 	}
340 	auio.uio_iov = iov;
341 	auio.uio_iovcnt = iovcnt;
342 	auio.uio_segflg = UIO_USERSPACE;
343 	if ((error = copyin(uap->iovp, iov, iovlen)))
344 		goto finish;
345 
346 	for (i = 0; i < iovcnt; i++) {
347 		if (iov->iov_len > MMAXOPTIONLEN) {
348 			error = EINVAL;
349 			goto finish;
350 		}
351 		iov++;
352 	}
353 	error = vfs_nmount(td, SCARG(uap, flags), &auio);
354 finish:
355 	if (needfree != NULL)
356 		free(needfree, M_TEMP);
357 	return (error);
358 }
359 
360 int
361 kernel_mount(iovp, iovcnt, flags)
362 	struct iovec *iovp;
363 	unsigned int iovcnt;
364 	int flags;
365 {
366 	struct uio auio;
367 	int error;
368 
369 	/*
370 	 * Check that we have an even number of iovec's
371 	 * and that we have at least two options.
372 	 */
373 	if ((iovcnt & 1) || (iovcnt < 4))
374 		return (EINVAL);
375 
376 	auio.uio_iov = iovp;
377 	auio.uio_iovcnt = iovcnt;
378 	auio.uio_segflg = UIO_SYSSPACE;
379 
380 	error = vfs_nmount(curthread, flags, &auio);
381 	return (error);
382 }
383 
384 int
385 kernel_vmount(int flags, ...)
386 {
387 	struct iovec *iovp;
388 	struct uio auio;
389 	va_list ap;
390 	unsigned int iovcnt, iovlen, len;
391 	const char *cp;
392 	char *buf, *pos;
393 	size_t n;
394 	int error, i;
395 
396 	len = 0;
397 	va_start(ap, flags);
398 	for (iovcnt = 0; (cp = va_arg(ap, const char *)) != NULL; iovcnt++)
399 		len += strlen(cp) + 1;
400 	va_end(ap);
401 
402 	if (iovcnt < 4 || iovcnt & 1)
403 		return (EINVAL);
404 
405 	iovlen = iovcnt * sizeof (struct iovec);
406 	MALLOC(iovp, struct iovec *, iovlen, M_MOUNT, M_WAITOK);
407 	MALLOC(buf, char *, len, M_MOUNT, M_WAITOK);
408 	pos = buf;
409 	va_start(ap, flags);
410 	for (i = 0; i < iovcnt; i++) {
411 		cp = va_arg(ap, const char *);
412 		copystr(cp, pos, len - (pos - buf), &n);
413 		iovp[i].iov_base = pos;
414 		iovp[i].iov_len = n;
415 		pos += n;
416 	}
417 	va_end(ap);
418 
419 	auio.uio_iov = iovp;
420 	auio.uio_iovcnt = iovcnt;
421 	auio.uio_segflg = UIO_SYSSPACE;
422 
423 	error = vfs_nmount(curthread, flags, &auio);
424 	FREE(iovp, M_MOUNT);
425 	FREE(buf, M_MOUNT);
426 	return (error);
427 }
428 
429 /*
430  * vfs_nmount(): actually attempt a filesystem mount.
431  */
432 static int
433 vfs_nmount(td, fsflags, fsoptions)
434 	struct thread *td;
435 	int fsflags;		/* Flags common to all filesystems. */
436 	struct uio *fsoptions;	/* Options local to the filesystem. */
437 {
438 	linker_file_t lf;
439 	struct vnode *vp;
440 	struct mount *mp;
441 	struct vfsconf *vfsp;
442 	struct vfsoptlist *optlist;
443 	char *fstype, *fspath;
444 	int error, flag = 0, kern_flag = 0;
445 	int fstypelen, fspathlen;
446 	struct vattr va;
447 	struct nameidata nd;
448 
449 	error = vfs_buildopts(fsoptions, &optlist);
450 	if (error)
451 		return (error);
452 
453 	/*
454 	 * We need these two options before the others,
455 	 * and they are mandatory for any filesystem.
456 	 * Ensure they are NUL terminated as well.
457 	 */
458 	fstypelen = 0;
459 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
460 	if (error || fstype[fstypelen - 1] != '\0') {
461 		error = EINVAL;
462 		goto bad;
463 	}
464 	fspathlen = 0;
465 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
466 	if (error || fspath[fspathlen - 1] != '\0') {
467 		error = EINVAL;
468 		goto bad;
469 	}
470 
471 	/*
472 	 * Be ultra-paranoid about making sure the type and fspath
473 	 * variables will fit in our mp buffers, including the
474 	 * terminating NUL.
475 	 */
476 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
477 		error = ENAMETOOLONG;
478 		goto bad;
479 	}
480 
481 	if (usermount == 0) {
482 	       	error = suser(td);
483 		if (error)
484 			goto bad;
485 	}
486 	/*
487 	 * Do not allow NFS export by non-root users.
488 	 */
489 	if (fsflags & MNT_EXPORTED) {
490 		error = suser(td);
491 		if (error)
492 			goto bad;
493 	}
494 	/*
495 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users.
496 	 */
497 	if (suser(td))
498 		fsflags |= MNT_NOSUID | MNT_NODEV;
499 	/*
500 	 * Get vnode to be covered
501 	 */
502 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
503 	if ((error = namei(&nd)) != 0)
504 		goto bad;
505 	NDFREE(&nd, NDF_ONLY_PNBUF);
506 	vp = nd.ni_vp;
507 	if (fsflags & MNT_UPDATE) {
508 		if ((vp->v_vflag & VV_ROOT) == 0) {
509 			vput(vp);
510 			error = EINVAL;
511 			goto bad;
512 		}
513 		mp = vp->v_mount;
514 		flag = mp->mnt_flag;
515 		kern_flag = mp->mnt_kern_flag;
516 		/*
517 		 * We only allow the filesystem to be reloaded if it
518 		 * is currently mounted read-only.
519 		 */
520 		if ((fsflags & MNT_RELOAD) &&
521 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
522 			vput(vp);
523 			error = EOPNOTSUPP;	/* Needs translation */
524 			goto bad;
525 		}
526 		/*
527 		 * Only root, or the user that did the original mount is
528 		 * permitted to update it.
529 		 */
530 		if (mp->mnt_stat.f_owner != td->td_ucred->cr_uid) {
531 			error = suser(td);
532 			if (error) {
533 				vput(vp);
534 				goto bad;
535 			}
536 		}
537 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
538 			vput(vp);
539 			error = EBUSY;
540 			goto bad;
541 		}
542 		VI_LOCK(vp);
543 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
544 		    vp->v_mountedhere != NULL) {
545 			VI_UNLOCK(vp);
546 			vfs_unbusy(mp, td);
547 			vput(vp);
548 			error = EBUSY;
549 			goto bad;
550 		}
551 		vp->v_iflag |= VI_MOUNT;
552 		VI_UNLOCK(vp);
553 		mp->mnt_flag |= fsflags &
554 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT);
555 		VOP_UNLOCK(vp, 0, td);
556 		mp->mnt_optnew = optlist;
557 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
558 		goto update;
559 	}
560 	/*
561 	 * If the user is not root, ensure that they own the directory
562 	 * onto which we are attempting to mount.
563 	 */
564 	error = VOP_GETATTR(vp, &va, td->td_ucred, td);
565 	if (error) {
566 		vput(vp);
567 		goto bad;
568 	}
569 	if (va.va_uid != td->td_ucred->cr_uid) {
570 		error = suser(td);
571 		if (error) {
572 			vput(vp);
573 			goto bad;
574 		}
575 	}
576 	if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) {
577 		vput(vp);
578 		goto bad;
579 	}
580 	if (vp->v_type != VDIR) {
581 		vput(vp);
582 		error = ENOTDIR;
583 		goto bad;
584 	}
585 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
586 		if (!strcmp(vfsp->vfc_name, fstype))
587 			break;
588 	if (vfsp == NULL) {
589 		/* Only load modules for root (very important!). */
590 		error = suser(td);
591 		if (error) {
592 			vput(vp);
593 			goto bad;
594 		}
595 		error = securelevel_gt(td->td_ucred, 0);
596 		if (error) {
597 			vput(vp);
598 			goto bad;
599 		}
600 		error = linker_load_module(NULL, fstype, NULL, NULL, &lf);
601 		if (error || lf == NULL) {
602 			vput(vp);
603 			if (lf == NULL)
604 				error = ENODEV;
605 			goto bad;
606 		}
607 		lf->userrefs++;
608 		/* Look up again to see if the VFS was loaded. */
609 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
610 			if (!strcmp(vfsp->vfc_name, fstype))
611 				break;
612 		if (vfsp == NULL) {
613 			lf->userrefs--;
614 			linker_file_unload(lf);
615 			vput(vp);
616 			error = ENODEV;
617 			goto bad;
618 		}
619 	}
620 	VI_LOCK(vp);
621 	if ((vp->v_iflag & VI_MOUNT) != 0 ||
622 	    vp->v_mountedhere != NULL) {
623 		VI_UNLOCK(vp);
624 		vput(vp);
625 		error = EBUSY;
626 		goto bad;
627 	}
628 	vp->v_iflag |= VI_MOUNT;
629 	VI_UNLOCK(vp);
630 
631 	/*
632 	 * Allocate and initialize the filesystem.
633 	 */
634 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
635 	TAILQ_INIT(&mp->mnt_nvnodelist);
636 	TAILQ_INIT(&mp->mnt_reservedvnlist);
637 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
638 	(void)vfs_busy(mp, LK_NOWAIT, 0, td);
639 	mp->mnt_op = vfsp->vfc_vfsops;
640 	mp->mnt_vfc = vfsp;
641 	vfsp->vfc_refcount++;
642 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
643 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
644 	strncpy(mp->mnt_stat.f_fstypename, fstype, MFSNAMELEN);
645 	mp->mnt_vnodecovered = vp;
646 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
647 	strncpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
648 	mp->mnt_iosize_max = DFLTPHYS;
649 #ifdef MAC
650 	mac_init_mount(mp);
651 	mac_create_mount(td->td_ucred, mp);
652 #endif
653 	VOP_UNLOCK(vp, 0, td);
654 	mp->mnt_optnew = optlist;	/* XXXMAC: should this be above? */
655 
656 update:
657 	/*
658 	 * Check if the fs implements the new VFS_NMOUNT()
659 	 * function, since the new system call was used.
660 	 */
661 	if (mp->mnt_op->vfs_mount != NULL) {
662 		printf("%s doesn't support the new mount syscall\n",
663 		    mp->mnt_vfc->vfc_name);
664 		VI_LOCK(vp);
665 		vp->v_iflag &= ~VI_MOUNT;
666 		VI_UNLOCK(vp);
667 		if (mp->mnt_flag & MNT_UPDATE)
668 			vfs_unbusy(mp, td);
669 		else {
670 			mp->mnt_vfc->vfc_refcount--;
671 			vfs_unbusy(mp, td);
672 #ifdef MAC
673 			mac_destroy_mount(mp);
674 #endif
675 			free(mp, M_MOUNT);
676 		}
677 		vrele(vp);
678 		error = EOPNOTSUPP;
679 		goto bad;
680 	}
681 
682 	/*
683 	 * Set the mount level flags.
684 	 */
685 	if (fsflags & MNT_RDONLY)
686 		mp->mnt_flag |= MNT_RDONLY;
687 	else if (mp->mnt_flag & MNT_RDONLY)
688 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
689 	mp->mnt_flag &=~ MNT_UPDATEMASK;
690 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE);
691 	/*
692 	 * Mount the filesystem.
693 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
694 	 * get.  No freeing of cn_pnbuf.
695 	 */
696 	error = VFS_NMOUNT(mp, &nd, td);
697 	if (!error) {
698 		if (mp->mnt_opt != NULL)
699 			vfs_freeopts(mp->mnt_opt);
700 		mp->mnt_opt = mp->mnt_optnew;
701 	}
702 	/*
703 	 * Prevent external consumers of mount
704 	 * options to read mnt_optnew.
705 	 */
706 	mp->mnt_optnew = NULL;
707 	if (mp->mnt_flag & MNT_UPDATE) {
708 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
709 			mp->mnt_flag &= ~MNT_RDONLY;
710 		mp->mnt_flag &=~
711 		    (MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
712 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
713 		if (error) {
714 			mp->mnt_flag = flag;
715 			mp->mnt_kern_flag = kern_flag;
716 		}
717 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
718 			if (mp->mnt_syncer == NULL)
719 				error = vfs_allocate_syncvnode(mp);
720 		} else {
721 			if (mp->mnt_syncer != NULL)
722 				vrele(mp->mnt_syncer);
723 			mp->mnt_syncer = NULL;
724 		}
725 		vfs_unbusy(mp, td);
726 		VI_LOCK(vp);
727 		vp->v_iflag &= ~VI_MOUNT;
728 		VI_UNLOCK(vp);
729 		vrele(vp);
730 		return (error);
731 	}
732 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
733 	/*
734 	 * Put the new filesystem on the mount list after root.
735 	 */
736 	cache_purge(vp);
737 	if (!error) {
738 		struct vnode *newdp;
739 
740 		VI_LOCK(vp);
741 		vp->v_iflag &= ~VI_MOUNT;
742 		vp->v_mountedhere = mp;
743 		VI_UNLOCK(vp);
744 		mtx_lock(&mountlist_mtx);
745 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
746 		mtx_unlock(&mountlist_mtx);
747 		if (VFS_ROOT(mp, &newdp))
748 			panic("mount: lost mount");
749 		checkdirs(vp, newdp);
750 		vput(newdp);
751 		VOP_UNLOCK(vp, 0, td);
752 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
753 			error = vfs_allocate_syncvnode(mp);
754 		vfs_unbusy(mp, td);
755 		if ((error = VFS_START(mp, 0, td)) != 0) {
756 			vrele(vp);
757 			goto bad;
758 		}
759 	} else {
760 		VI_LOCK(vp);
761 		vp->v_iflag &= ~VI_MOUNT;
762 		VI_UNLOCK(vp);
763 		mp->mnt_vfc->vfc_refcount--;
764 		vfs_unbusy(mp, td);
765 #ifdef MAC
766 		mac_destroy_mount(mp);
767 #endif
768 		free(mp, M_MOUNT);
769 		vput(vp);
770 		goto bad;
771 	}
772 	return (0);
773 bad:
774 	vfs_freeopts(optlist);
775 	return (error);
776 }
777 
778 /*
779  * Old mount API.
780  */
781 #ifndef _SYS_SYSPROTO_H_
782 struct mount_args {
783 	char	*type;
784 	char	*path;
785 	int	flags;
786 	caddr_t	data;
787 };
788 #endif
789 /* ARGSUSED */
790 int
791 mount(td, uap)
792 	struct thread *td;
793 	struct mount_args /* {
794 		syscallarg(char *) type;
795 		syscallarg(char *) path;
796 		syscallarg(int) flags;
797 		syscallarg(caddr_t) data;
798 	} */ *uap;
799 {
800 	char *fstype;
801 	char *fspath;
802 	int error;
803 
804 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
805 	fspath = malloc(MNAMELEN, M_TEMP, M_WAITOK);
806 
807 	/*
808 	 * vfs_mount() actually takes a kernel string for `type' and
809 	 * `path' now, so extract them.
810 	 */
811 	error = copyinstr(SCARG(uap, type), fstype, MFSNAMELEN, NULL);
812 	if (error)
813 		goto finish;
814 	error = copyinstr(SCARG(uap, path), fspath, MNAMELEN, NULL);
815 	if (error)
816 		goto finish;
817 	error = vfs_mount(td, fstype, fspath, SCARG(uap, flags),
818 	    SCARG(uap, data));
819 finish:
820 	free(fstype, M_TEMP);
821 	free(fspath, M_TEMP);
822 	return (error);
823 }
824 
825 /*
826  * vfs_mount(): actually attempt a filesystem mount.
827  *
828  * This routine is designed to be a "generic" entry point for routines
829  * that wish to mount a filesystem. All parameters except `fsdata' are
830  * pointers into kernel space. `fsdata' is currently still a pointer
831  * into userspace.
832  */
833 int
834 vfs_mount(td, fstype, fspath, fsflags, fsdata)
835 	struct thread *td;
836 	const char *fstype;
837 	char *fspath;
838 	int fsflags;
839 	void *fsdata;
840 {
841 	linker_file_t lf;
842 	struct vnode *vp;
843 	struct mount *mp;
844 	struct vfsconf *vfsp;
845 	int error, flag = 0, kern_flag = 0;
846 	struct vattr va;
847 	struct nameidata nd;
848 
849 	/*
850 	 * Be ultra-paranoid about making sure the type and fspath
851 	 * variables will fit in our mp buffers, including the
852 	 * terminating NUL.
853 	 */
854 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
855 		return (ENAMETOOLONG);
856 
857 	if (usermount == 0) {
858 		error = suser(td);
859 		if (error)
860 			return (error);
861 	}
862 	/*
863 	 * Do not allow NFS export by non-root users.
864 	 */
865 	if (fsflags & MNT_EXPORTED) {
866 		error = suser(td);
867 		if (error)
868 			return (error);
869 	}
870 	/*
871 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users.
872 	 */
873 	if (suser(td))
874 		fsflags |= MNT_NOSUID | MNT_NODEV;
875 	/*
876 	 * Get vnode to be covered
877 	 */
878 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td);
879 	if ((error = namei(&nd)) != 0)
880 		return (error);
881 	NDFREE(&nd, NDF_ONLY_PNBUF);
882 	vp = nd.ni_vp;
883 	if (fsflags & MNT_UPDATE) {
884 		if ((vp->v_vflag & VV_ROOT) == 0) {
885 			vput(vp);
886 			return (EINVAL);
887 		}
888 		mp = vp->v_mount;
889 		flag = mp->mnt_flag;
890 		kern_flag = mp->mnt_kern_flag;
891 		/*
892 		 * We only allow the filesystem to be reloaded if it
893 		 * is currently mounted read-only.
894 		 */
895 		if ((fsflags & MNT_RELOAD) &&
896 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
897 			vput(vp);
898 			return (EOPNOTSUPP);	/* Needs translation */
899 		}
900 		/*
901 		 * Only root, or the user that did the original mount is
902 		 * permitted to update it.
903 		 */
904 		if (mp->mnt_stat.f_owner != td->td_ucred->cr_uid) {
905 			error = suser(td);
906 			if (error) {
907 				vput(vp);
908 				return (error);
909 			}
910 		}
911 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
912 			vput(vp);
913 			return (EBUSY);
914 		}
915 		VI_LOCK(vp);
916 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
917 		    vp->v_mountedhere != NULL) {
918 			VI_UNLOCK(vp);
919 			vfs_unbusy(mp, td);
920 			vput(vp);
921 			return (EBUSY);
922 		}
923 		vp->v_iflag |= VI_MOUNT;
924 		VI_UNLOCK(vp);
925 		mp->mnt_flag |= fsflags &
926 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT);
927 		VOP_UNLOCK(vp, 0, td);
928 		goto update;
929 	}
930 	/*
931 	 * If the user is not root, ensure that they own the directory
932 	 * onto which we are attempting to mount.
933 	 */
934 	error = VOP_GETATTR(vp, &va, td->td_ucred, td);
935 	if (error) {
936 		vput(vp);
937 		return (error);
938 	}
939 	if (va.va_uid != td->td_ucred->cr_uid) {
940 		error = suser(td);
941 		if (error) {
942 			vput(vp);
943 			return (error);
944 		}
945 	}
946 	if ((error = vinvalbuf(vp, V_SAVE, td->td_ucred, td, 0, 0)) != 0) {
947 		vput(vp);
948 		return (error);
949 	}
950 	if (vp->v_type != VDIR) {
951 		vput(vp);
952 		return (ENOTDIR);
953 	}
954 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
955 		if (!strcmp(vfsp->vfc_name, fstype))
956 			break;
957 	if (vfsp == NULL) {
958 		/* Only load modules for root (very important!). */
959 		error = suser(td);
960 		if (error) {
961 			vput(vp);
962 			return (error);
963 		}
964 		error = securelevel_gt(td->td_ucred, 0);
965 		if (error) {
966 			vput(vp);
967 			return (error);
968 		}
969 		error = linker_load_module(NULL, fstype, NULL, NULL, &lf);
970 		if (error || lf == NULL) {
971 			vput(vp);
972 			if (lf == NULL)
973 				error = ENODEV;
974 			return (error);
975 		}
976 		lf->userrefs++;
977 		/* Look up again to see if the VFS was loaded. */
978 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
979 			if (!strcmp(vfsp->vfc_name, fstype))
980 				break;
981 		if (vfsp == NULL) {
982 			lf->userrefs--;
983 			linker_file_unload(lf);
984 			vput(vp);
985 			return (ENODEV);
986 		}
987 	}
988 	VI_LOCK(vp);
989 	if ((vp->v_iflag & VI_MOUNT) != 0 ||
990 	    vp->v_mountedhere != NULL) {
991 		VI_UNLOCK(vp);
992 		vput(vp);
993 		return (EBUSY);
994 	}
995 	vp->v_iflag |= VI_MOUNT;
996 	VI_UNLOCK(vp);
997 
998 	/*
999 	 * Allocate and initialize the filesystem.
1000 	 */
1001 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
1002 	TAILQ_INIT(&mp->mnt_nvnodelist);
1003 	TAILQ_INIT(&mp->mnt_reservedvnlist);
1004 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
1005 	(void)vfs_busy(mp, LK_NOWAIT, 0, td);
1006 	mp->mnt_op = vfsp->vfc_vfsops;
1007 	mp->mnt_vfc = vfsp;
1008 	vfsp->vfc_refcount++;
1009 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
1010 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
1011 	strncpy(mp->mnt_stat.f_fstypename, fstype, MFSNAMELEN);
1012 	mp->mnt_vnodecovered = vp;
1013 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
1014 	strncpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
1015 	mp->mnt_iosize_max = DFLTPHYS;
1016 #ifdef MAC
1017 	mac_init_mount(mp);
1018 	mac_create_mount(td->td_ucred, mp);
1019 #endif
1020 	VOP_UNLOCK(vp, 0, td);
1021 update:
1022 	/*
1023 	 * Check if the fs implements the old VFS_MOUNT()
1024 	 * function, since the old system call was used.
1025 	 */
1026 	if (mp->mnt_op->vfs_mount == NULL) {
1027 		printf("%s doesn't support the old mount syscall\n",
1028 		    mp->mnt_vfc->vfc_name);
1029 		VI_LOCK(vp);
1030 		vp->v_iflag &= ~VI_MOUNT;
1031 		VI_UNLOCK(vp);
1032 		if (mp->mnt_flag & MNT_UPDATE)
1033 			vfs_unbusy(mp, td);
1034 		else {
1035 			mp->mnt_vfc->vfc_refcount--;
1036 			vfs_unbusy(mp, td);
1037 #ifdef MAC
1038 			mac_destroy_mount(mp);
1039 #endif
1040 			free(mp, M_MOUNT);
1041 		}
1042 		vrele(vp);
1043 		return (EOPNOTSUPP);
1044 	}
1045 
1046 	/*
1047 	 * Set the mount level flags.
1048 	 */
1049 	if (fsflags & MNT_RDONLY)
1050 		mp->mnt_flag |= MNT_RDONLY;
1051 	else if (mp->mnt_flag & MNT_RDONLY)
1052 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
1053 	mp->mnt_flag &=~ MNT_UPDATEMASK;
1054 	mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE);
1055 	/*
1056 	 * Mount the filesystem.
1057 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1058 	 * get.  No freeing of cn_pnbuf.
1059 	 */
1060 	error = VFS_MOUNT(mp, fspath, fsdata, &nd, td);
1061 	if (mp->mnt_flag & MNT_UPDATE) {
1062 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
1063 			mp->mnt_flag &= ~MNT_RDONLY;
1064 		mp->mnt_flag &=~
1065 		    (MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT);
1066 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
1067 		if (error) {
1068 			mp->mnt_flag = flag;
1069 			mp->mnt_kern_flag = kern_flag;
1070 		}
1071 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1072 			if (mp->mnt_syncer == NULL)
1073 				error = vfs_allocate_syncvnode(mp);
1074 		} else {
1075 			if (mp->mnt_syncer != NULL)
1076 				vrele(mp->mnt_syncer);
1077 			mp->mnt_syncer = NULL;
1078 		}
1079 		vfs_unbusy(mp, td);
1080 		VI_LOCK(vp);
1081 		vp->v_iflag &= ~VI_MOUNT;
1082 		VI_UNLOCK(vp);
1083 		vrele(vp);
1084 		return (error);
1085 	}
1086 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1087 	/*
1088 	 * Put the new filesystem on the mount list after root.
1089 	 */
1090 	cache_purge(vp);
1091 	if (!error) {
1092 		struct vnode *newdp;
1093 
1094 		mp_fixme("Does interlock protect mounted here or not?");
1095 		VI_LOCK(vp);
1096 		vp->v_iflag &= ~VI_MOUNT;
1097 		vp->v_mountedhere = mp;
1098 		VI_UNLOCK(vp);
1099 		mtx_lock(&mountlist_mtx);
1100 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1101 		mtx_unlock(&mountlist_mtx);
1102 		if (VFS_ROOT(mp, &newdp))
1103 			panic("mount: lost mount");
1104 		checkdirs(vp, newdp);
1105 		vput(newdp);
1106 		VOP_UNLOCK(vp, 0, td);
1107 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1108 			error = vfs_allocate_syncvnode(mp);
1109 		vfs_unbusy(mp, td);
1110 		if ((error = VFS_START(mp, 0, td)) != 0)
1111 			vrele(vp);
1112 	} else {
1113 		VI_LOCK(vp);
1114 		vp->v_iflag &= ~VI_MOUNT;
1115 		VI_UNLOCK(vp);
1116 		mp->mnt_vfc->vfc_refcount--;
1117 		vfs_unbusy(mp, td);
1118 #ifdef MAC
1119 		mac_destroy_mount(mp);
1120 #endif
1121 		free(mp, M_MOUNT);
1122 		vput(vp);
1123 	}
1124 	return (error);
1125 }
1126 
1127 /*
1128  * Scan all active processes to see if any of them have a current
1129  * or root directory of `olddp'. If so, replace them with the new
1130  * mount point.
1131  */
1132 static void
1133 checkdirs(olddp, newdp)
1134 	struct vnode *olddp, *newdp;
1135 {
1136 	struct filedesc *fdp;
1137 	struct proc *p;
1138 	int nrele;
1139 
1140 	if (olddp->v_usecount == 1)
1141 		return;
1142 	sx_slock(&allproc_lock);
1143 	LIST_FOREACH(p, &allproc, p_list) {
1144 		PROC_LOCK(p);
1145 		fdp = p->p_fd;
1146 		if (fdp == NULL) {
1147 			PROC_UNLOCK(p);
1148 			continue;
1149 		}
1150 		nrele = 0;
1151 		FILEDESC_LOCK(fdp);
1152 		if (fdp->fd_cdir == olddp) {
1153 			VREF(newdp);
1154 			fdp->fd_cdir = newdp;
1155 			nrele++;
1156 		}
1157 		if (fdp->fd_rdir == olddp) {
1158 			VREF(newdp);
1159 			fdp->fd_rdir = newdp;
1160 			nrele++;
1161 		}
1162 		FILEDESC_UNLOCK(fdp);
1163 		PROC_UNLOCK(p);
1164 		while (nrele--)
1165 			vrele(olddp);
1166 	}
1167 	sx_sunlock(&allproc_lock);
1168 	if (rootvnode == olddp) {
1169 		vrele(rootvnode);
1170 		VREF(newdp);
1171 		rootvnode = newdp;
1172 	}
1173 }
1174 
1175 /*
1176  * Unmount a filesystem.
1177  *
1178  * Note: unmount takes a path to the vnode mounted on as argument,
1179  * not special file (as before).
1180  */
1181 #ifndef _SYS_SYSPROTO_H_
1182 struct unmount_args {
1183 	char	*path;
1184 	int	flags;
1185 };
1186 #endif
1187 /* ARGSUSED */
1188 int
1189 unmount(td, uap)
1190 	struct thread *td;
1191 	register struct unmount_args /* {
1192 		syscallarg(char *) path;
1193 		syscallarg(int) flags;
1194 	} */ *uap;
1195 {
1196 	register struct vnode *vp;
1197 	struct mount *mp;
1198 	int error;
1199 	struct nameidata nd;
1200 
1201 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE,
1202 	    SCARG(uap, path), td);
1203 	if ((error = namei(&nd)) != 0)
1204 		return (error);
1205 	vp = nd.ni_vp;
1206 	NDFREE(&nd, NDF_ONLY_PNBUF);
1207 	mp = vp->v_mount;
1208 
1209 	/*
1210 	 * Only root, or the user that did the original mount is
1211 	 * permitted to unmount this filesystem.
1212 	 */
1213 	if (mp->mnt_stat.f_owner != td->td_ucred->cr_uid) {
1214 		error = suser(td);
1215 		if (error) {
1216 			vput(vp);
1217 			return (error);
1218 		}
1219 	}
1220 
1221 	/*
1222 	 * Don't allow unmounting the root filesystem.
1223 	 */
1224 	if (mp->mnt_flag & MNT_ROOTFS) {
1225 		vput(vp);
1226 		return (EINVAL);
1227 	}
1228 
1229 	/*
1230 	 * Must be the root of the filesystem
1231 	 */
1232 	if ((vp->v_vflag & VV_ROOT) == 0) {
1233 		vput(vp);
1234 		return (EINVAL);
1235 	}
1236 	vput(vp);
1237 	return (dounmount(mp, SCARG(uap, flags), td));
1238 }
1239 
1240 /*
1241  * Do the actual filesystem unmount.
1242  */
1243 int
1244 dounmount(mp, flags, td)
1245 	struct mount *mp;
1246 	int flags;
1247 	struct thread *td;
1248 {
1249 	struct vnode *coveredvp, *fsrootvp;
1250 	int error;
1251 	int async_flag;
1252 
1253 	mtx_lock(&mountlist_mtx);
1254 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1255 		mtx_unlock(&mountlist_mtx);
1256 		return (EBUSY);
1257 	}
1258 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1259 	/* Allow filesystems to detect that a forced unmount is in progress. */
1260 	if (flags & MNT_FORCE)
1261 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1262 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1263 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), &mountlist_mtx, td);
1264 	if (error) {
1265 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1266 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1267 			wakeup(mp);
1268 		return (error);
1269 	}
1270 	vn_start_write(NULL, &mp, V_WAIT);
1271 
1272 	if (mp->mnt_flag & MNT_EXPUBLIC)
1273 		vfs_setpublicfs(NULL, NULL, NULL);
1274 
1275 	vfs_msync(mp, MNT_WAIT);
1276 	async_flag = mp->mnt_flag & MNT_ASYNC;
1277 	mp->mnt_flag &=~ MNT_ASYNC;
1278 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1279 	if (mp->mnt_syncer != NULL)
1280 		vrele(mp->mnt_syncer);
1281 	/* Move process cdir/rdir refs on fs root to underlying vnode. */
1282 	if (VFS_ROOT(mp, &fsrootvp) == 0) {
1283 		if (mp->mnt_vnodecovered != NULL)
1284 			checkdirs(fsrootvp, mp->mnt_vnodecovered);
1285 		if (fsrootvp == rootvnode) {
1286 			vrele(rootvnode);
1287 			rootvnode = NULL;
1288 		}
1289 		vput(fsrootvp);
1290 	}
1291 	if (((mp->mnt_flag & MNT_RDONLY) ||
1292 	     (error = VFS_SYNC(mp, MNT_WAIT, td->td_ucred, td)) == 0) ||
1293 	    (flags & MNT_FORCE)) {
1294 		error = VFS_UNMOUNT(mp, flags, td);
1295 	}
1296 	vn_finished_write(mp);
1297 	if (error) {
1298 		/* Undo cdir/rdir and rootvnode changes made above. */
1299 		if (VFS_ROOT(mp, &fsrootvp) == 0) {
1300 			if (mp->mnt_vnodecovered != NULL)
1301 				checkdirs(mp->mnt_vnodecovered, fsrootvp);
1302 			if (rootvnode == NULL) {
1303 				rootvnode = fsrootvp;
1304 				vref(rootvnode);
1305 			}
1306 			vput(fsrootvp);
1307 		}
1308 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL)
1309 			(void) vfs_allocate_syncvnode(mp);
1310 		mtx_lock(&mountlist_mtx);
1311 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1312 		mp->mnt_flag |= async_flag;
1313 		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK,
1314 		    &mountlist_mtx, td);
1315 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1316 			wakeup(mp);
1317 		return (error);
1318 	}
1319 	mtx_lock(&mountlist_mtx);
1320 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1321 	if ((coveredvp = mp->mnt_vnodecovered) != NULL)
1322 		coveredvp->v_mountedhere = NULL;
1323 	mp->mnt_vfc->vfc_refcount--;
1324 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
1325 		panic("unmount: dangling vnode");
1326 	lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK, &mountlist_mtx, td);
1327 	lockdestroy(&mp->mnt_lock);
1328 	if (coveredvp != NULL)
1329 		vrele(coveredvp);
1330 	if (mp->mnt_kern_flag & MNTK_MWAIT)
1331 		wakeup(mp);
1332 #ifdef MAC
1333 	mac_destroy_mount(mp);
1334 #endif
1335 	if (mp->mnt_op->vfs_mount == NULL)
1336 		vfs_freeopts(mp->mnt_opt);
1337 	free(mp, M_MOUNT);
1338 	return (0);
1339 }
1340 
1341 /*
1342  * Lookup a filesystem type, and if found allocate and initialize
1343  * a mount structure for it.
1344  *
1345  * Devname is usually updated by mount(8) after booting.
1346  */
1347 int
1348 vfs_rootmountalloc(fstypename, devname, mpp)
1349 	char *fstypename;
1350 	char *devname;
1351 	struct mount **mpp;
1352 {
1353 	struct thread *td = curthread;	/* XXX */
1354 	struct vfsconf *vfsp;
1355 	struct mount *mp;
1356 
1357 	if (fstypename == NULL)
1358 		return (ENODEV);
1359 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1360 		if (!strcmp(vfsp->vfc_name, fstypename))
1361 			break;
1362 	if (vfsp == NULL)
1363 		return (ENODEV);
1364 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
1365 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
1366 	(void)vfs_busy(mp, LK_NOWAIT, 0, td);
1367 	TAILQ_INIT(&mp->mnt_nvnodelist);
1368 	TAILQ_INIT(&mp->mnt_reservedvnlist);
1369 	mp->mnt_vfc = vfsp;
1370 	mp->mnt_op = vfsp->vfc_vfsops;
1371 	mp->mnt_flag = MNT_RDONLY;
1372 	mp->mnt_vnodecovered = NULLVP;
1373 	vfsp->vfc_refcount++;
1374 	mp->mnt_iosize_max = DFLTPHYS;
1375 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
1376 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
1377 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
1378 	mp->mnt_stat.f_mntonname[0] = '/';
1379 	mp->mnt_stat.f_mntonname[1] = 0;
1380 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
1381 #ifdef MAC
1382 	mac_init_mount(mp);
1383 	mac_create_mount(td->td_ucred, mp);
1384 #endif
1385 	*mpp = mp;
1386 	return (0);
1387 }
1388 
1389 /*
1390  * Find and mount the root filesystem
1391  */
1392 void
1393 vfs_mountroot(void)
1394 {
1395 	char		*cp;
1396 	int		i, error;
1397 
1398 	/*
1399 	 * The root filesystem information is compiled in, and we are
1400 	 * booted with instructions to use it.
1401 	 */
1402 #ifdef ROOTDEVNAME
1403 	if ((boothowto & RB_DFLTROOT) &&
1404 	    !vfs_mountroot_try(ROOTDEVNAME))
1405 		return;
1406 #endif
1407 	/*
1408 	 * We are booted with instructions to prompt for the root filesystem,
1409 	 * or to use the compiled-in default when it doesn't exist.
1410 	 */
1411 	if (boothowto & (RB_DFLTROOT | RB_ASKNAME)) {
1412 		if (!vfs_mountroot_ask())
1413 			return;
1414 	}
1415 
1416 	/*
1417 	 * We've been given the generic "use CDROM as root" flag.  This is
1418 	 * necessary because one media may be used in many different
1419 	 * devices, so we need to search for them.
1420 	 */
1421 	if (boothowto & RB_CDROM) {
1422 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1423 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1424 				return;
1425 		}
1426 	}
1427 
1428 	/*
1429 	 * Try to use the value read by the loader from /etc/fstab, or
1430 	 * supplied via some other means.  This is the preferred
1431 	 * mechanism.
1432 	 */
1433 	if ((cp = getenv("vfs.root.mountfrom")) != NULL) {
1434 		error = vfs_mountroot_try(cp);
1435 		freeenv(cp);
1436 		if (!error)
1437 			return;
1438 	}
1439 
1440 	/*
1441 	 * Try values that may have been computed by the machine-dependant
1442 	 * legacy code.
1443 	 */
1444 	if (!vfs_mountroot_try(rootdevnames[0]))
1445 		return;
1446 	if (!vfs_mountroot_try(rootdevnames[1]))
1447 		return;
1448 
1449 	/*
1450 	 * If we have a compiled-in default, and haven't already tried it, try
1451 	 * it now.
1452 	 */
1453 #ifdef ROOTDEVNAME
1454 	if (!(boothowto & RB_DFLTROOT))
1455 		if (!vfs_mountroot_try(ROOTDEVNAME))
1456 			return;
1457 #endif
1458 
1459 	/*
1460 	 * Everything so far has failed, prompt on the console if we haven't
1461 	 * already tried that.
1462 	 */
1463 	if (!(boothowto & (RB_DFLTROOT | RB_ASKNAME)) && !vfs_mountroot_ask())
1464 		return;
1465 	panic("Root mount failed, startup aborted.");
1466 }
1467 
1468 /*
1469  * Mount (mountfrom) as the root filesystem.
1470  */
1471 static int
1472 vfs_mountroot_try(char *mountfrom)
1473 {
1474         struct mount	*mp;
1475 	char		*vfsname, *path;
1476 	int		error;
1477 	char		patt[32];
1478 	int		s;
1479 
1480 	vfsname = NULL;
1481 	path    = NULL;
1482 	mp      = NULL;
1483 	error   = EINVAL;
1484 
1485 	if (mountfrom == NULL)
1486 		return(error);		/* don't complain */
1487 
1488 	s = splcam();			/* Overkill, but annoying without it */
1489 	printf("Mounting root from %s\n", mountfrom);
1490 	splx(s);
1491 
1492 	/* parse vfs name and path */
1493 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1494 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1495 	vfsname[0] = path[0] = 0;
1496 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1497 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1498 		goto done;
1499 
1500 	/* allocate a root mount */
1501 	error = vfs_rootmountalloc(vfsname, path[0] != 0 ? path : ROOTNAME,
1502 				   &mp);
1503 	if (error != 0) {
1504 		printf("Can't allocate root mount for filesystem '%s': %d\n",
1505 		       vfsname, error);
1506 		goto done;
1507 	}
1508 	mp->mnt_flag |= MNT_ROOTFS;
1509 
1510 	/* do our best to set rootdev */
1511 	if ((path[0] != 0) && setrootbyname(path))
1512 		printf("setrootbyname failed\n");
1513 
1514 	/* If the root device is a type "memory disk", mount RW */
1515 	if (rootdev != NODEV && devsw(rootdev) &&
1516 	    (devsw(rootdev)->d_flags & D_MEMDISK))
1517 		mp->mnt_flag &= ~MNT_RDONLY;
1518 
1519 	/*
1520 	 * Set the mount path to be something useful, because the
1521 	 * filesystem code isn't responsible now for initialising
1522 	 * f_mntonname unless they want to override the default
1523 	 * (which is `path'.)
1524 	 */
1525 	strncpy(mp->mnt_stat.f_mntonname, "/", MNAMELEN);
1526 
1527 	error = VFS_MOUNT(mp, NULL, NULL, NULL, curthread);
1528 
1529 done:
1530 	if (vfsname != NULL)
1531 		free(vfsname, M_MOUNT);
1532 	if (path != NULL)
1533 		free(path, M_MOUNT);
1534 	if (error != 0) {
1535 		if (mp != NULL) {
1536 			vfs_unbusy(mp, curthread);
1537 #ifdef MAC
1538 			mac_destroy_mount(mp);
1539 #endif
1540 			free(mp, M_MOUNT);
1541 		}
1542 		printf("Root mount failed: %d\n", error);
1543 	} else {
1544 
1545 		/* register with list of mounted filesystems */
1546 		mtx_lock(&mountlist_mtx);
1547 		TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1548 		mtx_unlock(&mountlist_mtx);
1549 
1550 		/* sanity check system clock against root fs timestamp */
1551 		inittodr(mp->mnt_time);
1552 		vfs_unbusy(mp, curthread);
1553 		error = VFS_START(mp, 0, curthread);
1554 	}
1555 	return(error);
1556 }
1557 
1558 /*
1559  * Spin prompting on the console for a suitable root filesystem
1560  */
1561 static int
1562 vfs_mountroot_ask(void)
1563 {
1564 	char name[128];
1565 	int i;
1566 	dev_t dev;
1567 
1568 	for(;;) {
1569 		printf("\nManual root filesystem specification:\n");
1570 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1571 #if defined(__i386__) || defined(__ia64__)
1572 		printf("                       eg. ufs:da0s1a\n");
1573 #else
1574 		printf("                       eg. ufs:da0a\n");
1575 #endif
1576 		printf("  ?                  List valid disk boot devices\n");
1577 		printf("  <empty line>       Abort manual input\n");
1578 		printf("\nmountroot> ");
1579 		gets(name);
1580 		if (name[0] == 0)
1581 			return(1);
1582 		if (name[0] == '?') {
1583 			printf("Possibly valid devices for 'ufs' root:\n");
1584 			for (i = 0; i < NUMCDEVSW; i++) {
1585 				dev = makedev(i, 0);
1586 				if (devsw(dev) != NULL)
1587 					printf(" \"%s\"", devsw(dev)->d_name);
1588 			}
1589 			printf("\n");
1590 			continue;
1591 		}
1592 		if (!vfs_mountroot_try(name))
1593 			return(0);
1594 	}
1595 }
1596 
1597 /*
1598  * Local helper function for vfs_mountroot_ask.
1599  */
1600 static void
1601 gets(char *cp)
1602 {
1603 	char *lp;
1604 	int c;
1605 
1606 	lp = cp;
1607 	for (;;) {
1608 		printf("%c", c = cngetc() & 0177);
1609 		switch (c) {
1610 		case -1:
1611 		case '\n':
1612 		case '\r':
1613 			*lp++ = '\0';
1614 			return;
1615 		case '\b':
1616 		case '\177':
1617 			if (lp > cp) {
1618 				printf(" \b");
1619 				lp--;
1620 			}
1621 			continue;
1622 		case '#':
1623 			lp--;
1624 			if (lp < cp)
1625 				lp = cp;
1626 			continue;
1627 		case '@':
1628 		case 'u' & 037:
1629 			lp = cp;
1630 			printf("%c", '\n');
1631 			continue;
1632 		default:
1633 			*lp++ = c;
1634 		}
1635 	}
1636 }
1637 
1638 /*
1639  * Convert a given name to the dev_t of the disk-like device
1640  * it refers to.
1641  */
1642 dev_t
1643 getdiskbyname(char *name) {
1644 	char *cp;
1645 	dev_t dev;
1646 
1647 	cp = name;
1648 	if (!bcmp(cp, "/dev/", 5))
1649 		cp += 5;
1650 
1651 	dev = NODEV;
1652 	EVENTHANDLER_INVOKE(dev_clone, cp, strlen(cp), &dev);
1653 	return (dev);
1654 }
1655 
1656 /*
1657  * Set rootdev to match (name), given that we expect it to
1658  * refer to a disk-like device.
1659  */
1660 static int
1661 setrootbyname(char *name)
1662 {
1663 	dev_t diskdev;
1664 
1665 	diskdev = getdiskbyname(name);
1666 	if (diskdev != NODEV) {
1667 		rootdev = diskdev;
1668 		return (0);
1669 	}
1670 
1671 	return (1);
1672 }
1673 
1674 /* Show the dev_t for a disk specified by name */
1675 #ifdef DDB
1676 DB_SHOW_COMMAND(disk, db_getdiskbyname)
1677 {
1678 	dev_t dev;
1679 
1680 	if (modif[0] == '\0') {
1681 		db_error("usage: show disk/devicename");
1682 		return;
1683 	}
1684 	dev = getdiskbyname(modif);
1685 	if (dev != NODEV)
1686 		db_printf("dev_t = %p\n", dev);
1687 	else
1688 		db_printf("No disk device matched.\n");
1689 }
1690 #endif
1691 
1692 /*
1693  * Get a mount option by its name.
1694  *
1695  * Return 0 if the option was found, ENOENT otherwise.
1696  * If len is non-NULL it will be filled with the length
1697  * of the option. If buf is non-NULL, it will be filled
1698  * with the address of the option.
1699  */
1700 int
1701 vfs_getopt(opts, name, buf, len)
1702 	struct vfsoptlist *opts;
1703 	const char *name;
1704 	void **buf;
1705 	int *len;
1706 {
1707 	struct vfsopt *opt;
1708 
1709 	TAILQ_FOREACH(opt, opts, link) {
1710 		if (strcmp(name, opt->name) == 0) {
1711 			if (len != NULL)
1712 				*len = opt->len;
1713 			if (buf != NULL)
1714 				*buf = opt->value;
1715 			return (0);
1716 		}
1717 	}
1718 	return (ENOENT);
1719 }
1720 
1721 /*
1722  * Find and copy a mount option.
1723  *
1724  * The size of the buffer has to be specified
1725  * in len, if it is not the same length as the
1726  * mount option, EINVAL is returned.
1727  * Returns ENOENT if the option is not found.
1728  */
1729 int
1730 vfs_copyopt(opts, name, dest, len)
1731 	struct vfsoptlist *opts;
1732 	const char *name;
1733 	void *dest;
1734 	int len;
1735 {
1736 	struct vfsopt *opt;
1737 
1738 	TAILQ_FOREACH(opt, opts, link) {
1739 		if (strcmp(name, opt->name) == 0) {
1740 			if (len != opt->len)
1741 				return (EINVAL);
1742 			bcopy(opt->value, dest, opt->len);
1743 			return (0);
1744 		}
1745 	}
1746 	return (ENOENT);
1747 }
1748