xref: /freebsd/sys/kern/vfs_mount.c (revision 721351876cd4d3a8a700f62d2061331fa951a488)
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/fcntl.h>
43 #include <sys/jail.h>
44 #include <sys/kernel.h>
45 #include <sys/libkern.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/priv.h>
51 #include <sys/proc.h>
52 #include <sys/filedesc.h>
53 #include <sys/reboot.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysproto.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/systm.h>
60 #include <sys/vnode.h>
61 #include <vm/uma.h>
62 
63 #include <geom/geom.h>
64 
65 #include <machine/stdarg.h>
66 
67 #include <security/audit/audit.h>
68 #include <security/mac/mac_framework.h>
69 
70 #include "opt_rootdevname.h"
71 #include "opt_mac.h"
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 int	vfs_mountroot_ask(void);
79 static int	vfs_mountroot_try(const char *mountfrom);
80 static int	vfs_donmount(struct thread *td, int fsflags,
81 		    struct uio *fsoptions);
82 static void	free_mntarg(struct mntarg *ma);
83 static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
84 
85 static int	usermount = 0;
86 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
87     "Unprivileged users may mount and unmount file systems");
88 
89 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
90 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
91 static uma_zone_t mount_zone;
92 
93 /* List of mounted filesystems. */
94 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
95 
96 /* For any iteration/modification of mountlist */
97 struct mtx mountlist_mtx;
98 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
99 
100 TAILQ_HEAD(vfsoptlist, vfsopt);
101 struct vfsopt {
102 	TAILQ_ENTRY(vfsopt) link;
103 	char	*name;
104 	void	*value;
105 	int	len;
106 };
107 
108 /*
109  * The vnode of the system's root (/ in the filesystem, without chroot
110  * active.)
111  */
112 struct vnode	*rootvnode;
113 
114 /*
115  * The root filesystem is detailed in the kernel environment variable
116  * vfs.root.mountfrom, which is expected to be in the general format
117  *
118  * <vfsname>:[<path>]
119  * vfsname   := the name of a VFS known to the kernel and capable
120  *              of being mounted as root
121  * path      := disk device name or other data used by the filesystem
122  *              to locate its physical store
123  */
124 
125 /*
126  * Global opts, taken by all filesystems
127  */
128 static const char *global_opts[] = {
129 	"errmsg",
130 	"fstype",
131 	"fspath",
132 	"ro",
133 	"rw",
134 	"nosuid",
135 	"noexec",
136 	"update",
137 	NULL
138 };
139 
140 /*
141  * The root specifiers we will try if RB_CDROM is specified.
142  */
143 static char *cdrom_rootdevnames[] = {
144 	"cd9660:cd0",
145 	"cd9660:acd0",
146 	NULL
147 };
148 
149 /* legacy find-root code */
150 char		*rootdevnames[2] = {NULL, NULL};
151 #ifndef ROOTDEVNAME
152 #  define ROOTDEVNAME NULL
153 #endif
154 static const char	*ctrootdevname = ROOTDEVNAME;
155 
156 /*
157  * ---------------------------------------------------------------------
158  * Functions for building and sanitizing the mount options
159  */
160 
161 /* Remove one mount option. */
162 static void
163 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
164 {
165 
166 	TAILQ_REMOVE(opts, opt, link);
167 	free(opt->name, M_MOUNT);
168 	if (opt->value != NULL)
169 		free(opt->value, M_MOUNT);
170 #ifdef INVARIANTS
171 	else if (opt->len != 0)
172 		panic("%s: mount option with NULL value but length != 0",
173 		    __func__);
174 #endif
175 	free(opt, M_MOUNT);
176 }
177 
178 /* Release all resources related to the mount options. */
179 void
180 vfs_freeopts(struct vfsoptlist *opts)
181 {
182 	struct vfsopt *opt;
183 
184 	while (!TAILQ_EMPTY(opts)) {
185 		opt = TAILQ_FIRST(opts);
186 		vfs_freeopt(opts, opt);
187 	}
188 	free(opts, M_MOUNT);
189 }
190 
191 void
192 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
193 {
194 	struct vfsopt *opt, *temp;
195 
196 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
197 		if (strcmp(opt->name, name) == 0)
198 			vfs_freeopt(opts, opt);
199 	}
200 }
201 
202 /*
203  * Check if options are equal (with or without the "no" prefix).
204  */
205 static int
206 vfs_equalopts(const char *opt1, const char *opt2)
207 {
208 
209 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
210 	if (strcmp(opt1, opt2) == 0)
211 		return (1);
212 	/* "noopt" vs. "opt" */
213 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
214 		return (1);
215 	/* "opt" vs. "noopt" */
216 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
217 		return (1);
218 	return (0);
219 }
220 
221 /*
222  * If a mount option is specified several times,
223  * (with or without the "no" prefix) only keep
224  * the last occurence of it.
225  */
226 static void
227 vfs_sanitizeopts(struct vfsoptlist *opts)
228 {
229 	struct vfsopt *opt, *opt2, *tmp;
230 
231 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
232 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
233 		while (opt2 != NULL) {
234 			if (vfs_equalopts(opt->name, opt2->name)) {
235 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
236 				vfs_freeopt(opts, opt2);
237 				opt2 = tmp;
238 			} else {
239 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
240 			}
241 		}
242 	}
243 }
244 
245 /*
246  * Build a linked list of mount options from a struct uio.
247  */
248 static int
249 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
250 {
251 	struct vfsoptlist *opts;
252 	struct vfsopt *opt;
253 	size_t memused;
254 	unsigned int i, iovcnt;
255 	int error, namelen, optlen;
256 
257 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
258 	TAILQ_INIT(opts);
259 	memused = 0;
260 	iovcnt = auio->uio_iovcnt;
261 	for (i = 0; i < iovcnt; i += 2) {
262 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
263 		namelen = auio->uio_iov[i].iov_len;
264 		optlen = auio->uio_iov[i + 1].iov_len;
265 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
266 		opt->value = NULL;
267 		opt->len = 0;
268 
269 		/*
270 		 * Do this early, so jumps to "bad" will free the current
271 		 * option.
272 		 */
273 		TAILQ_INSERT_TAIL(opts, opt, link);
274 		memused += sizeof(struct vfsopt) + optlen + namelen;
275 
276 		/*
277 		 * Avoid consuming too much memory, and attempts to overflow
278 		 * memused.
279 		 */
280 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
281 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
282 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
283 			error = EINVAL;
284 			goto bad;
285 		}
286 
287 		if (auio->uio_segflg == UIO_SYSSPACE) {
288 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
289 		} else {
290 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
291 			    namelen);
292 			if (error)
293 				goto bad;
294 		}
295 		/* Ensure names are null-terminated strings. */
296 		if (opt->name[namelen - 1] != '\0') {
297 			error = EINVAL;
298 			goto bad;
299 		}
300 		if (optlen != 0) {
301 			opt->len = optlen;
302 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
303 			if (auio->uio_segflg == UIO_SYSSPACE) {
304 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
305 				    optlen);
306 			} else {
307 				error = copyin(auio->uio_iov[i + 1].iov_base,
308 				    opt->value, optlen);
309 				if (error)
310 					goto bad;
311 			}
312 		}
313 	}
314 	vfs_sanitizeopts(opts);
315 	*options = opts;
316 	return (0);
317 bad:
318 	vfs_freeopts(opts);
319 	return (error);
320 }
321 
322 /*
323  * Merge the old mount options with the new ones passed
324  * in the MNT_UPDATE case.
325  *
326  * XXX This function will keep a "nofoo" option in the
327  *     new options if there is no matching "foo" option
328  *     to be cancelled in the old options.  This is a bug
329  *     if the option's canonical name is "foo".  E.g., "noro"
330  *     shouldn't end up in the mount point's active options,
331  *     but it can.
332  */
333 static void
334 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
335 {
336 	struct vfsopt *opt, *opt2, *new;
337 
338 	TAILQ_FOREACH(opt, opts, link) {
339 		/*
340 		 * Check that this option hasn't been redefined
341 		 * nor cancelled with a "no" mount option.
342 		 */
343 		opt2 = TAILQ_FIRST(toopts);
344 		while (opt2 != NULL) {
345 			if (strcmp(opt2->name, opt->name) == 0)
346 				goto next;
347 			if (strncmp(opt2->name, "no", 2) == 0 &&
348 			    strcmp(opt2->name + 2, opt->name) == 0) {
349 				vfs_freeopt(toopts, opt2);
350 				goto next;
351 			}
352 			opt2 = TAILQ_NEXT(opt2, link);
353 		}
354 		/* We want this option, duplicate it. */
355 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
356 		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
357 		strcpy(new->name, opt->name);
358 		if (opt->len != 0) {
359 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
360 			bcopy(opt->value, new->value, opt->len);
361 		} else {
362 			new->value = NULL;
363 		}
364 		new->len = opt->len;
365 		TAILQ_INSERT_TAIL(toopts, new, link);
366 next:
367 		continue;
368 	}
369 }
370 
371 /*
372  * Mount a filesystem.
373  */
374 int
375 nmount(td, uap)
376 	struct thread *td;
377 	struct nmount_args /* {
378 		struct iovec *iovp;
379 		unsigned int iovcnt;
380 		int flags;
381 	} */ *uap;
382 {
383 	struct uio *auio;
384 	struct iovec *iov;
385 	unsigned int i;
386 	int error;
387 	u_int iovcnt;
388 
389 	AUDIT_ARG(fflags, uap->flags);
390 
391 	/*
392 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
393 	 * userspace to set this flag, but we must filter it out if we want
394 	 * MNT_UPDATE on the root file system to work.
395 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
396 	 */
397 	uap->flags &= ~MNT_ROOTFS;
398 
399 	iovcnt = uap->iovcnt;
400 	/*
401 	 * Check that we have an even number of iovec's
402 	 * and that we have at least two options.
403 	 */
404 	if ((iovcnt & 1) || (iovcnt < 4))
405 		return (EINVAL);
406 
407 	error = copyinuio(uap->iovp, iovcnt, &auio);
408 	if (error)
409 		return (error);
410 	iov = auio->uio_iov;
411 	for (i = 0; i < iovcnt; i++) {
412 		if (iov->iov_len > MMAXOPTIONLEN) {
413 			free(auio, M_IOV);
414 			return (EINVAL);
415 		}
416 		iov++;
417 	}
418 	error = vfs_donmount(td, uap->flags, auio);
419 
420 	free(auio, M_IOV);
421 	return (error);
422 }
423 
424 /*
425  * ---------------------------------------------------------------------
426  * Various utility functions
427  */
428 
429 void
430 vfs_ref(struct mount *mp)
431 {
432 
433 	MNT_ILOCK(mp);
434 	MNT_REF(mp);
435 	MNT_IUNLOCK(mp);
436 }
437 
438 void
439 vfs_rel(struct mount *mp)
440 {
441 
442 	MNT_ILOCK(mp);
443 	MNT_REL(mp);
444 	MNT_IUNLOCK(mp);
445 }
446 
447 static int
448 mount_init(void *mem, int size, int flags)
449 {
450 	struct mount *mp;
451 
452 	mp = (struct mount *)mem;
453 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
454 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
455 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
456 	return (0);
457 }
458 
459 static void
460 mount_fini(void *mem, int size)
461 {
462 	struct mount *mp;
463 
464 	mp = (struct mount *)mem;
465 	lockdestroy(&mp->mnt_explock);
466 	lockdestroy(&mp->mnt_lock);
467 	mtx_destroy(&mp->mnt_mtx);
468 }
469 
470 /*
471  * Allocate and initialize the mount point struct.
472  */
473 struct mount *
474 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
475     const char *fspath, struct thread *td)
476 {
477 	struct mount *mp;
478 
479 	mp = uma_zalloc(mount_zone, M_WAITOK);
480 	bzero(&mp->mnt_startzero,
481 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
482 	TAILQ_INIT(&mp->mnt_nvnodelist);
483 	mp->mnt_nvnodelistsize = 0;
484 	mp->mnt_ref = 0;
485 	(void) vfs_busy(mp, LK_NOWAIT, 0, td);
486 	mp->mnt_op = vfsp->vfc_vfsops;
487 	mp->mnt_vfc = vfsp;
488 	vfsp->vfc_refcount++;	/* XXX Unlocked */
489 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
490 	mp->mnt_gen++;
491 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
492 	mp->mnt_vnodecovered = vp;
493 	mp->mnt_cred = crdup(td->td_ucred);
494 	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
495 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
496 	mp->mnt_iosize_max = DFLTPHYS;
497 #ifdef MAC
498 	mac_mount_init(mp);
499 	mac_mount_create(td->td_ucred, mp);
500 #endif
501 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
502 	return (mp);
503 }
504 
505 /*
506  * Destroy the mount struct previously allocated by vfs_mount_alloc().
507  */
508 void
509 vfs_mount_destroy(struct mount *mp)
510 {
511 	int i;
512 
513 	MNT_ILOCK(mp);
514 	for (i = 0; mp->mnt_ref && i < 3; i++)
515 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
516 	/*
517 	 * This will always cause a 3 second delay in rebooting due to
518 	 * refs on the root mountpoint that never go away.  Most of these
519 	 * are held by init which never exits.
520 	 */
521 	if (i == 3 && (!rebooting || bootverbose))
522 		printf("Mount point %s had %d dangling refs\n",
523 		    mp->mnt_stat.f_mntonname, mp->mnt_ref);
524 	if (mp->mnt_holdcnt != 0) {
525 		printf("Waiting for mount point to be unheld\n");
526 		while (mp->mnt_holdcnt != 0) {
527 			mp->mnt_holdcntwaiters++;
528 			msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
529 			       PZERO, "mntdestroy", 0);
530 			mp->mnt_holdcntwaiters--;
531 		}
532 		printf("mount point unheld\n");
533 	}
534 	if (mp->mnt_writeopcount > 0) {
535 		printf("Waiting for mount point write ops\n");
536 		while (mp->mnt_writeopcount > 0) {
537 			mp->mnt_kern_flag |= MNTK_SUSPEND;
538 			msleep(&mp->mnt_writeopcount,
539 			       MNT_MTX(mp),
540 			       PZERO, "mntdestroy2", 0);
541 		}
542 		printf("mount point write ops completed\n");
543 	}
544 	if (mp->mnt_secondary_writes > 0) {
545 		printf("Waiting for mount point secondary write ops\n");
546 		while (mp->mnt_secondary_writes > 0) {
547 			mp->mnt_kern_flag |= MNTK_SUSPEND;
548 			msleep(&mp->mnt_secondary_writes,
549 			       MNT_MTX(mp),
550 			       PZERO, "mntdestroy3", 0);
551 		}
552 		printf("mount point secondary write ops completed\n");
553 	}
554 	MNT_IUNLOCK(mp);
555 	mp->mnt_vfc->vfc_refcount--;
556 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
557 		struct vnode *vp;
558 
559 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
560 			vprint("", vp);
561 		panic("unmount: dangling vnode");
562 	}
563 	MNT_ILOCK(mp);
564 	if (mp->mnt_kern_flag & MNTK_MWAIT)
565 		wakeup(mp);
566 	if (mp->mnt_writeopcount != 0)
567 		panic("vfs_mount_destroy: nonzero writeopcount");
568 	if (mp->mnt_secondary_writes != 0)
569 		panic("vfs_mount_destroy: nonzero secondary_writes");
570 	if (mp->mnt_nvnodelistsize != 0)
571 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
572 	mp->mnt_writeopcount = -1000;
573 	mp->mnt_nvnodelistsize = -1000;
574 	mp->mnt_secondary_writes = -1000;
575 	MNT_IUNLOCK(mp);
576 #ifdef MAC
577 	mac_mount_destroy(mp);
578 #endif
579 	if (mp->mnt_opt != NULL)
580 		vfs_freeopts(mp->mnt_opt);
581 	crfree(mp->mnt_cred);
582 	uma_zfree(mount_zone, mp);
583 }
584 
585 static int
586 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
587 {
588 	struct vfsoptlist *optlist;
589 	struct vfsopt *opt, *noro_opt;
590 	char *fstype, *fspath, *errmsg;
591 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
592 	int has_rw, has_noro;
593 
594 	errmsg = NULL;
595 	errmsg_len = 0;
596 	errmsg_pos = -1;
597 	has_rw = 0;
598 	has_noro = 0;
599 
600 	error = vfs_buildopts(fsoptions, &optlist);
601 	if (error)
602 		return (error);
603 
604 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
605 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
606 
607 	/*
608 	 * We need these two options before the others,
609 	 * and they are mandatory for any filesystem.
610 	 * Ensure they are NUL terminated as well.
611 	 */
612 	fstypelen = 0;
613 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
614 	if (error || fstype[fstypelen - 1] != '\0') {
615 		error = EINVAL;
616 		if (errmsg != NULL)
617 			strncpy(errmsg, "Invalid fstype", errmsg_len);
618 		goto bail;
619 	}
620 	fspathlen = 0;
621 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
622 	if (error || fspath[fspathlen - 1] != '\0') {
623 		error = EINVAL;
624 		if (errmsg != NULL)
625 			strncpy(errmsg, "Invalid fspath", errmsg_len);
626 		goto bail;
627 	}
628 
629 	/*
630 	 * We need to see if we have the "update" option
631 	 * before we call vfs_domount(), since vfs_domount() has special
632 	 * logic based on MNT_UPDATE.  This is very important
633 	 * when we want to update the root filesystem.
634 	 */
635 	TAILQ_FOREACH(opt, optlist, link) {
636 		if (strcmp(opt->name, "update") == 0)
637 			fsflags |= MNT_UPDATE;
638 		else if (strcmp(opt->name, "async") == 0)
639 			fsflags |= MNT_ASYNC;
640 		else if (strcmp(opt->name, "force") == 0)
641 			fsflags |= MNT_FORCE;
642 		else if (strcmp(opt->name, "multilabel") == 0)
643 			fsflags |= MNT_MULTILABEL;
644 		else if (strcmp(opt->name, "noasync") == 0)
645 			fsflags &= ~MNT_ASYNC;
646 		else if (strcmp(opt->name, "noatime") == 0)
647 			fsflags |= MNT_NOATIME;
648 		else if (strcmp(opt->name, "atime") == 0) {
649 			free(opt->name, M_MOUNT);
650 			opt->name = strdup("nonoatime", M_MOUNT);
651 		}
652 		else if (strcmp(opt->name, "noclusterr") == 0)
653 			fsflags |= MNT_NOCLUSTERR;
654 		else if (strcmp(opt->name, "clusterr") == 0) {
655 			free(opt->name, M_MOUNT);
656 			opt->name = strdup("nonoclusterr", M_MOUNT);
657 		}
658 		else if (strcmp(opt->name, "noclusterw") == 0)
659 			fsflags |= MNT_NOCLUSTERW;
660 		else if (strcmp(opt->name, "clusterw") == 0) {
661 			free(opt->name, M_MOUNT);
662 			opt->name = strdup("nonoclusterw", M_MOUNT);
663 		}
664 		else if (strcmp(opt->name, "noexec") == 0)
665 			fsflags |= MNT_NOEXEC;
666 		else if (strcmp(opt->name, "exec") == 0) {
667 			free(opt->name, M_MOUNT);
668 			opt->name = strdup("nonoexec", M_MOUNT);
669 		}
670 		else if (strcmp(opt->name, "nosuid") == 0)
671 			fsflags |= MNT_NOSUID;
672 		else if (strcmp(opt->name, "suid") == 0) {
673 			free(opt->name, M_MOUNT);
674 			opt->name = strdup("nonosuid", M_MOUNT);
675 		}
676 		else if (strcmp(opt->name, "nosymfollow") == 0)
677 			fsflags |= MNT_NOSYMFOLLOW;
678 		else if (strcmp(opt->name, "symfollow") == 0) {
679 			free(opt->name, M_MOUNT);
680 			opt->name = strdup("nonosymfollow", M_MOUNT);
681 		}
682 		else if (strcmp(opt->name, "noro") == 0) {
683 			fsflags &= ~MNT_RDONLY;
684 			has_noro = 1;
685 		}
686 		else if (strcmp(opt->name, "rw") == 0) {
687 			fsflags &= ~MNT_RDONLY;
688 			has_rw = 1;
689 		}
690 		else if (strcmp(opt->name, "ro") == 0)
691 			fsflags |= MNT_RDONLY;
692 		else if (strcmp(opt->name, "rdonly") == 0) {
693 			free(opt->name, M_MOUNT);
694 			opt->name = strdup("ro", M_MOUNT);
695 			fsflags |= MNT_RDONLY;
696 		}
697 		else if (strcmp(opt->name, "suiddir") == 0)
698 			fsflags |= MNT_SUIDDIR;
699 		else if (strcmp(opt->name, "sync") == 0)
700 			fsflags |= MNT_SYNCHRONOUS;
701 		else if (strcmp(opt->name, "union") == 0)
702 			fsflags |= MNT_UNION;
703 	}
704 
705 	/*
706 	 * If "rw" was specified as a mount option, and we
707 	 * are trying to update a mount-point from "ro" to "rw",
708 	 * we need a mount option "noro", since in vfs_mergeopts(),
709 	 * "noro" will cancel "ro", but "rw" will not do anything.
710 	 */
711 	if (has_rw && !has_noro) {
712 		noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
713 		noro_opt->name = strdup("noro", M_MOUNT);
714 		noro_opt->value = NULL;
715 		noro_opt->len = 0;
716 		TAILQ_INSERT_TAIL(optlist, noro_opt, link);
717 	}
718 
719 	/*
720 	 * Be ultra-paranoid about making sure the type and fspath
721 	 * variables will fit in our mp buffers, including the
722 	 * terminating NUL.
723 	 */
724 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
725 		error = ENAMETOOLONG;
726 		goto bail;
727 	}
728 
729 	mtx_lock(&Giant);
730 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
731 	mtx_unlock(&Giant);
732 bail:
733 	/* copyout the errmsg */
734 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
735 	    && errmsg_len > 0 && errmsg != NULL) {
736 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
737 			bcopy(errmsg,
738 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
739 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
740 		} else {
741 			copyout(errmsg,
742 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
743 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
744 		}
745 	}
746 
747 	if (error != 0)
748 		vfs_freeopts(optlist);
749 	return (error);
750 }
751 
752 /*
753  * Old mount API.
754  */
755 #ifndef _SYS_SYSPROTO_H_
756 struct mount_args {
757 	char	*type;
758 	char	*path;
759 	int	flags;
760 	caddr_t	data;
761 };
762 #endif
763 /* ARGSUSED */
764 int
765 mount(td, uap)
766 	struct thread *td;
767 	struct mount_args /* {
768 		char *type;
769 		char *path;
770 		int flags;
771 		caddr_t data;
772 	} */ *uap;
773 {
774 	char *fstype;
775 	struct vfsconf *vfsp = NULL;
776 	struct mntarg *ma = NULL;
777 	int error;
778 
779 	AUDIT_ARG(fflags, uap->flags);
780 
781 	/*
782 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
783 	 * userspace to set this flag, but we must filter it out if we want
784 	 * MNT_UPDATE on the root file system to work.
785 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
786 	 */
787 	uap->flags &= ~MNT_ROOTFS;
788 
789 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
790 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
791 	if (error) {
792 		free(fstype, M_TEMP);
793 		return (error);
794 	}
795 
796 	AUDIT_ARG(text, fstype);
797 	mtx_lock(&Giant);
798 	vfsp = vfs_byname_kld(fstype, td, &error);
799 	free(fstype, M_TEMP);
800 	if (vfsp == NULL) {
801 		mtx_unlock(&Giant);
802 		return (ENOENT);
803 	}
804 	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
805 		mtx_unlock(&Giant);
806 		return (EOPNOTSUPP);
807 	}
808 
809 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
810 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
811 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
812 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
813 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
814 
815 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
816 	mtx_unlock(&Giant);
817 	return (error);
818 }
819 
820 
821 /*
822  * vfs_domount(): actually attempt a filesystem mount.
823  */
824 static int
825 vfs_domount(
826 	struct thread *td,	/* Calling thread. */
827 	const char *fstype,	/* Filesystem type. */
828 	char *fspath,		/* Mount path. */
829 	int fsflags,		/* Flags common to all filesystems. */
830 	void *fsdata		/* Options local to the filesystem. */
831 	)
832 {
833 	struct vnode *vp;
834 	struct mount *mp;
835 	struct vfsconf *vfsp;
836 	struct export_args export;
837 	int error, flag = 0;
838 	struct vattr va;
839 	struct nameidata nd;
840 
841 	mtx_assert(&Giant, MA_OWNED);
842 	/*
843 	 * Be ultra-paranoid about making sure the type and fspath
844 	 * variables will fit in our mp buffers, including the
845 	 * terminating NUL.
846 	 */
847 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
848 		return (ENAMETOOLONG);
849 
850 	if (jailed(td->td_ucred) || usermount == 0) {
851 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
852 			return (error);
853 	}
854 
855 	/*
856 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
857 	 */
858 	if (fsflags & MNT_EXPORTED) {
859 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
860 		if (error)
861 			return (error);
862 	}
863 	if (fsflags & MNT_SUIDDIR) {
864 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
865 		if (error)
866 			return (error);
867 	}
868 	/*
869 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
870 	 */
871 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
872 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
873 			fsflags |= MNT_NOSUID | MNT_USER;
874 	}
875 
876 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
877 	vfsp = NULL;
878 	if ((fsflags & MNT_UPDATE) == 0) {
879 		/* Don't try to load KLDs if we're mounting the root. */
880 		if (fsflags & MNT_ROOTFS)
881 			vfsp = vfs_byname(fstype);
882 		else
883 			vfsp = vfs_byname_kld(fstype, td, &error);
884 		if (vfsp == NULL)
885 			return (ENODEV);
886 		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
887 			return (EPERM);
888 	}
889 	/*
890 	 * Get vnode to be covered
891 	 */
892 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
893 	    fspath, td);
894 	if ((error = namei(&nd)) != 0)
895 		return (error);
896 	NDFREE(&nd, NDF_ONLY_PNBUF);
897 	vp = nd.ni_vp;
898 	if (fsflags & MNT_UPDATE) {
899 		if ((vp->v_vflag & VV_ROOT) == 0) {
900 			vput(vp);
901 			return (EINVAL);
902 		}
903 		mp = vp->v_mount;
904 		MNT_ILOCK(mp);
905 		flag = mp->mnt_flag;
906 		/*
907 		 * We only allow the filesystem to be reloaded if it
908 		 * is currently mounted read-only.
909 		 */
910 		if ((fsflags & MNT_RELOAD) &&
911 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
912 			MNT_IUNLOCK(mp);
913 			vput(vp);
914 			return (EOPNOTSUPP);	/* Needs translation */
915 		}
916 		MNT_IUNLOCK(mp);
917 		/*
918 		 * Only privileged root, or (if MNT_USER is set) the user that
919 		 * did the original mount is permitted to update it.
920 		 */
921 		error = vfs_suser(mp, td);
922 		if (error) {
923 			vput(vp);
924 			return (error);
925 		}
926 		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
927 			vput(vp);
928 			return (EBUSY);
929 		}
930 		VI_LOCK(vp);
931 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
932 		    vp->v_mountedhere != NULL) {
933 			VI_UNLOCK(vp);
934 			vfs_unbusy(mp, td);
935 			vput(vp);
936 			return (EBUSY);
937 		}
938 		vp->v_iflag |= VI_MOUNT;
939 		VI_UNLOCK(vp);
940 		MNT_ILOCK(mp);
941 		mp->mnt_flag |= fsflags &
942 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
943 		MNT_IUNLOCK(mp);
944 		VOP_UNLOCK(vp, 0);
945 		mp->mnt_optnew = fsdata;
946 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
947 	} else {
948 		/*
949 		 * If the user is not root, ensure that they own the directory
950 		 * onto which we are attempting to mount.
951 		 */
952 		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
953 		if (error) {
954 			vput(vp);
955 			return (error);
956 		}
957 		if (va.va_uid != td->td_ucred->cr_uid) {
958 			error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
959 			    0);
960 			if (error) {
961 				vput(vp);
962 				return (error);
963 			}
964 		}
965 		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
966 		if (error != 0) {
967 			vput(vp);
968 			return (error);
969 		}
970 		if (vp->v_type != VDIR) {
971 			vput(vp);
972 			return (ENOTDIR);
973 		}
974 		VI_LOCK(vp);
975 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
976 		    vp->v_mountedhere != NULL) {
977 			VI_UNLOCK(vp);
978 			vput(vp);
979 			return (EBUSY);
980 		}
981 		vp->v_iflag |= VI_MOUNT;
982 		VI_UNLOCK(vp);
983 
984 		/*
985 		 * Allocate and initialize the filesystem.
986 		 */
987 		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
988 		VOP_UNLOCK(vp, 0);
989 
990 		/* XXXMAC: pass to vfs_mount_alloc? */
991 		mp->mnt_optnew = fsdata;
992 	}
993 
994 	/*
995 	 * Set the mount level flags.
996 	 */
997 	MNT_ILOCK(mp);
998 	mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
999 		(fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
1000 			    MNT_RDONLY));
1001 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1002 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1003 	MNT_IUNLOCK(mp);
1004 	/*
1005 	 * Mount the filesystem.
1006 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1007 	 * get.  No freeing of cn_pnbuf.
1008 	 */
1009         error = VFS_MOUNT(mp, td);
1010 
1011 	/*
1012 	 * Process the export option only if we are
1013 	 * updating mount options.
1014 	 */
1015 	if (!error && (fsflags & MNT_UPDATE)) {
1016 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
1017 		    sizeof(export)) == 0)
1018 			error = vfs_export(mp, &export);
1019 	}
1020 
1021 	if (!error) {
1022 		if (mp->mnt_opt != NULL)
1023 			vfs_freeopts(mp->mnt_opt);
1024 		mp->mnt_opt = mp->mnt_optnew;
1025 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
1026 	}
1027 	/*
1028 	 * Prevent external consumers of mount options from reading
1029 	 * mnt_optnew.
1030 	*/
1031 	mp->mnt_optnew = NULL;
1032 	if (mp->mnt_flag & MNT_UPDATE) {
1033 		MNT_ILOCK(mp);
1034 		if (error)
1035 			mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1036 				(flag & ~MNT_QUOTA);
1037 		else
1038 			mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD |
1039 					  MNT_FORCE | MNT_SNAPSHOT);
1040 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1041 			mp->mnt_kern_flag |= MNTK_ASYNC;
1042 		else
1043 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1044 		MNT_IUNLOCK(mp);
1045 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1046 			if (mp->mnt_syncer == NULL)
1047 				error = vfs_allocate_syncvnode(mp);
1048 		} else {
1049 			if (mp->mnt_syncer != NULL)
1050 				vrele(mp->mnt_syncer);
1051 			mp->mnt_syncer = NULL;
1052 		}
1053 		vfs_unbusy(mp, td);
1054 		VI_LOCK(vp);
1055 		vp->v_iflag &= ~VI_MOUNT;
1056 		VI_UNLOCK(vp);
1057 		vrele(vp);
1058 		return (error);
1059 	}
1060 	MNT_ILOCK(mp);
1061 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1062 		mp->mnt_kern_flag |= MNTK_ASYNC;
1063 	else
1064 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1065 	MNT_IUNLOCK(mp);
1066 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1067 	/*
1068 	 * Put the new filesystem on the mount list after root.
1069 	 */
1070 	cache_purge(vp);
1071 	if (!error) {
1072 		struct vnode *newdp;
1073 
1074 		VI_LOCK(vp);
1075 		vp->v_iflag &= ~VI_MOUNT;
1076 		VI_UNLOCK(vp);
1077 		vp->v_mountedhere = mp;
1078 		mtx_lock(&mountlist_mtx);
1079 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1080 		mtx_unlock(&mountlist_mtx);
1081 		vfs_event_signal(NULL, VQ_MOUNT, 0);
1082 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1083 			panic("mount: lost mount");
1084 		mountcheckdirs(vp, newdp);
1085 		vput(newdp);
1086 		VOP_UNLOCK(vp, 0);
1087 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1088 			error = vfs_allocate_syncvnode(mp);
1089 		vfs_unbusy(mp, td);
1090 		if (error)
1091 			vrele(vp);
1092 	} else {
1093 		VI_LOCK(vp);
1094 		vp->v_iflag &= ~VI_MOUNT;
1095 		VI_UNLOCK(vp);
1096 		vfs_unbusy(mp, td);
1097 		vfs_mount_destroy(mp);
1098 		vput(vp);
1099 	}
1100 	return (error);
1101 }
1102 
1103 /*
1104  * Unmount a filesystem.
1105  *
1106  * Note: unmount takes a path to the vnode mounted on as argument, not
1107  * special file (as before).
1108  */
1109 #ifndef _SYS_SYSPROTO_H_
1110 struct unmount_args {
1111 	char	*path;
1112 	int	flags;
1113 };
1114 #endif
1115 /* ARGSUSED */
1116 int
1117 unmount(td, uap)
1118 	struct thread *td;
1119 	register struct unmount_args /* {
1120 		char *path;
1121 		int flags;
1122 	} */ *uap;
1123 {
1124 	struct mount *mp;
1125 	char *pathbuf;
1126 	int error, id0, id1;
1127 
1128 	if (jailed(td->td_ucred) || usermount == 0) {
1129 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1130 		if (error)
1131 			return (error);
1132 	}
1133 
1134 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1135 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1136 	if (error) {
1137 		free(pathbuf, M_TEMP);
1138 		return (error);
1139 	}
1140 	AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1141 	mtx_lock(&Giant);
1142 	if (uap->flags & MNT_BYFSID) {
1143 		/* Decode the filesystem ID. */
1144 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1145 			mtx_unlock(&Giant);
1146 			free(pathbuf, M_TEMP);
1147 			return (EINVAL);
1148 		}
1149 
1150 		mtx_lock(&mountlist_mtx);
1151 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1152 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1153 			    mp->mnt_stat.f_fsid.val[1] == id1)
1154 				break;
1155 		}
1156 		mtx_unlock(&mountlist_mtx);
1157 	} else {
1158 		mtx_lock(&mountlist_mtx);
1159 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1160 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1161 				break;
1162 		}
1163 		mtx_unlock(&mountlist_mtx);
1164 	}
1165 	free(pathbuf, M_TEMP);
1166 	if (mp == NULL) {
1167 		/*
1168 		 * Previously we returned ENOENT for a nonexistent path and
1169 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1170 		 * now, so in the !MNT_BYFSID case return the more likely
1171 		 * EINVAL for compatibility.
1172 		 */
1173 		mtx_unlock(&Giant);
1174 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1175 	}
1176 
1177 	/*
1178 	 * Don't allow unmounting the root filesystem.
1179 	 */
1180 	if (mp->mnt_flag & MNT_ROOTFS) {
1181 		mtx_unlock(&Giant);
1182 		return (EINVAL);
1183 	}
1184 	error = dounmount(mp, uap->flags, td);
1185 	mtx_unlock(&Giant);
1186 	return (error);
1187 }
1188 
1189 /*
1190  * Do the actual filesystem unmount.
1191  */
1192 int
1193 dounmount(mp, flags, td)
1194 	struct mount *mp;
1195 	int flags;
1196 	struct thread *td;
1197 {
1198 	struct vnode *coveredvp, *fsrootvp;
1199 	int error;
1200 	int async_flag;
1201 	int mnt_gen_r;
1202 
1203 	mtx_assert(&Giant, MA_OWNED);
1204 
1205 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1206 		mnt_gen_r = mp->mnt_gen;
1207 		VI_LOCK(coveredvp);
1208 		vholdl(coveredvp);
1209 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1210 		vdrop(coveredvp);
1211 		/*
1212 		 * Check for mp being unmounted while waiting for the
1213 		 * covered vnode lock.
1214 		 */
1215 		if (coveredvp->v_mountedhere != mp ||
1216 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1217 			VOP_UNLOCK(coveredvp, 0);
1218 			return (EBUSY);
1219 		}
1220 	}
1221 	/*
1222 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1223 	 * original mount is permitted to unmount this filesystem.
1224 	 */
1225 	error = vfs_suser(mp, td);
1226 	if (error) {
1227 		if (coveredvp)
1228 			VOP_UNLOCK(coveredvp, 0);
1229 		return (error);
1230 	}
1231 
1232 	MNT_ILOCK(mp);
1233 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1234 		MNT_IUNLOCK(mp);
1235 		if (coveredvp)
1236 			VOP_UNLOCK(coveredvp, 0);
1237 		return (EBUSY);
1238 	}
1239 	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1240 	/* Allow filesystems to detect that a forced unmount is in progress. */
1241 	if (flags & MNT_FORCE)
1242 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1243 	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1244 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp));
1245 	if (error) {
1246 		MNT_ILOCK(mp);
1247 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ |
1248 		    MNTK_UNMOUNTF);
1249 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1250 			wakeup(mp);
1251 		MNT_IUNLOCK(mp);
1252 		if (coveredvp)
1253 			VOP_UNLOCK(coveredvp, 0);
1254 		return (error);
1255 	}
1256 	vn_start_write(NULL, &mp, V_WAIT);
1257 
1258 	if (mp->mnt_flag & MNT_EXPUBLIC)
1259 		vfs_setpublicfs(NULL, NULL, NULL);
1260 
1261 	vfs_msync(mp, MNT_WAIT);
1262 	MNT_ILOCK(mp);
1263 	async_flag = mp->mnt_flag & MNT_ASYNC;
1264 	mp->mnt_flag &= ~MNT_ASYNC;
1265 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1266 	MNT_IUNLOCK(mp);
1267 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1268 	if (mp->mnt_syncer != NULL)
1269 		vrele(mp->mnt_syncer);
1270 	/*
1271 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1272 	 * vnode to the covered vnode.  For non-forced unmounts we want
1273 	 * such references to cause an EBUSY error.
1274 	 */
1275 	if ((flags & MNT_FORCE) &&
1276 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1277 		if (mp->mnt_vnodecovered != NULL)
1278 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1279 		if (fsrootvp == rootvnode) {
1280 			vrele(rootvnode);
1281 			rootvnode = NULL;
1282 		}
1283 		vput(fsrootvp);
1284 	}
1285 	if (((mp->mnt_flag & MNT_RDONLY) ||
1286 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1287 	    (flags & MNT_FORCE)) {
1288 		error = VFS_UNMOUNT(mp, flags, td);
1289 	}
1290 	vn_finished_write(mp);
1291 	/*
1292 	 * If we failed to flush the dirty blocks for this mount point,
1293 	 * undo all the cdir/rdir and rootvnode changes we made above.
1294 	 * Unless we failed to do so because the device is reporting that
1295 	 * it doesn't exist anymore.
1296 	 */
1297 	if (error && error != ENXIO) {
1298 		if ((flags & MNT_FORCE) &&
1299 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1300 			if (mp->mnt_vnodecovered != NULL)
1301 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1302 			if (rootvnode == NULL) {
1303 				rootvnode = fsrootvp;
1304 				vref(rootvnode);
1305 			}
1306 			vput(fsrootvp);
1307 		}
1308 		MNT_ILOCK(mp);
1309 		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1310 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) {
1311 			MNT_IUNLOCK(mp);
1312 			(void) vfs_allocate_syncvnode(mp);
1313 			MNT_ILOCK(mp);
1314 		}
1315 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1316 		mp->mnt_flag |= async_flag;
1317 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1318 			mp->mnt_kern_flag |= MNTK_ASYNC;
1319 		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1320 		if (mp->mnt_kern_flag & MNTK_MWAIT)
1321 			wakeup(mp);
1322 		MNT_IUNLOCK(mp);
1323 		if (coveredvp)
1324 			VOP_UNLOCK(coveredvp, 0);
1325 		return (error);
1326 	}
1327 	mtx_lock(&mountlist_mtx);
1328 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1329 	mtx_unlock(&mountlist_mtx);
1330 	if (coveredvp != NULL) {
1331 		coveredvp->v_mountedhere = NULL;
1332 		vput(coveredvp);
1333 	}
1334 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1335 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1336 	vfs_mount_destroy(mp);
1337 	return (0);
1338 }
1339 
1340 /*
1341  * ---------------------------------------------------------------------
1342  * Mounting of root filesystem
1343  *
1344  */
1345 
1346 struct root_hold_token {
1347 	const char			*who;
1348 	LIST_ENTRY(root_hold_token)	list;
1349 };
1350 
1351 static LIST_HEAD(, root_hold_token)	root_holds =
1352     LIST_HEAD_INITIALIZER(&root_holds);
1353 
1354 static int root_mount_complete;
1355 
1356 /*
1357  * Hold root mount.
1358  */
1359 struct root_hold_token *
1360 root_mount_hold(const char *identifier)
1361 {
1362 	struct root_hold_token *h;
1363 
1364 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1365 	h->who = identifier;
1366 	mtx_lock(&mountlist_mtx);
1367 	LIST_INSERT_HEAD(&root_holds, h, list);
1368 	mtx_unlock(&mountlist_mtx);
1369 	return (h);
1370 }
1371 
1372 /*
1373  * Release root mount.
1374  */
1375 void
1376 root_mount_rel(struct root_hold_token *h)
1377 {
1378 
1379 	mtx_lock(&mountlist_mtx);
1380 	LIST_REMOVE(h, list);
1381 	wakeup(&root_holds);
1382 	mtx_unlock(&mountlist_mtx);
1383 	free(h, M_DEVBUF);
1384 }
1385 
1386 /*
1387  * Wait for all subsystems to release root mount.
1388  */
1389 static void
1390 root_mount_prepare(void)
1391 {
1392 	struct root_hold_token *h;
1393 
1394 	for (;;) {
1395 		DROP_GIANT();
1396 		g_waitidle();
1397 		PICKUP_GIANT();
1398 		mtx_lock(&mountlist_mtx);
1399 		if (LIST_EMPTY(&root_holds)) {
1400 			mtx_unlock(&mountlist_mtx);
1401 			break;
1402 		}
1403 		printf("Root mount waiting for:");
1404 		LIST_FOREACH(h, &root_holds, list)
1405 			printf(" %s", h->who);
1406 		printf("\n");
1407 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1408 		    hz);
1409 	}
1410 }
1411 
1412 /*
1413  * Root was mounted, share the good news.
1414  */
1415 static void
1416 root_mount_done(void)
1417 {
1418 
1419 	/*
1420 	 * Use a mutex to prevent the wakeup being missed and waiting for
1421 	 * an extra 1 second sleep.
1422 	 */
1423 	mtx_lock(&mountlist_mtx);
1424 	root_mount_complete = 1;
1425 	wakeup(&root_mount_complete);
1426 	mtx_unlock(&mountlist_mtx);
1427 }
1428 
1429 /*
1430  * Return true if root is already mounted.
1431  */
1432 int
1433 root_mounted(void)
1434 {
1435 
1436 	/* No mutex is acquired here because int stores are atomic. */
1437 	return (root_mount_complete);
1438 }
1439 
1440 /*
1441  * Wait until root is mounted.
1442  */
1443 void
1444 root_mount_wait(void)
1445 {
1446 
1447 	/*
1448 	 * Panic on an obvious deadlock - the function can't be called from
1449 	 * a thread which is doing the whole SYSINIT stuff.
1450 	 */
1451 	KASSERT(curthread->td_proc->p_pid != 0,
1452 	    ("root_mount_wait: cannot be called from the swapper thread"));
1453 	mtx_lock(&mountlist_mtx);
1454 	while (!root_mount_complete) {
1455 		msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
1456 		    hz);
1457 	}
1458 	mtx_unlock(&mountlist_mtx);
1459 }
1460 
1461 static void
1462 set_rootvnode(struct thread *td)
1463 {
1464 	struct proc *p;
1465 
1466 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1467 		panic("Cannot find root vnode");
1468 
1469 	p = td->td_proc;
1470 	FILEDESC_XLOCK(p->p_fd);
1471 
1472 	if (p->p_fd->fd_cdir != NULL)
1473 		vrele(p->p_fd->fd_cdir);
1474 	p->p_fd->fd_cdir = rootvnode;
1475 	VREF(rootvnode);
1476 
1477 	if (p->p_fd->fd_rdir != NULL)
1478 		vrele(p->p_fd->fd_rdir);
1479 	p->p_fd->fd_rdir = rootvnode;
1480 	VREF(rootvnode);
1481 
1482 	FILEDESC_XUNLOCK(p->p_fd);
1483 
1484 	VOP_UNLOCK(rootvnode, 0);
1485 
1486 	EVENTHANDLER_INVOKE(mountroot);
1487 }
1488 
1489 /*
1490  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1491  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1492  */
1493 
1494 static void
1495 devfs_first(void)
1496 {
1497 	struct thread *td = curthread;
1498 	struct vfsoptlist *opts;
1499 	struct vfsconf *vfsp;
1500 	struct mount *mp = NULL;
1501 	int error;
1502 
1503 	vfsp = vfs_byname("devfs");
1504 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1505 	if (vfsp == NULL)
1506 		return;
1507 
1508 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1509 
1510 	error = VFS_MOUNT(mp, td);
1511 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1512 	if (error)
1513 		return;
1514 
1515 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1516 	TAILQ_INIT(opts);
1517 	mp->mnt_opt = opts;
1518 
1519 	mtx_lock(&mountlist_mtx);
1520 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1521 	mtx_unlock(&mountlist_mtx);
1522 
1523 	set_rootvnode(td);
1524 
1525 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1526 	if (error)
1527 		printf("kern_symlink /dev -> / returns %d\n", error);
1528 }
1529 
1530 /*
1531  * Surgically move our devfs to be mounted on /dev.
1532  */
1533 
1534 static void
1535 devfs_fixup(struct thread *td)
1536 {
1537 	struct nameidata nd;
1538 	int error;
1539 	struct vnode *vp, *dvp;
1540 	struct mount *mp;
1541 
1542 	/* Remove our devfs mount from the mountlist and purge the cache */
1543 	mtx_lock(&mountlist_mtx);
1544 	mp = TAILQ_FIRST(&mountlist);
1545 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1546 	mtx_unlock(&mountlist_mtx);
1547 	cache_purgevfs(mp);
1548 
1549 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1550 	VI_LOCK(dvp);
1551 	dvp->v_iflag &= ~VI_MOUNT;
1552 	VI_UNLOCK(dvp);
1553 	dvp->v_mountedhere = NULL;
1554 
1555 	/* Set up the real rootvnode, and purge the cache */
1556 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1557 	set_rootvnode(td);
1558 	cache_purgevfs(rootvnode->v_mount);
1559 
1560 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1561 	error = namei(&nd);
1562 	if (error) {
1563 		printf("Lookup of /dev for devfs, error: %d\n", error);
1564 		return;
1565 	}
1566 	NDFREE(&nd, NDF_ONLY_PNBUF);
1567 	vp = nd.ni_vp;
1568 	if (vp->v_type != VDIR) {
1569 		vput(vp);
1570 	}
1571 	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1572 	if (error) {
1573 		vput(vp);
1574 	}
1575 	cache_purge(vp);
1576 	mp->mnt_vnodecovered = vp;
1577 	vp->v_mountedhere = mp;
1578 	mtx_lock(&mountlist_mtx);
1579 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1580 	mtx_unlock(&mountlist_mtx);
1581 	VOP_UNLOCK(vp, 0);
1582 	vput(dvp);
1583 	vfs_unbusy(mp, td);
1584 
1585 	/* Unlink the no longer needed /dev/dev -> / symlink */
1586 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1587 }
1588 
1589 /*
1590  * Report errors during filesystem mounting.
1591  */
1592 void
1593 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1594 {
1595 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1596 	va_list ap;
1597 	int error, len;
1598 	char *errmsg;
1599 
1600 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1601 	if (error || errmsg == NULL || len <= 0)
1602 		return;
1603 
1604 	va_start(ap, fmt);
1605 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1606 	va_end(ap);
1607 }
1608 
1609 /*
1610  * Find and mount the root filesystem
1611  */
1612 void
1613 vfs_mountroot(void)
1614 {
1615 	char *cp;
1616 	int error, i, asked = 0;
1617 
1618 	root_mount_prepare();
1619 
1620 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1621 	    NULL, NULL, mount_init, mount_fini,
1622 	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1623 	devfs_first();
1624 
1625 	/*
1626 	 * We are booted with instructions to prompt for the root filesystem.
1627 	 */
1628 	if (boothowto & RB_ASKNAME) {
1629 		if (!vfs_mountroot_ask())
1630 			goto mounted;
1631 		asked = 1;
1632 	}
1633 
1634 	/*
1635 	 * The root filesystem information is compiled in, and we are
1636 	 * booted with instructions to use it.
1637 	 */
1638 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1639 		if (!vfs_mountroot_try(ctrootdevname))
1640 			goto mounted;
1641 		ctrootdevname = NULL;
1642 	}
1643 
1644 	/*
1645 	 * We've been given the generic "use CDROM as root" flag.  This is
1646 	 * necessary because one media may be used in many different
1647 	 * devices, so we need to search for them.
1648 	 */
1649 	if (boothowto & RB_CDROM) {
1650 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1651 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1652 				goto mounted;
1653 		}
1654 	}
1655 
1656 	/*
1657 	 * Try to use the value read by the loader from /etc/fstab, or
1658 	 * supplied via some other means.  This is the preferred
1659 	 * mechanism.
1660 	 */
1661 	cp = getenv("vfs.root.mountfrom");
1662 	if (cp != NULL) {
1663 		error = vfs_mountroot_try(cp);
1664 		freeenv(cp);
1665 		if (!error)
1666 			goto mounted;
1667 	}
1668 
1669 	/*
1670 	 * Try values that may have been computed by code during boot
1671 	 */
1672 	if (!vfs_mountroot_try(rootdevnames[0]))
1673 		goto mounted;
1674 	if (!vfs_mountroot_try(rootdevnames[1]))
1675 		goto mounted;
1676 
1677 	/*
1678 	 * If we (still) have a compiled-in default, try it.
1679 	 */
1680 	if (ctrootdevname != NULL)
1681 		if (!vfs_mountroot_try(ctrootdevname))
1682 			goto mounted;
1683 	/*
1684 	 * Everything so far has failed, prompt on the console if we haven't
1685 	 * already tried that.
1686 	 */
1687 	if (!asked)
1688 		if (!vfs_mountroot_ask())
1689 			goto mounted;
1690 
1691 	panic("Root mount failed, startup aborted.");
1692 
1693 mounted:
1694 	root_mount_done();
1695 }
1696 
1697 /*
1698  * Mount (mountfrom) as the root filesystem.
1699  */
1700 static int
1701 vfs_mountroot_try(const char *mountfrom)
1702 {
1703 	struct mount	*mp;
1704 	char		*vfsname, *path;
1705 	time_t		timebase;
1706 	int		error;
1707 	char		patt[32];
1708 
1709 	vfsname = NULL;
1710 	path    = NULL;
1711 	mp      = NULL;
1712 	error   = EINVAL;
1713 
1714 	if (mountfrom == NULL)
1715 		return (error);		/* don't complain */
1716 	printf("Trying to mount root from %s\n", mountfrom);
1717 
1718 	/* parse vfs name and path */
1719 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1720 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1721 	vfsname[0] = path[0] = 0;
1722 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1723 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1724 		goto out;
1725 
1726 	if (path[0] == '\0')
1727 		strcpy(path, ROOTNAME);
1728 
1729 	error = kernel_vmount(
1730 	    MNT_RDONLY | MNT_ROOTFS,
1731 	    "fstype", vfsname,
1732 	    "fspath", "/",
1733 	    "from", path,
1734 	    NULL);
1735 	if (error == 0) {
1736 		/*
1737 		 * We mount devfs prior to mounting the / FS, so the first
1738 		 * entry will typically be devfs.
1739 		 */
1740 		mp = TAILQ_FIRST(&mountlist);
1741 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1742 
1743 		/*
1744 		 * Iterate over all currently mounted file systems and use
1745 		 * the time stamp found to check and/or initialize the RTC.
1746 		 * Typically devfs has no time stamp and the only other FS
1747 		 * is the actual / FS.
1748 		 * Call inittodr() only once and pass it the largest of the
1749 		 * timestamps we encounter.
1750 		 */
1751 		timebase = 0;
1752 		do {
1753 			if (mp->mnt_time > timebase)
1754 				timebase = mp->mnt_time;
1755 			mp = TAILQ_NEXT(mp, mnt_list);
1756 		} while (mp != NULL);
1757 		inittodr(timebase);
1758 
1759 		devfs_fixup(curthread);
1760 	}
1761 out:
1762 	free(path, M_MOUNT);
1763 	free(vfsname, M_MOUNT);
1764 	return (error);
1765 }
1766 
1767 /*
1768  * ---------------------------------------------------------------------
1769  * Interactive root filesystem selection code.
1770  */
1771 
1772 static int
1773 vfs_mountroot_ask(void)
1774 {
1775 	char name[128];
1776 
1777 	for(;;) {
1778 		printf("\nManual root filesystem specification:\n");
1779 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1780 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1781 		printf("                       eg. ufs:da0s1a\n");
1782 #else
1783 		printf("                       eg. ufs:/dev/da0a\n");
1784 #endif
1785 		printf("  ?                  List valid disk boot devices\n");
1786 		printf("  <empty line>       Abort manual input\n");
1787 		printf("\nmountroot> ");
1788 		gets(name, sizeof(name), 1);
1789 		if (name[0] == '\0')
1790 			return (1);
1791 		if (name[0] == '?') {
1792 			printf("\nList of GEOM managed disk devices:\n  ");
1793 			g_dev_print();
1794 			continue;
1795 		}
1796 		if (!vfs_mountroot_try(name))
1797 			return (0);
1798 	}
1799 }
1800 
1801 /*
1802  * ---------------------------------------------------------------------
1803  * Functions for querying mount options/arguments from filesystems.
1804  */
1805 
1806 /*
1807  * Check that no unknown options are given
1808  */
1809 int
1810 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1811 {
1812 	struct vfsopt *opt;
1813 	char errmsg[255];
1814 	const char **t, *p, *q;
1815 	int ret = 0;
1816 
1817 	TAILQ_FOREACH(opt, opts, link) {
1818 		p = opt->name;
1819 		q = NULL;
1820 		if (p[0] == 'n' && p[1] == 'o')
1821 			q = p + 2;
1822 		for(t = global_opts; *t != NULL; t++) {
1823 			if (strcmp(*t, p) == 0)
1824 				break;
1825 			if (q != NULL) {
1826 				if (strcmp(*t, q) == 0)
1827 					break;
1828 			}
1829 		}
1830 		if (*t != NULL)
1831 			continue;
1832 		for(t = legal; *t != NULL; t++) {
1833 			if (strcmp(*t, p) == 0)
1834 				break;
1835 			if (q != NULL) {
1836 				if (strcmp(*t, q) == 0)
1837 					break;
1838 			}
1839 		}
1840 		if (*t != NULL)
1841 			continue;
1842 		sprintf(errmsg, "mount option <%s> is unknown", p);
1843 		printf("%s\n", errmsg);
1844 		ret = EINVAL;
1845 	}
1846 	if (ret != 0) {
1847 		TAILQ_FOREACH(opt, opts, link) {
1848 			if (strcmp(opt->name, "errmsg") == 0) {
1849 				strncpy((char *)opt->value, errmsg, opt->len);
1850 			}
1851 		}
1852 	}
1853 	return (ret);
1854 }
1855 
1856 /*
1857  * Get a mount option by its name.
1858  *
1859  * Return 0 if the option was found, ENOENT otherwise.
1860  * If len is non-NULL it will be filled with the length
1861  * of the option. If buf is non-NULL, it will be filled
1862  * with the address of the option.
1863  */
1864 int
1865 vfs_getopt(opts, name, buf, len)
1866 	struct vfsoptlist *opts;
1867 	const char *name;
1868 	void **buf;
1869 	int *len;
1870 {
1871 	struct vfsopt *opt;
1872 
1873 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1874 
1875 	TAILQ_FOREACH(opt, opts, link) {
1876 		if (strcmp(name, opt->name) == 0) {
1877 			if (len != NULL)
1878 				*len = opt->len;
1879 			if (buf != NULL)
1880 				*buf = opt->value;
1881 			return (0);
1882 		}
1883 	}
1884 	return (ENOENT);
1885 }
1886 
1887 static int
1888 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1889 {
1890 	struct vfsopt *opt;
1891 	int i;
1892 
1893 	if (opts == NULL)
1894 		return (-1);
1895 
1896 	i = 0;
1897 	TAILQ_FOREACH(opt, opts, link) {
1898 		if (strcmp(name, opt->name) == 0)
1899 			return (i);
1900 		++i;
1901 	}
1902 	return (-1);
1903 }
1904 
1905 char *
1906 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1907 {
1908 	struct vfsopt *opt;
1909 
1910 	*error = 0;
1911 	TAILQ_FOREACH(opt, opts, link) {
1912 		if (strcmp(name, opt->name) != 0)
1913 			continue;
1914 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1915 			*error = EINVAL;
1916 			return (NULL);
1917 		}
1918 		return (opt->value);
1919 	}
1920 	*error = ENOENT;
1921 	return (NULL);
1922 }
1923 
1924 int
1925 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1926 {
1927 	struct vfsopt *opt;
1928 
1929 	TAILQ_FOREACH(opt, opts, link) {
1930 		if (strcmp(name, opt->name) == 0) {
1931 			if (w != NULL)
1932 				*w |= val;
1933 			return (1);
1934 		}
1935 	}
1936 	if (w != NULL)
1937 		*w &= ~val;
1938 	return (0);
1939 }
1940 
1941 int
1942 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1943 {
1944 	va_list ap;
1945 	struct vfsopt *opt;
1946 	int ret;
1947 
1948 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1949 
1950 	TAILQ_FOREACH(opt, opts, link) {
1951 		if (strcmp(name, opt->name) != 0)
1952 			continue;
1953 		if (opt->len == 0 || opt->value == NULL)
1954 			return (0);
1955 		if (((char *)opt->value)[opt->len - 1] != '\0')
1956 			return (0);
1957 		va_start(ap, fmt);
1958 		ret = vsscanf(opt->value, fmt, ap);
1959 		va_end(ap);
1960 		return (ret);
1961 	}
1962 	return (0);
1963 }
1964 
1965 /*
1966  * Find and copy a mount option.
1967  *
1968  * The size of the buffer has to be specified
1969  * in len, if it is not the same length as the
1970  * mount option, EINVAL is returned.
1971  * Returns ENOENT if the option is not found.
1972  */
1973 int
1974 vfs_copyopt(opts, name, dest, len)
1975 	struct vfsoptlist *opts;
1976 	const char *name;
1977 	void *dest;
1978 	int len;
1979 {
1980 	struct vfsopt *opt;
1981 
1982 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1983 
1984 	TAILQ_FOREACH(opt, opts, link) {
1985 		if (strcmp(name, opt->name) == 0) {
1986 			if (len != opt->len)
1987 				return (EINVAL);
1988 			bcopy(opt->value, dest, opt->len);
1989 			return (0);
1990 		}
1991 	}
1992 	return (ENOENT);
1993 }
1994 
1995 /*
1996  * This is a helper function for filesystems to traverse their
1997  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1998  */
1999 
2000 struct vnode *
2001 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
2002 {
2003 	struct vnode *vp;
2004 
2005 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2006 
2007 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2008 	if ((*mvp)->v_yield++ == 500) {
2009 		MNT_IUNLOCK(mp);
2010 		(*mvp)->v_yield = 0;
2011 		uio_yield();
2012 		MNT_ILOCK(mp);
2013 	}
2014 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
2015 	while (vp != NULL && vp->v_type == VMARKER)
2016 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2017 
2018 	/* Check if we are done */
2019 	if (vp == NULL) {
2020 		__mnt_vnode_markerfree(mvp, mp);
2021 		return (NULL);
2022 	}
2023 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2024 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2025 	return (vp);
2026 }
2027 
2028 struct vnode *
2029 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
2030 {
2031 	struct vnode *vp;
2032 
2033 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2034 
2035 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2036 	while (vp != NULL && vp->v_type == VMARKER)
2037 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2038 
2039 	/* Check if we are done */
2040 	if (vp == NULL) {
2041 		*mvp = NULL;
2042 		return (NULL);
2043 	}
2044 	mp->mnt_holdcnt++;
2045 	MNT_IUNLOCK(mp);
2046 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
2047 				       M_VNODE_MARKER,
2048 				       M_WAITOK | M_ZERO);
2049 	MNT_ILOCK(mp);
2050 	(*mvp)->v_type = VMARKER;
2051 
2052 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2053 	while (vp != NULL && vp->v_type == VMARKER)
2054 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2055 
2056 	/* Check if we are done */
2057 	if (vp == NULL) {
2058 		MNT_IUNLOCK(mp);
2059 		free(*mvp, M_VNODE_MARKER);
2060 		MNT_ILOCK(mp);
2061 		*mvp = NULL;
2062 		mp->mnt_holdcnt--;
2063 		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2064 			wakeup(&mp->mnt_holdcnt);
2065 		return (NULL);
2066 	}
2067 	mp->mnt_markercnt++;
2068 	(*mvp)->v_mount = mp;
2069 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2070 	return (vp);
2071 }
2072 
2073 
2074 void
2075 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
2076 {
2077 
2078 	if (*mvp == NULL)
2079 		return;
2080 
2081 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2082 
2083 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2084 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2085 	MNT_IUNLOCK(mp);
2086 	free(*mvp, M_VNODE_MARKER);
2087 	MNT_ILOCK(mp);
2088 	*mvp = NULL;
2089 
2090 	mp->mnt_markercnt--;
2091 	mp->mnt_holdcnt--;
2092 	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2093 		wakeup(&mp->mnt_holdcnt);
2094 }
2095 
2096 
2097 int
2098 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
2099 {
2100 	int error;
2101 
2102 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
2103 	if (sbp != &mp->mnt_stat)
2104 		*sbp = mp->mnt_stat;
2105 	return (error);
2106 }
2107 
2108 void
2109 vfs_mountedfrom(struct mount *mp, const char *from)
2110 {
2111 
2112 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2113 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2114 	    sizeof mp->mnt_stat.f_mntfromname);
2115 }
2116 
2117 /*
2118  * ---------------------------------------------------------------------
2119  * This is the api for building mount args and mounting filesystems from
2120  * inside the kernel.
2121  *
2122  * The API works by accumulation of individual args.  First error is
2123  * latched.
2124  *
2125  * XXX: should be documented in new manpage kernel_mount(9)
2126  */
2127 
2128 /* A memory allocation which must be freed when we are done */
2129 struct mntaarg {
2130 	SLIST_ENTRY(mntaarg)	next;
2131 };
2132 
2133 /* The header for the mount arguments */
2134 struct mntarg {
2135 	struct iovec *v;
2136 	int len;
2137 	int error;
2138 	SLIST_HEAD(, mntaarg)	list;
2139 };
2140 
2141 /*
2142  * Add a boolean argument.
2143  *
2144  * flag is the boolean value.
2145  * name must start with "no".
2146  */
2147 struct mntarg *
2148 mount_argb(struct mntarg *ma, int flag, const char *name)
2149 {
2150 
2151 	KASSERT(name[0] == 'n' && name[1] == 'o',
2152 	    ("mount_argb(...,%s): name must start with 'no'", name));
2153 
2154 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2155 }
2156 
2157 /*
2158  * Add an argument printf style
2159  */
2160 struct mntarg *
2161 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2162 {
2163 	va_list ap;
2164 	struct mntaarg *maa;
2165 	struct sbuf *sb;
2166 	int len;
2167 
2168 	if (ma == NULL) {
2169 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2170 		SLIST_INIT(&ma->list);
2171 	}
2172 	if (ma->error)
2173 		return (ma);
2174 
2175 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2176 	    M_MOUNT, M_WAITOK);
2177 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2178 	ma->v[ma->len].iov_len = strlen(name) + 1;
2179 	ma->len++;
2180 
2181 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
2182 	va_start(ap, fmt);
2183 	sbuf_vprintf(sb, fmt, ap);
2184 	va_end(ap);
2185 	sbuf_finish(sb);
2186 	len = sbuf_len(sb) + 1;
2187 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2188 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2189 	bcopy(sbuf_data(sb), maa + 1, len);
2190 	sbuf_delete(sb);
2191 
2192 	ma->v[ma->len].iov_base = maa + 1;
2193 	ma->v[ma->len].iov_len = len;
2194 	ma->len++;
2195 
2196 	return (ma);
2197 }
2198 
2199 /*
2200  * Add an argument which is a userland string.
2201  */
2202 struct mntarg *
2203 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2204 {
2205 	struct mntaarg *maa;
2206 	char *tbuf;
2207 
2208 	if (val == NULL)
2209 		return (ma);
2210 	if (ma == NULL) {
2211 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2212 		SLIST_INIT(&ma->list);
2213 	}
2214 	if (ma->error)
2215 		return (ma);
2216 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2217 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2218 	tbuf = (void *)(maa + 1);
2219 	ma->error = copyinstr(val, tbuf, len, NULL);
2220 	return (mount_arg(ma, name, tbuf, -1));
2221 }
2222 
2223 /*
2224  * Plain argument.
2225  *
2226  * If length is -1, treat value as a C string.
2227  */
2228 struct mntarg *
2229 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2230 {
2231 
2232 	if (ma == NULL) {
2233 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2234 		SLIST_INIT(&ma->list);
2235 	}
2236 	if (ma->error)
2237 		return (ma);
2238 
2239 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2240 	    M_MOUNT, M_WAITOK);
2241 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2242 	ma->v[ma->len].iov_len = strlen(name) + 1;
2243 	ma->len++;
2244 
2245 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2246 	if (len < 0)
2247 		ma->v[ma->len].iov_len = strlen(val) + 1;
2248 	else
2249 		ma->v[ma->len].iov_len = len;
2250 	ma->len++;
2251 	return (ma);
2252 }
2253 
2254 /*
2255  * Free a mntarg structure
2256  */
2257 static void
2258 free_mntarg(struct mntarg *ma)
2259 {
2260 	struct mntaarg *maa;
2261 
2262 	while (!SLIST_EMPTY(&ma->list)) {
2263 		maa = SLIST_FIRST(&ma->list);
2264 		SLIST_REMOVE_HEAD(&ma->list, next);
2265 		free(maa, M_MOUNT);
2266 	}
2267 	free(ma->v, M_MOUNT);
2268 	free(ma, M_MOUNT);
2269 }
2270 
2271 /*
2272  * Mount a filesystem
2273  */
2274 int
2275 kernel_mount(struct mntarg *ma, int flags)
2276 {
2277 	struct uio auio;
2278 	int error;
2279 
2280 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2281 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2282 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2283 
2284 	auio.uio_iov = ma->v;
2285 	auio.uio_iovcnt = ma->len;
2286 	auio.uio_segflg = UIO_SYSSPACE;
2287 
2288 	error = ma->error;
2289 	if (!error)
2290 		error = vfs_donmount(curthread, flags, &auio);
2291 	free_mntarg(ma);
2292 	return (error);
2293 }
2294 
2295 /*
2296  * A printflike function to mount a filesystem.
2297  */
2298 int
2299 kernel_vmount(int flags, ...)
2300 {
2301 	struct mntarg *ma = NULL;
2302 	va_list ap;
2303 	const char *cp;
2304 	const void *vp;
2305 	int error;
2306 
2307 	va_start(ap, flags);
2308 	for (;;) {
2309 		cp = va_arg(ap, const char *);
2310 		if (cp == NULL)
2311 			break;
2312 		vp = va_arg(ap, const void *);
2313 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2314 	}
2315 	va_end(ap);
2316 
2317 	error = kernel_mount(ma, flags);
2318 	return (error);
2319 }
2320