vfs_mount.c revision 172930
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: head/sys/kern/vfs_mount.c 172930 2007-10-24 19:04:04Z rwatson $");
39
40#include <sys/param.h>
41#include <sys/conf.h>
42#include <sys/clock.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_ddb.h"
72#include "opt_mac.h"
73
74#ifdef DDB
75#include <ddb/ddb.h>
76#endif
77
78#define	ROOTNAME		"root_device"
79#define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
80
81static int	vfs_domount(struct thread *td, const char *fstype,
82		    char *fspath, int fsflags, void *fsdata);
83static int	vfs_mountroot_ask(void);
84static int	vfs_mountroot_try(const char *mountfrom);
85static int	vfs_donmount(struct thread *td, int fsflags,
86		    struct uio *fsoptions);
87static void	free_mntarg(struct mntarg *ma);
88static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
89
90static int	usermount = 0;
91SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
92    "Unprivileged users may mount and unmount file systems");
93
94MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
95MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
96static uma_zone_t mount_zone;
97
98/* List of mounted filesystems. */
99struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
100
101/* For any iteration/modification of mountlist */
102struct mtx mountlist_mtx;
103MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
104
105TAILQ_HEAD(vfsoptlist, vfsopt);
106struct vfsopt {
107	TAILQ_ENTRY(vfsopt) link;
108	char	*name;
109	void	*value;
110	int	len;
111};
112
113/*
114 * The vnode of the system's root (/ in the filesystem, without chroot
115 * active.)
116 */
117struct vnode	*rootvnode;
118
119/*
120 * The root filesystem is detailed in the kernel environment variable
121 * vfs.root.mountfrom, which is expected to be in the general format
122 *
123 * <vfsname>:[<path>]
124 * vfsname   := the name of a VFS known to the kernel and capable
125 *              of being mounted as root
126 * path      := disk device name or other data used by the filesystem
127 *              to locate its physical store
128 */
129
130/*
131 * Global opts, taken by all filesystems
132 */
133static const char *global_opts[] = {
134	"errmsg",
135	"fstype",
136	"fspath",
137	"rdonly",
138	"ro",
139	"rw",
140	"nosuid",
141	"noexec",
142	"update",
143	NULL
144};
145
146/*
147 * The root specifiers we will try if RB_CDROM is specified.
148 */
149static char *cdrom_rootdevnames[] = {
150	"cd9660:cd0",
151	"cd9660:acd0",
152	NULL
153};
154
155/* legacy find-root code */
156char		*rootdevnames[2] = {NULL, NULL};
157#ifndef ROOTDEVNAME
158#  define ROOTDEVNAME NULL
159#endif
160static const char	*ctrootdevname = ROOTDEVNAME;
161
162/*
163 * ---------------------------------------------------------------------
164 * Functions for building and sanitizing the mount options
165 */
166
167/* Remove one mount option. */
168static void
169vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
170{
171
172	TAILQ_REMOVE(opts, opt, link);
173	free(opt->name, M_MOUNT);
174	if (opt->value != NULL)
175		free(opt->value, M_MOUNT);
176#ifdef INVARIANTS
177	else if (opt->len != 0)
178		panic("%s: mount option with NULL value but length != 0",
179		    __func__);
180#endif
181	free(opt, M_MOUNT);
182}
183
184/* Release all resources related to the mount options. */
185void
186vfs_freeopts(struct vfsoptlist *opts)
187{
188	struct vfsopt *opt;
189
190	while (!TAILQ_EMPTY(opts)) {
191		opt = TAILQ_FIRST(opts);
192		vfs_freeopt(opts, opt);
193	}
194	free(opts, M_MOUNT);
195}
196
197void
198vfs_deleteopt(struct vfsoptlist *opts, const char *name)
199{
200	struct vfsopt *opt, *temp;
201
202	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
203		if (strcmp(opt->name, name) == 0)
204			vfs_freeopt(opts, opt);
205	}
206}
207
208/*
209 * Check if options are equal (with or without the "no" prefix).
210 */
211static int
212vfs_equalopts(const char *opt1, const char *opt2)
213{
214
215	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
216	if (strcmp(opt1, opt2) == 0)
217		return (1);
218	/* "noopt" vs. "opt" */
219	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
220		return (1);
221	/* "opt" vs. "noopt" */
222	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
223		return (1);
224	return (0);
225}
226
227/*
228 * If a mount option is specified several times,
229 * (with or without the "no" prefix) only keep
230 * the last occurence of it.
231 */
232static void
233vfs_sanitizeopts(struct vfsoptlist *opts)
234{
235	struct vfsopt *opt, *opt2, *tmp;
236
237	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
238		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
239		while (opt2 != NULL) {
240			if (vfs_equalopts(opt->name, opt2->name)) {
241				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
242				vfs_freeopt(opts, opt2);
243				opt2 = tmp;
244			} else {
245				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
246			}
247		}
248	}
249}
250
251/*
252 * Build a linked list of mount options from a struct uio.
253 */
254static int
255vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
256{
257	struct vfsoptlist *opts;
258	struct vfsopt *opt;
259	size_t memused;
260	unsigned int i, iovcnt;
261	int error, namelen, optlen;
262
263	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
264	TAILQ_INIT(opts);
265	memused = 0;
266	iovcnt = auio->uio_iovcnt;
267	for (i = 0; i < iovcnt; i += 2) {
268		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
269		namelen = auio->uio_iov[i].iov_len;
270		optlen = auio->uio_iov[i + 1].iov_len;
271		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
272		opt->value = NULL;
273		opt->len = 0;
274
275		/*
276		 * Do this early, so jumps to "bad" will free the current
277		 * option.
278		 */
279		TAILQ_INSERT_TAIL(opts, opt, link);
280		memused += sizeof(struct vfsopt) + optlen + namelen;
281
282		/*
283		 * Avoid consuming too much memory, and attempts to overflow
284		 * memused.
285		 */
286		if (memused > VFS_MOUNTARG_SIZE_MAX ||
287		    optlen > VFS_MOUNTARG_SIZE_MAX ||
288		    namelen > VFS_MOUNTARG_SIZE_MAX) {
289			error = EINVAL;
290			goto bad;
291		}
292
293		if (auio->uio_segflg == UIO_SYSSPACE) {
294			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
295		} else {
296			error = copyin(auio->uio_iov[i].iov_base, opt->name,
297			    namelen);
298			if (error)
299				goto bad;
300		}
301		/* Ensure names are null-terminated strings. */
302		if (opt->name[namelen - 1] != '\0') {
303			error = EINVAL;
304			goto bad;
305		}
306		if (optlen != 0) {
307			opt->len = optlen;
308			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
309			if (auio->uio_segflg == UIO_SYSSPACE) {
310				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
311				    optlen);
312			} else {
313				error = copyin(auio->uio_iov[i + 1].iov_base,
314				    opt->value, optlen);
315				if (error)
316					goto bad;
317			}
318		}
319	}
320	vfs_sanitizeopts(opts);
321	*options = opts;
322	return (0);
323bad:
324	vfs_freeopts(opts);
325	return (error);
326}
327
328/*
329 * Merge the old mount options with the new ones passed
330 * in the MNT_UPDATE case.
331 */
332static void
333vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
334{
335	struct vfsopt *opt, *opt2, *new;
336
337	TAILQ_FOREACH(opt, opts, link) {
338		/*
339		 * Check that this option hasn't been redefined
340		 * nor cancelled with a "no" mount option.
341		 */
342		opt2 = TAILQ_FIRST(toopts);
343		while (opt2 != NULL) {
344			if (strcmp(opt2->name, opt->name) == 0)
345				goto next;
346			if (strncmp(opt2->name, "no", 2) == 0 &&
347			    strcmp(opt2->name + 2, opt->name) == 0) {
348				vfs_freeopt(toopts, opt2);
349				goto next;
350			}
351			opt2 = TAILQ_NEXT(opt2, link);
352		}
353		/* We want this option, duplicate it. */
354		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
355		new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
356		strcpy(new->name, opt->name);
357		if (opt->len != 0) {
358			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
359			bcopy(opt->value, new->value, opt->len);
360		} else {
361			new->value = NULL;
362		}
363		new->len = opt->len;
364		TAILQ_INSERT_TAIL(toopts, new, link);
365next:
366		continue;
367	}
368}
369
370/*
371 * Mount a filesystem.
372 */
373int
374nmount(td, uap)
375	struct thread *td;
376	struct nmount_args /* {
377		struct iovec *iovp;
378		unsigned int iovcnt;
379		int flags;
380	} */ *uap;
381{
382	struct uio *auio;
383	struct iovec *iov;
384	unsigned int i;
385	int error;
386	u_int iovcnt;
387
388	AUDIT_ARG(fflags, uap->flags);
389
390	/* Kick out MNT_ROOTFS early as it is legal internally */
391	if (uap->flags & MNT_ROOTFS)
392		return (EINVAL);
393
394	iovcnt = uap->iovcnt;
395	/*
396	 * Check that we have an even number of iovec's
397	 * and that we have at least two options.
398	 */
399	if ((iovcnt & 1) || (iovcnt < 4))
400		return (EINVAL);
401
402	error = copyinuio(uap->iovp, iovcnt, &auio);
403	if (error)
404		return (error);
405	iov = auio->uio_iov;
406	for (i = 0; i < iovcnt; i++) {
407		if (iov->iov_len > MMAXOPTIONLEN) {
408			free(auio, M_IOV);
409			return (EINVAL);
410		}
411		iov++;
412	}
413	error = vfs_donmount(td, uap->flags, auio);
414
415	free(auio, M_IOV);
416	return (error);
417}
418
419/*
420 * ---------------------------------------------------------------------
421 * Various utility functions
422 */
423
424void
425vfs_ref(struct mount *mp)
426{
427
428	MNT_ILOCK(mp);
429	MNT_REF(mp);
430	MNT_IUNLOCK(mp);
431}
432
433void
434vfs_rel(struct mount *mp)
435{
436
437	MNT_ILOCK(mp);
438	MNT_REL(mp);
439	MNT_IUNLOCK(mp);
440}
441
442static int
443mount_init(void *mem, int size, int flags)
444{
445	struct mount *mp;
446
447	mp = (struct mount *)mem;
448	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
449	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
450	return (0);
451}
452
453static void
454mount_fini(void *mem, int size)
455{
456	struct mount *mp;
457
458	mp = (struct mount *)mem;
459	lockdestroy(&mp->mnt_lock);
460	mtx_destroy(&mp->mnt_mtx);
461}
462
463/*
464 * Allocate and initialize the mount point struct.
465 */
466struct mount *
467vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
468    const char *fspath, struct thread *td)
469{
470	struct mount *mp;
471
472	mp = uma_zalloc(mount_zone, M_WAITOK);
473	bzero(&mp->mnt_startzero,
474	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
475	TAILQ_INIT(&mp->mnt_nvnodelist);
476	mp->mnt_nvnodelistsize = 0;
477	mp->mnt_ref = 0;
478	(void) vfs_busy(mp, LK_NOWAIT, 0, td);
479	mp->mnt_op = vfsp->vfc_vfsops;
480	mp->mnt_vfc = vfsp;
481	vfsp->vfc_refcount++;	/* XXX Unlocked */
482	mp->mnt_stat.f_type = vfsp->vfc_typenum;
483	mp->mnt_gen++;
484	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
485	mp->mnt_vnodecovered = vp;
486	mp->mnt_cred = crdup(td->td_ucred);
487	mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
488	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
489	mp->mnt_iosize_max = DFLTPHYS;
490#ifdef MAC
491	mac_mount_init(mp);
492	mac_mount_create(td->td_ucred, mp);
493#endif
494	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
495	return (mp);
496}
497
498/*
499 * Destroy the mount struct previously allocated by vfs_mount_alloc().
500 */
501void
502vfs_mount_destroy(struct mount *mp)
503{
504	int i;
505
506	MNT_ILOCK(mp);
507	for (i = 0; mp->mnt_ref && i < 3; i++)
508		msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
509	/*
510	 * This will always cause a 3 second delay in rebooting due to
511	 * refs on the root mountpoint that never go away.  Most of these
512	 * are held by init which never exits.
513	 */
514	if (i == 3 && (!rebooting || bootverbose))
515		printf("Mount point %s had %d dangling refs\n",
516		    mp->mnt_stat.f_mntonname, mp->mnt_ref);
517	if (mp->mnt_holdcnt != 0) {
518		printf("Waiting for mount point to be unheld\n");
519		while (mp->mnt_holdcnt != 0) {
520			mp->mnt_holdcntwaiters++;
521			msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
522			       PZERO, "mntdestroy", 0);
523			mp->mnt_holdcntwaiters--;
524		}
525		printf("mount point unheld\n");
526	}
527	if (mp->mnt_writeopcount > 0) {
528		printf("Waiting for mount point write ops\n");
529		while (mp->mnt_writeopcount > 0) {
530			mp->mnt_kern_flag |= MNTK_SUSPEND;
531			msleep(&mp->mnt_writeopcount,
532			       MNT_MTX(mp),
533			       PZERO, "mntdestroy2", 0);
534		}
535		printf("mount point write ops completed\n");
536	}
537	if (mp->mnt_secondary_writes > 0) {
538		printf("Waiting for mount point secondary write ops\n");
539		while (mp->mnt_secondary_writes > 0) {
540			mp->mnt_kern_flag |= MNTK_SUSPEND;
541			msleep(&mp->mnt_secondary_writes,
542			       MNT_MTX(mp),
543			       PZERO, "mntdestroy3", 0);
544		}
545		printf("mount point secondary write ops completed\n");
546	}
547	MNT_IUNLOCK(mp);
548	mp->mnt_vfc->vfc_refcount--;
549	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
550		struct vnode *vp;
551
552		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
553			vprint("", vp);
554		panic("unmount: dangling vnode");
555	}
556	MNT_ILOCK(mp);
557	if (mp->mnt_kern_flag & MNTK_MWAIT)
558		wakeup(mp);
559	if (mp->mnt_writeopcount != 0)
560		panic("vfs_mount_destroy: nonzero writeopcount");
561	if (mp->mnt_secondary_writes != 0)
562		panic("vfs_mount_destroy: nonzero secondary_writes");
563	if (mp->mnt_nvnodelistsize != 0)
564		panic("vfs_mount_destroy: nonzero nvnodelistsize");
565	mp->mnt_writeopcount = -1000;
566	mp->mnt_nvnodelistsize = -1000;
567	mp->mnt_secondary_writes = -1000;
568	MNT_IUNLOCK(mp);
569#ifdef MAC
570	mac_mount_destroy(mp);
571#endif
572	if (mp->mnt_opt != NULL)
573		vfs_freeopts(mp->mnt_opt);
574	crfree(mp->mnt_cred);
575	uma_zfree(mount_zone, mp);
576}
577
578static int
579vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
580{
581	struct vfsoptlist *optlist;
582	struct vfsopt *opt, *noro_opt;
583	char *fstype, *fspath, *errmsg;
584	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
585	int has_rw, has_noro;
586
587	errmsg = NULL;
588	errmsg_len = 0;
589	errmsg_pos = -1;
590	has_rw = 0;
591	has_noro = 0;
592
593	error = vfs_buildopts(fsoptions, &optlist);
594	if (error)
595		return (error);
596
597	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
598		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
599
600	/*
601	 * We need these two options before the others,
602	 * and they are mandatory for any filesystem.
603	 * Ensure they are NUL terminated as well.
604	 */
605	fstypelen = 0;
606	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
607	if (error || fstype[fstypelen - 1] != '\0') {
608		error = EINVAL;
609		if (errmsg != NULL)
610			strncpy(errmsg, "Invalid fstype", errmsg_len);
611		goto bail;
612	}
613	fspathlen = 0;
614	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
615	if (error || fspath[fspathlen - 1] != '\0') {
616		error = EINVAL;
617		if (errmsg != NULL)
618			strncpy(errmsg, "Invalid fspath", errmsg_len);
619		goto bail;
620	}
621
622	/*
623	 * We need to see if we have the "update" option
624	 * before we call vfs_domount(), since vfs_domount() has special
625	 * logic based on MNT_UPDATE.  This is very important
626	 * when we want to update the root filesystem.
627	 */
628	TAILQ_FOREACH(opt, optlist, link) {
629		if (strcmp(opt->name, "update") == 0)
630			fsflags |= MNT_UPDATE;
631		else if (strcmp(opt->name, "async") == 0)
632			fsflags |= MNT_ASYNC;
633		else if (strcmp(opt->name, "force") == 0)
634			fsflags |= MNT_FORCE;
635		else if (strcmp(opt->name, "multilabel") == 0)
636			fsflags |= MNT_MULTILABEL;
637		else if (strcmp(opt->name, "noasync") == 0)
638			fsflags &= ~MNT_ASYNC;
639		else if (strcmp(opt->name, "noatime") == 0)
640			fsflags |= MNT_NOATIME;
641		else if (strcmp(opt->name, "atime") == 0) {
642			free(opt->name, M_MOUNT);
643			opt->name = strdup("nonoatime", M_MOUNT);
644		}
645		else if (strcmp(opt->name, "noclusterr") == 0)
646			fsflags |= MNT_NOCLUSTERR;
647		else if (strcmp(opt->name, "clusterr") == 0) {
648			free(opt->name, M_MOUNT);
649			opt->name = strdup("nonoclusterr", M_MOUNT);
650		}
651		else if (strcmp(opt->name, "noclusterw") == 0)
652			fsflags |= MNT_NOCLUSTERW;
653		else if (strcmp(opt->name, "clusterw") == 0) {
654			free(opt->name, M_MOUNT);
655			opt->name = strdup("nonoclusterw", M_MOUNT);
656		}
657		else if (strcmp(opt->name, "noexec") == 0)
658			fsflags |= MNT_NOEXEC;
659		else if (strcmp(opt->name, "exec") == 0) {
660			free(opt->name, M_MOUNT);
661			opt->name = strdup("nonoexec", M_MOUNT);
662		}
663		else if (strcmp(opt->name, "nosuid") == 0)
664			fsflags |= MNT_NOSUID;
665		else if (strcmp(opt->name, "suid") == 0) {
666			free(opt->name, M_MOUNT);
667			opt->name = strdup("nonosuid", M_MOUNT);
668		}
669		else if (strcmp(opt->name, "nosymfollow") == 0)
670			fsflags |= MNT_NOSYMFOLLOW;
671		else if (strcmp(opt->name, "symfollow") == 0) {
672			free(opt->name, M_MOUNT);
673			opt->name = strdup("nonosymfollow", M_MOUNT);
674		}
675		else if (strcmp(opt->name, "noro") == 0) {
676			fsflags &= ~MNT_RDONLY;
677			has_noro = 1;
678		}
679		else if (strcmp(opt->name, "rw") == 0) {
680			fsflags &= ~MNT_RDONLY;
681			has_rw = 1;
682		}
683		else if (strcmp(opt->name, "ro") == 0 ||
684		    strcmp(opt->name, "rdonly") == 0)
685			fsflags |= MNT_RDONLY;
686		else if (strcmp(opt->name, "snapshot") == 0)
687			fsflags |= MNT_SNAPSHOT;
688		else if (strcmp(opt->name, "suiddir") == 0)
689			fsflags |= MNT_SUIDDIR;
690		else if (strcmp(opt->name, "sync") == 0)
691			fsflags |= MNT_SYNCHRONOUS;
692		else if (strcmp(opt->name, "union") == 0)
693			fsflags |= MNT_UNION;
694	}
695
696	/*
697	 * If "rw" was specified as a mount option, and we
698	 * are trying to update a mount-point from "ro" to "rw",
699	 * we need a mount option "noro", since in vfs_mergeopts(),
700	 * "noro" will cancel "ro", but "rw" will not do anything.
701	 */
702	if (has_rw && !has_noro) {
703		noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
704		noro_opt->name = strdup("noro", M_MOUNT);
705		noro_opt->value = NULL;
706		noro_opt->len = 0;
707		TAILQ_INSERT_TAIL(optlist, noro_opt, link);
708	}
709
710	/*
711	 * Be ultra-paranoid about making sure the type and fspath
712	 * variables will fit in our mp buffers, including the
713	 * terminating NUL.
714	 */
715	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
716		error = ENAMETOOLONG;
717		goto bail;
718	}
719
720	mtx_lock(&Giant);
721	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
722	mtx_unlock(&Giant);
723bail:
724	/* copyout the errmsg */
725	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
726	    && errmsg_len > 0 && errmsg != NULL) {
727		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
728			bcopy(errmsg,
729			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
730			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
731		} else {
732			copyout(errmsg,
733			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
734			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
735		}
736	}
737
738	if (error != 0)
739		vfs_freeopts(optlist);
740	return (error);
741}
742
743/*
744 * Old mount API.
745 */
746#ifndef _SYS_SYSPROTO_H_
747struct mount_args {
748	char	*type;
749	char	*path;
750	int	flags;
751	caddr_t	data;
752};
753#endif
754/* ARGSUSED */
755int
756mount(td, uap)
757	struct thread *td;
758	struct mount_args /* {
759		char *type;
760		char *path;
761		int flags;
762		caddr_t data;
763	} */ *uap;
764{
765	char *fstype;
766	struct vfsconf *vfsp = NULL;
767	struct mntarg *ma = NULL;
768	int error;
769
770	AUDIT_ARG(fflags, uap->flags);
771
772	/* Kick out MNT_ROOTFS early as it is legal internally */
773	uap->flags &= ~MNT_ROOTFS;
774
775	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
776	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
777	if (error) {
778		free(fstype, M_TEMP);
779		return (error);
780	}
781
782	AUDIT_ARG(text, fstype);
783	mtx_lock(&Giant);
784	vfsp = vfs_byname_kld(fstype, td, &error);
785	free(fstype, M_TEMP);
786	if (vfsp == NULL) {
787		mtx_unlock(&Giant);
788		return (ENOENT);
789	}
790	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
791		mtx_unlock(&Giant);
792		return (EOPNOTSUPP);
793	}
794
795	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
796	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
797	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
798	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
799	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
800
801	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
802	mtx_unlock(&Giant);
803	return (error);
804}
805
806
807/*
808 * vfs_domount(): actually attempt a filesystem mount.
809 */
810static int
811vfs_domount(
812	struct thread *td,	/* Calling thread. */
813	const char *fstype,	/* Filesystem type. */
814	char *fspath,		/* Mount path. */
815	int fsflags,		/* Flags common to all filesystems. */
816	void *fsdata		/* Options local to the filesystem. */
817	)
818{
819	struct vnode *vp;
820	struct mount *mp;
821	struct vfsconf *vfsp;
822	struct export_args export;
823	int error, flag = 0;
824	struct vattr va;
825	struct nameidata nd;
826
827	mtx_assert(&Giant, MA_OWNED);
828	/*
829	 * Be ultra-paranoid about making sure the type and fspath
830	 * variables will fit in our mp buffers, including the
831	 * terminating NUL.
832	 */
833	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
834		return (ENAMETOOLONG);
835
836	if (jailed(td->td_ucred) || usermount == 0) {
837		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
838			return (error);
839	}
840
841	/*
842	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
843	 */
844	if (fsflags & MNT_EXPORTED) {
845		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
846		if (error)
847			return (error);
848	}
849	if (fsflags & MNT_SUIDDIR) {
850		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
851		if (error)
852			return (error);
853	}
854	/*
855	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
856	 */
857	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
858		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
859			fsflags |= MNT_NOSUID | MNT_USER;
860	}
861
862	/* Load KLDs before we lock the covered vnode to avoid reversals. */
863	vfsp = NULL;
864	if ((fsflags & MNT_UPDATE) == 0) {
865		/* Don't try to load KLDs if we're mounting the root. */
866		if (fsflags & MNT_ROOTFS)
867			vfsp = vfs_byname(fstype);
868		else
869			vfsp = vfs_byname_kld(fstype, td, &error);
870		if (vfsp == NULL)
871			return (ENODEV);
872		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
873			return (EPERM);
874	}
875	/*
876	 * Get vnode to be covered
877	 */
878	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
879	    fspath, td);
880	if ((error = namei(&nd)) != 0)
881		return (error);
882	NDFREE(&nd, NDF_ONLY_PNBUF);
883	vp = nd.ni_vp;
884	if (fsflags & MNT_UPDATE) {
885		if ((vp->v_vflag & VV_ROOT) == 0) {
886			vput(vp);
887			return (EINVAL);
888		}
889		mp = vp->v_mount;
890		MNT_ILOCK(mp);
891		flag = mp->mnt_flag;
892		/*
893		 * We only allow the filesystem to be reloaded if it
894		 * is currently mounted read-only.
895		 */
896		if ((fsflags & MNT_RELOAD) &&
897		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
898			MNT_IUNLOCK(mp);
899			vput(vp);
900			return (EOPNOTSUPP);	/* Needs translation */
901		}
902		MNT_IUNLOCK(mp);
903		/*
904		 * Only privileged root, or (if MNT_USER is set) the user that
905		 * did the original mount is permitted to update it.
906		 */
907		error = vfs_suser(mp, td);
908		if (error) {
909			vput(vp);
910			return (error);
911		}
912		if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
913			vput(vp);
914			return (EBUSY);
915		}
916		VI_LOCK(vp);
917		if ((vp->v_iflag & VI_MOUNT) != 0 ||
918		    vp->v_mountedhere != NULL) {
919			VI_UNLOCK(vp);
920			vfs_unbusy(mp, td);
921			vput(vp);
922			return (EBUSY);
923		}
924		vp->v_iflag |= VI_MOUNT;
925		VI_UNLOCK(vp);
926		MNT_ILOCK(mp);
927		mp->mnt_flag |= fsflags &
928		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
929		MNT_IUNLOCK(mp);
930		VOP_UNLOCK(vp, 0, td);
931		mp->mnt_optnew = fsdata;
932		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
933	} else {
934		/*
935		 * If the user is not root, ensure that they own the directory
936		 * onto which we are attempting to mount.
937		 */
938		error = VOP_GETATTR(vp, &va, td->td_ucred, td);
939		if (error) {
940			vput(vp);
941			return (error);
942		}
943		if (va.va_uid != td->td_ucred->cr_uid) {
944			error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
945			    0);
946			if (error) {
947				vput(vp);
948				return (error);
949			}
950		}
951		error = vinvalbuf(vp, V_SAVE, td, 0, 0);
952		if (error != 0) {
953			vput(vp);
954			return (error);
955		}
956		if (vp->v_type != VDIR) {
957			vput(vp);
958			return (ENOTDIR);
959		}
960		VI_LOCK(vp);
961		if ((vp->v_iflag & VI_MOUNT) != 0 ||
962		    vp->v_mountedhere != NULL) {
963			VI_UNLOCK(vp);
964			vput(vp);
965			return (EBUSY);
966		}
967		vp->v_iflag |= VI_MOUNT;
968		VI_UNLOCK(vp);
969
970		/*
971		 * Allocate and initialize the filesystem.
972		 */
973		mp = vfs_mount_alloc(vp, vfsp, fspath, td);
974		VOP_UNLOCK(vp, 0, td);
975
976		/* XXXMAC: pass to vfs_mount_alloc? */
977		mp->mnt_optnew = fsdata;
978	}
979
980	/*
981	 * Set the mount level flags.
982	 */
983	MNT_ILOCK(mp);
984	mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
985		(fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
986			    MNT_RDONLY));
987	if ((mp->mnt_flag & MNT_ASYNC) == 0)
988		mp->mnt_kern_flag &= ~MNTK_ASYNC;
989	MNT_IUNLOCK(mp);
990	/*
991	 * Mount the filesystem.
992	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
993	 * get.  No freeing of cn_pnbuf.
994	 */
995        error = VFS_MOUNT(mp, td);
996
997	/*
998	 * Process the export option only if we are
999	 * updating mount options.
1000	 */
1001	if (!error && (fsflags & MNT_UPDATE)) {
1002		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
1003		    sizeof(export)) == 0)
1004			error = vfs_export(mp, &export);
1005	}
1006
1007	if (!error) {
1008		if (mp->mnt_opt != NULL)
1009			vfs_freeopts(mp->mnt_opt);
1010		mp->mnt_opt = mp->mnt_optnew;
1011		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
1012	}
1013	/*
1014	 * Prevent external consumers of mount options from reading
1015	 * mnt_optnew.
1016	*/
1017	mp->mnt_optnew = NULL;
1018	if (mp->mnt_flag & MNT_UPDATE) {
1019		MNT_ILOCK(mp);
1020		if (error)
1021			mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1022				(flag & ~MNT_QUOTA);
1023		else
1024			mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD |
1025					  MNT_FORCE | MNT_SNAPSHOT);
1026		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1027			mp->mnt_kern_flag |= MNTK_ASYNC;
1028		else
1029			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1030		MNT_IUNLOCK(mp);
1031		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1032			if (mp->mnt_syncer == NULL)
1033				error = vfs_allocate_syncvnode(mp);
1034		} else {
1035			if (mp->mnt_syncer != NULL)
1036				vrele(mp->mnt_syncer);
1037			mp->mnt_syncer = NULL;
1038		}
1039		vfs_unbusy(mp, td);
1040		VI_LOCK(vp);
1041		vp->v_iflag &= ~VI_MOUNT;
1042		VI_UNLOCK(vp);
1043		vrele(vp);
1044		return (error);
1045	}
1046	MNT_ILOCK(mp);
1047	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1048		mp->mnt_kern_flag |= MNTK_ASYNC;
1049	else
1050		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1051	MNT_IUNLOCK(mp);
1052	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1053	/*
1054	 * Put the new filesystem on the mount list after root.
1055	 */
1056	cache_purge(vp);
1057	if (!error) {
1058		struct vnode *newdp;
1059
1060		VI_LOCK(vp);
1061		vp->v_iflag &= ~VI_MOUNT;
1062		VI_UNLOCK(vp);
1063		vp->v_mountedhere = mp;
1064		mtx_lock(&mountlist_mtx);
1065		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1066		mtx_unlock(&mountlist_mtx);
1067		vfs_event_signal(NULL, VQ_MOUNT, 0);
1068		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1069			panic("mount: lost mount");
1070		mountcheckdirs(vp, newdp);
1071		vput(newdp);
1072		VOP_UNLOCK(vp, 0, td);
1073		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1074			error = vfs_allocate_syncvnode(mp);
1075		vfs_unbusy(mp, td);
1076		if (error)
1077			vrele(vp);
1078	} else {
1079		VI_LOCK(vp);
1080		vp->v_iflag &= ~VI_MOUNT;
1081		VI_UNLOCK(vp);
1082		vfs_unbusy(mp, td);
1083		vfs_mount_destroy(mp);
1084		vput(vp);
1085	}
1086	return (error);
1087}
1088
1089/*
1090 * Unmount a filesystem.
1091 *
1092 * Note: unmount takes a path to the vnode mounted on as argument, not
1093 * special file (as before).
1094 */
1095#ifndef _SYS_SYSPROTO_H_
1096struct unmount_args {
1097	char	*path;
1098	int	flags;
1099};
1100#endif
1101/* ARGSUSED */
1102int
1103unmount(td, uap)
1104	struct thread *td;
1105	register struct unmount_args /* {
1106		char *path;
1107		int flags;
1108	} */ *uap;
1109{
1110	struct mount *mp;
1111	char *pathbuf;
1112	int error, id0, id1;
1113
1114	if (jailed(td->td_ucred) || usermount == 0) {
1115		error = priv_check(td, PRIV_VFS_UNMOUNT);
1116		if (error)
1117			return (error);
1118	}
1119
1120	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1121	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1122	if (error) {
1123		free(pathbuf, M_TEMP);
1124		return (error);
1125	}
1126	AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1127	mtx_lock(&Giant);
1128	if (uap->flags & MNT_BYFSID) {
1129		/* Decode the filesystem ID. */
1130		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1131			mtx_unlock(&Giant);
1132			free(pathbuf, M_TEMP);
1133			return (EINVAL);
1134		}
1135
1136		mtx_lock(&mountlist_mtx);
1137		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1138			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1139			    mp->mnt_stat.f_fsid.val[1] == id1)
1140				break;
1141		}
1142		mtx_unlock(&mountlist_mtx);
1143	} else {
1144		mtx_lock(&mountlist_mtx);
1145		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1146			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1147				break;
1148		}
1149		mtx_unlock(&mountlist_mtx);
1150	}
1151	free(pathbuf, M_TEMP);
1152	if (mp == NULL) {
1153		/*
1154		 * Previously we returned ENOENT for a nonexistent path and
1155		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1156		 * now, so in the !MNT_BYFSID case return the more likely
1157		 * EINVAL for compatibility.
1158		 */
1159		mtx_unlock(&Giant);
1160		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1161	}
1162
1163	/*
1164	 * Don't allow unmounting the root filesystem.
1165	 */
1166	if (mp->mnt_flag & MNT_ROOTFS) {
1167		mtx_unlock(&Giant);
1168		return (EINVAL);
1169	}
1170	error = dounmount(mp, uap->flags, td);
1171	mtx_unlock(&Giant);
1172	return (error);
1173}
1174
1175/*
1176 * Do the actual filesystem unmount.
1177 */
1178int
1179dounmount(mp, flags, td)
1180	struct mount *mp;
1181	int flags;
1182	struct thread *td;
1183{
1184	struct vnode *coveredvp, *fsrootvp;
1185	int error;
1186	int async_flag;
1187	int mnt_gen_r;
1188
1189	mtx_assert(&Giant, MA_OWNED);
1190
1191	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1192		mnt_gen_r = mp->mnt_gen;
1193		VI_LOCK(coveredvp);
1194		vholdl(coveredvp);
1195		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY, td);
1196		vdrop(coveredvp);
1197		/*
1198		 * Check for mp being unmounted while waiting for the
1199		 * covered vnode lock.
1200		 */
1201		if (coveredvp->v_mountedhere != mp ||
1202		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1203			VOP_UNLOCK(coveredvp, 0, td);
1204			return (EBUSY);
1205		}
1206	}
1207	/*
1208	 * Only privileged root, or (if MNT_USER is set) the user that did the
1209	 * original mount is permitted to unmount this filesystem.
1210	 */
1211	error = vfs_suser(mp, td);
1212	if (error) {
1213		if (coveredvp)
1214			VOP_UNLOCK(coveredvp, 0, td);
1215		return (error);
1216	}
1217
1218	MNT_ILOCK(mp);
1219	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1220		MNT_IUNLOCK(mp);
1221		if (coveredvp)
1222			VOP_UNLOCK(coveredvp, 0, td);
1223		return (EBUSY);
1224	}
1225	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1226	/* Allow filesystems to detect that a forced unmount is in progress. */
1227	if (flags & MNT_FORCE)
1228		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1229	error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1230	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td);
1231	if (error) {
1232		MNT_ILOCK(mp);
1233		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ |
1234		    MNTK_UNMOUNTF);
1235		if (mp->mnt_kern_flag & MNTK_MWAIT)
1236			wakeup(mp);
1237		MNT_IUNLOCK(mp);
1238		if (coveredvp)
1239			VOP_UNLOCK(coveredvp, 0, td);
1240		return (error);
1241	}
1242	vn_start_write(NULL, &mp, V_WAIT);
1243
1244	if (mp->mnt_flag & MNT_EXPUBLIC)
1245		vfs_setpublicfs(NULL, NULL, NULL);
1246
1247	vfs_msync(mp, MNT_WAIT);
1248	MNT_ILOCK(mp);
1249	async_flag = mp->mnt_flag & MNT_ASYNC;
1250	mp->mnt_flag &= ~MNT_ASYNC;
1251	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1252	MNT_IUNLOCK(mp);
1253	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1254	if (mp->mnt_syncer != NULL)
1255		vrele(mp->mnt_syncer);
1256	/*
1257	 * For forced unmounts, move process cdir/rdir refs on the fs root
1258	 * vnode to the covered vnode.  For non-forced unmounts we want
1259	 * such references to cause an EBUSY error.
1260	 */
1261	if ((flags & MNT_FORCE) &&
1262	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1263		if (mp->mnt_vnodecovered != NULL)
1264			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1265		if (fsrootvp == rootvnode) {
1266			vrele(rootvnode);
1267			rootvnode = NULL;
1268		}
1269		vput(fsrootvp);
1270	}
1271	if (((mp->mnt_flag & MNT_RDONLY) ||
1272	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1273	    (flags & MNT_FORCE)) {
1274		error = VFS_UNMOUNT(mp, flags, td);
1275	}
1276	vn_finished_write(mp);
1277	if (error) {
1278		/* Undo cdir/rdir and rootvnode changes made above. */
1279		if ((flags & MNT_FORCE) &&
1280		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1281			if (mp->mnt_vnodecovered != NULL)
1282				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1283			if (rootvnode == NULL) {
1284				rootvnode = fsrootvp;
1285				vref(rootvnode);
1286			}
1287			vput(fsrootvp);
1288		}
1289		MNT_ILOCK(mp);
1290		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1291		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) {
1292			MNT_IUNLOCK(mp);
1293			(void) vfs_allocate_syncvnode(mp);
1294			MNT_ILOCK(mp);
1295		}
1296		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1297		mp->mnt_flag |= async_flag;
1298		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1299			mp->mnt_kern_flag |= MNTK_ASYNC;
1300		lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1301		if (mp->mnt_kern_flag & MNTK_MWAIT)
1302			wakeup(mp);
1303		MNT_IUNLOCK(mp);
1304		if (coveredvp)
1305			VOP_UNLOCK(coveredvp, 0, td);
1306		return (error);
1307	}
1308	mtx_lock(&mountlist_mtx);
1309	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1310	mtx_unlock(&mountlist_mtx);
1311	if (coveredvp != NULL) {
1312		coveredvp->v_mountedhere = NULL;
1313		vput(coveredvp);
1314	}
1315	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1316	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
1317	vfs_mount_destroy(mp);
1318	return (0);
1319}
1320
1321/*
1322 * ---------------------------------------------------------------------
1323 * Mounting of root filesystem
1324 *
1325 */
1326
1327struct root_hold_token {
1328	const char			*who;
1329	LIST_ENTRY(root_hold_token)	list;
1330};
1331
1332static LIST_HEAD(, root_hold_token)	root_holds =
1333    LIST_HEAD_INITIALIZER(&root_holds);
1334
1335static int root_mount_complete;
1336
1337/*
1338 * Hold root mount.
1339 */
1340struct root_hold_token *
1341root_mount_hold(const char *identifier)
1342{
1343	struct root_hold_token *h;
1344
1345	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1346	h->who = identifier;
1347	mtx_lock(&mountlist_mtx);
1348	LIST_INSERT_HEAD(&root_holds, h, list);
1349	mtx_unlock(&mountlist_mtx);
1350	return (h);
1351}
1352
1353/*
1354 * Release root mount.
1355 */
1356void
1357root_mount_rel(struct root_hold_token *h)
1358{
1359
1360	mtx_lock(&mountlist_mtx);
1361	LIST_REMOVE(h, list);
1362	wakeup(&root_holds);
1363	mtx_unlock(&mountlist_mtx);
1364	free(h, M_DEVBUF);
1365}
1366
1367/*
1368 * Wait for all subsystems to release root mount.
1369 */
1370static void
1371root_mount_prepare(void)
1372{
1373	struct root_hold_token *h;
1374
1375	for (;;) {
1376		DROP_GIANT();
1377		g_waitidle();
1378		PICKUP_GIANT();
1379		mtx_lock(&mountlist_mtx);
1380		if (LIST_EMPTY(&root_holds)) {
1381			mtx_unlock(&mountlist_mtx);
1382			break;
1383		}
1384		printf("Root mount waiting for:");
1385		LIST_FOREACH(h, &root_holds, list)
1386			printf(" %s", h->who);
1387		printf("\n");
1388		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1389		    hz);
1390	}
1391}
1392
1393/*
1394 * Root was mounted, share the good news.
1395 */
1396static void
1397root_mount_done(void)
1398{
1399
1400	/*
1401	 * Use a mutex to prevent the wakeup being missed and waiting for
1402	 * an extra 1 second sleep.
1403	 */
1404	mtx_lock(&mountlist_mtx);
1405	root_mount_complete = 1;
1406	wakeup(&root_mount_complete);
1407	mtx_unlock(&mountlist_mtx);
1408}
1409
1410/*
1411 * Return true if root is already mounted.
1412 */
1413int
1414root_mounted(void)
1415{
1416
1417	/* No mutex is acquired here because int stores are atomic. */
1418	return (root_mount_complete);
1419}
1420
1421/*
1422 * Wait until root is mounted.
1423 */
1424void
1425root_mount_wait(void)
1426{
1427
1428	/*
1429	 * Panic on an obvious deadlock - the function can't be called from
1430	 * a thread which is doing the whole SYSINIT stuff.
1431	 */
1432	KASSERT(curthread->td_proc->p_pid != 0,
1433	    ("root_mount_wait: cannot be called from the swapper thread"));
1434	mtx_lock(&mountlist_mtx);
1435	while (!root_mount_complete) {
1436		msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
1437		    hz);
1438	}
1439	mtx_unlock(&mountlist_mtx);
1440}
1441
1442static void
1443set_rootvnode(struct thread *td)
1444{
1445	struct proc *p;
1446
1447	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1448		panic("Cannot find root vnode");
1449
1450	p = td->td_proc;
1451	FILEDESC_SLOCK(p->p_fd);
1452
1453	if (p->p_fd->fd_cdir != NULL)
1454		vrele(p->p_fd->fd_cdir);
1455	p->p_fd->fd_cdir = rootvnode;
1456	VREF(rootvnode);
1457
1458	if (p->p_fd->fd_rdir != NULL)
1459		vrele(p->p_fd->fd_rdir);
1460	p->p_fd->fd_rdir = rootvnode;
1461	VREF(rootvnode);
1462
1463	FILEDESC_SUNLOCK(p->p_fd);
1464
1465	VOP_UNLOCK(rootvnode, 0, td);
1466}
1467
1468/*
1469 * Mount /devfs as our root filesystem, but do not put it on the mountlist
1470 * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1471 */
1472
1473static void
1474devfs_first(void)
1475{
1476	struct thread *td = curthread;
1477	struct vfsoptlist *opts;
1478	struct vfsconf *vfsp;
1479	struct mount *mp = NULL;
1480	int error;
1481
1482	vfsp = vfs_byname("devfs");
1483	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1484	if (vfsp == NULL)
1485		return;
1486
1487	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1488
1489	error = VFS_MOUNT(mp, td);
1490	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1491	if (error)
1492		return;
1493
1494	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1495	TAILQ_INIT(opts);
1496	mp->mnt_opt = opts;
1497
1498	mtx_lock(&mountlist_mtx);
1499	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1500	mtx_unlock(&mountlist_mtx);
1501
1502	set_rootvnode(td);
1503
1504	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1505	if (error)
1506		printf("kern_symlink /dev -> / returns %d\n", error);
1507}
1508
1509/*
1510 * Surgically move our devfs to be mounted on /dev.
1511 */
1512
1513static void
1514devfs_fixup(struct thread *td)
1515{
1516	struct nameidata nd;
1517	int error;
1518	struct vnode *vp, *dvp;
1519	struct mount *mp;
1520
1521	/* Remove our devfs mount from the mountlist and purge the cache */
1522	mtx_lock(&mountlist_mtx);
1523	mp = TAILQ_FIRST(&mountlist);
1524	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1525	mtx_unlock(&mountlist_mtx);
1526	cache_purgevfs(mp);
1527
1528	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1529	VI_LOCK(dvp);
1530	dvp->v_iflag &= ~VI_MOUNT;
1531	VI_UNLOCK(dvp);
1532	dvp->v_mountedhere = NULL;
1533
1534	/* Set up the real rootvnode, and purge the cache */
1535	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1536	set_rootvnode(td);
1537	cache_purgevfs(rootvnode->v_mount);
1538
1539	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1540	error = namei(&nd);
1541	if (error) {
1542		printf("Lookup of /dev for devfs, error: %d\n", error);
1543		return;
1544	}
1545	NDFREE(&nd, NDF_ONLY_PNBUF);
1546	vp = nd.ni_vp;
1547	if (vp->v_type != VDIR) {
1548		vput(vp);
1549	}
1550	error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1551	if (error) {
1552		vput(vp);
1553	}
1554	cache_purge(vp);
1555	mp->mnt_vnodecovered = vp;
1556	vp->v_mountedhere = mp;
1557	mtx_lock(&mountlist_mtx);
1558	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1559	mtx_unlock(&mountlist_mtx);
1560	VOP_UNLOCK(vp, 0, td);
1561	vput(dvp);
1562	vfs_unbusy(mp, td);
1563
1564	/* Unlink the no longer needed /dev/dev -> / symlink */
1565	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1566}
1567
1568/*
1569 * Report errors during filesystem mounting.
1570 */
1571void
1572vfs_mount_error(struct mount *mp, const char *fmt, ...)
1573{
1574	struct vfsoptlist *moptlist = mp->mnt_optnew;
1575	va_list ap;
1576	int error, len;
1577	char *errmsg;
1578
1579	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1580	if (error || errmsg == NULL || len <= 0)
1581		return;
1582
1583	va_start(ap, fmt);
1584	vsnprintf(errmsg, (size_t)len, fmt, ap);
1585	va_end(ap);
1586}
1587
1588/*
1589 * Find and mount the root filesystem
1590 */
1591void
1592vfs_mountroot(void)
1593{
1594	char *cp;
1595	int error, i, asked = 0;
1596
1597	root_mount_prepare();
1598
1599	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1600	    NULL, NULL, mount_init, mount_fini,
1601	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1602	devfs_first();
1603
1604	/*
1605	 * We are booted with instructions to prompt for the root filesystem.
1606	 */
1607	if (boothowto & RB_ASKNAME) {
1608		if (!vfs_mountroot_ask())
1609			goto mounted;
1610		asked = 1;
1611	}
1612
1613	/*
1614	 * The root filesystem information is compiled in, and we are
1615	 * booted with instructions to use it.
1616	 */
1617	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1618		if (!vfs_mountroot_try(ctrootdevname))
1619			goto mounted;
1620		ctrootdevname = NULL;
1621	}
1622
1623	/*
1624	 * We've been given the generic "use CDROM as root" flag.  This is
1625	 * necessary because one media may be used in many different
1626	 * devices, so we need to search for them.
1627	 */
1628	if (boothowto & RB_CDROM) {
1629		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1630			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1631				goto mounted;
1632		}
1633	}
1634
1635	/*
1636	 * Try to use the value read by the loader from /etc/fstab, or
1637	 * supplied via some other means.  This is the preferred
1638	 * mechanism.
1639	 */
1640	cp = getenv("vfs.root.mountfrom");
1641	if (cp != NULL) {
1642		error = vfs_mountroot_try(cp);
1643		freeenv(cp);
1644		if (!error)
1645			goto mounted;
1646	}
1647
1648	/*
1649	 * Try values that may have been computed by code during boot
1650	 */
1651	if (!vfs_mountroot_try(rootdevnames[0]))
1652		goto mounted;
1653	if (!vfs_mountroot_try(rootdevnames[1]))
1654		goto mounted;
1655
1656	/*
1657	 * If we (still) have a compiled-in default, try it.
1658	 */
1659	if (ctrootdevname != NULL)
1660		if (!vfs_mountroot_try(ctrootdevname))
1661			goto mounted;
1662	/*
1663	 * Everything so far has failed, prompt on the console if we haven't
1664	 * already tried that.
1665	 */
1666	if (!asked)
1667		if (!vfs_mountroot_ask())
1668			goto mounted;
1669
1670	panic("Root mount failed, startup aborted.");
1671
1672mounted:
1673	root_mount_done();
1674}
1675
1676/*
1677 * Mount (mountfrom) as the root filesystem.
1678 */
1679static int
1680vfs_mountroot_try(const char *mountfrom)
1681{
1682	struct mount	*mp;
1683	char		*vfsname, *path;
1684	time_t		timebase;
1685	int		error;
1686	char		patt[32];
1687
1688	vfsname = NULL;
1689	path    = NULL;
1690	mp      = NULL;
1691	error   = EINVAL;
1692
1693	if (mountfrom == NULL)
1694		return (error);		/* don't complain */
1695	printf("Trying to mount root from %s\n", mountfrom);
1696
1697	/* parse vfs name and path */
1698	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1699	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1700	vfsname[0] = path[0] = 0;
1701	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1702	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1703		goto out;
1704
1705	if (path[0] == '\0')
1706		strcpy(path, ROOTNAME);
1707
1708	error = kernel_vmount(
1709	    MNT_RDONLY | MNT_ROOTFS,
1710	    "fstype", vfsname,
1711	    "fspath", "/",
1712	    "from", path,
1713	    NULL);
1714	if (error == 0) {
1715		/*
1716		 * We mount devfs prior to mounting the / FS, so the first
1717		 * entry will typically be devfs.
1718		 */
1719		mp = TAILQ_FIRST(&mountlist);
1720		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1721
1722		/*
1723		 * Iterate over all currently mounted file systems and use
1724		 * the time stamp found to check and/or initialize the RTC.
1725		 * Typically devfs has no time stamp and the only other FS
1726		 * is the actual / FS.
1727		 * Call inittodr() only once and pass it the largest of the
1728		 * timestamps we encounter.
1729		 */
1730		timebase = 0;
1731		do {
1732			if (mp->mnt_time > timebase)
1733				timebase = mp->mnt_time;
1734			mp = TAILQ_NEXT(mp, mnt_list);
1735		} while (mp != NULL);
1736		inittodr(timebase);
1737
1738		devfs_fixup(curthread);
1739	}
1740out:
1741	free(path, M_MOUNT);
1742	free(vfsname, M_MOUNT);
1743	return (error);
1744}
1745
1746/*
1747 * ---------------------------------------------------------------------
1748 * Interactive root filesystem selection code.
1749 */
1750
1751static int
1752vfs_mountroot_ask(void)
1753{
1754	char name[128];
1755
1756	for(;;) {
1757		printf("\nManual root filesystem specification:\n");
1758		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1759#if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1760		printf("                       eg. ufs:da0s1a\n");
1761#else
1762		printf("                       eg. ufs:/dev/da0a\n");
1763#endif
1764		printf("  ?                  List valid disk boot devices\n");
1765		printf("  <empty line>       Abort manual input\n");
1766		printf("\nmountroot> ");
1767		gets(name, sizeof(name), 1);
1768		if (name[0] == '\0')
1769			return (1);
1770		if (name[0] == '?') {
1771			printf("\nList of GEOM managed disk devices:\n  ");
1772			g_dev_print();
1773			continue;
1774		}
1775		if (!vfs_mountroot_try(name))
1776			return (0);
1777	}
1778}
1779
1780/*
1781 * ---------------------------------------------------------------------
1782 * Functions for querying mount options/arguments from filesystems.
1783 */
1784
1785/*
1786 * Check that no unknown options are given
1787 */
1788int
1789vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1790{
1791	struct vfsopt *opt;
1792	char errmsg[255];
1793	const char **t, *p, *q;
1794	int ret = 0;
1795
1796	TAILQ_FOREACH(opt, opts, link) {
1797		p = opt->name;
1798		q = NULL;
1799		if (p[0] == 'n' && p[1] == 'o')
1800			q = p + 2;
1801		for(t = global_opts; *t != NULL; t++) {
1802			if (strcmp(*t, p) == 0)
1803				break;
1804			if (q != NULL) {
1805				if (strcmp(*t, q) == 0)
1806					break;
1807			}
1808		}
1809		if (*t != NULL)
1810			continue;
1811		for(t = legal; *t != NULL; t++) {
1812			if (strcmp(*t, p) == 0)
1813				break;
1814			if (q != NULL) {
1815				if (strcmp(*t, q) == 0)
1816					break;
1817			}
1818		}
1819		if (*t != NULL)
1820			continue;
1821		sprintf(errmsg, "mount option <%s> is unknown", p);
1822		printf("%s\n", errmsg);
1823		ret = EINVAL;
1824	}
1825	if (ret != 0) {
1826		TAILQ_FOREACH(opt, opts, link) {
1827			if (strcmp(opt->name, "errmsg") == 0) {
1828				strncpy((char *)opt->value, errmsg, opt->len);
1829			}
1830		}
1831	}
1832	return (ret);
1833}
1834
1835/*
1836 * Get a mount option by its name.
1837 *
1838 * Return 0 if the option was found, ENOENT otherwise.
1839 * If len is non-NULL it will be filled with the length
1840 * of the option. If buf is non-NULL, it will be filled
1841 * with the address of the option.
1842 */
1843int
1844vfs_getopt(opts, name, buf, len)
1845	struct vfsoptlist *opts;
1846	const char *name;
1847	void **buf;
1848	int *len;
1849{
1850	struct vfsopt *opt;
1851
1852	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1853
1854	TAILQ_FOREACH(opt, opts, link) {
1855		if (strcmp(name, opt->name) == 0) {
1856			if (len != NULL)
1857				*len = opt->len;
1858			if (buf != NULL)
1859				*buf = opt->value;
1860			return (0);
1861		}
1862	}
1863	return (ENOENT);
1864}
1865
1866static int
1867vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1868{
1869	struct vfsopt *opt;
1870	int i;
1871
1872	if (opts == NULL)
1873		return (-1);
1874
1875	i = 0;
1876	TAILQ_FOREACH(opt, opts, link) {
1877		if (strcmp(name, opt->name) == 0)
1878			return (i);
1879		++i;
1880	}
1881	return (-1);
1882}
1883
1884char *
1885vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1886{
1887	struct vfsopt *opt;
1888
1889	*error = 0;
1890	TAILQ_FOREACH(opt, opts, link) {
1891		if (strcmp(name, opt->name) != 0)
1892			continue;
1893		if (((char *)opt->value)[opt->len - 1] != '\0') {
1894			*error = EINVAL;
1895			return (NULL);
1896		}
1897		return (opt->value);
1898	}
1899	*error = ENOENT;
1900	return (NULL);
1901}
1902
1903int
1904vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1905{
1906	struct vfsopt *opt;
1907
1908	TAILQ_FOREACH(opt, opts, link) {
1909		if (strcmp(name, opt->name) == 0) {
1910			if (w != NULL)
1911				*w |= val;
1912			return (1);
1913		}
1914	}
1915	if (w != NULL)
1916		*w &= ~val;
1917	return (0);
1918}
1919
1920int
1921vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1922{
1923	va_list ap;
1924	struct vfsopt *opt;
1925	int ret;
1926
1927	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1928
1929	TAILQ_FOREACH(opt, opts, link) {
1930		if (strcmp(name, opt->name) != 0)
1931			continue;
1932		if (((char *)opt->value)[opt->len - 1] != '\0')
1933			return (0);
1934		va_start(ap, fmt);
1935		ret = vsscanf(opt->value, fmt, ap);
1936		va_end(ap);
1937		return (ret);
1938	}
1939	return (0);
1940}
1941
1942/*
1943 * Find and copy a mount option.
1944 *
1945 * The size of the buffer has to be specified
1946 * in len, if it is not the same length as the
1947 * mount option, EINVAL is returned.
1948 * Returns ENOENT if the option is not found.
1949 */
1950int
1951vfs_copyopt(opts, name, dest, len)
1952	struct vfsoptlist *opts;
1953	const char *name;
1954	void *dest;
1955	int len;
1956{
1957	struct vfsopt *opt;
1958
1959	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1960
1961	TAILQ_FOREACH(opt, opts, link) {
1962		if (strcmp(name, opt->name) == 0) {
1963			if (len != opt->len)
1964				return (EINVAL);
1965			bcopy(opt->value, dest, opt->len);
1966			return (0);
1967		}
1968	}
1969	return (ENOENT);
1970}
1971
1972/*
1973 * This is a helper function for filesystems to traverse their
1974 * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1975 */
1976
1977struct vnode *
1978__mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1979{
1980	struct vnode *vp;
1981
1982	mtx_assert(MNT_MTX(mp), MA_OWNED);
1983
1984	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1985	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1986	while (vp != NULL && vp->v_type == VMARKER)
1987		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1988
1989	/* Check if we are done */
1990	if (vp == NULL) {
1991		__mnt_vnode_markerfree(mvp, mp);
1992		return (NULL);
1993	}
1994	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1995	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1996	return (vp);
1997}
1998
1999struct vnode *
2000__mnt_vnode_first(struct vnode **mvp, struct mount *mp)
2001{
2002	struct vnode *vp;
2003
2004	mtx_assert(MNT_MTX(mp), MA_OWNED);
2005
2006	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2007	while (vp != NULL && vp->v_type == VMARKER)
2008		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2009
2010	/* Check if we are done */
2011	if (vp == NULL) {
2012		*mvp = NULL;
2013		return (NULL);
2014	}
2015	mp->mnt_holdcnt++;
2016	MNT_IUNLOCK(mp);
2017	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
2018				       M_VNODE_MARKER,
2019				       M_WAITOK | M_ZERO);
2020	MNT_ILOCK(mp);
2021	(*mvp)->v_type = VMARKER;
2022
2023	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2024	while (vp != NULL && vp->v_type == VMARKER)
2025		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2026
2027	/* Check if we are done */
2028	if (vp == NULL) {
2029		MNT_IUNLOCK(mp);
2030		free(*mvp, M_VNODE_MARKER);
2031		MNT_ILOCK(mp);
2032		*mvp = NULL;
2033		mp->mnt_holdcnt--;
2034		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2035			wakeup(&mp->mnt_holdcnt);
2036		return (NULL);
2037	}
2038	mp->mnt_markercnt++;
2039	(*mvp)->v_mount = mp;
2040	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2041	return (vp);
2042}
2043
2044
2045void
2046__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
2047{
2048
2049	if (*mvp == NULL)
2050		return;
2051
2052	mtx_assert(MNT_MTX(mp), MA_OWNED);
2053
2054	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2055	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2056	MNT_IUNLOCK(mp);
2057	free(*mvp, M_VNODE_MARKER);
2058	MNT_ILOCK(mp);
2059	*mvp = NULL;
2060
2061	mp->mnt_markercnt--;
2062	mp->mnt_holdcnt--;
2063	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2064		wakeup(&mp->mnt_holdcnt);
2065}
2066
2067
2068int
2069__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
2070{
2071	int error;
2072
2073	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
2074	if (sbp != &mp->mnt_stat)
2075		*sbp = mp->mnt_stat;
2076	return (error);
2077}
2078
2079void
2080vfs_mountedfrom(struct mount *mp, const char *from)
2081{
2082
2083	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2084	strlcpy(mp->mnt_stat.f_mntfromname, from,
2085	    sizeof mp->mnt_stat.f_mntfromname);
2086}
2087
2088/*
2089 * ---------------------------------------------------------------------
2090 * This is the api for building mount args and mounting filesystems from
2091 * inside the kernel.
2092 *
2093 * The API works by accumulation of individual args.  First error is
2094 * latched.
2095 *
2096 * XXX: should be documented in new manpage kernel_mount(9)
2097 */
2098
2099/* A memory allocation which must be freed when we are done */
2100struct mntaarg {
2101	SLIST_ENTRY(mntaarg)	next;
2102};
2103
2104/* The header for the mount arguments */
2105struct mntarg {
2106	struct iovec *v;
2107	int len;
2108	int error;
2109	SLIST_HEAD(, mntaarg)	list;
2110};
2111
2112/*
2113 * Add a boolean argument.
2114 *
2115 * flag is the boolean value.
2116 * name must start with "no".
2117 */
2118struct mntarg *
2119mount_argb(struct mntarg *ma, int flag, const char *name)
2120{
2121
2122	KASSERT(name[0] == 'n' && name[1] == 'o',
2123	    ("mount_argb(...,%s): name must start with 'no'", name));
2124
2125	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2126}
2127
2128/*
2129 * Add an argument printf style
2130 */
2131struct mntarg *
2132mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2133{
2134	va_list ap;
2135	struct mntaarg *maa;
2136	struct sbuf *sb;
2137	int len;
2138
2139	if (ma == NULL) {
2140		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2141		SLIST_INIT(&ma->list);
2142	}
2143	if (ma->error)
2144		return (ma);
2145
2146	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2147	    M_MOUNT, M_WAITOK);
2148	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2149	ma->v[ma->len].iov_len = strlen(name) + 1;
2150	ma->len++;
2151
2152	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
2153	va_start(ap, fmt);
2154	sbuf_vprintf(sb, fmt, ap);
2155	va_end(ap);
2156	sbuf_finish(sb);
2157	len = sbuf_len(sb) + 1;
2158	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2159	SLIST_INSERT_HEAD(&ma->list, maa, next);
2160	bcopy(sbuf_data(sb), maa + 1, len);
2161	sbuf_delete(sb);
2162
2163	ma->v[ma->len].iov_base = maa + 1;
2164	ma->v[ma->len].iov_len = len;
2165	ma->len++;
2166
2167	return (ma);
2168}
2169
2170/*
2171 * Add an argument which is a userland string.
2172 */
2173struct mntarg *
2174mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2175{
2176	struct mntaarg *maa;
2177	char *tbuf;
2178
2179	if (val == NULL)
2180		return (ma);
2181	if (ma == NULL) {
2182		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2183		SLIST_INIT(&ma->list);
2184	}
2185	if (ma->error)
2186		return (ma);
2187	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2188	SLIST_INSERT_HEAD(&ma->list, maa, next);
2189	tbuf = (void *)(maa + 1);
2190	ma->error = copyinstr(val, tbuf, len, NULL);
2191	return (mount_arg(ma, name, tbuf, -1));
2192}
2193
2194/*
2195 * Plain argument.
2196 *
2197 * If length is -1, use printf.
2198 */
2199struct mntarg *
2200mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2201{
2202
2203	if (ma == NULL) {
2204		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2205		SLIST_INIT(&ma->list);
2206	}
2207	if (ma->error)
2208		return (ma);
2209
2210	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2211	    M_MOUNT, M_WAITOK);
2212	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2213	ma->v[ma->len].iov_len = strlen(name) + 1;
2214	ma->len++;
2215
2216	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2217	if (len < 0)
2218		ma->v[ma->len].iov_len = strlen(val) + 1;
2219	else
2220		ma->v[ma->len].iov_len = len;
2221	ma->len++;
2222	return (ma);
2223}
2224
2225/*
2226 * Free a mntarg structure
2227 */
2228static void
2229free_mntarg(struct mntarg *ma)
2230{
2231	struct mntaarg *maa;
2232
2233	while (!SLIST_EMPTY(&ma->list)) {
2234		maa = SLIST_FIRST(&ma->list);
2235		SLIST_REMOVE_HEAD(&ma->list, next);
2236		free(maa, M_MOUNT);
2237	}
2238	free(ma->v, M_MOUNT);
2239	free(ma, M_MOUNT);
2240}
2241
2242/*
2243 * Mount a filesystem
2244 */
2245int
2246kernel_mount(struct mntarg *ma, int flags)
2247{
2248	struct uio auio;
2249	int error;
2250
2251	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2252	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2253	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2254
2255	auio.uio_iov = ma->v;
2256	auio.uio_iovcnt = ma->len;
2257	auio.uio_segflg = UIO_SYSSPACE;
2258
2259	error = ma->error;
2260	if (!error)
2261		error = vfs_donmount(curthread, flags, &auio);
2262	free_mntarg(ma);
2263	return (error);
2264}
2265
2266/*
2267 * A printflike function to mount a filesystem.
2268 */
2269int
2270kernel_vmount(int flags, ...)
2271{
2272	struct mntarg *ma = NULL;
2273	va_list ap;
2274	const char *cp;
2275	const void *vp;
2276	int error;
2277
2278	va_start(ap, flags);
2279	for (;;) {
2280		cp = va_arg(ap, const char *);
2281		if (cp == NULL)
2282			break;
2283		vp = va_arg(ap, const void *);
2284		ma = mount_arg(ma, cp, vp, -1);
2285	}
2286	va_end(ap);
2287
2288	error = kernel_mount(ma, flags);
2289	return (error);
2290}
2291