xref: /freebsd/sys/kern/vfs_mount.c (revision 9162f64b58d01ec01481d60b6cdc06ffd8e8c7fc)
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 void	free_mntarg(struct mntarg *ma);
81 static int	vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
82 
83 static int	usermount = 0;
84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
85     "Unprivileged users may mount and unmount file systems");
86 
87 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
88 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
89 static uma_zone_t mount_zone;
90 
91 /* List of mounted filesystems. */
92 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
93 
94 /* For any iteration/modification of mountlist */
95 struct mtx mountlist_mtx;
96 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
97 
98 TAILQ_HEAD(vfsoptlist, vfsopt);
99 struct vfsopt {
100 	TAILQ_ENTRY(vfsopt) link;
101 	char	*name;
102 	void	*value;
103 	int	len;
104 };
105 
106 /*
107  * The vnode of the system's root (/ in the filesystem, without chroot
108  * active.)
109  */
110 struct vnode	*rootvnode;
111 
112 /*
113  * The root filesystem is detailed in the kernel environment variable
114  * vfs.root.mountfrom, which is expected to be in the general format
115  *
116  * <vfsname>:[<path>]
117  * vfsname   := the name of a VFS known to the kernel and capable
118  *              of being mounted as root
119  * path      := disk device name or other data used by the filesystem
120  *              to locate its physical store
121  */
122 
123 /*
124  * Global opts, taken by all filesystems
125  */
126 static const char *global_opts[] = {
127 	"errmsg",
128 	"fstype",
129 	"fspath",
130 	"ro",
131 	"rw",
132 	"nosuid",
133 	"noexec",
134 	NULL
135 };
136 
137 /*
138  * The root specifiers we will try if RB_CDROM is specified.
139  */
140 static char *cdrom_rootdevnames[] = {
141 	"cd9660:cd0",
142 	"cd9660:acd0",
143 	NULL
144 };
145 
146 /* legacy find-root code */
147 char		*rootdevnames[2] = {NULL, NULL};
148 #ifndef ROOTDEVNAME
149 #  define ROOTDEVNAME NULL
150 #endif
151 static const char	*ctrootdevname = ROOTDEVNAME;
152 
153 /*
154  * ---------------------------------------------------------------------
155  * Functions for building and sanitizing the mount options
156  */
157 
158 /* Remove one mount option. */
159 static void
160 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
161 {
162 
163 	TAILQ_REMOVE(opts, opt, link);
164 	free(opt->name, M_MOUNT);
165 	if (opt->value != NULL)
166 		free(opt->value, M_MOUNT);
167 #ifdef INVARIANTS
168 	else if (opt->len != 0)
169 		panic("%s: mount option with NULL value but length != 0",
170 		    __func__);
171 #endif
172 	free(opt, M_MOUNT);
173 }
174 
175 /* Release all resources related to the mount options. */
176 void
177 vfs_freeopts(struct vfsoptlist *opts)
178 {
179 	struct vfsopt *opt;
180 
181 	while (!TAILQ_EMPTY(opts)) {
182 		opt = TAILQ_FIRST(opts);
183 		vfs_freeopt(opts, opt);
184 	}
185 	free(opts, M_MOUNT);
186 }
187 
188 void
189 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
190 {
191 	struct vfsopt *opt, *temp;
192 
193 	if (opts == NULL)
194 		return;
195 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
196 		if (strcmp(opt->name, name) == 0)
197 			vfs_freeopt(opts, opt);
198 	}
199 }
200 
201 /*
202  * Check if options are equal (with or without the "no" prefix).
203  */
204 static int
205 vfs_equalopts(const char *opt1, const char *opt2)
206 {
207 
208 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
209 	if (strcmp(opt1, opt2) == 0)
210 		return (1);
211 	/* "noopt" vs. "opt" */
212 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
213 		return (1);
214 	/* "opt" vs. "noopt" */
215 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
216 		return (1);
217 	return (0);
218 }
219 
220 /*
221  * If a mount option is specified several times,
222  * (with or without the "no" prefix) only keep
223  * the last occurence of it.
224  */
225 static void
226 vfs_sanitizeopts(struct vfsoptlist *opts)
227 {
228 	struct vfsopt *opt, *opt2, *tmp;
229 
230 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
231 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
232 		while (opt2 != NULL) {
233 			if (vfs_equalopts(opt->name, opt2->name)) {
234 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
235 				vfs_freeopt(opts, opt2);
236 				opt2 = tmp;
237 			} else {
238 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
239 			}
240 		}
241 	}
242 }
243 
244 /*
245  * Build a linked list of mount options from a struct uio.
246  */
247 static int
248 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
249 {
250 	struct vfsoptlist *opts;
251 	struct vfsopt *opt;
252 	size_t memused;
253 	unsigned int i, iovcnt;
254 	int error, namelen, optlen;
255 
256 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
257 	TAILQ_INIT(opts);
258 	memused = 0;
259 	iovcnt = auio->uio_iovcnt;
260 	for (i = 0; i < iovcnt; i += 2) {
261 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
262 		namelen = auio->uio_iov[i].iov_len;
263 		optlen = auio->uio_iov[i + 1].iov_len;
264 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
265 		opt->value = NULL;
266 		opt->len = 0;
267 
268 		/*
269 		 * Do this early, so jumps to "bad" will free the current
270 		 * option.
271 		 */
272 		TAILQ_INSERT_TAIL(opts, opt, link);
273 		memused += sizeof(struct vfsopt) + optlen + namelen;
274 
275 		/*
276 		 * Avoid consuming too much memory, and attempts to overflow
277 		 * memused.
278 		 */
279 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
280 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
281 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
282 			error = EINVAL;
283 			goto bad;
284 		}
285 
286 		if (auio->uio_segflg == UIO_SYSSPACE) {
287 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
288 		} else {
289 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
290 			    namelen);
291 			if (error)
292 				goto bad;
293 		}
294 		/* Ensure names are null-terminated strings. */
295 		if (opt->name[namelen - 1] != '\0') {
296 			error = EINVAL;
297 			goto bad;
298 		}
299 		if (optlen != 0) {
300 			opt->len = optlen;
301 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
302 			if (auio->uio_segflg == UIO_SYSSPACE) {
303 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
304 				    optlen);
305 			} else {
306 				error = copyin(auio->uio_iov[i + 1].iov_base,
307 				    opt->value, optlen);
308 				if (error)
309 					goto bad;
310 			}
311 		}
312 	}
313 	vfs_sanitizeopts(opts);
314 	*options = opts;
315 	return (0);
316 bad:
317 	vfs_freeopts(opts);
318 	return (error);
319 }
320 
321 /*
322  * Merge the old mount options with the new ones passed
323  * in the MNT_UPDATE case.
324  *
325  * XXX This function will keep a "nofoo" option in the
326  *     new options if there is no matching "foo" option
327  *     to be cancelled in the old options.  This is a bug
328  *     if the option's canonical name is "foo".  E.g., "noro"
329  *     shouldn't end up in the mount point's active options,
330  *     but it can.
331  */
332 static void
333 vfs_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);
365 next:
366 		continue;
367 	}
368 }
369 
370 /*
371  * Mount a filesystem.
372  */
373 int
374 nmount(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 	/*
391 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
392 	 * userspace to set this flag, but we must filter it out if we want
393 	 * MNT_UPDATE on the root file system to work.
394 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
395 	 */
396 	uap->flags &= ~MNT_ROOTFS;
397 
398 	iovcnt = uap->iovcnt;
399 	/*
400 	 * Check that we have an even number of iovec's
401 	 * and that we have at least two options.
402 	 */
403 	if ((iovcnt & 1) || (iovcnt < 4))
404 		return (EINVAL);
405 
406 	error = copyinuio(uap->iovp, iovcnt, &auio);
407 	if (error)
408 		return (error);
409 	iov = auio->uio_iov;
410 	for (i = 0; i < iovcnt; i++) {
411 		if (iov->iov_len > MMAXOPTIONLEN) {
412 			free(auio, M_IOV);
413 			return (EINVAL);
414 		}
415 		iov++;
416 	}
417 	error = vfs_donmount(td, uap->flags, auio);
418 
419 	free(auio, M_IOV);
420 	return (error);
421 }
422 
423 /*
424  * ---------------------------------------------------------------------
425  * Various utility functions
426  */
427 
428 void
429 vfs_ref(struct mount *mp)
430 {
431 
432 	MNT_ILOCK(mp);
433 	MNT_REF(mp);
434 	MNT_IUNLOCK(mp);
435 }
436 
437 void
438 vfs_rel(struct mount *mp)
439 {
440 
441 	MNT_ILOCK(mp);
442 	MNT_REL(mp);
443 	MNT_IUNLOCK(mp);
444 }
445 
446 static int
447 mount_init(void *mem, int size, int flags)
448 {
449 	struct mount *mp;
450 
451 	mp = (struct mount *)mem;
452 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
453 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
454 	return (0);
455 }
456 
457 static void
458 mount_fini(void *mem, int size)
459 {
460 	struct mount *mp;
461 
462 	mp = (struct mount *)mem;
463 	lockdestroy(&mp->mnt_explock);
464 	mtx_destroy(&mp->mnt_mtx);
465 }
466 
467 /*
468  * Allocate and initialize the mount point struct.
469  */
470 struct mount *
471 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
472     struct ucred *cred)
473 {
474 	struct mount *mp;
475 
476 	mp = uma_zalloc(mount_zone, M_WAITOK);
477 	bzero(&mp->mnt_startzero,
478 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
479 	TAILQ_INIT(&mp->mnt_nvnodelist);
480 	mp->mnt_nvnodelistsize = 0;
481 	mp->mnt_ref = 0;
482 	(void) vfs_busy(mp, MBF_NOWAIT);
483 	mp->mnt_op = vfsp->vfc_vfsops;
484 	mp->mnt_vfc = vfsp;
485 	vfsp->vfc_refcount++;	/* XXX Unlocked */
486 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
487 	mp->mnt_gen++;
488 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
489 	mp->mnt_vnodecovered = vp;
490 	mp->mnt_cred = crdup(cred);
491 	mp->mnt_stat.f_owner = cred->cr_uid;
492 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
493 	mp->mnt_iosize_max = DFLTPHYS;
494 #ifdef MAC
495 	mac_mount_init(mp);
496 	mac_mount_create(cred, mp);
497 #endif
498 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
499 	return (mp);
500 }
501 
502 /*
503  * Destroy the mount struct previously allocated by vfs_mount_alloc().
504  */
505 void
506 vfs_mount_destroy(struct mount *mp)
507 {
508 
509 	MNT_ILOCK(mp);
510 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
511 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
512 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
513 		wakeup(mp);
514 	}
515 	while (mp->mnt_ref)
516 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
517 	KASSERT(mp->mnt_ref == 0,
518 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
519 	    __FILE__, __LINE__));
520 	if (mp->mnt_writeopcount != 0)
521 		panic("vfs_mount_destroy: nonzero writeopcount");
522 	if (mp->mnt_secondary_writes != 0)
523 		panic("vfs_mount_destroy: nonzero secondary_writes");
524 	mp->mnt_vfc->vfc_refcount--;
525 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
526 		struct vnode *vp;
527 
528 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
529 			vprint("", vp);
530 		panic("unmount: dangling vnode");
531 	}
532 	if (mp->mnt_nvnodelistsize != 0)
533 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
534 	if (mp->mnt_lockref != 0)
535 		panic("vfs_mount_destroy: nonzero lock refcount");
536 	MNT_IUNLOCK(mp);
537 #ifdef MAC
538 	mac_mount_destroy(mp);
539 #endif
540 	if (mp->mnt_opt != NULL)
541 		vfs_freeopts(mp->mnt_opt);
542 	crfree(mp->mnt_cred);
543 	uma_zfree(mount_zone, mp);
544 }
545 
546 int
547 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
548 {
549 	struct vfsoptlist *optlist;
550 	struct vfsopt *opt, *noro_opt, *tmp_opt;
551 	char *fstype, *fspath, *errmsg;
552 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
553 	int has_rw, has_noro;
554 
555 	errmsg = fspath = NULL;
556 	errmsg_len = has_noro = has_rw = fspathlen = 0;
557 	errmsg_pos = -1;
558 
559 	error = vfs_buildopts(fsoptions, &optlist);
560 	if (error)
561 		return (error);
562 
563 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
564 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
565 
566 	/*
567 	 * We need these two options before the others,
568 	 * and they are mandatory for any filesystem.
569 	 * Ensure they are NUL terminated as well.
570 	 */
571 	fstypelen = 0;
572 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
573 	if (error || fstype[fstypelen - 1] != '\0') {
574 		error = EINVAL;
575 		if (errmsg != NULL)
576 			strncpy(errmsg, "Invalid fstype", errmsg_len);
577 		goto bail;
578 	}
579 	fspathlen = 0;
580 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
581 	if (error || fspath[fspathlen - 1] != '\0') {
582 		error = EINVAL;
583 		if (errmsg != NULL)
584 			strncpy(errmsg, "Invalid fspath", errmsg_len);
585 		goto bail;
586 	}
587 
588 	/*
589 	 * We need to see if we have the "update" option
590 	 * before we call vfs_domount(), since vfs_domount() has special
591 	 * logic based on MNT_UPDATE.  This is very important
592 	 * when we want to update the root filesystem.
593 	 */
594 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
595 		if (strcmp(opt->name, "update") == 0) {
596 			fsflags |= MNT_UPDATE;
597 			vfs_freeopt(optlist, opt);
598 		}
599 		else if (strcmp(opt->name, "async") == 0)
600 			fsflags |= MNT_ASYNC;
601 		else if (strcmp(opt->name, "force") == 0) {
602 			fsflags |= MNT_FORCE;
603 			vfs_freeopt(optlist, opt);
604 		}
605 		else if (strcmp(opt->name, "reload") == 0) {
606 			fsflags |= MNT_RELOAD;
607 			vfs_freeopt(optlist, opt);
608 		}
609 		else if (strcmp(opt->name, "multilabel") == 0)
610 			fsflags |= MNT_MULTILABEL;
611 		else if (strcmp(opt->name, "noasync") == 0)
612 			fsflags &= ~MNT_ASYNC;
613 		else if (strcmp(opt->name, "noatime") == 0)
614 			fsflags |= MNT_NOATIME;
615 		else if (strcmp(opt->name, "atime") == 0) {
616 			free(opt->name, M_MOUNT);
617 			opt->name = strdup("nonoatime", M_MOUNT);
618 		}
619 		else if (strcmp(opt->name, "noclusterr") == 0)
620 			fsflags |= MNT_NOCLUSTERR;
621 		else if (strcmp(opt->name, "clusterr") == 0) {
622 			free(opt->name, M_MOUNT);
623 			opt->name = strdup("nonoclusterr", M_MOUNT);
624 		}
625 		else if (strcmp(opt->name, "noclusterw") == 0)
626 			fsflags |= MNT_NOCLUSTERW;
627 		else if (strcmp(opt->name, "clusterw") == 0) {
628 			free(opt->name, M_MOUNT);
629 			opt->name = strdup("nonoclusterw", M_MOUNT);
630 		}
631 		else if (strcmp(opt->name, "noexec") == 0)
632 			fsflags |= MNT_NOEXEC;
633 		else if (strcmp(opt->name, "exec") == 0) {
634 			free(opt->name, M_MOUNT);
635 			opt->name = strdup("nonoexec", M_MOUNT);
636 		}
637 		else if (strcmp(opt->name, "nosuid") == 0)
638 			fsflags |= MNT_NOSUID;
639 		else if (strcmp(opt->name, "suid") == 0) {
640 			free(opt->name, M_MOUNT);
641 			opt->name = strdup("nonosuid", M_MOUNT);
642 		}
643 		else if (strcmp(opt->name, "nosymfollow") == 0)
644 			fsflags |= MNT_NOSYMFOLLOW;
645 		else if (strcmp(opt->name, "symfollow") == 0) {
646 			free(opt->name, M_MOUNT);
647 			opt->name = strdup("nonosymfollow", M_MOUNT);
648 		}
649 		else if (strcmp(opt->name, "noro") == 0) {
650 			fsflags &= ~MNT_RDONLY;
651 			has_noro = 1;
652 		}
653 		else if (strcmp(opt->name, "rw") == 0) {
654 			fsflags &= ~MNT_RDONLY;
655 			has_rw = 1;
656 		}
657 		else if (strcmp(opt->name, "ro") == 0)
658 			fsflags |= MNT_RDONLY;
659 		else if (strcmp(opt->name, "rdonly") == 0) {
660 			free(opt->name, M_MOUNT);
661 			opt->name = strdup("ro", M_MOUNT);
662 			fsflags |= MNT_RDONLY;
663 		}
664 		else if (strcmp(opt->name, "suiddir") == 0)
665 			fsflags |= MNT_SUIDDIR;
666 		else if (strcmp(opt->name, "sync") == 0)
667 			fsflags |= MNT_SYNCHRONOUS;
668 		else if (strcmp(opt->name, "union") == 0)
669 			fsflags |= MNT_UNION;
670 	}
671 
672 	/*
673 	 * If "rw" was specified as a mount option, and we
674 	 * are trying to update a mount-point from "ro" to "rw",
675 	 * we need a mount option "noro", since in vfs_mergeopts(),
676 	 * "noro" will cancel "ro", but "rw" will not do anything.
677 	 */
678 	if (has_rw && !has_noro) {
679 		noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
680 		noro_opt->name = strdup("noro", M_MOUNT);
681 		noro_opt->value = NULL;
682 		noro_opt->len = 0;
683 		TAILQ_INSERT_TAIL(optlist, noro_opt, link);
684 	}
685 
686 	/*
687 	 * Be ultra-paranoid about making sure the type and fspath
688 	 * variables will fit in our mp buffers, including the
689 	 * terminating NUL.
690 	 */
691 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
692 		error = ENAMETOOLONG;
693 		goto bail;
694 	}
695 
696 	mtx_lock(&Giant);
697 	error = vfs_domount(td, fstype, fspath, fsflags, optlist);
698 	mtx_unlock(&Giant);
699 bail:
700 	/* copyout the errmsg */
701 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
702 	    && errmsg_len > 0 && errmsg != NULL) {
703 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
704 			bcopy(errmsg,
705 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
706 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
707 		} else {
708 			copyout(errmsg,
709 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
710 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
711 		}
712 	}
713 
714 	if (error != 0)
715 		vfs_freeopts(optlist);
716 	return (error);
717 }
718 
719 /*
720  * Old mount API.
721  */
722 #ifndef _SYS_SYSPROTO_H_
723 struct mount_args {
724 	char	*type;
725 	char	*path;
726 	int	flags;
727 	caddr_t	data;
728 };
729 #endif
730 /* ARGSUSED */
731 int
732 mount(td, uap)
733 	struct thread *td;
734 	struct mount_args /* {
735 		char *type;
736 		char *path;
737 		int flags;
738 		caddr_t data;
739 	} */ *uap;
740 {
741 	char *fstype;
742 	struct vfsconf *vfsp = NULL;
743 	struct mntarg *ma = NULL;
744 	int error;
745 
746 	AUDIT_ARG(fflags, uap->flags);
747 
748 	/*
749 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
750 	 * userspace to set this flag, but we must filter it out if we want
751 	 * MNT_UPDATE on the root file system to work.
752 	 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
753 	 */
754 	uap->flags &= ~MNT_ROOTFS;
755 
756 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
757 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
758 	if (error) {
759 		free(fstype, M_TEMP);
760 		return (error);
761 	}
762 
763 	AUDIT_ARG(text, fstype);
764 	mtx_lock(&Giant);
765 	vfsp = vfs_byname_kld(fstype, td, &error);
766 	free(fstype, M_TEMP);
767 	if (vfsp == NULL) {
768 		mtx_unlock(&Giant);
769 		return (ENOENT);
770 	}
771 	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
772 		mtx_unlock(&Giant);
773 		return (EOPNOTSUPP);
774 	}
775 
776 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
777 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
778 	ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
779 	ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
780 	ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
781 
782 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
783 	mtx_unlock(&Giant);
784 	return (error);
785 }
786 
787 
788 /*
789  * vfs_domount(): actually attempt a filesystem mount.
790  */
791 static int
792 vfs_domount(
793 	struct thread *td,	/* Calling thread. */
794 	const char *fstype,	/* Filesystem type. */
795 	char *fspath,		/* Mount path. */
796 	int fsflags,		/* Flags common to all filesystems. */
797 	void *fsdata		/* Options local to the filesystem. */
798 	)
799 {
800 	struct vnode *vp;
801 	struct mount *mp;
802 	struct vfsconf *vfsp;
803 	struct oexport_args oexport;
804 	struct export_args export;
805 	int error, flag = 0;
806 	struct vattr va;
807 	struct nameidata nd;
808 
809 	mtx_assert(&Giant, MA_OWNED);
810 	/*
811 	 * Be ultra-paranoid about making sure the type and fspath
812 	 * variables will fit in our mp buffers, including the
813 	 * terminating NUL.
814 	 */
815 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
816 		return (ENAMETOOLONG);
817 
818 	if (jailed(td->td_ucred) || usermount == 0) {
819 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
820 			return (error);
821 	}
822 
823 	/*
824 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
825 	 */
826 	if (fsflags & MNT_EXPORTED) {
827 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
828 		if (error)
829 			return (error);
830 	}
831 	if (fsflags & MNT_SUIDDIR) {
832 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
833 		if (error)
834 			return (error);
835 	}
836 	/*
837 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
838 	 */
839 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
840 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
841 			fsflags |= MNT_NOSUID | MNT_USER;
842 	}
843 
844 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
845 	vfsp = NULL;
846 	if ((fsflags & MNT_UPDATE) == 0) {
847 		/* Don't try to load KLDs if we're mounting the root. */
848 		if (fsflags & MNT_ROOTFS)
849 			vfsp = vfs_byname(fstype);
850 		else
851 			vfsp = vfs_byname_kld(fstype, td, &error);
852 		if (vfsp == NULL)
853 			return (ENODEV);
854 		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
855 			return (EPERM);
856 	}
857 	/*
858 	 * Get vnode to be covered
859 	 */
860 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
861 	    fspath, td);
862 	if ((error = namei(&nd)) != 0)
863 		return (error);
864 	NDFREE(&nd, NDF_ONLY_PNBUF);
865 	vp = nd.ni_vp;
866 	if (fsflags & MNT_UPDATE) {
867 		if ((vp->v_vflag & VV_ROOT) == 0) {
868 			vput(vp);
869 			return (EINVAL);
870 		}
871 		mp = vp->v_mount;
872 		MNT_ILOCK(mp);
873 		flag = mp->mnt_flag;
874 		/*
875 		 * We only allow the filesystem to be reloaded if it
876 		 * is currently mounted read-only.
877 		 */
878 		if ((fsflags & MNT_RELOAD) &&
879 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
880 			MNT_IUNLOCK(mp);
881 			vput(vp);
882 			return (EOPNOTSUPP);	/* Needs translation */
883 		}
884 		MNT_IUNLOCK(mp);
885 		/*
886 		 * Only privileged root, or (if MNT_USER is set) the user that
887 		 * did the original mount is permitted to update it.
888 		 */
889 		error = vfs_suser(mp, td);
890 		if (error) {
891 			vput(vp);
892 			return (error);
893 		}
894 		if (vfs_busy(mp, MBF_NOWAIT)) {
895 			vput(vp);
896 			return (EBUSY);
897 		}
898 		VI_LOCK(vp);
899 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
900 		    vp->v_mountedhere != NULL) {
901 			VI_UNLOCK(vp);
902 			vfs_unbusy(mp);
903 			vput(vp);
904 			return (EBUSY);
905 		}
906 		vp->v_iflag |= VI_MOUNT;
907 		VI_UNLOCK(vp);
908 		MNT_ILOCK(mp);
909 		mp->mnt_flag |= fsflags &
910 		    (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
911 		MNT_IUNLOCK(mp);
912 		VOP_UNLOCK(vp, 0);
913 		mp->mnt_optnew = fsdata;
914 		vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
915 	} else {
916 		/*
917 		 * If the user is not root, ensure that they own the directory
918 		 * onto which we are attempting to mount.
919 		 */
920 		error = VOP_GETATTR(vp, &va, td->td_ucred);
921 		if (error) {
922 			vput(vp);
923 			return (error);
924 		}
925 		if (va.va_uid != td->td_ucred->cr_uid) {
926 			error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
927 			    0);
928 			if (error) {
929 				vput(vp);
930 				return (error);
931 			}
932 		}
933 		error = vinvalbuf(vp, V_SAVE, 0, 0);
934 		if (error != 0) {
935 			vput(vp);
936 			return (error);
937 		}
938 		if (vp->v_type != VDIR) {
939 			vput(vp);
940 			return (ENOTDIR);
941 		}
942 		VI_LOCK(vp);
943 		if ((vp->v_iflag & VI_MOUNT) != 0 ||
944 		    vp->v_mountedhere != NULL) {
945 			VI_UNLOCK(vp);
946 			vput(vp);
947 			return (EBUSY);
948 		}
949 		vp->v_iflag |= VI_MOUNT;
950 		VI_UNLOCK(vp);
951 
952 		/*
953 		 * Allocate and initialize the filesystem.
954 		 */
955 		mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
956 		VOP_UNLOCK(vp, 0);
957 
958 		/* XXXMAC: pass to vfs_mount_alloc? */
959 		mp->mnt_optnew = fsdata;
960 	}
961 
962 	/*
963 	 * Set the mount level flags.
964 	 */
965 	MNT_ILOCK(mp);
966 	mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
967 		(fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
968 			    MNT_RDONLY));
969 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
970 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
971 	MNT_IUNLOCK(mp);
972 	/*
973 	 * Mount the filesystem.
974 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
975 	 * get.  No freeing of cn_pnbuf.
976 	 */
977         error = VFS_MOUNT(mp, td);
978 
979 	/*
980 	 * Process the export option only if we are
981 	 * updating mount options.
982 	 */
983 	if (!error && (fsflags & MNT_UPDATE)) {
984 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
985 		    sizeof(export)) == 0)
986 			error = vfs_export(mp, &export);
987 		else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport,
988 			sizeof(oexport)) == 0) {
989 			export.ex_flags = oexport.ex_flags;
990 			export.ex_root = oexport.ex_root;
991 			export.ex_anon = oexport.ex_anon;
992 			export.ex_addr = oexport.ex_addr;
993 			export.ex_addrlen = oexport.ex_addrlen;
994 			export.ex_mask = oexport.ex_mask;
995 			export.ex_masklen = oexport.ex_masklen;
996 			export.ex_indexfile = oexport.ex_indexfile;
997 			export.ex_numsecflavors = 0;
998 			error = vfs_export(mp, &export);
999 		}
1000 	}
1001 
1002 	if (!error) {
1003 		if (mp->mnt_opt != NULL)
1004 			vfs_freeopts(mp->mnt_opt);
1005 		mp->mnt_opt = mp->mnt_optnew;
1006 		(void)VFS_STATFS(mp, &mp->mnt_stat, td);
1007 	}
1008 	/*
1009 	 * Prevent external consumers of mount options from reading
1010 	 * mnt_optnew.
1011 	*/
1012 	mp->mnt_optnew = NULL;
1013 	if (mp->mnt_flag & MNT_UPDATE) {
1014 		MNT_ILOCK(mp);
1015 		if (error)
1016 			mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1017 				(flag & ~MNT_QUOTA);
1018 		else
1019 			mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD |
1020 					  MNT_FORCE | MNT_SNAPSHOT);
1021 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1022 			mp->mnt_kern_flag |= MNTK_ASYNC;
1023 		else
1024 			mp->mnt_kern_flag &= ~MNTK_ASYNC;
1025 		MNT_IUNLOCK(mp);
1026 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1027 			if (mp->mnt_syncer == NULL)
1028 				error = vfs_allocate_syncvnode(mp);
1029 		} else {
1030 			if (mp->mnt_syncer != NULL)
1031 				vrele(mp->mnt_syncer);
1032 			mp->mnt_syncer = NULL;
1033 		}
1034 		vfs_unbusy(mp);
1035 		VI_LOCK(vp);
1036 		vp->v_iflag &= ~VI_MOUNT;
1037 		VI_UNLOCK(vp);
1038 		vrele(vp);
1039 		return (error);
1040 	}
1041 	MNT_ILOCK(mp);
1042 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1043 		mp->mnt_kern_flag |= MNTK_ASYNC;
1044 	else
1045 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1046 	MNT_IUNLOCK(mp);
1047 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1048 	/*
1049 	 * Put the new filesystem on the mount list after root.
1050 	 */
1051 	cache_purge(vp);
1052 	if (!error) {
1053 		struct vnode *newdp;
1054 
1055 		VI_LOCK(vp);
1056 		vp->v_iflag &= ~VI_MOUNT;
1057 		VI_UNLOCK(vp);
1058 		vp->v_mountedhere = mp;
1059 		mtx_lock(&mountlist_mtx);
1060 		TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1061 		mtx_unlock(&mountlist_mtx);
1062 		vfs_event_signal(NULL, VQ_MOUNT, 0);
1063 		if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1064 			panic("mount: lost mount");
1065 		mountcheckdirs(vp, newdp);
1066 		vput(newdp);
1067 		VOP_UNLOCK(vp, 0);
1068 		if ((mp->mnt_flag & MNT_RDONLY) == 0)
1069 			error = vfs_allocate_syncvnode(mp);
1070 		vfs_unbusy(mp);
1071 		if (error)
1072 			vrele(vp);
1073 	} else {
1074 		VI_LOCK(vp);
1075 		vp->v_iflag &= ~VI_MOUNT;
1076 		VI_UNLOCK(vp);
1077 		vfs_unbusy(mp);
1078 		vfs_mount_destroy(mp);
1079 		vput(vp);
1080 	}
1081 	return (error);
1082 }
1083 
1084 /*
1085  * Unmount a filesystem.
1086  *
1087  * Note: unmount takes a path to the vnode mounted on as argument, not
1088  * special file (as before).
1089  */
1090 #ifndef _SYS_SYSPROTO_H_
1091 struct unmount_args {
1092 	char	*path;
1093 	int	flags;
1094 };
1095 #endif
1096 /* ARGSUSED */
1097 int
1098 unmount(td, uap)
1099 	struct thread *td;
1100 	register struct unmount_args /* {
1101 		char *path;
1102 		int flags;
1103 	} */ *uap;
1104 {
1105 	struct mount *mp;
1106 	char *pathbuf;
1107 	int error, id0, id1;
1108 
1109 	if (jailed(td->td_ucred) || usermount == 0) {
1110 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1111 		if (error)
1112 			return (error);
1113 	}
1114 
1115 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1116 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1117 	if (error) {
1118 		free(pathbuf, M_TEMP);
1119 		return (error);
1120 	}
1121 	AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1122 	mtx_lock(&Giant);
1123 	if (uap->flags & MNT_BYFSID) {
1124 		/* Decode the filesystem ID. */
1125 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1126 			mtx_unlock(&Giant);
1127 			free(pathbuf, M_TEMP);
1128 			return (EINVAL);
1129 		}
1130 
1131 		mtx_lock(&mountlist_mtx);
1132 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1133 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1134 			    mp->mnt_stat.f_fsid.val[1] == id1)
1135 				break;
1136 		}
1137 		mtx_unlock(&mountlist_mtx);
1138 	} else {
1139 		mtx_lock(&mountlist_mtx);
1140 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1141 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1142 				break;
1143 		}
1144 		mtx_unlock(&mountlist_mtx);
1145 	}
1146 	free(pathbuf, M_TEMP);
1147 	if (mp == NULL) {
1148 		/*
1149 		 * Previously we returned ENOENT for a nonexistent path and
1150 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1151 		 * now, so in the !MNT_BYFSID case return the more likely
1152 		 * EINVAL for compatibility.
1153 		 */
1154 		mtx_unlock(&Giant);
1155 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1156 	}
1157 
1158 	/*
1159 	 * Don't allow unmounting the root filesystem.
1160 	 */
1161 	if (mp->mnt_flag & MNT_ROOTFS) {
1162 		mtx_unlock(&Giant);
1163 		return (EINVAL);
1164 	}
1165 	error = dounmount(mp, uap->flags, td);
1166 	mtx_unlock(&Giant);
1167 	return (error);
1168 }
1169 
1170 /*
1171  * Do the actual filesystem unmount.
1172  */
1173 int
1174 dounmount(mp, flags, td)
1175 	struct mount *mp;
1176 	int flags;
1177 	struct thread *td;
1178 {
1179 	struct vnode *coveredvp, *fsrootvp;
1180 	int error;
1181 	int async_flag;
1182 	int mnt_gen_r;
1183 
1184 	mtx_assert(&Giant, MA_OWNED);
1185 
1186 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1187 		mnt_gen_r = mp->mnt_gen;
1188 		VI_LOCK(coveredvp);
1189 		vholdl(coveredvp);
1190 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1191 		vdrop(coveredvp);
1192 		/*
1193 		 * Check for mp being unmounted while waiting for the
1194 		 * covered vnode lock.
1195 		 */
1196 		if (coveredvp->v_mountedhere != mp ||
1197 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1198 			VOP_UNLOCK(coveredvp, 0);
1199 			return (EBUSY);
1200 		}
1201 	}
1202 	/*
1203 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1204 	 * original mount is permitted to unmount this filesystem.
1205 	 */
1206 	error = vfs_suser(mp, td);
1207 	if (error) {
1208 		if (coveredvp)
1209 			VOP_UNLOCK(coveredvp, 0);
1210 		return (error);
1211 	}
1212 
1213 	MNT_ILOCK(mp);
1214 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1215 		MNT_IUNLOCK(mp);
1216 		if (coveredvp)
1217 			VOP_UNLOCK(coveredvp, 0);
1218 		return (EBUSY);
1219 	}
1220 	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1221 	/* Allow filesystems to detect that a forced unmount is in progress. */
1222 	if (flags & MNT_FORCE)
1223 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1224 	error = 0;
1225 	if (mp->mnt_lockref) {
1226 		if ((flags & MNT_FORCE) == 0) {
1227 			mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ |
1228 			    MNTK_UNMOUNTF);
1229 			if (mp->mnt_kern_flag & MNTK_MWAIT) {
1230 				mp->mnt_kern_flag &= ~MNTK_MWAIT;
1231 				wakeup(mp);
1232 			}
1233 			MNT_IUNLOCK(mp);
1234 			if (coveredvp)
1235 				VOP_UNLOCK(coveredvp, 0);
1236 			return (EBUSY);
1237 		}
1238 		mp->mnt_kern_flag |= MNTK_DRAINING;
1239 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1240 		    "mount drain", 0);
1241 	}
1242 	MNT_IUNLOCK(mp);
1243 	KASSERT(mp->mnt_lockref == 0,
1244 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1245 	    __func__, __FILE__, __LINE__));
1246 	KASSERT(error == 0,
1247 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1248 	    __func__, __FILE__, __LINE__));
1249 	vn_start_write(NULL, &mp, V_WAIT);
1250 
1251 	if (mp->mnt_flag & MNT_EXPUBLIC)
1252 		vfs_setpublicfs(NULL, NULL, NULL);
1253 
1254 	vfs_msync(mp, MNT_WAIT);
1255 	MNT_ILOCK(mp);
1256 	async_flag = mp->mnt_flag & MNT_ASYNC;
1257 	mp->mnt_flag &= ~MNT_ASYNC;
1258 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1259 	MNT_IUNLOCK(mp);
1260 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1261 	if (mp->mnt_syncer != NULL)
1262 		vrele(mp->mnt_syncer);
1263 	/*
1264 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1265 	 * vnode to the covered vnode.  For non-forced unmounts we want
1266 	 * such references to cause an EBUSY error.
1267 	 */
1268 	if ((flags & MNT_FORCE) &&
1269 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1270 		if (mp->mnt_vnodecovered != NULL)
1271 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1272 		if (fsrootvp == rootvnode) {
1273 			vrele(rootvnode);
1274 			rootvnode = NULL;
1275 		}
1276 		vput(fsrootvp);
1277 	}
1278 	if (((mp->mnt_flag & MNT_RDONLY) ||
1279 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1280 	    (flags & MNT_FORCE)) {
1281 		error = VFS_UNMOUNT(mp, flags, td);
1282 	}
1283 	vn_finished_write(mp);
1284 	/*
1285 	 * If we failed to flush the dirty blocks for this mount point,
1286 	 * undo all the cdir/rdir and rootvnode changes we made above.
1287 	 * Unless we failed to do so because the device is reporting that
1288 	 * it doesn't exist anymore.
1289 	 */
1290 	if (error && error != ENXIO) {
1291 		if ((flags & MNT_FORCE) &&
1292 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1293 			if (mp->mnt_vnodecovered != NULL)
1294 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1295 			if (rootvnode == NULL) {
1296 				rootvnode = fsrootvp;
1297 				vref(rootvnode);
1298 			}
1299 			vput(fsrootvp);
1300 		}
1301 		MNT_ILOCK(mp);
1302 		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1303 		if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) {
1304 			MNT_IUNLOCK(mp);
1305 			(void) vfs_allocate_syncvnode(mp);
1306 			MNT_ILOCK(mp);
1307 		}
1308 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1309 		mp->mnt_flag |= async_flag;
1310 		if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1311 			mp->mnt_kern_flag |= MNTK_ASYNC;
1312 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1313 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1314 			wakeup(mp);
1315 		}
1316 		MNT_IUNLOCK(mp);
1317 		if (coveredvp)
1318 			VOP_UNLOCK(coveredvp, 0);
1319 		return (error);
1320 	}
1321 	mtx_lock(&mountlist_mtx);
1322 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1323 	mtx_unlock(&mountlist_mtx);
1324 	if (coveredvp != NULL) {
1325 		coveredvp->v_mountedhere = NULL;
1326 		vput(coveredvp);
1327 	}
1328 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1329 	vfs_mount_destroy(mp);
1330 	return (0);
1331 }
1332 
1333 /*
1334  * ---------------------------------------------------------------------
1335  * Mounting of root filesystem
1336  *
1337  */
1338 
1339 struct root_hold_token {
1340 	const char			*who;
1341 	LIST_ENTRY(root_hold_token)	list;
1342 };
1343 
1344 static LIST_HEAD(, root_hold_token)	root_holds =
1345     LIST_HEAD_INITIALIZER(&root_holds);
1346 
1347 static int root_mount_complete;
1348 
1349 /*
1350  * Hold root mount.
1351  */
1352 struct root_hold_token *
1353 root_mount_hold(const char *identifier)
1354 {
1355 	struct root_hold_token *h;
1356 
1357 	h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1358 	h->who = identifier;
1359 	mtx_lock(&mountlist_mtx);
1360 	LIST_INSERT_HEAD(&root_holds, h, list);
1361 	mtx_unlock(&mountlist_mtx);
1362 	return (h);
1363 }
1364 
1365 /*
1366  * Release root mount.
1367  */
1368 void
1369 root_mount_rel(struct root_hold_token *h)
1370 {
1371 
1372 	mtx_lock(&mountlist_mtx);
1373 	LIST_REMOVE(h, list);
1374 	wakeup(&root_holds);
1375 	mtx_unlock(&mountlist_mtx);
1376 	free(h, M_DEVBUF);
1377 }
1378 
1379 /*
1380  * Wait for all subsystems to release root mount.
1381  */
1382 static void
1383 root_mount_prepare(void)
1384 {
1385 	struct root_hold_token *h;
1386 
1387 	for (;;) {
1388 		DROP_GIANT();
1389 		g_waitidle();
1390 		PICKUP_GIANT();
1391 		mtx_lock(&mountlist_mtx);
1392 		if (LIST_EMPTY(&root_holds)) {
1393 			mtx_unlock(&mountlist_mtx);
1394 			break;
1395 		}
1396 		printf("Root mount waiting for:");
1397 		LIST_FOREACH(h, &root_holds, list)
1398 			printf(" %s", h->who);
1399 		printf("\n");
1400 		msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1401 		    hz);
1402 	}
1403 }
1404 
1405 /*
1406  * Root was mounted, share the good news.
1407  */
1408 static void
1409 root_mount_done(void)
1410 {
1411 
1412 	/*
1413 	 * Use a mutex to prevent the wakeup being missed and waiting for
1414 	 * an extra 1 second sleep.
1415 	 */
1416 	mtx_lock(&mountlist_mtx);
1417 	root_mount_complete = 1;
1418 	wakeup(&root_mount_complete);
1419 	mtx_unlock(&mountlist_mtx);
1420 }
1421 
1422 /*
1423  * Return true if root is already mounted.
1424  */
1425 int
1426 root_mounted(void)
1427 {
1428 
1429 	/* No mutex is acquired here because int stores are atomic. */
1430 	return (root_mount_complete);
1431 }
1432 
1433 /*
1434  * Wait until root is mounted.
1435  */
1436 void
1437 root_mount_wait(void)
1438 {
1439 
1440 	/*
1441 	 * Panic on an obvious deadlock - the function can't be called from
1442 	 * a thread which is doing the whole SYSINIT stuff.
1443 	 */
1444 	KASSERT(curthread->td_proc->p_pid != 0,
1445 	    ("root_mount_wait: cannot be called from the swapper thread"));
1446 	mtx_lock(&mountlist_mtx);
1447 	while (!root_mount_complete) {
1448 		msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
1449 		    hz);
1450 	}
1451 	mtx_unlock(&mountlist_mtx);
1452 }
1453 
1454 static void
1455 set_rootvnode(struct thread *td)
1456 {
1457 	struct proc *p;
1458 
1459 	if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1460 		panic("Cannot find root vnode");
1461 
1462 	p = td->td_proc;
1463 	FILEDESC_XLOCK(p->p_fd);
1464 
1465 	if (p->p_fd->fd_cdir != NULL)
1466 		vrele(p->p_fd->fd_cdir);
1467 	p->p_fd->fd_cdir = rootvnode;
1468 	VREF(rootvnode);
1469 
1470 	if (p->p_fd->fd_rdir != NULL)
1471 		vrele(p->p_fd->fd_rdir);
1472 	p->p_fd->fd_rdir = rootvnode;
1473 	VREF(rootvnode);
1474 
1475 	FILEDESC_XUNLOCK(p->p_fd);
1476 
1477 	VOP_UNLOCK(rootvnode, 0);
1478 
1479 	EVENTHANDLER_INVOKE(mountroot);
1480 }
1481 
1482 /*
1483  * Mount /devfs as our root filesystem, but do not put it on the mountlist
1484  * yet.  Create a /dev -> / symlink so that absolute pathnames will lookup.
1485  */
1486 
1487 static void
1488 devfs_first(void)
1489 {
1490 	struct thread *td = curthread;
1491 	struct vfsoptlist *opts;
1492 	struct vfsconf *vfsp;
1493 	struct mount *mp = NULL;
1494 	int error;
1495 
1496 	vfsp = vfs_byname("devfs");
1497 	KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1498 	if (vfsp == NULL)
1499 		return;
1500 
1501 	mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred);
1502 
1503 	error = VFS_MOUNT(mp, td);
1504 	KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1505 	if (error)
1506 		return;
1507 
1508 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1509 	TAILQ_INIT(opts);
1510 	mp->mnt_opt = opts;
1511 
1512 	mtx_lock(&mountlist_mtx);
1513 	TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1514 	mtx_unlock(&mountlist_mtx);
1515 
1516 	set_rootvnode(td);
1517 
1518 	error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1519 	if (error)
1520 		printf("kern_symlink /dev -> / returns %d\n", error);
1521 }
1522 
1523 /*
1524  * Surgically move our devfs to be mounted on /dev.
1525  */
1526 
1527 static void
1528 devfs_fixup(struct thread *td)
1529 {
1530 	struct nameidata nd;
1531 	int error;
1532 	struct vnode *vp, *dvp;
1533 	struct mount *mp;
1534 
1535 	/* Remove our devfs mount from the mountlist and purge the cache */
1536 	mtx_lock(&mountlist_mtx);
1537 	mp = TAILQ_FIRST(&mountlist);
1538 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1539 	mtx_unlock(&mountlist_mtx);
1540 	cache_purgevfs(mp);
1541 
1542 	VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1543 	VI_LOCK(dvp);
1544 	dvp->v_iflag &= ~VI_MOUNT;
1545 	VI_UNLOCK(dvp);
1546 	dvp->v_mountedhere = NULL;
1547 
1548 	/* Set up the real rootvnode, and purge the cache */
1549 	TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1550 	set_rootvnode(td);
1551 	cache_purgevfs(rootvnode->v_mount);
1552 
1553 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1554 	error = namei(&nd);
1555 	if (error) {
1556 		printf("Lookup of /dev for devfs, error: %d\n", error);
1557 		return;
1558 	}
1559 	NDFREE(&nd, NDF_ONLY_PNBUF);
1560 	vp = nd.ni_vp;
1561 	if (vp->v_type != VDIR) {
1562 		vput(vp);
1563 	}
1564 	error = vinvalbuf(vp, V_SAVE, 0, 0);
1565 	if (error) {
1566 		vput(vp);
1567 	}
1568 	cache_purge(vp);
1569 	mp->mnt_vnodecovered = vp;
1570 	vp->v_mountedhere = mp;
1571 	mtx_lock(&mountlist_mtx);
1572 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1573 	mtx_unlock(&mountlist_mtx);
1574 	VOP_UNLOCK(vp, 0);
1575 	vput(dvp);
1576 	vfs_unbusy(mp);
1577 
1578 	/* Unlink the no longer needed /dev/dev -> / symlink */
1579 	kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1580 }
1581 
1582 /*
1583  * Report errors during filesystem mounting.
1584  */
1585 void
1586 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1587 {
1588 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1589 	va_list ap;
1590 	int error, len;
1591 	char *errmsg;
1592 
1593 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1594 	if (error || errmsg == NULL || len <= 0)
1595 		return;
1596 
1597 	va_start(ap, fmt);
1598 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1599 	va_end(ap);
1600 }
1601 
1602 /*
1603  * Find and mount the root filesystem
1604  */
1605 void
1606 vfs_mountroot(void)
1607 {
1608 	char *cp;
1609 	int error, i, asked = 0;
1610 
1611 	root_mount_prepare();
1612 
1613 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1614 	    NULL, NULL, mount_init, mount_fini,
1615 	    UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1616 	devfs_first();
1617 
1618 	/*
1619 	 * We are booted with instructions to prompt for the root filesystem.
1620 	 */
1621 	if (boothowto & RB_ASKNAME) {
1622 		if (!vfs_mountroot_ask())
1623 			goto mounted;
1624 		asked = 1;
1625 	}
1626 
1627 	/*
1628 	 * The root filesystem information is compiled in, and we are
1629 	 * booted with instructions to use it.
1630 	 */
1631 	if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1632 		if (!vfs_mountroot_try(ctrootdevname))
1633 			goto mounted;
1634 		ctrootdevname = NULL;
1635 	}
1636 
1637 	/*
1638 	 * We've been given the generic "use CDROM as root" flag.  This is
1639 	 * necessary because one media may be used in many different
1640 	 * devices, so we need to search for them.
1641 	 */
1642 	if (boothowto & RB_CDROM) {
1643 		for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1644 			if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1645 				goto mounted;
1646 		}
1647 	}
1648 
1649 	/*
1650 	 * Try to use the value read by the loader from /etc/fstab, or
1651 	 * supplied via some other means.  This is the preferred
1652 	 * mechanism.
1653 	 */
1654 	cp = getenv("vfs.root.mountfrom");
1655 	if (cp != NULL) {
1656 		error = vfs_mountroot_try(cp);
1657 		freeenv(cp);
1658 		if (!error)
1659 			goto mounted;
1660 	}
1661 
1662 	/*
1663 	 * Try values that may have been computed by code during boot
1664 	 */
1665 	if (!vfs_mountroot_try(rootdevnames[0]))
1666 		goto mounted;
1667 	if (!vfs_mountroot_try(rootdevnames[1]))
1668 		goto mounted;
1669 
1670 	/*
1671 	 * If we (still) have a compiled-in default, try it.
1672 	 */
1673 	if (ctrootdevname != NULL)
1674 		if (!vfs_mountroot_try(ctrootdevname))
1675 			goto mounted;
1676 	/*
1677 	 * Everything so far has failed, prompt on the console if we haven't
1678 	 * already tried that.
1679 	 */
1680 	if (!asked)
1681 		if (!vfs_mountroot_ask())
1682 			goto mounted;
1683 
1684 	panic("Root mount failed, startup aborted.");
1685 
1686 mounted:
1687 	root_mount_done();
1688 }
1689 
1690 /*
1691  * Mount (mountfrom) as the root filesystem.
1692  */
1693 static int
1694 vfs_mountroot_try(const char *mountfrom)
1695 {
1696 	struct mount	*mp;
1697 	char		*vfsname, *path;
1698 	time_t		timebase;
1699 	int		error;
1700 	char		patt[32];
1701 
1702 	vfsname = NULL;
1703 	path    = NULL;
1704 	mp      = NULL;
1705 	error   = EINVAL;
1706 
1707 	if (mountfrom == NULL)
1708 		return (error);		/* don't complain */
1709 	printf("Trying to mount root from %s\n", mountfrom);
1710 
1711 	/* parse vfs name and path */
1712 	vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1713 	path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1714 	vfsname[0] = path[0] = 0;
1715 	sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1716 	if (sscanf(mountfrom, patt, vfsname, path) < 1)
1717 		goto out;
1718 
1719 	if (path[0] == '\0')
1720 		strcpy(path, ROOTNAME);
1721 
1722 	error = kernel_vmount(
1723 	    MNT_RDONLY | MNT_ROOTFS,
1724 	    "fstype", vfsname,
1725 	    "fspath", "/",
1726 	    "from", path,
1727 	    NULL);
1728 	if (error == 0) {
1729 		/*
1730 		 * We mount devfs prior to mounting the / FS, so the first
1731 		 * entry will typically be devfs.
1732 		 */
1733 		mp = TAILQ_FIRST(&mountlist);
1734 		KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1735 
1736 		/*
1737 		 * Iterate over all currently mounted file systems and use
1738 		 * the time stamp found to check and/or initialize the RTC.
1739 		 * Typically devfs has no time stamp and the only other FS
1740 		 * is the actual / FS.
1741 		 * Call inittodr() only once and pass it the largest of the
1742 		 * timestamps we encounter.
1743 		 */
1744 		timebase = 0;
1745 		do {
1746 			if (mp->mnt_time > timebase)
1747 				timebase = mp->mnt_time;
1748 			mp = TAILQ_NEXT(mp, mnt_list);
1749 		} while (mp != NULL);
1750 		inittodr(timebase);
1751 
1752 		devfs_fixup(curthread);
1753 	}
1754 out:
1755 	free(path, M_MOUNT);
1756 	free(vfsname, M_MOUNT);
1757 	return (error);
1758 }
1759 
1760 /*
1761  * ---------------------------------------------------------------------
1762  * Interactive root filesystem selection code.
1763  */
1764 
1765 static int
1766 vfs_mountroot_ask(void)
1767 {
1768 	char name[128];
1769 
1770 	for(;;) {
1771 		printf("\nManual root filesystem specification:\n");
1772 		printf("  <fstype>:<device>  Mount <device> using filesystem <fstype>\n");
1773 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1774 		printf("                       eg. ufs:da0s1a\n");
1775 #else
1776 		printf("                       eg. ufs:/dev/da0a\n");
1777 #endif
1778 		printf("  ?                  List valid disk boot devices\n");
1779 		printf("  <empty line>       Abort manual input\n");
1780 		printf("\nmountroot> ");
1781 		gets(name, sizeof(name), 1);
1782 		if (name[0] == '\0')
1783 			return (1);
1784 		if (name[0] == '?') {
1785 			printf("\nList of GEOM managed disk devices:\n  ");
1786 			g_dev_print();
1787 			continue;
1788 		}
1789 		if (!vfs_mountroot_try(name))
1790 			return (0);
1791 	}
1792 }
1793 
1794 /*
1795  * ---------------------------------------------------------------------
1796  * Functions for querying mount options/arguments from filesystems.
1797  */
1798 
1799 /*
1800  * Check that no unknown options are given
1801  */
1802 int
1803 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1804 {
1805 	struct vfsopt *opt;
1806 	char errmsg[255];
1807 	const char **t, *p, *q;
1808 	int ret = 0;
1809 
1810 	TAILQ_FOREACH(opt, opts, link) {
1811 		p = opt->name;
1812 		q = NULL;
1813 		if (p[0] == 'n' && p[1] == 'o')
1814 			q = p + 2;
1815 		for(t = global_opts; *t != NULL; t++) {
1816 			if (strcmp(*t, p) == 0)
1817 				break;
1818 			if (q != NULL) {
1819 				if (strcmp(*t, q) == 0)
1820 					break;
1821 			}
1822 		}
1823 		if (*t != NULL)
1824 			continue;
1825 		for(t = legal; *t != NULL; t++) {
1826 			if (strcmp(*t, p) == 0)
1827 				break;
1828 			if (q != NULL) {
1829 				if (strcmp(*t, q) == 0)
1830 					break;
1831 			}
1832 		}
1833 		if (*t != NULL)
1834 			continue;
1835 		snprintf(errmsg, sizeof(errmsg),
1836 		    "mount option <%s> is unknown", p);
1837 		printf("%s\n", errmsg);
1838 		ret = EINVAL;
1839 	}
1840 	if (ret != 0) {
1841 		TAILQ_FOREACH(opt, opts, link) {
1842 			if (strcmp(opt->name, "errmsg") == 0) {
1843 				strncpy((char *)opt->value, errmsg, opt->len);
1844 			}
1845 		}
1846 	}
1847 	return (ret);
1848 }
1849 
1850 /*
1851  * Get a mount option by its name.
1852  *
1853  * Return 0 if the option was found, ENOENT otherwise.
1854  * If len is non-NULL it will be filled with the length
1855  * of the option. If buf is non-NULL, it will be filled
1856  * with the address of the option.
1857  */
1858 int
1859 vfs_getopt(opts, name, buf, len)
1860 	struct vfsoptlist *opts;
1861 	const char *name;
1862 	void **buf;
1863 	int *len;
1864 {
1865 	struct vfsopt *opt;
1866 
1867 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1868 
1869 	TAILQ_FOREACH(opt, opts, link) {
1870 		if (strcmp(name, opt->name) == 0) {
1871 			if (len != NULL)
1872 				*len = opt->len;
1873 			if (buf != NULL)
1874 				*buf = opt->value;
1875 			return (0);
1876 		}
1877 	}
1878 	return (ENOENT);
1879 }
1880 
1881 static int
1882 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1883 {
1884 	struct vfsopt *opt;
1885 	int i;
1886 
1887 	if (opts == NULL)
1888 		return (-1);
1889 
1890 	i = 0;
1891 	TAILQ_FOREACH(opt, opts, link) {
1892 		if (strcmp(name, opt->name) == 0)
1893 			return (i);
1894 		++i;
1895 	}
1896 	return (-1);
1897 }
1898 
1899 char *
1900 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1901 {
1902 	struct vfsopt *opt;
1903 
1904 	*error = 0;
1905 	TAILQ_FOREACH(opt, opts, link) {
1906 		if (strcmp(name, opt->name) != 0)
1907 			continue;
1908 		if (((char *)opt->value)[opt->len - 1] != '\0') {
1909 			*error = EINVAL;
1910 			return (NULL);
1911 		}
1912 		return (opt->value);
1913 	}
1914 	*error = ENOENT;
1915 	return (NULL);
1916 }
1917 
1918 int
1919 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1920 {
1921 	struct vfsopt *opt;
1922 
1923 	TAILQ_FOREACH(opt, opts, link) {
1924 		if (strcmp(name, opt->name) == 0) {
1925 			if (w != NULL)
1926 				*w |= val;
1927 			return (1);
1928 		}
1929 	}
1930 	if (w != NULL)
1931 		*w &= ~val;
1932 	return (0);
1933 }
1934 
1935 int
1936 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1937 {
1938 	va_list ap;
1939 	struct vfsopt *opt;
1940 	int ret;
1941 
1942 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1943 
1944 	TAILQ_FOREACH(opt, opts, link) {
1945 		if (strcmp(name, opt->name) != 0)
1946 			continue;
1947 		if (opt->len == 0 || opt->value == NULL)
1948 			return (0);
1949 		if (((char *)opt->value)[opt->len - 1] != '\0')
1950 			return (0);
1951 		va_start(ap, fmt);
1952 		ret = vsscanf(opt->value, fmt, ap);
1953 		va_end(ap);
1954 		return (ret);
1955 	}
1956 	return (0);
1957 }
1958 
1959 /*
1960  * Find and copy a mount option.
1961  *
1962  * The size of the buffer has to be specified
1963  * in len, if it is not the same length as the
1964  * mount option, EINVAL is returned.
1965  * Returns ENOENT if the option is not found.
1966  */
1967 int
1968 vfs_copyopt(opts, name, dest, len)
1969 	struct vfsoptlist *opts;
1970 	const char *name;
1971 	void *dest;
1972 	int len;
1973 {
1974 	struct vfsopt *opt;
1975 
1976 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1977 
1978 	TAILQ_FOREACH(opt, opts, link) {
1979 		if (strcmp(name, opt->name) == 0) {
1980 			if (len != opt->len)
1981 				return (EINVAL);
1982 			bcopy(opt->value, dest, opt->len);
1983 			return (0);
1984 		}
1985 	}
1986 	return (ENOENT);
1987 }
1988 
1989 /*
1990  * This is a helper function for filesystems to traverse their
1991  * vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h
1992  */
1993 
1994 struct vnode *
1995 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1996 {
1997 	struct vnode *vp;
1998 
1999 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2000 
2001 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2002 	if ((*mvp)->v_yield++ == 500) {
2003 		MNT_IUNLOCK(mp);
2004 		(*mvp)->v_yield = 0;
2005 		uio_yield();
2006 		MNT_ILOCK(mp);
2007 	}
2008 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
2009 	while (vp != NULL && vp->v_type == VMARKER)
2010 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2011 
2012 	/* Check if we are done */
2013 	if (vp == NULL) {
2014 		__mnt_vnode_markerfree(mvp, mp);
2015 		return (NULL);
2016 	}
2017 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2018 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2019 	return (vp);
2020 }
2021 
2022 struct vnode *
2023 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
2024 {
2025 	struct vnode *vp;
2026 
2027 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2028 
2029 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2030 	while (vp != NULL && vp->v_type == VMARKER)
2031 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2032 
2033 	/* Check if we are done */
2034 	if (vp == NULL) {
2035 		*mvp = NULL;
2036 		return (NULL);
2037 	}
2038 	MNT_REF(mp);
2039 	MNT_IUNLOCK(mp);
2040 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
2041 				       M_VNODE_MARKER,
2042 				       M_WAITOK | M_ZERO);
2043 	MNT_ILOCK(mp);
2044 	(*mvp)->v_type = VMARKER;
2045 
2046 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2047 	while (vp != NULL && vp->v_type == VMARKER)
2048 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
2049 
2050 	/* Check if we are done */
2051 	if (vp == NULL) {
2052 		MNT_IUNLOCK(mp);
2053 		free(*mvp, M_VNODE_MARKER);
2054 		MNT_ILOCK(mp);
2055 		*mvp = NULL;
2056 		MNT_REL(mp);
2057 		return (NULL);
2058 	}
2059 	(*mvp)->v_mount = mp;
2060 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2061 	return (vp);
2062 }
2063 
2064 
2065 void
2066 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
2067 {
2068 
2069 	if (*mvp == NULL)
2070 		return;
2071 
2072 	mtx_assert(MNT_MTX(mp), MA_OWNED);
2073 
2074 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2075 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2076 	MNT_IUNLOCK(mp);
2077 	free(*mvp, M_VNODE_MARKER);
2078 	MNT_ILOCK(mp);
2079 	*mvp = NULL;
2080 	MNT_REL(mp);
2081 }
2082 
2083 
2084 int
2085 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
2086 {
2087 	int error;
2088 
2089 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
2090 	if (sbp != &mp->mnt_stat)
2091 		*sbp = mp->mnt_stat;
2092 	return (error);
2093 }
2094 
2095 void
2096 vfs_mountedfrom(struct mount *mp, const char *from)
2097 {
2098 
2099 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2100 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2101 	    sizeof mp->mnt_stat.f_mntfromname);
2102 }
2103 
2104 /*
2105  * ---------------------------------------------------------------------
2106  * This is the api for building mount args and mounting filesystems from
2107  * inside the kernel.
2108  *
2109  * The API works by accumulation of individual args.  First error is
2110  * latched.
2111  *
2112  * XXX: should be documented in new manpage kernel_mount(9)
2113  */
2114 
2115 /* A memory allocation which must be freed when we are done */
2116 struct mntaarg {
2117 	SLIST_ENTRY(mntaarg)	next;
2118 };
2119 
2120 /* The header for the mount arguments */
2121 struct mntarg {
2122 	struct iovec *v;
2123 	int len;
2124 	int error;
2125 	SLIST_HEAD(, mntaarg)	list;
2126 };
2127 
2128 /*
2129  * Add a boolean argument.
2130  *
2131  * flag is the boolean value.
2132  * name must start with "no".
2133  */
2134 struct mntarg *
2135 mount_argb(struct mntarg *ma, int flag, const char *name)
2136 {
2137 
2138 	KASSERT(name[0] == 'n' && name[1] == 'o',
2139 	    ("mount_argb(...,%s): name must start with 'no'", name));
2140 
2141 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2142 }
2143 
2144 /*
2145  * Add an argument printf style
2146  */
2147 struct mntarg *
2148 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2149 {
2150 	va_list ap;
2151 	struct mntaarg *maa;
2152 	struct sbuf *sb;
2153 	int len;
2154 
2155 	if (ma == NULL) {
2156 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2157 		SLIST_INIT(&ma->list);
2158 	}
2159 	if (ma->error)
2160 		return (ma);
2161 
2162 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2163 	    M_MOUNT, M_WAITOK);
2164 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2165 	ma->v[ma->len].iov_len = strlen(name) + 1;
2166 	ma->len++;
2167 
2168 	sb = sbuf_new_auto();
2169 	va_start(ap, fmt);
2170 	sbuf_vprintf(sb, fmt, ap);
2171 	va_end(ap);
2172 	sbuf_finish(sb);
2173 	len = sbuf_len(sb) + 1;
2174 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2175 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2176 	bcopy(sbuf_data(sb), maa + 1, len);
2177 	sbuf_delete(sb);
2178 
2179 	ma->v[ma->len].iov_base = maa + 1;
2180 	ma->v[ma->len].iov_len = len;
2181 	ma->len++;
2182 
2183 	return (ma);
2184 }
2185 
2186 /*
2187  * Add an argument which is a userland string.
2188  */
2189 struct mntarg *
2190 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2191 {
2192 	struct mntaarg *maa;
2193 	char *tbuf;
2194 
2195 	if (val == NULL)
2196 		return (ma);
2197 	if (ma == NULL) {
2198 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2199 		SLIST_INIT(&ma->list);
2200 	}
2201 	if (ma->error)
2202 		return (ma);
2203 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2204 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2205 	tbuf = (void *)(maa + 1);
2206 	ma->error = copyinstr(val, tbuf, len, NULL);
2207 	return (mount_arg(ma, name, tbuf, -1));
2208 }
2209 
2210 /*
2211  * Plain argument.
2212  *
2213  * If length is -1, treat value as a C string.
2214  */
2215 struct mntarg *
2216 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2217 {
2218 
2219 	if (ma == NULL) {
2220 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2221 		SLIST_INIT(&ma->list);
2222 	}
2223 	if (ma->error)
2224 		return (ma);
2225 
2226 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2227 	    M_MOUNT, M_WAITOK);
2228 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2229 	ma->v[ma->len].iov_len = strlen(name) + 1;
2230 	ma->len++;
2231 
2232 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2233 	if (len < 0)
2234 		ma->v[ma->len].iov_len = strlen(val) + 1;
2235 	else
2236 		ma->v[ma->len].iov_len = len;
2237 	ma->len++;
2238 	return (ma);
2239 }
2240 
2241 /*
2242  * Free a mntarg structure
2243  */
2244 static void
2245 free_mntarg(struct mntarg *ma)
2246 {
2247 	struct mntaarg *maa;
2248 
2249 	while (!SLIST_EMPTY(&ma->list)) {
2250 		maa = SLIST_FIRST(&ma->list);
2251 		SLIST_REMOVE_HEAD(&ma->list, next);
2252 		free(maa, M_MOUNT);
2253 	}
2254 	free(ma->v, M_MOUNT);
2255 	free(ma, M_MOUNT);
2256 }
2257 
2258 /*
2259  * Mount a filesystem
2260  */
2261 int
2262 kernel_mount(struct mntarg *ma, int flags)
2263 {
2264 	struct uio auio;
2265 	int error;
2266 
2267 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2268 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2269 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2270 
2271 	auio.uio_iov = ma->v;
2272 	auio.uio_iovcnt = ma->len;
2273 	auio.uio_segflg = UIO_SYSSPACE;
2274 
2275 	error = ma->error;
2276 	if (!error)
2277 		error = vfs_donmount(curthread, flags, &auio);
2278 	free_mntarg(ma);
2279 	return (error);
2280 }
2281 
2282 /*
2283  * A printflike function to mount a filesystem.
2284  */
2285 int
2286 kernel_vmount(int flags, ...)
2287 {
2288 	struct mntarg *ma = NULL;
2289 	va_list ap;
2290 	const char *cp;
2291 	const void *vp;
2292 	int error;
2293 
2294 	va_start(ap, flags);
2295 	for (;;) {
2296 		cp = va_arg(ap, const char *);
2297 		if (cp == NULL)
2298 			break;
2299 		vp = va_arg(ap, const void *);
2300 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2301 	}
2302 	va_end(ap);
2303 
2304 	error = kernel_mount(ma, flags);
2305 	return (error);
2306 }
2307