xref: /freebsd/sys/kern/vfs_mount.c (revision 641a6cfb86023499caafe26a4d821a0b885cf00b)
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 "opt_vfs_allow_nonmpsafe.h"
41 
42 #include <sys/param.h>
43 #include <sys/conf.h>
44 #include <sys/fcntl.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/libkern.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/filedesc.h>
55 #include <sys/reboot.h>
56 #include <sys/sbuf.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysproto.h>
59 #include <sys/sx.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/systm.h>
63 #include <sys/vnode.h>
64 #include <vm/uma.h>
65 
66 #include <geom/geom.h>
67 
68 #include <machine/stdarg.h>
69 
70 #include <security/audit/audit.h>
71 #include <security/mac/mac_framework.h>
72 
73 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
74 
75 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
76 		    uint64_t fsflags, struct vfsoptlist **optlist);
77 static void	free_mntarg(struct mntarg *ma);
78 
79 static int	usermount = 0;
80 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
81     "Unprivileged users may mount and unmount file systems");
82 
83 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
84 static uma_zone_t mount_zone;
85 
86 /* List of mounted filesystems. */
87 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
88 
89 /* For any iteration/modification of mountlist */
90 struct mtx mountlist_mtx;
91 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
92 
93 /*
94  * Global opts, taken by all filesystems
95  */
96 static const char *global_opts[] = {
97 	"errmsg",
98 	"fstype",
99 	"fspath",
100 	"ro",
101 	"rw",
102 	"nosuid",
103 	"noexec",
104 	NULL
105 };
106 
107 static int
108 mount_init(void *mem, int size, int flags)
109 {
110 	struct mount *mp;
111 
112 	mp = (struct mount *)mem;
113 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
114 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
115 	return (0);
116 }
117 
118 static void
119 mount_fini(void *mem, int size)
120 {
121 	struct mount *mp;
122 
123 	mp = (struct mount *)mem;
124 	lockdestroy(&mp->mnt_explock);
125 	mtx_destroy(&mp->mnt_mtx);
126 }
127 
128 static void
129 vfs_mount_init(void *dummy __unused)
130 {
131 
132 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
133 	    NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
134 }
135 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
136 
137 /*
138  * ---------------------------------------------------------------------
139  * Functions for building and sanitizing the mount options
140  */
141 
142 /* Remove one mount option. */
143 static void
144 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
145 {
146 
147 	TAILQ_REMOVE(opts, opt, link);
148 	free(opt->name, M_MOUNT);
149 	if (opt->value != NULL)
150 		free(opt->value, M_MOUNT);
151 	free(opt, M_MOUNT);
152 }
153 
154 /* Release all resources related to the mount options. */
155 void
156 vfs_freeopts(struct vfsoptlist *opts)
157 {
158 	struct vfsopt *opt;
159 
160 	while (!TAILQ_EMPTY(opts)) {
161 		opt = TAILQ_FIRST(opts);
162 		vfs_freeopt(opts, opt);
163 	}
164 	free(opts, M_MOUNT);
165 }
166 
167 void
168 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
169 {
170 	struct vfsopt *opt, *temp;
171 
172 	if (opts == NULL)
173 		return;
174 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
175 		if (strcmp(opt->name, name) == 0)
176 			vfs_freeopt(opts, opt);
177 	}
178 }
179 
180 static int
181 vfs_isopt_ro(const char *opt)
182 {
183 
184 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
185 	    strcmp(opt, "norw") == 0)
186 		return (1);
187 	return (0);
188 }
189 
190 static int
191 vfs_isopt_rw(const char *opt)
192 {
193 
194 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
195 		return (1);
196 	return (0);
197 }
198 
199 /*
200  * Check if options are equal (with or without the "no" prefix).
201  */
202 static int
203 vfs_equalopts(const char *opt1, const char *opt2)
204 {
205 	char *p;
206 
207 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
208 	if (strcmp(opt1, opt2) == 0)
209 		return (1);
210 	/* "noopt" vs. "opt" */
211 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
212 		return (1);
213 	/* "opt" vs. "noopt" */
214 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
215 		return (1);
216 	while ((p = strchr(opt1, '.')) != NULL &&
217 	    !strncmp(opt1, opt2, ++p - opt1)) {
218 		opt2 += p - opt1;
219 		opt1 = p;
220 		/* "foo.noopt" vs. "foo.opt" */
221 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
222 			return (1);
223 		/* "foo.opt" vs. "foo.noopt" */
224 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
225 			return (1);
226 	}
227 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
228 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
229 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
230 		return (1);
231 	return (0);
232 }
233 
234 /*
235  * If a mount option is specified several times,
236  * (with or without the "no" prefix) only keep
237  * the last occurence of it.
238  */
239 static void
240 vfs_sanitizeopts(struct vfsoptlist *opts)
241 {
242 	struct vfsopt *opt, *opt2, *tmp;
243 
244 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
245 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
246 		while (opt2 != NULL) {
247 			if (vfs_equalopts(opt->name, opt2->name)) {
248 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
249 				vfs_freeopt(opts, opt2);
250 				opt2 = tmp;
251 			} else {
252 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
253 			}
254 		}
255 	}
256 }
257 
258 /*
259  * Build a linked list of mount options from a struct uio.
260  */
261 int
262 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
263 {
264 	struct vfsoptlist *opts;
265 	struct vfsopt *opt;
266 	size_t memused, namelen, optlen;
267 	unsigned int i, iovcnt;
268 	int error;
269 
270 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
271 	TAILQ_INIT(opts);
272 	memused = 0;
273 	iovcnt = auio->uio_iovcnt;
274 	for (i = 0; i < iovcnt; i += 2) {
275 		namelen = auio->uio_iov[i].iov_len;
276 		optlen = auio->uio_iov[i + 1].iov_len;
277 		memused += sizeof(struct vfsopt) + optlen + namelen;
278 		/*
279 		 * Avoid consuming too much memory, and attempts to overflow
280 		 * memused.
281 		 */
282 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
283 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
284 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
285 			error = EINVAL;
286 			goto bad;
287 		}
288 
289 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
290 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
291 		opt->value = NULL;
292 		opt->len = 0;
293 		opt->pos = i / 2;
294 		opt->seen = 0;
295 
296 		/*
297 		 * Do this early, so jumps to "bad" will free the current
298 		 * option.
299 		 */
300 		TAILQ_INSERT_TAIL(opts, opt, link);
301 
302 		if (auio->uio_segflg == UIO_SYSSPACE) {
303 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
304 		} else {
305 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
306 			    namelen);
307 			if (error)
308 				goto bad;
309 		}
310 		/* Ensure names are null-terminated strings. */
311 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
312 			error = EINVAL;
313 			goto bad;
314 		}
315 		if (optlen != 0) {
316 			opt->len = optlen;
317 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
318 			if (auio->uio_segflg == UIO_SYSSPACE) {
319 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
320 				    optlen);
321 			} else {
322 				error = copyin(auio->uio_iov[i + 1].iov_base,
323 				    opt->value, optlen);
324 				if (error)
325 					goto bad;
326 			}
327 		}
328 	}
329 	vfs_sanitizeopts(opts);
330 	*options = opts;
331 	return (0);
332 bad:
333 	vfs_freeopts(opts);
334 	return (error);
335 }
336 
337 /*
338  * Merge the old mount options with the new ones passed
339  * in the MNT_UPDATE case.
340  *
341  * XXX: This function will keep a "nofoo" option in the new
342  * options.  E.g, if the option's canonical name is "foo",
343  * "nofoo" ends up in the mount point's active options.
344  */
345 static void
346 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
347 {
348 	struct vfsopt *opt, *new;
349 
350 	TAILQ_FOREACH(opt, oldopts, link) {
351 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
352 		new->name = strdup(opt->name, M_MOUNT);
353 		if (opt->len != 0) {
354 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
355 			bcopy(opt->value, new->value, opt->len);
356 		} else
357 			new->value = NULL;
358 		new->len = opt->len;
359 		new->seen = opt->seen;
360 		TAILQ_INSERT_HEAD(toopts, new, link);
361 	}
362 	vfs_sanitizeopts(toopts);
363 }
364 
365 /*
366  * Mount a filesystem.
367  */
368 int
369 sys_nmount(td, uap)
370 	struct thread *td;
371 	struct nmount_args /* {
372 		struct iovec *iovp;
373 		unsigned int iovcnt;
374 		int flags;
375 	} */ *uap;
376 {
377 	struct uio *auio;
378 	int error;
379 	u_int iovcnt;
380 	uint64_t flags;
381 
382 	/*
383 	 * Mount flags are now 64-bits. On 32-bit archtectures only
384 	 * 32-bits are passed in, but from here on everything handles
385 	 * 64-bit flags correctly.
386 	 */
387 	flags = uap->flags;
388 
389 	AUDIT_ARG_FFLAGS(flags);
390 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
391 	    uap->iovp, uap->iovcnt, flags);
392 
393 	/*
394 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
395 	 * userspace to set this flag, but we must filter it out if we want
396 	 * MNT_UPDATE on the root file system to work.
397 	 * MNT_ROOTFS should only be set by the kernel when mounting its
398 	 * root file system.
399 	 */
400 	flags &= ~MNT_ROOTFS;
401 
402 	iovcnt = uap->iovcnt;
403 	/*
404 	 * Check that we have an even number of iovec's
405 	 * and that we have at least two options.
406 	 */
407 	if ((iovcnt & 1) || (iovcnt < 4)) {
408 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
409 		    uap->iovcnt);
410 		return (EINVAL);
411 	}
412 
413 	error = copyinuio(uap->iovp, iovcnt, &auio);
414 	if (error) {
415 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
416 		    __func__, error);
417 		return (error);
418 	}
419 	error = vfs_donmount(td, flags, auio);
420 
421 	free(auio, M_IOV);
422 	return (error);
423 }
424 
425 /*
426  * ---------------------------------------------------------------------
427  * Various utility functions
428  */
429 
430 void
431 vfs_ref(struct mount *mp)
432 {
433 
434 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
435 	MNT_ILOCK(mp);
436 	MNT_REF(mp);
437 	MNT_IUNLOCK(mp);
438 }
439 
440 void
441 vfs_rel(struct mount *mp)
442 {
443 
444 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
445 	MNT_ILOCK(mp);
446 	MNT_REL(mp);
447 	MNT_IUNLOCK(mp);
448 }
449 
450 /*
451  * Allocate and initialize the mount point struct.
452  */
453 struct mount *
454 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
455     struct ucred *cred)
456 {
457 	struct mount *mp;
458 
459 	mp = uma_zalloc(mount_zone, M_WAITOK);
460 	bzero(&mp->mnt_startzero,
461 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
462 	TAILQ_INIT(&mp->mnt_nvnodelist);
463 	mp->mnt_nvnodelistsize = 0;
464 	TAILQ_INIT(&mp->mnt_activevnodelist);
465 	mp->mnt_activevnodelistsize = 0;
466 	mp->mnt_ref = 0;
467 	(void) vfs_busy(mp, MBF_NOWAIT);
468 	mp->mnt_op = vfsp->vfc_vfsops;
469 	mp->mnt_vfc = vfsp;
470 	vfsp->vfc_refcount++;	/* XXX Unlocked */
471 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
472 	mp->mnt_gen++;
473 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
474 	mp->mnt_vnodecovered = vp;
475 	mp->mnt_cred = crdup(cred);
476 	mp->mnt_stat.f_owner = cred->cr_uid;
477 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
478 	mp->mnt_iosize_max = DFLTPHYS;
479 #ifdef MAC
480 	mac_mount_init(mp);
481 	mac_mount_create(cred, mp);
482 #endif
483 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
484 	return (mp);
485 }
486 
487 /*
488  * Destroy the mount struct previously allocated by vfs_mount_alloc().
489  */
490 void
491 vfs_mount_destroy(struct mount *mp)
492 {
493 
494 	MNT_ILOCK(mp);
495 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
496 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
497 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
498 		wakeup(mp);
499 	}
500 	while (mp->mnt_ref)
501 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
502 	KASSERT(mp->mnt_ref == 0,
503 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
504 	    __FILE__, __LINE__));
505 	if (mp->mnt_writeopcount != 0)
506 		panic("vfs_mount_destroy: nonzero writeopcount");
507 	if (mp->mnt_secondary_writes != 0)
508 		panic("vfs_mount_destroy: nonzero secondary_writes");
509 	mp->mnt_vfc->vfc_refcount--;
510 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
511 		struct vnode *vp;
512 
513 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
514 			vprint("", vp);
515 		panic("unmount: dangling vnode");
516 	}
517 	if (mp->mnt_nvnodelistsize != 0)
518 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
519 	if (mp->mnt_activevnodelistsize != 0)
520 		panic("vfs_mount_destroy: nonzero activevnodelistsize");
521 	if (mp->mnt_lockref != 0)
522 		panic("vfs_mount_destroy: nonzero lock refcount");
523 	MNT_IUNLOCK(mp);
524 #ifdef MAC
525 	mac_mount_destroy(mp);
526 #endif
527 	if (mp->mnt_opt != NULL)
528 		vfs_freeopts(mp->mnt_opt);
529 	crfree(mp->mnt_cred);
530 	uma_zfree(mount_zone, mp);
531 }
532 
533 int
534 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
535 {
536 	struct vfsoptlist *optlist;
537 	struct vfsopt *opt, *tmp_opt;
538 	char *fstype, *fspath, *errmsg;
539 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
540 
541 	errmsg = fspath = NULL;
542 	errmsg_len = fspathlen = 0;
543 	errmsg_pos = -1;
544 
545 	error = vfs_buildopts(fsoptions, &optlist);
546 	if (error)
547 		return (error);
548 
549 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
550 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
551 
552 	/*
553 	 * We need these two options before the others,
554 	 * and they are mandatory for any filesystem.
555 	 * Ensure they are NUL terminated as well.
556 	 */
557 	fstypelen = 0;
558 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
559 	if (error || fstype[fstypelen - 1] != '\0') {
560 		error = EINVAL;
561 		if (errmsg != NULL)
562 			strncpy(errmsg, "Invalid fstype", errmsg_len);
563 		goto bail;
564 	}
565 	fspathlen = 0;
566 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
567 	if (error || fspath[fspathlen - 1] != '\0') {
568 		error = EINVAL;
569 		if (errmsg != NULL)
570 			strncpy(errmsg, "Invalid fspath", errmsg_len);
571 		goto bail;
572 	}
573 
574 	/*
575 	 * We need to see if we have the "update" option
576 	 * before we call vfs_domount(), since vfs_domount() has special
577 	 * logic based on MNT_UPDATE.  This is very important
578 	 * when we want to update the root filesystem.
579 	 */
580 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
581 		if (strcmp(opt->name, "update") == 0) {
582 			fsflags |= MNT_UPDATE;
583 			vfs_freeopt(optlist, opt);
584 		}
585 		else if (strcmp(opt->name, "async") == 0)
586 			fsflags |= MNT_ASYNC;
587 		else if (strcmp(opt->name, "force") == 0) {
588 			fsflags |= MNT_FORCE;
589 			vfs_freeopt(optlist, opt);
590 		}
591 		else if (strcmp(opt->name, "reload") == 0) {
592 			fsflags |= MNT_RELOAD;
593 			vfs_freeopt(optlist, opt);
594 		}
595 		else if (strcmp(opt->name, "multilabel") == 0)
596 			fsflags |= MNT_MULTILABEL;
597 		else if (strcmp(opt->name, "noasync") == 0)
598 			fsflags &= ~MNT_ASYNC;
599 		else if (strcmp(opt->name, "noatime") == 0)
600 			fsflags |= MNT_NOATIME;
601 		else if (strcmp(opt->name, "atime") == 0) {
602 			free(opt->name, M_MOUNT);
603 			opt->name = strdup("nonoatime", M_MOUNT);
604 		}
605 		else if (strcmp(opt->name, "noclusterr") == 0)
606 			fsflags |= MNT_NOCLUSTERR;
607 		else if (strcmp(opt->name, "clusterr") == 0) {
608 			free(opt->name, M_MOUNT);
609 			opt->name = strdup("nonoclusterr", M_MOUNT);
610 		}
611 		else if (strcmp(opt->name, "noclusterw") == 0)
612 			fsflags |= MNT_NOCLUSTERW;
613 		else if (strcmp(opt->name, "clusterw") == 0) {
614 			free(opt->name, M_MOUNT);
615 			opt->name = strdup("nonoclusterw", M_MOUNT);
616 		}
617 		else if (strcmp(opt->name, "noexec") == 0)
618 			fsflags |= MNT_NOEXEC;
619 		else if (strcmp(opt->name, "exec") == 0) {
620 			free(opt->name, M_MOUNT);
621 			opt->name = strdup("nonoexec", M_MOUNT);
622 		}
623 		else if (strcmp(opt->name, "nosuid") == 0)
624 			fsflags |= MNT_NOSUID;
625 		else if (strcmp(opt->name, "suid") == 0) {
626 			free(opt->name, M_MOUNT);
627 			opt->name = strdup("nonosuid", M_MOUNT);
628 		}
629 		else if (strcmp(opt->name, "nosymfollow") == 0)
630 			fsflags |= MNT_NOSYMFOLLOW;
631 		else if (strcmp(opt->name, "symfollow") == 0) {
632 			free(opt->name, M_MOUNT);
633 			opt->name = strdup("nonosymfollow", M_MOUNT);
634 		}
635 		else if (strcmp(opt->name, "noro") == 0)
636 			fsflags &= ~MNT_RDONLY;
637 		else if (strcmp(opt->name, "rw") == 0)
638 			fsflags &= ~MNT_RDONLY;
639 		else if (strcmp(opt->name, "ro") == 0)
640 			fsflags |= MNT_RDONLY;
641 		else if (strcmp(opt->name, "rdonly") == 0) {
642 			free(opt->name, M_MOUNT);
643 			opt->name = strdup("ro", M_MOUNT);
644 			fsflags |= MNT_RDONLY;
645 		}
646 		else if (strcmp(opt->name, "suiddir") == 0)
647 			fsflags |= MNT_SUIDDIR;
648 		else if (strcmp(opt->name, "sync") == 0)
649 			fsflags |= MNT_SYNCHRONOUS;
650 		else if (strcmp(opt->name, "union") == 0)
651 			fsflags |= MNT_UNION;
652 	}
653 
654 	/*
655 	 * Be ultra-paranoid about making sure the type and fspath
656 	 * variables will fit in our mp buffers, including the
657 	 * terminating NUL.
658 	 */
659 	if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
660 		error = ENAMETOOLONG;
661 		goto bail;
662 	}
663 
664 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
665 bail:
666 	/* copyout the errmsg */
667 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
668 	    && errmsg_len > 0 && errmsg != NULL) {
669 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
670 			bcopy(errmsg,
671 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
672 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
673 		} else {
674 			copyout(errmsg,
675 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
676 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
677 		}
678 	}
679 
680 	if (optlist != NULL)
681 		vfs_freeopts(optlist);
682 	return (error);
683 }
684 
685 /*
686  * Old mount API.
687  */
688 #ifndef _SYS_SYSPROTO_H_
689 struct mount_args {
690 	char	*type;
691 	char	*path;
692 	int	flags;
693 	caddr_t	data;
694 };
695 #endif
696 /* ARGSUSED */
697 int
698 sys_mount(td, uap)
699 	struct thread *td;
700 	struct mount_args /* {
701 		char *type;
702 		char *path;
703 		int flags;
704 		caddr_t data;
705 	} */ *uap;
706 {
707 	char *fstype;
708 	struct vfsconf *vfsp = NULL;
709 	struct mntarg *ma = NULL;
710 	uint64_t flags;
711 	int error;
712 
713 	/*
714 	 * Mount flags are now 64-bits. On 32-bit archtectures only
715 	 * 32-bits are passed in, but from here on everything handles
716 	 * 64-bit flags correctly.
717 	 */
718 	flags = uap->flags;
719 
720 	AUDIT_ARG_FFLAGS(flags);
721 
722 	/*
723 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
724 	 * userspace to set this flag, but we must filter it out if we want
725 	 * MNT_UPDATE on the root file system to work.
726 	 * MNT_ROOTFS should only be set by the kernel when mounting its
727 	 * root file system.
728 	 */
729 	flags &= ~MNT_ROOTFS;
730 
731 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
732 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
733 	if (error) {
734 		free(fstype, M_TEMP);
735 		return (error);
736 	}
737 
738 	AUDIT_ARG_TEXT(fstype);
739 	mtx_lock(&Giant);
740 	vfsp = vfs_byname_kld(fstype, td, &error);
741 	free(fstype, M_TEMP);
742 	if (vfsp == NULL) {
743 		mtx_unlock(&Giant);
744 		return (ENOENT);
745 	}
746 	if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
747 		mtx_unlock(&Giant);
748 		return (EOPNOTSUPP);
749 	}
750 
751 	ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
752 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
753 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
754 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
755 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
756 
757 	error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags);
758 	mtx_unlock(&Giant);
759 	return (error);
760 }
761 
762 /*
763  * vfs_domount_first(): first file system mount (not update)
764  */
765 static int
766 vfs_domount_first(
767 	struct thread *td,		/* Calling thread. */
768 	struct vfsconf *vfsp,		/* File system type. */
769 	char *fspath,			/* Mount path. */
770 	struct vnode *vp,		/* Vnode to be covered. */
771 	uint64_t fsflags,		/* Flags common to all filesystems. */
772 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
773 	)
774 {
775 	struct vattr va;
776 	struct mount *mp;
777 	struct vnode *newdp;
778 	int error;
779 
780 	mtx_assert(&Giant, MA_OWNED);
781 	ASSERT_VOP_ELOCKED(vp, __func__);
782 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
783 
784 	/*
785 	 * If the user is not root, ensure that they own the directory
786 	 * onto which we are attempting to mount.
787 	 */
788 	error = VOP_GETATTR(vp, &va, td->td_ucred);
789 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
790 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0);
791 	if (error == 0)
792 		error = vinvalbuf(vp, V_SAVE, 0, 0);
793 	if (error == 0 && vp->v_type != VDIR)
794 		error = ENOTDIR;
795 	if (error == 0) {
796 		VI_LOCK(vp);
797 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
798 			vp->v_iflag |= VI_MOUNT;
799 		else
800 			error = EBUSY;
801 		VI_UNLOCK(vp);
802 	}
803 	if (error != 0) {
804 		vput(vp);
805 		return (error);
806 	}
807 	VOP_UNLOCK(vp, 0);
808 
809 	/* Allocate and initialize the filesystem. */
810 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
811 	/* XXXMAC: pass to vfs_mount_alloc? */
812 	mp->mnt_optnew = *optlist;
813 	/* Set the mount level flags. */
814 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
815 
816 	/*
817 	 * Mount the filesystem.
818 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
819 	 * get.  No freeing of cn_pnbuf.
820 	 */
821 	error = VFS_MOUNT(mp);
822 #ifndef VFS_ALLOW_NONMPSAFE
823 	if (error == 0 && VFS_NEEDSGIANT(mp)) {
824 		(void)VFS_UNMOUNT(mp, fsflags);
825 		error = ENXIO;
826 		printf("%s: Mounting non-MPSAFE fs (%s) is disabled\n",
827 		    __func__, mp->mnt_vfc->vfc_name);
828 	}
829 #endif
830 	if (error != 0) {
831 		vfs_unbusy(mp);
832 		vfs_mount_destroy(mp);
833 		VI_LOCK(vp);
834 		vp->v_iflag &= ~VI_MOUNT;
835 		VI_UNLOCK(vp);
836 		vrele(vp);
837 		return (error);
838 	}
839 #ifdef VFS_ALLOW_NONMPSAFE
840 	if (VFS_NEEDSGIANT(mp))
841 		printf("%s: Mounting non-MPSAFE fs (%s) is deprecated\n",
842 		    __func__, mp->mnt_vfc->vfc_name);
843 #endif
844 
845 	if (mp->mnt_opt != NULL)
846 		vfs_freeopts(mp->mnt_opt);
847 	mp->mnt_opt = mp->mnt_optnew;
848 	*optlist = NULL;
849 	(void)VFS_STATFS(mp, &mp->mnt_stat);
850 
851 	/*
852 	 * Prevent external consumers of mount options from reading mnt_optnew.
853 	 */
854 	mp->mnt_optnew = NULL;
855 
856 	MNT_ILOCK(mp);
857 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
858 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
859 		mp->mnt_kern_flag |= MNTK_ASYNC;
860 	else
861 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
862 	MNT_IUNLOCK(mp);
863 
864 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
865 	cache_purge(vp);
866 	VI_LOCK(vp);
867 	vp->v_iflag &= ~VI_MOUNT;
868 	VI_UNLOCK(vp);
869 	vp->v_mountedhere = mp;
870 	/* Place the new filesystem at the end of the mount list. */
871 	mtx_lock(&mountlist_mtx);
872 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
873 	mtx_unlock(&mountlist_mtx);
874 	vfs_event_signal(NULL, VQ_MOUNT, 0);
875 	if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp))
876 		panic("mount: lost mount");
877 	VOP_UNLOCK(newdp, 0);
878 	VOP_UNLOCK(vp, 0);
879 	mountcheckdirs(vp, newdp);
880 	vrele(newdp);
881 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
882 		vfs_allocate_syncvnode(mp);
883 	vfs_unbusy(mp);
884 	return (0);
885 }
886 
887 /*
888  * vfs_domount_update(): update of mounted file system
889  */
890 static int
891 vfs_domount_update(
892 	struct thread *td,		/* Calling thread. */
893 	struct vnode *vp,		/* Mount point vnode. */
894 	uint64_t fsflags,		/* Flags common to all filesystems. */
895 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
896 	)
897 {
898 	struct oexport_args oexport;
899 	struct export_args export;
900 	struct mount *mp;
901 	int error, export_error;
902 	uint64_t flag;
903 
904 	mtx_assert(&Giant, MA_OWNED);
905 	ASSERT_VOP_ELOCKED(vp, __func__);
906 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
907 
908 	if ((vp->v_vflag & VV_ROOT) == 0) {
909 		vput(vp);
910 		return (EINVAL);
911 	}
912 	mp = vp->v_mount;
913 	/*
914 	 * We only allow the filesystem to be reloaded if it
915 	 * is currently mounted read-only.
916 	 */
917 	flag = mp->mnt_flag;
918 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
919 		vput(vp);
920 		return (EOPNOTSUPP);	/* Needs translation */
921 	}
922 	/*
923 	 * Only privileged root, or (if MNT_USER is set) the user that
924 	 * did the original mount is permitted to update it.
925 	 */
926 	error = vfs_suser(mp, td);
927 	if (error != 0) {
928 		vput(vp);
929 		return (error);
930 	}
931 	if (vfs_busy(mp, MBF_NOWAIT)) {
932 		vput(vp);
933 		return (EBUSY);
934 	}
935 	VI_LOCK(vp);
936 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
937 		VI_UNLOCK(vp);
938 		vfs_unbusy(mp);
939 		vput(vp);
940 		return (EBUSY);
941 	}
942 	vp->v_iflag |= VI_MOUNT;
943 	VI_UNLOCK(vp);
944 	VOP_UNLOCK(vp, 0);
945 
946 	MNT_ILOCK(mp);
947 	mp->mnt_flag &= ~MNT_UPDATEMASK;
948 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
949 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
950 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
951 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
952 	MNT_IUNLOCK(mp);
953 	mp->mnt_optnew = *optlist;
954 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
955 
956 	/*
957 	 * Mount the filesystem.
958 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
959 	 * get.  No freeing of cn_pnbuf.
960 	 */
961 	error = VFS_MOUNT(mp);
962 
963 	export_error = 0;
964 	if (error == 0) {
965 		/* Process the export option. */
966 		if (vfs_copyopt(mp->mnt_optnew, "export", &export,
967 		    sizeof(export)) == 0) {
968 			export_error = vfs_export(mp, &export);
969 		} else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport,
970 		    sizeof(oexport)) == 0) {
971 			export.ex_flags = oexport.ex_flags;
972 			export.ex_root = oexport.ex_root;
973 			export.ex_anon = oexport.ex_anon;
974 			export.ex_addr = oexport.ex_addr;
975 			export.ex_addrlen = oexport.ex_addrlen;
976 			export.ex_mask = oexport.ex_mask;
977 			export.ex_masklen = oexport.ex_masklen;
978 			export.ex_indexfile = oexport.ex_indexfile;
979 			export.ex_numsecflavors = 0;
980 			export_error = vfs_export(mp, &export);
981 		}
982 	}
983 
984 	MNT_ILOCK(mp);
985 	if (error == 0) {
986 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
987 		    MNT_SNAPSHOT);
988 	} else {
989 		/*
990 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
991 		 * because it is not part of MNT_UPDATEMASK, but it could have
992 		 * changed in the meantime if quotactl(2) was called.
993 		 * All in all we want current value of MNT_QUOTA, not the old
994 		 * one.
995 		 */
996 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
997 	}
998 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
999 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1000 		mp->mnt_kern_flag |= MNTK_ASYNC;
1001 	else
1002 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1003 	MNT_IUNLOCK(mp);
1004 
1005 	if (error != 0)
1006 		goto end;
1007 
1008 	if (mp->mnt_opt != NULL)
1009 		vfs_freeopts(mp->mnt_opt);
1010 	mp->mnt_opt = mp->mnt_optnew;
1011 	*optlist = NULL;
1012 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1013 	/*
1014 	 * Prevent external consumers of mount options from reading
1015 	 * mnt_optnew.
1016 	 */
1017 	mp->mnt_optnew = NULL;
1018 
1019 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1020 		vfs_allocate_syncvnode(mp);
1021 	else
1022 		vfs_deallocate_syncvnode(mp);
1023 end:
1024 	vfs_unbusy(mp);
1025 	VI_LOCK(vp);
1026 	vp->v_iflag &= ~VI_MOUNT;
1027 	VI_UNLOCK(vp);
1028 	vrele(vp);
1029 	return (error != 0 ? error : export_error);
1030 }
1031 
1032 /*
1033  * vfs_domount(): actually attempt a filesystem mount.
1034  */
1035 static int
1036 vfs_domount(
1037 	struct thread *td,		/* Calling thread. */
1038 	const char *fstype,		/* Filesystem type. */
1039 	char *fspath,			/* Mount path. */
1040 	uint64_t fsflags,		/* Flags common to all filesystems. */
1041 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1042 	)
1043 {
1044 	struct vfsconf *vfsp;
1045 	struct nameidata nd;
1046 	struct vnode *vp;
1047 	char *pathbuf;
1048 	int error;
1049 
1050 	/*
1051 	 * Be ultra-paranoid about making sure the type and fspath
1052 	 * variables will fit in our mp buffers, including the
1053 	 * terminating NUL.
1054 	 */
1055 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1056 		return (ENAMETOOLONG);
1057 
1058 	if (jailed(td->td_ucred) || usermount == 0) {
1059 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1060 			return (error);
1061 	}
1062 
1063 	/*
1064 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1065 	 */
1066 	if (fsflags & MNT_EXPORTED) {
1067 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1068 		if (error)
1069 			return (error);
1070 	}
1071 	if (fsflags & MNT_SUIDDIR) {
1072 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1073 		if (error)
1074 			return (error);
1075 	}
1076 	/*
1077 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1078 	 */
1079 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1080 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1081 			fsflags |= MNT_NOSUID | MNT_USER;
1082 	}
1083 
1084 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1085 	vfsp = NULL;
1086 	if ((fsflags & MNT_UPDATE) == 0) {
1087 		/* Don't try to load KLDs if we're mounting the root. */
1088 		if (fsflags & MNT_ROOTFS)
1089 			vfsp = vfs_byname(fstype);
1090 		else
1091 			vfsp = vfs_byname_kld(fstype, td, &error);
1092 		if (vfsp == NULL)
1093 			return (ENODEV);
1094 		if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
1095 			return (EPERM);
1096 	}
1097 
1098 	/*
1099 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1100 	 */
1101 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | MPSAFE | AUDITVNODE1,
1102 	    UIO_SYSSPACE, fspath, td);
1103 	error = namei(&nd);
1104 	if (error != 0)
1105 		return (error);
1106 	if (!NDHASGIANT(&nd))
1107 		mtx_lock(&Giant);
1108 	NDFREE(&nd, NDF_ONLY_PNBUF);
1109 	vp = nd.ni_vp;
1110 	if ((fsflags & MNT_UPDATE) == 0) {
1111 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1112 		strcpy(pathbuf, fspath);
1113 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1114 		/* debug.disablefullpath == 1 results in ENODEV */
1115 		if (error == 0 || error == ENODEV) {
1116 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1117 			    fsflags, optlist);
1118 		}
1119 		free(pathbuf, M_TEMP);
1120 	} else
1121 		error = vfs_domount_update(td, vp, fsflags, optlist);
1122 	mtx_unlock(&Giant);
1123 
1124 	ASSERT_VI_UNLOCKED(vp, __func__);
1125 	ASSERT_VOP_UNLOCKED(vp, __func__);
1126 
1127 	return (error);
1128 }
1129 
1130 /*
1131  * Unmount a filesystem.
1132  *
1133  * Note: unmount takes a path to the vnode mounted on as argument, not
1134  * special file (as before).
1135  */
1136 #ifndef _SYS_SYSPROTO_H_
1137 struct unmount_args {
1138 	char	*path;
1139 	int	flags;
1140 };
1141 #endif
1142 /* ARGSUSED */
1143 int
1144 sys_unmount(td, uap)
1145 	struct thread *td;
1146 	register struct unmount_args /* {
1147 		char *path;
1148 		int flags;
1149 	} */ *uap;
1150 {
1151 	struct nameidata nd;
1152 	struct mount *mp;
1153 	char *pathbuf;
1154 	int error, id0, id1, vfslocked;
1155 
1156 	AUDIT_ARG_VALUE(uap->flags);
1157 	if (jailed(td->td_ucred) || usermount == 0) {
1158 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1159 		if (error)
1160 			return (error);
1161 	}
1162 
1163 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1164 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1165 	if (error) {
1166 		free(pathbuf, M_TEMP);
1167 		return (error);
1168 	}
1169 	mtx_lock(&Giant);
1170 	if (uap->flags & MNT_BYFSID) {
1171 		AUDIT_ARG_TEXT(pathbuf);
1172 		/* Decode the filesystem ID. */
1173 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1174 			mtx_unlock(&Giant);
1175 			free(pathbuf, M_TEMP);
1176 			return (EINVAL);
1177 		}
1178 
1179 		mtx_lock(&mountlist_mtx);
1180 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1181 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1182 			    mp->mnt_stat.f_fsid.val[1] == id1)
1183 				break;
1184 		}
1185 		mtx_unlock(&mountlist_mtx);
1186 	} else {
1187 		AUDIT_ARG_UPATH1(td, pathbuf);
1188 		/*
1189 		 * Try to find global path for path argument.
1190 		 */
1191 		NDINIT(&nd, LOOKUP,
1192 		    FOLLOW | LOCKLEAF | MPSAFE | AUDITVNODE1,
1193 		    UIO_SYSSPACE, pathbuf, td);
1194 		if (namei(&nd) == 0) {
1195 			vfslocked = NDHASGIANT(&nd);
1196 			NDFREE(&nd, NDF_ONLY_PNBUF);
1197 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1198 			    MNAMELEN);
1199 			if (error == 0 || error == ENODEV)
1200 				vput(nd.ni_vp);
1201 			VFS_UNLOCK_GIANT(vfslocked);
1202 		}
1203 		mtx_lock(&mountlist_mtx);
1204 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1205 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1206 				break;
1207 		}
1208 		mtx_unlock(&mountlist_mtx);
1209 	}
1210 	free(pathbuf, M_TEMP);
1211 	if (mp == NULL) {
1212 		/*
1213 		 * Previously we returned ENOENT for a nonexistent path and
1214 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1215 		 * now, so in the !MNT_BYFSID case return the more likely
1216 		 * EINVAL for compatibility.
1217 		 */
1218 		mtx_unlock(&Giant);
1219 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1220 	}
1221 
1222 	/*
1223 	 * Don't allow unmounting the root filesystem.
1224 	 */
1225 	if (mp->mnt_flag & MNT_ROOTFS) {
1226 		mtx_unlock(&Giant);
1227 		return (EINVAL);
1228 	}
1229 	error = dounmount(mp, uap->flags, td);
1230 	mtx_unlock(&Giant);
1231 	return (error);
1232 }
1233 
1234 /*
1235  * Do the actual filesystem unmount.
1236  */
1237 int
1238 dounmount(mp, flags, td)
1239 	struct mount *mp;
1240 	int flags;
1241 	struct thread *td;
1242 {
1243 	struct vnode *coveredvp, *fsrootvp;
1244 	int error;
1245 	uint64_t async_flag;
1246 	int mnt_gen_r;
1247 
1248 	mtx_assert(&Giant, MA_OWNED);
1249 
1250 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1251 		mnt_gen_r = mp->mnt_gen;
1252 		VI_LOCK(coveredvp);
1253 		vholdl(coveredvp);
1254 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1255 		vdrop(coveredvp);
1256 		/*
1257 		 * Check for mp being unmounted while waiting for the
1258 		 * covered vnode lock.
1259 		 */
1260 		if (coveredvp->v_mountedhere != mp ||
1261 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1262 			VOP_UNLOCK(coveredvp, 0);
1263 			return (EBUSY);
1264 		}
1265 	}
1266 	/*
1267 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1268 	 * original mount is permitted to unmount this filesystem.
1269 	 */
1270 	error = vfs_suser(mp, td);
1271 	if (error) {
1272 		if (coveredvp)
1273 			VOP_UNLOCK(coveredvp, 0);
1274 		return (error);
1275 	}
1276 
1277 	MNT_ILOCK(mp);
1278 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1279 		MNT_IUNLOCK(mp);
1280 		if (coveredvp)
1281 			VOP_UNLOCK(coveredvp, 0);
1282 		return (EBUSY);
1283 	}
1284 	mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1285 	/* Allow filesystems to detect that a forced unmount is in progress. */
1286 	if (flags & MNT_FORCE)
1287 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1288 	error = 0;
1289 	if (mp->mnt_lockref) {
1290 		mp->mnt_kern_flag |= MNTK_DRAINING;
1291 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1292 		    "mount drain", 0);
1293 	}
1294 	MNT_IUNLOCK(mp);
1295 	KASSERT(mp->mnt_lockref == 0,
1296 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1297 	    __func__, __FILE__, __LINE__));
1298 	KASSERT(error == 0,
1299 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1300 	    __func__, __FILE__, __LINE__));
1301 	vn_start_write(NULL, &mp, V_WAIT);
1302 
1303 	if (mp->mnt_flag & MNT_EXPUBLIC)
1304 		vfs_setpublicfs(NULL, NULL, NULL);
1305 
1306 	vfs_msync(mp, MNT_WAIT);
1307 	MNT_ILOCK(mp);
1308 	async_flag = mp->mnt_flag & MNT_ASYNC;
1309 	mp->mnt_flag &= ~MNT_ASYNC;
1310 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1311 	MNT_IUNLOCK(mp);
1312 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
1313 	vfs_deallocate_syncvnode(mp);
1314 	/*
1315 	 * For forced unmounts, move process cdir/rdir refs on the fs root
1316 	 * vnode to the covered vnode.  For non-forced unmounts we want
1317 	 * such references to cause an EBUSY error.
1318 	 */
1319 	if ((flags & MNT_FORCE) &&
1320 	    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1321 		if (mp->mnt_vnodecovered != NULL)
1322 			mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1323 		if (fsrootvp == rootvnode) {
1324 			vrele(rootvnode);
1325 			rootvnode = NULL;
1326 		}
1327 		vput(fsrootvp);
1328 	}
1329 	if (((mp->mnt_flag & MNT_RDONLY) ||
1330 	     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0)
1331 		error = VFS_UNMOUNT(mp, flags);
1332 	vn_finished_write(mp);
1333 	/*
1334 	 * If we failed to flush the dirty blocks for this mount point,
1335 	 * undo all the cdir/rdir and rootvnode changes we made above.
1336 	 * Unless we failed to do so because the device is reporting that
1337 	 * it doesn't exist anymore.
1338 	 */
1339 	if (error && error != ENXIO) {
1340 		if ((flags & MNT_FORCE) &&
1341 		    VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) {
1342 			if (mp->mnt_vnodecovered != NULL)
1343 				mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1344 			if (rootvnode == NULL) {
1345 				rootvnode = fsrootvp;
1346 				vref(rootvnode);
1347 			}
1348 			vput(fsrootvp);
1349 		}
1350 		MNT_ILOCK(mp);
1351 		mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1352 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1353 			MNT_IUNLOCK(mp);
1354 			vfs_allocate_syncvnode(mp);
1355 			MNT_ILOCK(mp);
1356 		}
1357 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1358 		mp->mnt_flag |= async_flag;
1359 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1360 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1361 			mp->mnt_kern_flag |= MNTK_ASYNC;
1362 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1363 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1364 			wakeup(mp);
1365 		}
1366 		MNT_IUNLOCK(mp);
1367 		if (coveredvp)
1368 			VOP_UNLOCK(coveredvp, 0);
1369 		return (error);
1370 	}
1371 	mtx_lock(&mountlist_mtx);
1372 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1373 	mtx_unlock(&mountlist_mtx);
1374 	if (coveredvp != NULL) {
1375 		coveredvp->v_mountedhere = NULL;
1376 		vput(coveredvp);
1377 	}
1378 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1379 	vfs_mount_destroy(mp);
1380 	return (0);
1381 }
1382 
1383 /*
1384  * Report errors during filesystem mounting.
1385  */
1386 void
1387 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1388 {
1389 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1390 	va_list ap;
1391 	int error, len;
1392 	char *errmsg;
1393 
1394 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1395 	if (error || errmsg == NULL || len <= 0)
1396 		return;
1397 
1398 	va_start(ap, fmt);
1399 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1400 	va_end(ap);
1401 }
1402 
1403 void
1404 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1405 {
1406 	va_list ap;
1407 	int error, len;
1408 	char *errmsg;
1409 
1410 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1411 	if (error || errmsg == NULL || len <= 0)
1412 		return;
1413 
1414 	va_start(ap, fmt);
1415 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1416 	va_end(ap);
1417 }
1418 
1419 /*
1420  * ---------------------------------------------------------------------
1421  * Functions for querying mount options/arguments from filesystems.
1422  */
1423 
1424 /*
1425  * Check that no unknown options are given
1426  */
1427 int
1428 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1429 {
1430 	struct vfsopt *opt;
1431 	char errmsg[255];
1432 	const char **t, *p, *q;
1433 	int ret = 0;
1434 
1435 	TAILQ_FOREACH(opt, opts, link) {
1436 		p = opt->name;
1437 		q = NULL;
1438 		if (p[0] == 'n' && p[1] == 'o')
1439 			q = p + 2;
1440 		for(t = global_opts; *t != NULL; t++) {
1441 			if (strcmp(*t, p) == 0)
1442 				break;
1443 			if (q != NULL) {
1444 				if (strcmp(*t, q) == 0)
1445 					break;
1446 			}
1447 		}
1448 		if (*t != NULL)
1449 			continue;
1450 		for(t = legal; *t != NULL; t++) {
1451 			if (strcmp(*t, p) == 0)
1452 				break;
1453 			if (q != NULL) {
1454 				if (strcmp(*t, q) == 0)
1455 					break;
1456 			}
1457 		}
1458 		if (*t != NULL)
1459 			continue;
1460 		snprintf(errmsg, sizeof(errmsg),
1461 		    "mount option <%s> is unknown", p);
1462 		ret = EINVAL;
1463 	}
1464 	if (ret != 0) {
1465 		TAILQ_FOREACH(opt, opts, link) {
1466 			if (strcmp(opt->name, "errmsg") == 0) {
1467 				strncpy((char *)opt->value, errmsg, opt->len);
1468 				break;
1469 			}
1470 		}
1471 		if (opt == NULL)
1472 			printf("%s\n", errmsg);
1473 	}
1474 	return (ret);
1475 }
1476 
1477 /*
1478  * Get a mount option by its name.
1479  *
1480  * Return 0 if the option was found, ENOENT otherwise.
1481  * If len is non-NULL it will be filled with the length
1482  * of the option. If buf is non-NULL, it will be filled
1483  * with the address of the option.
1484  */
1485 int
1486 vfs_getopt(opts, name, buf, len)
1487 	struct vfsoptlist *opts;
1488 	const char *name;
1489 	void **buf;
1490 	int *len;
1491 {
1492 	struct vfsopt *opt;
1493 
1494 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1495 
1496 	TAILQ_FOREACH(opt, opts, link) {
1497 		if (strcmp(name, opt->name) == 0) {
1498 			opt->seen = 1;
1499 			if (len != NULL)
1500 				*len = opt->len;
1501 			if (buf != NULL)
1502 				*buf = opt->value;
1503 			return (0);
1504 		}
1505 	}
1506 	return (ENOENT);
1507 }
1508 
1509 int
1510 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1511 {
1512 	struct vfsopt *opt;
1513 
1514 	if (opts == NULL)
1515 		return (-1);
1516 
1517 	TAILQ_FOREACH(opt, opts, link) {
1518 		if (strcmp(name, opt->name) == 0) {
1519 			opt->seen = 1;
1520 			return (opt->pos);
1521 		}
1522 	}
1523 	return (-1);
1524 }
1525 
1526 int
1527 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1528 {
1529 	char *opt_value, *vtp;
1530 	quad_t iv;
1531 	int error, opt_len;
1532 
1533 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1534 	if (error != 0)
1535 		return (error);
1536 	if (opt_len == 0 || opt_value == NULL)
1537 		return (EINVAL);
1538 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1539 		return (EINVAL);
1540 	iv = strtoq(opt_value, &vtp, 0);
1541 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1542 		return (EINVAL);
1543 	if (iv < 0)
1544 		return (EINVAL);
1545 	switch (vtp[0]) {
1546 	case 't':
1547 	case 'T':
1548 		iv *= 1024;
1549 	case 'g':
1550 	case 'G':
1551 		iv *= 1024;
1552 	case 'm':
1553 	case 'M':
1554 		iv *= 1024;
1555 	case 'k':
1556 	case 'K':
1557 		iv *= 1024;
1558 	case '\0':
1559 		break;
1560 	default:
1561 		return (EINVAL);
1562 	}
1563 	*value = iv;
1564 
1565 	return (0);
1566 }
1567 
1568 char *
1569 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1570 {
1571 	struct vfsopt *opt;
1572 
1573 	*error = 0;
1574 	TAILQ_FOREACH(opt, opts, link) {
1575 		if (strcmp(name, opt->name) != 0)
1576 			continue;
1577 		opt->seen = 1;
1578 		if (opt->len == 0 ||
1579 		    ((char *)opt->value)[opt->len - 1] != '\0') {
1580 			*error = EINVAL;
1581 			return (NULL);
1582 		}
1583 		return (opt->value);
1584 	}
1585 	*error = ENOENT;
1586 	return (NULL);
1587 }
1588 
1589 int
1590 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1591 	uint64_t val)
1592 {
1593 	struct vfsopt *opt;
1594 
1595 	TAILQ_FOREACH(opt, opts, link) {
1596 		if (strcmp(name, opt->name) == 0) {
1597 			opt->seen = 1;
1598 			if (w != NULL)
1599 				*w |= val;
1600 			return (1);
1601 		}
1602 	}
1603 	if (w != NULL)
1604 		*w &= ~val;
1605 	return (0);
1606 }
1607 
1608 int
1609 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1610 {
1611 	va_list ap;
1612 	struct vfsopt *opt;
1613 	int ret;
1614 
1615 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1616 
1617 	TAILQ_FOREACH(opt, opts, link) {
1618 		if (strcmp(name, opt->name) != 0)
1619 			continue;
1620 		opt->seen = 1;
1621 		if (opt->len == 0 || opt->value == NULL)
1622 			return (0);
1623 		if (((char *)opt->value)[opt->len - 1] != '\0')
1624 			return (0);
1625 		va_start(ap, fmt);
1626 		ret = vsscanf(opt->value, fmt, ap);
1627 		va_end(ap);
1628 		return (ret);
1629 	}
1630 	return (0);
1631 }
1632 
1633 int
1634 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
1635 {
1636 	struct vfsopt *opt;
1637 
1638 	TAILQ_FOREACH(opt, opts, link) {
1639 		if (strcmp(name, opt->name) != 0)
1640 			continue;
1641 		opt->seen = 1;
1642 		if (opt->value == NULL)
1643 			opt->len = len;
1644 		else {
1645 			if (opt->len != len)
1646 				return (EINVAL);
1647 			bcopy(value, opt->value, len);
1648 		}
1649 		return (0);
1650 	}
1651 	return (ENOENT);
1652 }
1653 
1654 int
1655 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
1656 {
1657 	struct vfsopt *opt;
1658 
1659 	TAILQ_FOREACH(opt, opts, link) {
1660 		if (strcmp(name, opt->name) != 0)
1661 			continue;
1662 		opt->seen = 1;
1663 		if (opt->value == NULL)
1664 			opt->len = len;
1665 		else {
1666 			if (opt->len < len)
1667 				return (EINVAL);
1668 			opt->len = len;
1669 			bcopy(value, opt->value, len);
1670 		}
1671 		return (0);
1672 	}
1673 	return (ENOENT);
1674 }
1675 
1676 int
1677 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
1678 {
1679 	struct vfsopt *opt;
1680 
1681 	TAILQ_FOREACH(opt, opts, link) {
1682 		if (strcmp(name, opt->name) != 0)
1683 			continue;
1684 		opt->seen = 1;
1685 		if (opt->value == NULL)
1686 			opt->len = strlen(value) + 1;
1687 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
1688 			return (EINVAL);
1689 		return (0);
1690 	}
1691 	return (ENOENT);
1692 }
1693 
1694 /*
1695  * Find and copy a mount option.
1696  *
1697  * The size of the buffer has to be specified
1698  * in len, if it is not the same length as the
1699  * mount option, EINVAL is returned.
1700  * Returns ENOENT if the option is not found.
1701  */
1702 int
1703 vfs_copyopt(opts, name, dest, len)
1704 	struct vfsoptlist *opts;
1705 	const char *name;
1706 	void *dest;
1707 	int len;
1708 {
1709 	struct vfsopt *opt;
1710 
1711 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1712 
1713 	TAILQ_FOREACH(opt, opts, link) {
1714 		if (strcmp(name, opt->name) == 0) {
1715 			opt->seen = 1;
1716 			if (len != opt->len)
1717 				return (EINVAL);
1718 			bcopy(opt->value, dest, opt->len);
1719 			return (0);
1720 		}
1721 	}
1722 	return (ENOENT);
1723 }
1724 
1725 /*
1726  * These are helper functions for filesystems to traverse all
1727  * their vnodes.  See MNT_VNODE_FOREACH() in sys/mount.h.
1728  *
1729  * This interface has been deprecated in favor of MNT_VNODE_FOREACH_ALL.
1730  */
1731 
1732 MALLOC_DECLARE(M_VNODE_MARKER);
1733 
1734 struct vnode *
1735 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
1736 {
1737 	struct vnode *vp;
1738 
1739 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1740 
1741 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1742 	if (should_yield()) {
1743 		MNT_IUNLOCK(mp);
1744 		kern_yield(PRI_UNCHANGED);
1745 		MNT_ILOCK(mp);
1746 	}
1747 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
1748 	while (vp != NULL && vp->v_type == VMARKER)
1749 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1750 
1751 	/* Check if we are done */
1752 	if (vp == NULL) {
1753 		__mnt_vnode_markerfree(mvp, mp);
1754 		return (NULL);
1755 	}
1756 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1757 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1758 	return (vp);
1759 }
1760 
1761 struct vnode *
1762 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
1763 {
1764 	struct vnode *vp;
1765 
1766 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1767 
1768 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1769 	while (vp != NULL && vp->v_type == VMARKER)
1770 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1771 
1772 	/* Check if we are done */
1773 	if (vp == NULL) {
1774 		*mvp = NULL;
1775 		return (NULL);
1776 	}
1777 	MNT_REF(mp);
1778 	MNT_IUNLOCK(mp);
1779 	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
1780 				       M_VNODE_MARKER,
1781 				       M_WAITOK | M_ZERO);
1782 	MNT_ILOCK(mp);
1783 	(*mvp)->v_type = VMARKER;
1784 
1785 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1786 	while (vp != NULL && vp->v_type == VMARKER)
1787 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
1788 
1789 	/* Check if we are done */
1790 	if (vp == NULL) {
1791 		MNT_IUNLOCK(mp);
1792 		free(*mvp, M_VNODE_MARKER);
1793 		MNT_ILOCK(mp);
1794 		*mvp = NULL;
1795 		MNT_REL(mp);
1796 		return (NULL);
1797 	}
1798 	(*mvp)->v_mount = mp;
1799 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
1800 	return (vp);
1801 }
1802 
1803 
1804 void
1805 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
1806 {
1807 
1808 	if (*mvp == NULL)
1809 		return;
1810 
1811 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1812 
1813 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
1814 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
1815 	MNT_IUNLOCK(mp);
1816 	free(*mvp, M_VNODE_MARKER);
1817 	MNT_ILOCK(mp);
1818 	*mvp = NULL;
1819 	MNT_REL(mp);
1820 }
1821 
1822 int
1823 __vfs_statfs(struct mount *mp, struct statfs *sbp)
1824 {
1825 	int error;
1826 
1827 	error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat);
1828 	if (sbp != &mp->mnt_stat)
1829 		*sbp = mp->mnt_stat;
1830 	return (error);
1831 }
1832 
1833 void
1834 vfs_mountedfrom(struct mount *mp, const char *from)
1835 {
1836 
1837 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
1838 	strlcpy(mp->mnt_stat.f_mntfromname, from,
1839 	    sizeof mp->mnt_stat.f_mntfromname);
1840 }
1841 
1842 /*
1843  * ---------------------------------------------------------------------
1844  * This is the api for building mount args and mounting filesystems from
1845  * inside the kernel.
1846  *
1847  * The API works by accumulation of individual args.  First error is
1848  * latched.
1849  *
1850  * XXX: should be documented in new manpage kernel_mount(9)
1851  */
1852 
1853 /* A memory allocation which must be freed when we are done */
1854 struct mntaarg {
1855 	SLIST_ENTRY(mntaarg)	next;
1856 };
1857 
1858 /* The header for the mount arguments */
1859 struct mntarg {
1860 	struct iovec *v;
1861 	int len;
1862 	int error;
1863 	SLIST_HEAD(, mntaarg)	list;
1864 };
1865 
1866 /*
1867  * Add a boolean argument.
1868  *
1869  * flag is the boolean value.
1870  * name must start with "no".
1871  */
1872 struct mntarg *
1873 mount_argb(struct mntarg *ma, int flag, const char *name)
1874 {
1875 
1876 	KASSERT(name[0] == 'n' && name[1] == 'o',
1877 	    ("mount_argb(...,%s): name must start with 'no'", name));
1878 
1879 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
1880 }
1881 
1882 /*
1883  * Add an argument printf style
1884  */
1885 struct mntarg *
1886 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
1887 {
1888 	va_list ap;
1889 	struct mntaarg *maa;
1890 	struct sbuf *sb;
1891 	int len;
1892 
1893 	if (ma == NULL) {
1894 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1895 		SLIST_INIT(&ma->list);
1896 	}
1897 	if (ma->error)
1898 		return (ma);
1899 
1900 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1901 	    M_MOUNT, M_WAITOK);
1902 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1903 	ma->v[ma->len].iov_len = strlen(name) + 1;
1904 	ma->len++;
1905 
1906 	sb = sbuf_new_auto();
1907 	va_start(ap, fmt);
1908 	sbuf_vprintf(sb, fmt, ap);
1909 	va_end(ap);
1910 	sbuf_finish(sb);
1911 	len = sbuf_len(sb) + 1;
1912 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1913 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1914 	bcopy(sbuf_data(sb), maa + 1, len);
1915 	sbuf_delete(sb);
1916 
1917 	ma->v[ma->len].iov_base = maa + 1;
1918 	ma->v[ma->len].iov_len = len;
1919 	ma->len++;
1920 
1921 	return (ma);
1922 }
1923 
1924 /*
1925  * Add an argument which is a userland string.
1926  */
1927 struct mntarg *
1928 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
1929 {
1930 	struct mntaarg *maa;
1931 	char *tbuf;
1932 
1933 	if (val == NULL)
1934 		return (ma);
1935 	if (ma == NULL) {
1936 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1937 		SLIST_INIT(&ma->list);
1938 	}
1939 	if (ma->error)
1940 		return (ma);
1941 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
1942 	SLIST_INSERT_HEAD(&ma->list, maa, next);
1943 	tbuf = (void *)(maa + 1);
1944 	ma->error = copyinstr(val, tbuf, len, NULL);
1945 	return (mount_arg(ma, name, tbuf, -1));
1946 }
1947 
1948 /*
1949  * Plain argument.
1950  *
1951  * If length is -1, treat value as a C string.
1952  */
1953 struct mntarg *
1954 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
1955 {
1956 
1957 	if (ma == NULL) {
1958 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
1959 		SLIST_INIT(&ma->list);
1960 	}
1961 	if (ma->error)
1962 		return (ma);
1963 
1964 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
1965 	    M_MOUNT, M_WAITOK);
1966 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
1967 	ma->v[ma->len].iov_len = strlen(name) + 1;
1968 	ma->len++;
1969 
1970 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
1971 	if (len < 0)
1972 		ma->v[ma->len].iov_len = strlen(val) + 1;
1973 	else
1974 		ma->v[ma->len].iov_len = len;
1975 	ma->len++;
1976 	return (ma);
1977 }
1978 
1979 /*
1980  * Free a mntarg structure
1981  */
1982 static void
1983 free_mntarg(struct mntarg *ma)
1984 {
1985 	struct mntaarg *maa;
1986 
1987 	while (!SLIST_EMPTY(&ma->list)) {
1988 		maa = SLIST_FIRST(&ma->list);
1989 		SLIST_REMOVE_HEAD(&ma->list, next);
1990 		free(maa, M_MOUNT);
1991 	}
1992 	free(ma->v, M_MOUNT);
1993 	free(ma, M_MOUNT);
1994 }
1995 
1996 /*
1997  * Mount a filesystem
1998  */
1999 int
2000 kernel_mount(struct mntarg *ma, uint64_t flags)
2001 {
2002 	struct uio auio;
2003 	int error;
2004 
2005 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2006 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2007 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2008 
2009 	auio.uio_iov = ma->v;
2010 	auio.uio_iovcnt = ma->len;
2011 	auio.uio_segflg = UIO_SYSSPACE;
2012 
2013 	error = ma->error;
2014 	if (!error)
2015 		error = vfs_donmount(curthread, flags, &auio);
2016 	free_mntarg(ma);
2017 	return (error);
2018 }
2019 
2020 /*
2021  * A printflike function to mount a filesystem.
2022  */
2023 int
2024 kernel_vmount(int flags, ...)
2025 {
2026 	struct mntarg *ma = NULL;
2027 	va_list ap;
2028 	const char *cp;
2029 	const void *vp;
2030 	int error;
2031 
2032 	va_start(ap, flags);
2033 	for (;;) {
2034 		cp = va_arg(ap, const char *);
2035 		if (cp == NULL)
2036 			break;
2037 		vp = va_arg(ap, const void *);
2038 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2039 	}
2040 	va_end(ap);
2041 
2042 	error = kernel_mount(ma, flags);
2043 	return (error);
2044 }
2045 
2046 void
2047 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
2048 {
2049 
2050 	bcopy(oexp, exp, sizeof(*oexp));
2051 	exp->ex_numsecflavors = 0;
2052 }
2053