xref: /freebsd/sys/kern/vfs_mount.c (revision 0edc114ac0b998b06235da32bec24d55c10206cd)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1999-2004 Poul-Henning Kamp
5  * Copyright (c) 1999 Michael Smith
6  * Copyright (c) 1989, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/conf.h>
44 #include <sys/smp.h>
45 #include <sys/eventhandler.h>
46 #include <sys/fcntl.h>
47 #include <sys/jail.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/libkern.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/filedesc.h>
58 #include <sys/reboot.h>
59 #include <sys/sbuf.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/systm.h>
66 #include <sys/vnode.h>
67 #include <vm/uma.h>
68 
69 #include <geom/geom.h>
70 
71 #include <machine/stdarg.h>
72 
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
75 
76 #define	VFS_MOUNTARG_SIZE_MAX	(1024 * 64)
77 
78 static int	vfs_domount(struct thread *td, const char *fstype, char *fspath,
79 		    uint64_t fsflags, struct vfsoptlist **optlist);
80 static void	free_mntarg(struct mntarg *ma);
81 
82 static int	usermount = 0;
83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
84     "Unprivileged users may mount and unmount file systems");
85 
86 static bool	default_autoro = false;
87 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
88     "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
89 
90 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
91 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
92 static uma_zone_t mount_zone;
93 
94 /* List of mounted filesystems. */
95 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
96 
97 /* For any iteration/modification of mountlist */
98 struct mtx mountlist_mtx;
99 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
100 
101 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
102 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
103 
104 /*
105  * Global opts, taken by all filesystems
106  */
107 static const char *global_opts[] = {
108 	"errmsg",
109 	"fstype",
110 	"fspath",
111 	"ro",
112 	"rw",
113 	"nosuid",
114 	"noexec",
115 	NULL
116 };
117 
118 static int
119 mount_init(void *mem, int size, int flags)
120 {
121 	struct mount *mp;
122 
123 	mp = (struct mount *)mem;
124 	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
125 	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
126 	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
127 	mp->mnt_thread_in_ops_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
128 	    M_WAITOK | M_ZERO);
129 	mp->mnt_ref_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
130 	    M_WAITOK | M_ZERO);
131 	mp->mnt_lockref_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
132 	    M_WAITOK | M_ZERO);
133 	mp->mnt_writeopcount_pcpu = uma_zalloc_pcpu(pcpu_zone_int,
134 	    M_WAITOK | M_ZERO);
135 	mp->mnt_ref = 0;
136 	mp->mnt_vfs_ops = 1;
137 	return (0);
138 }
139 
140 static void
141 mount_fini(void *mem, int size)
142 {
143 	struct mount *mp;
144 
145 	mp = (struct mount *)mem;
146 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_writeopcount_pcpu);
147 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_lockref_pcpu);
148 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_ref_pcpu);
149 	uma_zfree_pcpu(pcpu_zone_int, mp->mnt_thread_in_ops_pcpu);
150 	lockdestroy(&mp->mnt_explock);
151 	mtx_destroy(&mp->mnt_listmtx);
152 	mtx_destroy(&mp->mnt_mtx);
153 }
154 
155 static void
156 vfs_mount_init(void *dummy __unused)
157 {
158 
159 	mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
160 	    NULL, mount_init, mount_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
161 }
162 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
163 
164 /*
165  * ---------------------------------------------------------------------
166  * Functions for building and sanitizing the mount options
167  */
168 
169 /* Remove one mount option. */
170 static void
171 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
172 {
173 
174 	TAILQ_REMOVE(opts, opt, link);
175 	free(opt->name, M_MOUNT);
176 	if (opt->value != NULL)
177 		free(opt->value, M_MOUNT);
178 	free(opt, M_MOUNT);
179 }
180 
181 /* Release all resources related to the mount options. */
182 void
183 vfs_freeopts(struct vfsoptlist *opts)
184 {
185 	struct vfsopt *opt;
186 
187 	while (!TAILQ_EMPTY(opts)) {
188 		opt = TAILQ_FIRST(opts);
189 		vfs_freeopt(opts, opt);
190 	}
191 	free(opts, M_MOUNT);
192 }
193 
194 void
195 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
196 {
197 	struct vfsopt *opt, *temp;
198 
199 	if (opts == NULL)
200 		return;
201 	TAILQ_FOREACH_SAFE(opt, opts, link, temp)  {
202 		if (strcmp(opt->name, name) == 0)
203 			vfs_freeopt(opts, opt);
204 	}
205 }
206 
207 static int
208 vfs_isopt_ro(const char *opt)
209 {
210 
211 	if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
212 	    strcmp(opt, "norw") == 0)
213 		return (1);
214 	return (0);
215 }
216 
217 static int
218 vfs_isopt_rw(const char *opt)
219 {
220 
221 	if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
222 		return (1);
223 	return (0);
224 }
225 
226 /*
227  * Check if options are equal (with or without the "no" prefix).
228  */
229 static int
230 vfs_equalopts(const char *opt1, const char *opt2)
231 {
232 	char *p;
233 
234 	/* "opt" vs. "opt" or "noopt" vs. "noopt" */
235 	if (strcmp(opt1, opt2) == 0)
236 		return (1);
237 	/* "noopt" vs. "opt" */
238 	if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
239 		return (1);
240 	/* "opt" vs. "noopt" */
241 	if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
242 		return (1);
243 	while ((p = strchr(opt1, '.')) != NULL &&
244 	    !strncmp(opt1, opt2, ++p - opt1)) {
245 		opt2 += p - opt1;
246 		opt1 = p;
247 		/* "foo.noopt" vs. "foo.opt" */
248 		if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
249 			return (1);
250 		/* "foo.opt" vs. "foo.noopt" */
251 		if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
252 			return (1);
253 	}
254 	/* "ro" / "rdonly" / "norw" / "rw" / "noro" */
255 	if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
256 	    (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
257 		return (1);
258 	return (0);
259 }
260 
261 /*
262  * If a mount option is specified several times,
263  * (with or without the "no" prefix) only keep
264  * the last occurrence of it.
265  */
266 static void
267 vfs_sanitizeopts(struct vfsoptlist *opts)
268 {
269 	struct vfsopt *opt, *opt2, *tmp;
270 
271 	TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
272 		opt2 = TAILQ_PREV(opt, vfsoptlist, link);
273 		while (opt2 != NULL) {
274 			if (vfs_equalopts(opt->name, opt2->name)) {
275 				tmp = TAILQ_PREV(opt2, vfsoptlist, link);
276 				vfs_freeopt(opts, opt2);
277 				opt2 = tmp;
278 			} else {
279 				opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
280 			}
281 		}
282 	}
283 }
284 
285 /*
286  * Build a linked list of mount options from a struct uio.
287  */
288 int
289 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
290 {
291 	struct vfsoptlist *opts;
292 	struct vfsopt *opt;
293 	size_t memused, namelen, optlen;
294 	unsigned int i, iovcnt;
295 	int error;
296 
297 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
298 	TAILQ_INIT(opts);
299 	memused = 0;
300 	iovcnt = auio->uio_iovcnt;
301 	for (i = 0; i < iovcnt; i += 2) {
302 		namelen = auio->uio_iov[i].iov_len;
303 		optlen = auio->uio_iov[i + 1].iov_len;
304 		memused += sizeof(struct vfsopt) + optlen + namelen;
305 		/*
306 		 * Avoid consuming too much memory, and attempts to overflow
307 		 * memused.
308 		 */
309 		if (memused > VFS_MOUNTARG_SIZE_MAX ||
310 		    optlen > VFS_MOUNTARG_SIZE_MAX ||
311 		    namelen > VFS_MOUNTARG_SIZE_MAX) {
312 			error = EINVAL;
313 			goto bad;
314 		}
315 
316 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
317 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
318 		opt->value = NULL;
319 		opt->len = 0;
320 		opt->pos = i / 2;
321 		opt->seen = 0;
322 
323 		/*
324 		 * Do this early, so jumps to "bad" will free the current
325 		 * option.
326 		 */
327 		TAILQ_INSERT_TAIL(opts, opt, link);
328 
329 		if (auio->uio_segflg == UIO_SYSSPACE) {
330 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
331 		} else {
332 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
333 			    namelen);
334 			if (error)
335 				goto bad;
336 		}
337 		/* Ensure names are null-terminated strings. */
338 		if (namelen == 0 || opt->name[namelen - 1] != '\0') {
339 			error = EINVAL;
340 			goto bad;
341 		}
342 		if (optlen != 0) {
343 			opt->len = optlen;
344 			opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
345 			if (auio->uio_segflg == UIO_SYSSPACE) {
346 				bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
347 				    optlen);
348 			} else {
349 				error = copyin(auio->uio_iov[i + 1].iov_base,
350 				    opt->value, optlen);
351 				if (error)
352 					goto bad;
353 			}
354 		}
355 	}
356 	vfs_sanitizeopts(opts);
357 	*options = opts;
358 	return (0);
359 bad:
360 	vfs_freeopts(opts);
361 	return (error);
362 }
363 
364 /*
365  * Merge the old mount options with the new ones passed
366  * in the MNT_UPDATE case.
367  *
368  * XXX: This function will keep a "nofoo" option in the new
369  * options.  E.g, if the option's canonical name is "foo",
370  * "nofoo" ends up in the mount point's active options.
371  */
372 static void
373 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
374 {
375 	struct vfsopt *opt, *new;
376 
377 	TAILQ_FOREACH(opt, oldopts, link) {
378 		new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
379 		new->name = strdup(opt->name, M_MOUNT);
380 		if (opt->len != 0) {
381 			new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
382 			bcopy(opt->value, new->value, opt->len);
383 		} else
384 			new->value = NULL;
385 		new->len = opt->len;
386 		new->seen = opt->seen;
387 		TAILQ_INSERT_HEAD(toopts, new, link);
388 	}
389 	vfs_sanitizeopts(toopts);
390 }
391 
392 /*
393  * Mount a filesystem.
394  */
395 #ifndef _SYS_SYSPROTO_H_
396 struct nmount_args {
397 	struct iovec *iovp;
398 	unsigned int iovcnt;
399 	int flags;
400 };
401 #endif
402 int
403 sys_nmount(struct thread *td, struct nmount_args *uap)
404 {
405 	struct uio *auio;
406 	int error;
407 	u_int iovcnt;
408 	uint64_t flags;
409 
410 	/*
411 	 * Mount flags are now 64-bits. On 32-bit archtectures only
412 	 * 32-bits are passed in, but from here on everything handles
413 	 * 64-bit flags correctly.
414 	 */
415 	flags = uap->flags;
416 
417 	AUDIT_ARG_FFLAGS(flags);
418 	CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
419 	    uap->iovp, uap->iovcnt, flags);
420 
421 	/*
422 	 * Filter out MNT_ROOTFS.  We do not want clients of nmount() in
423 	 * userspace to set this flag, but we must filter it out if we want
424 	 * MNT_UPDATE on the root file system to work.
425 	 * MNT_ROOTFS should only be set by the kernel when mounting its
426 	 * root file system.
427 	 */
428 	flags &= ~MNT_ROOTFS;
429 
430 	iovcnt = uap->iovcnt;
431 	/*
432 	 * Check that we have an even number of iovec's
433 	 * and that we have at least two options.
434 	 */
435 	if ((iovcnt & 1) || (iovcnt < 4)) {
436 		CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
437 		    uap->iovcnt);
438 		return (EINVAL);
439 	}
440 
441 	error = copyinuio(uap->iovp, iovcnt, &auio);
442 	if (error) {
443 		CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
444 		    __func__, error);
445 		return (error);
446 	}
447 	error = vfs_donmount(td, flags, auio);
448 
449 	free(auio, M_IOV);
450 	return (error);
451 }
452 
453 /*
454  * ---------------------------------------------------------------------
455  * Various utility functions
456  */
457 
458 void
459 vfs_ref(struct mount *mp)
460 {
461 
462 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
463 	if (vfs_op_thread_enter(mp)) {
464 		vfs_mp_count_add_pcpu(mp, ref, 1);
465 		vfs_op_thread_exit(mp);
466 		return;
467 	}
468 
469 	MNT_ILOCK(mp);
470 	MNT_REF(mp);
471 	MNT_IUNLOCK(mp);
472 }
473 
474 void
475 vfs_rel(struct mount *mp)
476 {
477 
478 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
479 	if (vfs_op_thread_enter(mp)) {
480 		vfs_mp_count_sub_pcpu(mp, ref, 1);
481 		vfs_op_thread_exit(mp);
482 		return;
483 	}
484 
485 	MNT_ILOCK(mp);
486 	MNT_REL(mp);
487 	MNT_IUNLOCK(mp);
488 }
489 
490 /*
491  * Allocate and initialize the mount point struct.
492  */
493 struct mount *
494 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
495     struct ucred *cred)
496 {
497 	struct mount *mp;
498 
499 	mp = uma_zalloc(mount_zone, M_WAITOK);
500 	bzero(&mp->mnt_startzero,
501 	    __rangeof(struct mount, mnt_startzero, mnt_endzero));
502 	TAILQ_INIT(&mp->mnt_nvnodelist);
503 	mp->mnt_nvnodelistsize = 0;
504 	TAILQ_INIT(&mp->mnt_activevnodelist);
505 	mp->mnt_activevnodelistsize = 0;
506 	TAILQ_INIT(&mp->mnt_tmpfreevnodelist);
507 	mp->mnt_tmpfreevnodelistsize = 0;
508 	if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
509 	    mp->mnt_writeopcount != 0)
510 		panic("%s: non-zero counters on new mp %p\n", __func__, mp);
511 	if (mp->mnt_vfs_ops != 1)
512 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
513 		    mp->mnt_vfs_ops);
514 	(void) vfs_busy(mp, MBF_NOWAIT);
515 	atomic_add_acq_int(&vfsp->vfc_refcount, 1);
516 	mp->mnt_op = vfsp->vfc_vfsops;
517 	mp->mnt_vfc = vfsp;
518 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
519 	mp->mnt_gen++;
520 	strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
521 	mp->mnt_vnodecovered = vp;
522 	mp->mnt_cred = crdup(cred);
523 	mp->mnt_stat.f_owner = cred->cr_uid;
524 	strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
525 	mp->mnt_iosize_max = DFLTPHYS;
526 #ifdef MAC
527 	mac_mount_init(mp);
528 	mac_mount_create(cred, mp);
529 #endif
530 	arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
531 	TAILQ_INIT(&mp->mnt_uppers);
532 	return (mp);
533 }
534 
535 /*
536  * Destroy the mount struct previously allocated by vfs_mount_alloc().
537  */
538 void
539 vfs_mount_destroy(struct mount *mp)
540 {
541 
542 	if (mp->mnt_vfs_ops == 0)
543 		panic("%s: entered with zero vfs_ops\n", __func__);
544 
545 	vfs_assert_mount_counters(mp);
546 
547 	MNT_ILOCK(mp);
548 	mp->mnt_kern_flag |= MNTK_REFEXPIRE;
549 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
550 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
551 		wakeup(mp);
552 	}
553 	while (mp->mnt_ref)
554 		msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
555 	KASSERT(mp->mnt_ref == 0,
556 	    ("%s: invalid refcount in the drain path @ %s:%d", __func__,
557 	    __FILE__, __LINE__));
558 	if (mp->mnt_writeopcount != 0)
559 		panic("vfs_mount_destroy: nonzero writeopcount");
560 	if (mp->mnt_secondary_writes != 0)
561 		panic("vfs_mount_destroy: nonzero secondary_writes");
562 	atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
563 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
564 		struct vnode *vp;
565 
566 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
567 			vn_printf(vp, "dangling vnode ");
568 		panic("unmount: dangling vnode");
569 	}
570 	KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
571 	if (mp->mnt_nvnodelistsize != 0)
572 		panic("vfs_mount_destroy: nonzero nvnodelistsize");
573 	if (mp->mnt_activevnodelistsize != 0)
574 		panic("vfs_mount_destroy: nonzero activevnodelistsize");
575 	if (mp->mnt_tmpfreevnodelistsize != 0)
576 		panic("vfs_mount_destroy: nonzero tmpfreevnodelistsize");
577 	if (mp->mnt_lockref != 0)
578 		panic("vfs_mount_destroy: nonzero lock refcount");
579 	MNT_IUNLOCK(mp);
580 
581 	if (mp->mnt_vfs_ops != 1)
582 		panic("%s: vfs_ops should be 1 but %d found\n", __func__,
583 		    mp->mnt_vfs_ops);
584 
585 	if (mp->mnt_vnodecovered != NULL)
586 		vrele(mp->mnt_vnodecovered);
587 #ifdef MAC
588 	mac_mount_destroy(mp);
589 #endif
590 	if (mp->mnt_opt != NULL)
591 		vfs_freeopts(mp->mnt_opt);
592 	crfree(mp->mnt_cred);
593 	uma_zfree(mount_zone, mp);
594 }
595 
596 static bool
597 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
598 {
599 	/* This is an upgrade of an exisiting mount. */
600 	if ((fsflags & MNT_UPDATE) != 0)
601 		return (false);
602 	/* This is already an R/O mount. */
603 	if ((fsflags & MNT_RDONLY) != 0)
604 		return (false);
605 
606 	switch (error) {
607 	case ENODEV:	/* generic, geom, ... */
608 	case EACCES:	/* cam/scsi, ... */
609 	case EROFS:	/* md, mmcsd, ... */
610 		/*
611 		 * These errors can be returned by the storage layer to signal
612 		 * that the media is read-only.  No harm in the R/O mount
613 		 * attempt if the error was returned for some other reason.
614 		 */
615 		return (true);
616 	default:
617 		return (false);
618 	}
619 }
620 
621 int
622 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
623 {
624 	struct vfsoptlist *optlist;
625 	struct vfsopt *opt, *tmp_opt;
626 	char *fstype, *fspath, *errmsg;
627 	int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
628 	bool autoro;
629 
630 	errmsg = fspath = NULL;
631 	errmsg_len = fspathlen = 0;
632 	errmsg_pos = -1;
633 	autoro = default_autoro;
634 
635 	error = vfs_buildopts(fsoptions, &optlist);
636 	if (error)
637 		return (error);
638 
639 	if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
640 		errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
641 
642 	/*
643 	 * We need these two options before the others,
644 	 * and they are mandatory for any filesystem.
645 	 * Ensure they are NUL terminated as well.
646 	 */
647 	fstypelen = 0;
648 	error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
649 	if (error || fstype[fstypelen - 1] != '\0') {
650 		error = EINVAL;
651 		if (errmsg != NULL)
652 			strncpy(errmsg, "Invalid fstype", errmsg_len);
653 		goto bail;
654 	}
655 	fspathlen = 0;
656 	error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
657 	if (error || fspath[fspathlen - 1] != '\0') {
658 		error = EINVAL;
659 		if (errmsg != NULL)
660 			strncpy(errmsg, "Invalid fspath", errmsg_len);
661 		goto bail;
662 	}
663 
664 	/*
665 	 * We need to see if we have the "update" option
666 	 * before we call vfs_domount(), since vfs_domount() has special
667 	 * logic based on MNT_UPDATE.  This is very important
668 	 * when we want to update the root filesystem.
669 	 */
670 	TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
671 		if (strcmp(opt->name, "update") == 0) {
672 			fsflags |= MNT_UPDATE;
673 			vfs_freeopt(optlist, opt);
674 		}
675 		else if (strcmp(opt->name, "async") == 0)
676 			fsflags |= MNT_ASYNC;
677 		else if (strcmp(opt->name, "force") == 0) {
678 			fsflags |= MNT_FORCE;
679 			vfs_freeopt(optlist, opt);
680 		}
681 		else if (strcmp(opt->name, "reload") == 0) {
682 			fsflags |= MNT_RELOAD;
683 			vfs_freeopt(optlist, opt);
684 		}
685 		else if (strcmp(opt->name, "multilabel") == 0)
686 			fsflags |= MNT_MULTILABEL;
687 		else if (strcmp(opt->name, "noasync") == 0)
688 			fsflags &= ~MNT_ASYNC;
689 		else if (strcmp(opt->name, "noatime") == 0)
690 			fsflags |= MNT_NOATIME;
691 		else if (strcmp(opt->name, "atime") == 0) {
692 			free(opt->name, M_MOUNT);
693 			opt->name = strdup("nonoatime", M_MOUNT);
694 		}
695 		else if (strcmp(opt->name, "noclusterr") == 0)
696 			fsflags |= MNT_NOCLUSTERR;
697 		else if (strcmp(opt->name, "clusterr") == 0) {
698 			free(opt->name, M_MOUNT);
699 			opt->name = strdup("nonoclusterr", M_MOUNT);
700 		}
701 		else if (strcmp(opt->name, "noclusterw") == 0)
702 			fsflags |= MNT_NOCLUSTERW;
703 		else if (strcmp(opt->name, "clusterw") == 0) {
704 			free(opt->name, M_MOUNT);
705 			opt->name = strdup("nonoclusterw", M_MOUNT);
706 		}
707 		else if (strcmp(opt->name, "noexec") == 0)
708 			fsflags |= MNT_NOEXEC;
709 		else if (strcmp(opt->name, "exec") == 0) {
710 			free(opt->name, M_MOUNT);
711 			opt->name = strdup("nonoexec", M_MOUNT);
712 		}
713 		else if (strcmp(opt->name, "nosuid") == 0)
714 			fsflags |= MNT_NOSUID;
715 		else if (strcmp(opt->name, "suid") == 0) {
716 			free(opt->name, M_MOUNT);
717 			opt->name = strdup("nonosuid", M_MOUNT);
718 		}
719 		else if (strcmp(opt->name, "nosymfollow") == 0)
720 			fsflags |= MNT_NOSYMFOLLOW;
721 		else if (strcmp(opt->name, "symfollow") == 0) {
722 			free(opt->name, M_MOUNT);
723 			opt->name = strdup("nonosymfollow", M_MOUNT);
724 		}
725 		else if (strcmp(opt->name, "noro") == 0) {
726 			fsflags &= ~MNT_RDONLY;
727 			autoro = false;
728 		}
729 		else if (strcmp(opt->name, "rw") == 0) {
730 			fsflags &= ~MNT_RDONLY;
731 			autoro = false;
732 		}
733 		else if (strcmp(opt->name, "ro") == 0) {
734 			fsflags |= MNT_RDONLY;
735 			autoro = false;
736 		}
737 		else if (strcmp(opt->name, "rdonly") == 0) {
738 			free(opt->name, M_MOUNT);
739 			opt->name = strdup("ro", M_MOUNT);
740 			fsflags |= MNT_RDONLY;
741 			autoro = false;
742 		}
743 		else if (strcmp(opt->name, "autoro") == 0) {
744 			vfs_freeopt(optlist, opt);
745 			autoro = true;
746 		}
747 		else if (strcmp(opt->name, "suiddir") == 0)
748 			fsflags |= MNT_SUIDDIR;
749 		else if (strcmp(opt->name, "sync") == 0)
750 			fsflags |= MNT_SYNCHRONOUS;
751 		else if (strcmp(opt->name, "union") == 0)
752 			fsflags |= MNT_UNION;
753 		else if (strcmp(opt->name, "automounted") == 0) {
754 			fsflags |= MNT_AUTOMOUNTED;
755 			vfs_freeopt(optlist, opt);
756 		}
757 	}
758 
759 	/*
760 	 * Be ultra-paranoid about making sure the type and fspath
761 	 * variables will fit in our mp buffers, including the
762 	 * terminating NUL.
763 	 */
764 	if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
765 		error = ENAMETOOLONG;
766 		goto bail;
767 	}
768 
769 	error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
770 
771 	/*
772 	 * See if we can mount in the read-only mode if the error code suggests
773 	 * that it could be possible and the mount options allow for that.
774 	 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
775 	 * overridden by "autoro".
776 	 */
777 	if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
778 		printf("%s: R/W mount failed, possibly R/O media,"
779 		    " trying R/O mount\n", __func__);
780 		fsflags |= MNT_RDONLY;
781 		error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
782 	}
783 bail:
784 	/* copyout the errmsg */
785 	if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
786 	    && errmsg_len > 0 && errmsg != NULL) {
787 		if (fsoptions->uio_segflg == UIO_SYSSPACE) {
788 			bcopy(errmsg,
789 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
790 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
791 		} else {
792 			copyout(errmsg,
793 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
794 			    fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
795 		}
796 	}
797 
798 	if (optlist != NULL)
799 		vfs_freeopts(optlist);
800 	return (error);
801 }
802 
803 /*
804  * Old mount API.
805  */
806 #ifndef _SYS_SYSPROTO_H_
807 struct mount_args {
808 	char	*type;
809 	char	*path;
810 	int	flags;
811 	caddr_t	data;
812 };
813 #endif
814 /* ARGSUSED */
815 int
816 sys_mount(struct thread *td, struct mount_args *uap)
817 {
818 	char *fstype;
819 	struct vfsconf *vfsp = NULL;
820 	struct mntarg *ma = NULL;
821 	uint64_t flags;
822 	int error;
823 
824 	/*
825 	 * Mount flags are now 64-bits. On 32-bit architectures only
826 	 * 32-bits are passed in, but from here on everything handles
827 	 * 64-bit flags correctly.
828 	 */
829 	flags = uap->flags;
830 
831 	AUDIT_ARG_FFLAGS(flags);
832 
833 	/*
834 	 * Filter out MNT_ROOTFS.  We do not want clients of mount() in
835 	 * userspace to set this flag, but we must filter it out if we want
836 	 * MNT_UPDATE on the root file system to work.
837 	 * MNT_ROOTFS should only be set by the kernel when mounting its
838 	 * root file system.
839 	 */
840 	flags &= ~MNT_ROOTFS;
841 
842 	fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
843 	error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
844 	if (error) {
845 		free(fstype, M_TEMP);
846 		return (error);
847 	}
848 
849 	AUDIT_ARG_TEXT(fstype);
850 	vfsp = vfs_byname_kld(fstype, td, &error);
851 	free(fstype, M_TEMP);
852 	if (vfsp == NULL)
853 		return (ENOENT);
854 	if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
855 	    vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
856 	    ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
857 	    vfsp->vfc_vfsops->vfs_cmount == NULL))
858 		return (EOPNOTSUPP);
859 
860 	ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
861 	ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
862 	ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
863 	ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
864 	ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
865 
866 	if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
867 		return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
868 	return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
869 }
870 
871 /*
872  * vfs_domount_first(): first file system mount (not update)
873  */
874 static int
875 vfs_domount_first(
876 	struct thread *td,		/* Calling thread. */
877 	struct vfsconf *vfsp,		/* File system type. */
878 	char *fspath,			/* Mount path. */
879 	struct vnode *vp,		/* Vnode to be covered. */
880 	uint64_t fsflags,		/* Flags common to all filesystems. */
881 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
882 	)
883 {
884 	struct vattr va;
885 	struct mount *mp;
886 	struct vnode *newdp;
887 	int error, error1;
888 
889 	ASSERT_VOP_ELOCKED(vp, __func__);
890 	KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
891 
892 	/*
893 	 * If the jail of the calling thread lacks permission for this type of
894 	 * file system, deny immediately.
895 	 */
896 	if (jailed(td->td_ucred) && !prison_allow(td->td_ucred,
897 	    vfsp->vfc_prison_flag)) {
898 		vput(vp);
899 		return (EPERM);
900 	}
901 
902 	/*
903 	 * If the user is not root, ensure that they own the directory
904 	 * onto which we are attempting to mount.
905 	 */
906 	error = VOP_GETATTR(vp, &va, td->td_ucred);
907 	if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
908 		error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
909 	if (error == 0)
910 		error = vinvalbuf(vp, V_SAVE, 0, 0);
911 	if (error == 0 && vp->v_type != VDIR)
912 		error = ENOTDIR;
913 	if (error == 0) {
914 		VI_LOCK(vp);
915 		if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
916 			vp->v_iflag |= VI_MOUNT;
917 		else
918 			error = EBUSY;
919 		VI_UNLOCK(vp);
920 	}
921 	if (error != 0) {
922 		vput(vp);
923 		return (error);
924 	}
925 	VOP_UNLOCK(vp, 0);
926 
927 	/* Allocate and initialize the filesystem. */
928 	mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
929 	/* XXXMAC: pass to vfs_mount_alloc? */
930 	mp->mnt_optnew = *optlist;
931 	/* Set the mount level flags. */
932 	mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
933 
934 	/*
935 	 * Mount the filesystem.
936 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
937 	 * get.  No freeing of cn_pnbuf.
938 	 */
939 	error1 = 0;
940 	if ((error = VFS_MOUNT(mp)) != 0 ||
941 	    (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
942 	    (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
943 		if (error1 != 0) {
944 			error = error1;
945 			if ((error1 = VFS_UNMOUNT(mp, 0)) != 0)
946 				printf("VFS_UNMOUNT returned %d\n", error1);
947 		}
948 		vfs_unbusy(mp);
949 		mp->mnt_vnodecovered = NULL;
950 		vfs_mount_destroy(mp);
951 		VI_LOCK(vp);
952 		vp->v_iflag &= ~VI_MOUNT;
953 		VI_UNLOCK(vp);
954 		vrele(vp);
955 		return (error);
956 	}
957 	VOP_UNLOCK(newdp, 0);
958 
959 	if (mp->mnt_opt != NULL)
960 		vfs_freeopts(mp->mnt_opt);
961 	mp->mnt_opt = mp->mnt_optnew;
962 	*optlist = NULL;
963 
964 	/*
965 	 * Prevent external consumers of mount options from reading mnt_optnew.
966 	 */
967 	mp->mnt_optnew = NULL;
968 
969 	MNT_ILOCK(mp);
970 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
971 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
972 		mp->mnt_kern_flag |= MNTK_ASYNC;
973 	else
974 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
975 	MNT_IUNLOCK(mp);
976 
977 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
978 	cache_purge(vp);
979 	VI_LOCK(vp);
980 	vp->v_iflag &= ~VI_MOUNT;
981 	VI_UNLOCK(vp);
982 	vp->v_mountedhere = mp;
983 	/* Place the new filesystem at the end of the mount list. */
984 	mtx_lock(&mountlist_mtx);
985 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
986 	mtx_unlock(&mountlist_mtx);
987 	vfs_event_signal(NULL, VQ_MOUNT, 0);
988 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
989 	VOP_UNLOCK(vp, 0);
990 	EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
991 	VOP_UNLOCK(newdp, 0);
992 	mountcheckdirs(vp, newdp);
993 	vrele(newdp);
994 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
995 		vfs_allocate_syncvnode(mp);
996 	vfs_op_exit(mp);
997 	vfs_unbusy(mp);
998 	return (0);
999 }
1000 
1001 /*
1002  * vfs_domount_update(): update of mounted file system
1003  */
1004 static int
1005 vfs_domount_update(
1006 	struct thread *td,		/* Calling thread. */
1007 	struct vnode *vp,		/* Mount point vnode. */
1008 	uint64_t fsflags,		/* Flags common to all filesystems. */
1009 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1010 	)
1011 {
1012 	struct export_args export;
1013 	void *bufp;
1014 	struct mount *mp;
1015 	int error, export_error, len;
1016 	uint64_t flag;
1017 
1018 	ASSERT_VOP_ELOCKED(vp, __func__);
1019 	KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1020 	mp = vp->v_mount;
1021 
1022 	if ((vp->v_vflag & VV_ROOT) == 0) {
1023 		if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1024 		    == 0)
1025 			error = EXDEV;
1026 		else
1027 			error = EINVAL;
1028 		vput(vp);
1029 		return (error);
1030 	}
1031 
1032 	/*
1033 	 * We only allow the filesystem to be reloaded if it
1034 	 * is currently mounted read-only.
1035 	 */
1036 	flag = mp->mnt_flag;
1037 	if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1038 		vput(vp);
1039 		return (EOPNOTSUPP);	/* Needs translation */
1040 	}
1041 	/*
1042 	 * Only privileged root, or (if MNT_USER is set) the user that
1043 	 * did the original mount is permitted to update it.
1044 	 */
1045 	error = vfs_suser(mp, td);
1046 	if (error != 0) {
1047 		vput(vp);
1048 		return (error);
1049 	}
1050 	if (vfs_busy(mp, MBF_NOWAIT)) {
1051 		vput(vp);
1052 		return (EBUSY);
1053 	}
1054 	VI_LOCK(vp);
1055 	if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1056 		VI_UNLOCK(vp);
1057 		vfs_unbusy(mp);
1058 		vput(vp);
1059 		return (EBUSY);
1060 	}
1061 	vp->v_iflag |= VI_MOUNT;
1062 	VI_UNLOCK(vp);
1063 	VOP_UNLOCK(vp, 0);
1064 
1065 	vfs_op_enter(mp);
1066 
1067 	MNT_ILOCK(mp);
1068 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1069 		MNT_IUNLOCK(mp);
1070 		error = EBUSY;
1071 		goto end;
1072 	}
1073 	mp->mnt_flag &= ~MNT_UPDATEMASK;
1074 	mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1075 	    MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1076 	if ((mp->mnt_flag & MNT_ASYNC) == 0)
1077 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1078 	MNT_IUNLOCK(mp);
1079 	mp->mnt_optnew = *optlist;
1080 	vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1081 
1082 	/*
1083 	 * Mount the filesystem.
1084 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1085 	 * get.  No freeing of cn_pnbuf.
1086 	 */
1087 	error = VFS_MOUNT(mp);
1088 
1089 	export_error = 0;
1090 	/* Process the export option. */
1091 	if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1092 	    &len) == 0) {
1093 		/* Assume that there is only 1 ABI for each length. */
1094 		switch (len) {
1095 		case (sizeof(struct oexport_args)):
1096 			bzero(&export, sizeof(export));
1097 			/* FALLTHROUGH */
1098 		case (sizeof(export)):
1099 			bcopy(bufp, &export, len);
1100 			export_error = vfs_export(mp, &export);
1101 			break;
1102 		default:
1103 			export_error = EINVAL;
1104 			break;
1105 		}
1106 	}
1107 
1108 	MNT_ILOCK(mp);
1109 	if (error == 0) {
1110 		mp->mnt_flag &=	~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1111 		    MNT_SNAPSHOT);
1112 	} else {
1113 		/*
1114 		 * If we fail, restore old mount flags. MNT_QUOTA is special,
1115 		 * because it is not part of MNT_UPDATEMASK, but it could have
1116 		 * changed in the meantime if quotactl(2) was called.
1117 		 * All in all we want current value of MNT_QUOTA, not the old
1118 		 * one.
1119 		 */
1120 		mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1121 	}
1122 	if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1123 	    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1124 		mp->mnt_kern_flag |= MNTK_ASYNC;
1125 	else
1126 		mp->mnt_kern_flag &= ~MNTK_ASYNC;
1127 	MNT_IUNLOCK(mp);
1128 
1129 	if (error != 0)
1130 		goto end;
1131 
1132 	if (mp->mnt_opt != NULL)
1133 		vfs_freeopts(mp->mnt_opt);
1134 	mp->mnt_opt = mp->mnt_optnew;
1135 	*optlist = NULL;
1136 	(void)VFS_STATFS(mp, &mp->mnt_stat);
1137 	/*
1138 	 * Prevent external consumers of mount options from reading
1139 	 * mnt_optnew.
1140 	 */
1141 	mp->mnt_optnew = NULL;
1142 
1143 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
1144 		vfs_allocate_syncvnode(mp);
1145 	else
1146 		vfs_deallocate_syncvnode(mp);
1147 end:
1148 	vfs_op_exit(mp);
1149 	vfs_unbusy(mp);
1150 	VI_LOCK(vp);
1151 	vp->v_iflag &= ~VI_MOUNT;
1152 	VI_UNLOCK(vp);
1153 	vrele(vp);
1154 	return (error != 0 ? error : export_error);
1155 }
1156 
1157 /*
1158  * vfs_domount(): actually attempt a filesystem mount.
1159  */
1160 static int
1161 vfs_domount(
1162 	struct thread *td,		/* Calling thread. */
1163 	const char *fstype,		/* Filesystem type. */
1164 	char *fspath,			/* Mount path. */
1165 	uint64_t fsflags,		/* Flags common to all filesystems. */
1166 	struct vfsoptlist **optlist	/* Options local to the filesystem. */
1167 	)
1168 {
1169 	struct vfsconf *vfsp;
1170 	struct nameidata nd;
1171 	struct vnode *vp;
1172 	char *pathbuf;
1173 	int error;
1174 
1175 	/*
1176 	 * Be ultra-paranoid about making sure the type and fspath
1177 	 * variables will fit in our mp buffers, including the
1178 	 * terminating NUL.
1179 	 */
1180 	if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1181 		return (ENAMETOOLONG);
1182 
1183 	if (jailed(td->td_ucred) || usermount == 0) {
1184 		if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1185 			return (error);
1186 	}
1187 
1188 	/*
1189 	 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1190 	 */
1191 	if (fsflags & MNT_EXPORTED) {
1192 		error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1193 		if (error)
1194 			return (error);
1195 	}
1196 	if (fsflags & MNT_SUIDDIR) {
1197 		error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1198 		if (error)
1199 			return (error);
1200 	}
1201 	/*
1202 	 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1203 	 */
1204 	if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1205 		if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1206 			fsflags |= MNT_NOSUID | MNT_USER;
1207 	}
1208 
1209 	/* Load KLDs before we lock the covered vnode to avoid reversals. */
1210 	vfsp = NULL;
1211 	if ((fsflags & MNT_UPDATE) == 0) {
1212 		/* Don't try to load KLDs if we're mounting the root. */
1213 		if (fsflags & MNT_ROOTFS)
1214 			vfsp = vfs_byname(fstype);
1215 		else
1216 			vfsp = vfs_byname_kld(fstype, td, &error);
1217 		if (vfsp == NULL)
1218 			return (ENODEV);
1219 	}
1220 
1221 	/*
1222 	 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1223 	 */
1224 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1225 	    UIO_SYSSPACE, fspath, td);
1226 	error = namei(&nd);
1227 	if (error != 0)
1228 		return (error);
1229 	NDFREE(&nd, NDF_ONLY_PNBUF);
1230 	vp = nd.ni_vp;
1231 	if ((fsflags & MNT_UPDATE) == 0) {
1232 		pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1233 		strcpy(pathbuf, fspath);
1234 		error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1235 		/* debug.disablefullpath == 1 results in ENODEV */
1236 		if (error == 0 || error == ENODEV) {
1237 			error = vfs_domount_first(td, vfsp, pathbuf, vp,
1238 			    fsflags, optlist);
1239 		}
1240 		free(pathbuf, M_TEMP);
1241 	} else
1242 		error = vfs_domount_update(td, vp, fsflags, optlist);
1243 
1244 	return (error);
1245 }
1246 
1247 /*
1248  * Unmount a filesystem.
1249  *
1250  * Note: unmount takes a path to the vnode mounted on as argument, not
1251  * special file (as before).
1252  */
1253 #ifndef _SYS_SYSPROTO_H_
1254 struct unmount_args {
1255 	char	*path;
1256 	int	flags;
1257 };
1258 #endif
1259 /* ARGSUSED */
1260 int
1261 sys_unmount(struct thread *td, struct unmount_args *uap)
1262 {
1263 	struct nameidata nd;
1264 	struct mount *mp;
1265 	char *pathbuf;
1266 	int error, id0, id1;
1267 
1268 	AUDIT_ARG_VALUE(uap->flags);
1269 	if (jailed(td->td_ucred) || usermount == 0) {
1270 		error = priv_check(td, PRIV_VFS_UNMOUNT);
1271 		if (error)
1272 			return (error);
1273 	}
1274 
1275 	pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1276 	error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1277 	if (error) {
1278 		free(pathbuf, M_TEMP);
1279 		return (error);
1280 	}
1281 	if (uap->flags & MNT_BYFSID) {
1282 		AUDIT_ARG_TEXT(pathbuf);
1283 		/* Decode the filesystem ID. */
1284 		if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1285 			free(pathbuf, M_TEMP);
1286 			return (EINVAL);
1287 		}
1288 
1289 		mtx_lock(&mountlist_mtx);
1290 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1291 			if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1292 			    mp->mnt_stat.f_fsid.val[1] == id1) {
1293 				vfs_ref(mp);
1294 				break;
1295 			}
1296 		}
1297 		mtx_unlock(&mountlist_mtx);
1298 	} else {
1299 		/*
1300 		 * Try to find global path for path argument.
1301 		 */
1302 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1303 		    UIO_SYSSPACE, pathbuf, td);
1304 		if (namei(&nd) == 0) {
1305 			NDFREE(&nd, NDF_ONLY_PNBUF);
1306 			error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1307 			    MNAMELEN);
1308 			if (error == 0 || error == ENODEV)
1309 				vput(nd.ni_vp);
1310 		}
1311 		mtx_lock(&mountlist_mtx);
1312 		TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1313 			if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1314 				vfs_ref(mp);
1315 				break;
1316 			}
1317 		}
1318 		mtx_unlock(&mountlist_mtx);
1319 	}
1320 	free(pathbuf, M_TEMP);
1321 	if (mp == NULL) {
1322 		/*
1323 		 * Previously we returned ENOENT for a nonexistent path and
1324 		 * EINVAL for a non-mountpoint.  We cannot tell these apart
1325 		 * now, so in the !MNT_BYFSID case return the more likely
1326 		 * EINVAL for compatibility.
1327 		 */
1328 		return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1329 	}
1330 
1331 	/*
1332 	 * Don't allow unmounting the root filesystem.
1333 	 */
1334 	if (mp->mnt_flag & MNT_ROOTFS) {
1335 		vfs_rel(mp);
1336 		return (EINVAL);
1337 	}
1338 	error = dounmount(mp, uap->flags, td);
1339 	return (error);
1340 }
1341 
1342 /*
1343  * Return error if any of the vnodes, ignoring the root vnode
1344  * and the syncer vnode, have non-zero usecount.
1345  *
1346  * This function is purely advisory - it can return false positives
1347  * and negatives.
1348  */
1349 static int
1350 vfs_check_usecounts(struct mount *mp)
1351 {
1352 	struct vnode *vp, *mvp;
1353 
1354 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1355 		if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1356 		    vp->v_usecount != 0) {
1357 			VI_UNLOCK(vp);
1358 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1359 			return (EBUSY);
1360 		}
1361 		VI_UNLOCK(vp);
1362 	}
1363 
1364 	return (0);
1365 }
1366 
1367 static void
1368 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1369 {
1370 
1371 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1372 	mp->mnt_kern_flag &= ~mntkflags;
1373 	if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1374 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
1375 		wakeup(mp);
1376 	}
1377 	vfs_op_exit_locked(mp);
1378 	MNT_IUNLOCK(mp);
1379 	if (coveredvp != NULL) {
1380 		VOP_UNLOCK(coveredvp, 0);
1381 		vdrop(coveredvp);
1382 	}
1383 	vn_finished_write(mp);
1384 }
1385 
1386 /*
1387  * There are various reference counters associated with the mount point.
1388  * Normally it is permitted to modify them without taking the mnt ilock,
1389  * but this behavior can be temporarily disabled if stable value is needed
1390  * or callers are expected to block (e.g. to not allow new users during
1391  * forced unmount).
1392  */
1393 void
1394 vfs_op_enter(struct mount *mp)
1395 {
1396 	int cpu;
1397 
1398 	MNT_ILOCK(mp);
1399 	mp->mnt_vfs_ops++;
1400 	if (mp->mnt_vfs_ops > 1) {
1401 		MNT_IUNLOCK(mp);
1402 		return;
1403 	}
1404 	/*
1405 	 * Paired with a fence in vfs_op_thread_enter(). See the comment
1406 	 * above it for details.
1407 	 */
1408 	atomic_thread_fence_seq_cst();
1409 	vfs_op_barrier_wait(mp);
1410 	/*
1411 	 * Paired with a fence in vfs_op_thread_exit().
1412 	 */
1413 	atomic_thread_fence_acq();
1414 	CPU_FOREACH(cpu) {
1415 		mp->mnt_ref +=
1416 		    zpcpu_replace_cpu(mp->mnt_ref_pcpu, 0, cpu);
1417 		mp->mnt_lockref +=
1418 		    zpcpu_replace_cpu(mp->mnt_lockref_pcpu, 0, cpu);
1419 		mp->mnt_writeopcount +=
1420 		    zpcpu_replace_cpu(mp->mnt_writeopcount_pcpu, 0, cpu);
1421 	}
1422 	MNT_IUNLOCK(mp);
1423 	vfs_assert_mount_counters(mp);
1424 }
1425 
1426 void
1427 vfs_op_exit_locked(struct mount *mp)
1428 {
1429 
1430 	mtx_assert(MNT_MTX(mp), MA_OWNED);
1431 
1432 	if (mp->mnt_vfs_ops <= 0)
1433 		panic("%s: invalid vfs_ops count %d for mp %p\n",
1434 		    __func__, mp->mnt_vfs_ops, mp);
1435 	mp->mnt_vfs_ops--;
1436 }
1437 
1438 void
1439 vfs_op_exit(struct mount *mp)
1440 {
1441 
1442 	MNT_ILOCK(mp);
1443 	vfs_op_exit_locked(mp);
1444 	MNT_IUNLOCK(mp);
1445 }
1446 
1447 /*
1448  * It is assumed the caller already posted at least an acquire barrier.
1449  */
1450 void
1451 vfs_op_barrier_wait(struct mount *mp)
1452 {
1453 	int *in_op;
1454 	int cpu;
1455 
1456 	CPU_FOREACH(cpu) {
1457 		in_op = zpcpu_get_cpu(mp->mnt_thread_in_ops_pcpu, cpu);
1458 		while (atomic_load_int(in_op))
1459 			cpu_spinwait();
1460 	}
1461 }
1462 
1463 #ifdef DIAGNOSTIC
1464 void
1465 vfs_assert_mount_counters(struct mount *mp)
1466 {
1467 	int cpu;
1468 
1469 	if (mp->mnt_vfs_ops == 0)
1470 		return;
1471 
1472 	CPU_FOREACH(cpu) {
1473 		if (*(int *)zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu) != 0 ||
1474 		    *(int *)zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu) != 0 ||
1475 		    *(int *)zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu) != 0)
1476 			vfs_dump_mount_counters(mp);
1477 	}
1478 }
1479 
1480 void
1481 vfs_dump_mount_counters(struct mount *mp)
1482 {
1483 	int cpu, *count;
1484 	int ref, lockref, writeopcount;
1485 
1486 	printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1487 
1488 	printf("        ref : ");
1489 	ref = mp->mnt_ref;
1490 	CPU_FOREACH(cpu) {
1491 		count = zpcpu_get_cpu(mp->mnt_ref_pcpu, cpu);
1492 		printf("%d ", *count);
1493 		ref += *count;
1494 	}
1495 	printf("\n");
1496 	printf("    lockref : ");
1497 	lockref = mp->mnt_lockref;
1498 	CPU_FOREACH(cpu) {
1499 		count = zpcpu_get_cpu(mp->mnt_lockref_pcpu, cpu);
1500 		printf("%d ", *count);
1501 		lockref += *count;
1502 	}
1503 	printf("\n");
1504 	printf("writeopcount: ");
1505 	writeopcount = mp->mnt_writeopcount;
1506 	CPU_FOREACH(cpu) {
1507 		count = zpcpu_get_cpu(mp->mnt_writeopcount_pcpu, cpu);
1508 		printf("%d ", *count);
1509 		writeopcount += *count;
1510 	}
1511 	printf("\n");
1512 
1513 	printf("counter       struct total\n");
1514 	printf("ref             %-5d  %-5d\n", mp->mnt_ref, ref);
1515 	printf("lockref         %-5d  %-5d\n", mp->mnt_lockref, lockref);
1516 	printf("writeopcount    %-5d  %-5d\n", mp->mnt_writeopcount, writeopcount);
1517 
1518 	panic("invalid counts on struct mount");
1519 }
1520 #endif
1521 
1522 int
1523 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1524 {
1525 	int *base, *pcpu;
1526 	int cpu, sum;
1527 
1528 	switch (which) {
1529 	case MNT_COUNT_REF:
1530 		base = &mp->mnt_ref;
1531 		pcpu = mp->mnt_ref_pcpu;
1532 		break;
1533 	case MNT_COUNT_LOCKREF:
1534 		base = &mp->mnt_lockref;
1535 		pcpu = mp->mnt_lockref_pcpu;
1536 		break;
1537 	case MNT_COUNT_WRITEOPCOUNT:
1538 		base = &mp->mnt_writeopcount;
1539 		pcpu = mp->mnt_writeopcount_pcpu;
1540 		break;
1541 	}
1542 
1543 	sum = *base;
1544 	CPU_FOREACH(cpu) {
1545 		sum += *(int *)zpcpu_get_cpu(pcpu, cpu);
1546 	}
1547 	return (sum);
1548 }
1549 
1550 /*
1551  * Do the actual filesystem unmount.
1552  */
1553 int
1554 dounmount(struct mount *mp, int flags, struct thread *td)
1555 {
1556 	struct vnode *coveredvp;
1557 	int error;
1558 	uint64_t async_flag;
1559 	int mnt_gen_r;
1560 
1561 	if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1562 		mnt_gen_r = mp->mnt_gen;
1563 		VI_LOCK(coveredvp);
1564 		vholdl(coveredvp);
1565 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1566 		/*
1567 		 * Check for mp being unmounted while waiting for the
1568 		 * covered vnode lock.
1569 		 */
1570 		if (coveredvp->v_mountedhere != mp ||
1571 		    coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1572 			VOP_UNLOCK(coveredvp, 0);
1573 			vdrop(coveredvp);
1574 			vfs_rel(mp);
1575 			return (EBUSY);
1576 		}
1577 	}
1578 
1579 	/*
1580 	 * Only privileged root, or (if MNT_USER is set) the user that did the
1581 	 * original mount is permitted to unmount this filesystem.
1582 	 */
1583 	error = vfs_suser(mp, td);
1584 	if (error != 0) {
1585 		if (coveredvp != NULL) {
1586 			VOP_UNLOCK(coveredvp, 0);
1587 			vdrop(coveredvp);
1588 		}
1589 		vfs_rel(mp);
1590 		return (error);
1591 	}
1592 
1593 	vfs_op_enter(mp);
1594 
1595 	vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1596 	MNT_ILOCK(mp);
1597 	if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1598 	    (mp->mnt_flag & MNT_UPDATE) != 0 ||
1599 	    !TAILQ_EMPTY(&mp->mnt_uppers)) {
1600 		dounmount_cleanup(mp, coveredvp, 0);
1601 		return (EBUSY);
1602 	}
1603 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
1604 	if (flags & MNT_NONBUSY) {
1605 		MNT_IUNLOCK(mp);
1606 		error = vfs_check_usecounts(mp);
1607 		MNT_ILOCK(mp);
1608 		if (error != 0) {
1609 			dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1610 			return (error);
1611 		}
1612 	}
1613 	/* Allow filesystems to detect that a forced unmount is in progress. */
1614 	if (flags & MNT_FORCE) {
1615 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1616 		MNT_IUNLOCK(mp);
1617 		/*
1618 		 * Must be done after setting MNTK_UNMOUNTF and before
1619 		 * waiting for mnt_lockref to become 0.
1620 		 */
1621 		VFS_PURGE(mp);
1622 		MNT_ILOCK(mp);
1623 	}
1624 	error = 0;
1625 	if (mp->mnt_lockref) {
1626 		mp->mnt_kern_flag |= MNTK_DRAINING;
1627 		error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1628 		    "mount drain", 0);
1629 	}
1630 	MNT_IUNLOCK(mp);
1631 	KASSERT(mp->mnt_lockref == 0,
1632 	    ("%s: invalid lock refcount in the drain path @ %s:%d",
1633 	    __func__, __FILE__, __LINE__));
1634 	KASSERT(error == 0,
1635 	    ("%s: invalid return value for msleep in the drain path @ %s:%d",
1636 	    __func__, __FILE__, __LINE__));
1637 
1638 	if (mp->mnt_flag & MNT_EXPUBLIC)
1639 		vfs_setpublicfs(NULL, NULL, NULL);
1640 
1641 	/*
1642 	 * From now, we can claim that the use reference on the
1643 	 * coveredvp is ours, and the ref can be released only by
1644 	 * successfull unmount by us, or left for later unmount
1645 	 * attempt.  The previously acquired hold reference is no
1646 	 * longer needed to protect the vnode from reuse.
1647 	 */
1648 	if (coveredvp != NULL)
1649 		vdrop(coveredvp);
1650 
1651 	vfs_msync(mp, MNT_WAIT);
1652 	MNT_ILOCK(mp);
1653 	async_flag = mp->mnt_flag & MNT_ASYNC;
1654 	mp->mnt_flag &= ~MNT_ASYNC;
1655 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
1656 	MNT_IUNLOCK(mp);
1657 	cache_purgevfs(mp, false); /* remove cache entries for this file sys */
1658 	vfs_deallocate_syncvnode(mp);
1659 	if ((mp->mnt_flag & MNT_RDONLY) != 0 || (flags & MNT_FORCE) != 0 ||
1660 	    (error = VFS_SYNC(mp, MNT_WAIT)) == 0)
1661 		error = VFS_UNMOUNT(mp, flags);
1662 	vn_finished_write(mp);
1663 	/*
1664 	 * If we failed to flush the dirty blocks for this mount point,
1665 	 * undo all the cdir/rdir and rootvnode changes we made above.
1666 	 * Unless we failed to do so because the device is reporting that
1667 	 * it doesn't exist anymore.
1668 	 */
1669 	if (error && error != ENXIO) {
1670 		MNT_ILOCK(mp);
1671 		if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1672 			MNT_IUNLOCK(mp);
1673 			vfs_allocate_syncvnode(mp);
1674 			MNT_ILOCK(mp);
1675 		}
1676 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1677 		mp->mnt_flag |= async_flag;
1678 		if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1679 		    (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1680 			mp->mnt_kern_flag |= MNTK_ASYNC;
1681 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
1682 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
1683 			wakeup(mp);
1684 		}
1685 		vfs_op_exit_locked(mp);
1686 		MNT_IUNLOCK(mp);
1687 		if (coveredvp)
1688 			VOP_UNLOCK(coveredvp, 0);
1689 		return (error);
1690 	}
1691 	mtx_lock(&mountlist_mtx);
1692 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
1693 	mtx_unlock(&mountlist_mtx);
1694 	EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1695 	if (coveredvp != NULL) {
1696 		coveredvp->v_mountedhere = NULL;
1697 		VOP_UNLOCK(coveredvp, 0);
1698 	}
1699 	vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1700 	if (rootvnode != NULL && mp == rootvnode->v_mount) {
1701 		vrele(rootvnode);
1702 		rootvnode = NULL;
1703 	}
1704 	if (mp == rootdevmp)
1705 		rootdevmp = NULL;
1706 	vfs_mount_destroy(mp);
1707 	return (0);
1708 }
1709 
1710 /*
1711  * Report errors during filesystem mounting.
1712  */
1713 void
1714 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1715 {
1716 	struct vfsoptlist *moptlist = mp->mnt_optnew;
1717 	va_list ap;
1718 	int error, len;
1719 	char *errmsg;
1720 
1721 	error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1722 	if (error || errmsg == NULL || len <= 0)
1723 		return;
1724 
1725 	va_start(ap, fmt);
1726 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1727 	va_end(ap);
1728 }
1729 
1730 void
1731 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1732 {
1733 	va_list ap;
1734 	int error, len;
1735 	char *errmsg;
1736 
1737 	error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1738 	if (error || errmsg == NULL || len <= 0)
1739 		return;
1740 
1741 	va_start(ap, fmt);
1742 	vsnprintf(errmsg, (size_t)len, fmt, ap);
1743 	va_end(ap);
1744 }
1745 
1746 /*
1747  * ---------------------------------------------------------------------
1748  * Functions for querying mount options/arguments from filesystems.
1749  */
1750 
1751 /*
1752  * Check that no unknown options are given
1753  */
1754 int
1755 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1756 {
1757 	struct vfsopt *opt;
1758 	char errmsg[255];
1759 	const char **t, *p, *q;
1760 	int ret = 0;
1761 
1762 	TAILQ_FOREACH(opt, opts, link) {
1763 		p = opt->name;
1764 		q = NULL;
1765 		if (p[0] == 'n' && p[1] == 'o')
1766 			q = p + 2;
1767 		for(t = global_opts; *t != NULL; t++) {
1768 			if (strcmp(*t, p) == 0)
1769 				break;
1770 			if (q != NULL) {
1771 				if (strcmp(*t, q) == 0)
1772 					break;
1773 			}
1774 		}
1775 		if (*t != NULL)
1776 			continue;
1777 		for(t = legal; *t != NULL; t++) {
1778 			if (strcmp(*t, p) == 0)
1779 				break;
1780 			if (q != NULL) {
1781 				if (strcmp(*t, q) == 0)
1782 					break;
1783 			}
1784 		}
1785 		if (*t != NULL)
1786 			continue;
1787 		snprintf(errmsg, sizeof(errmsg),
1788 		    "mount option <%s> is unknown", p);
1789 		ret = EINVAL;
1790 	}
1791 	if (ret != 0) {
1792 		TAILQ_FOREACH(opt, opts, link) {
1793 			if (strcmp(opt->name, "errmsg") == 0) {
1794 				strncpy((char *)opt->value, errmsg, opt->len);
1795 				break;
1796 			}
1797 		}
1798 		if (opt == NULL)
1799 			printf("%s\n", errmsg);
1800 	}
1801 	return (ret);
1802 }
1803 
1804 /*
1805  * Get a mount option by its name.
1806  *
1807  * Return 0 if the option was found, ENOENT otherwise.
1808  * If len is non-NULL it will be filled with the length
1809  * of the option. If buf is non-NULL, it will be filled
1810  * with the address of the option.
1811  */
1812 int
1813 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
1814 {
1815 	struct vfsopt *opt;
1816 
1817 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1818 
1819 	TAILQ_FOREACH(opt, opts, link) {
1820 		if (strcmp(name, opt->name) == 0) {
1821 			opt->seen = 1;
1822 			if (len != NULL)
1823 				*len = opt->len;
1824 			if (buf != NULL)
1825 				*buf = opt->value;
1826 			return (0);
1827 		}
1828 	}
1829 	return (ENOENT);
1830 }
1831 
1832 int
1833 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1834 {
1835 	struct vfsopt *opt;
1836 
1837 	if (opts == NULL)
1838 		return (-1);
1839 
1840 	TAILQ_FOREACH(opt, opts, link) {
1841 		if (strcmp(name, opt->name) == 0) {
1842 			opt->seen = 1;
1843 			return (opt->pos);
1844 		}
1845 	}
1846 	return (-1);
1847 }
1848 
1849 int
1850 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
1851 {
1852 	char *opt_value, *vtp;
1853 	quad_t iv;
1854 	int error, opt_len;
1855 
1856 	error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
1857 	if (error != 0)
1858 		return (error);
1859 	if (opt_len == 0 || opt_value == NULL)
1860 		return (EINVAL);
1861 	if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
1862 		return (EINVAL);
1863 	iv = strtoq(opt_value, &vtp, 0);
1864 	if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
1865 		return (EINVAL);
1866 	if (iv < 0)
1867 		return (EINVAL);
1868 	switch (vtp[0]) {
1869 	case 't': case 'T':
1870 		iv *= 1024;
1871 		/* FALLTHROUGH */
1872 	case 'g': case 'G':
1873 		iv *= 1024;
1874 		/* FALLTHROUGH */
1875 	case 'm': case 'M':
1876 		iv *= 1024;
1877 		/* FALLTHROUGH */
1878 	case 'k': case 'K':
1879 		iv *= 1024;
1880 	case '\0':
1881 		break;
1882 	default:
1883 		return (EINVAL);
1884 	}
1885 	*value = iv;
1886 
1887 	return (0);
1888 }
1889 
1890 char *
1891 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1892 {
1893 	struct vfsopt *opt;
1894 
1895 	*error = 0;
1896 	TAILQ_FOREACH(opt, opts, link) {
1897 		if (strcmp(name, opt->name) != 0)
1898 			continue;
1899 		opt->seen = 1;
1900 		if (opt->len == 0 ||
1901 		    ((char *)opt->value)[opt->len - 1] != '\0') {
1902 			*error = EINVAL;
1903 			return (NULL);
1904 		}
1905 		return (opt->value);
1906 	}
1907 	*error = ENOENT;
1908 	return (NULL);
1909 }
1910 
1911 int
1912 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
1913 	uint64_t val)
1914 {
1915 	struct vfsopt *opt;
1916 
1917 	TAILQ_FOREACH(opt, opts, link) {
1918 		if (strcmp(name, opt->name) == 0) {
1919 			opt->seen = 1;
1920 			if (w != NULL)
1921 				*w |= val;
1922 			return (1);
1923 		}
1924 	}
1925 	if (w != NULL)
1926 		*w &= ~val;
1927 	return (0);
1928 }
1929 
1930 int
1931 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1932 {
1933 	va_list ap;
1934 	struct vfsopt *opt;
1935 	int ret;
1936 
1937 	KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1938 
1939 	TAILQ_FOREACH(opt, opts, link) {
1940 		if (strcmp(name, opt->name) != 0)
1941 			continue;
1942 		opt->seen = 1;
1943 		if (opt->len == 0 || opt->value == NULL)
1944 			return (0);
1945 		if (((char *)opt->value)[opt->len - 1] != '\0')
1946 			return (0);
1947 		va_start(ap, fmt);
1948 		ret = vsscanf(opt->value, fmt, ap);
1949 		va_end(ap);
1950 		return (ret);
1951 	}
1952 	return (0);
1953 }
1954 
1955 int
1956 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
1957 {
1958 	struct vfsopt *opt;
1959 
1960 	TAILQ_FOREACH(opt, opts, link) {
1961 		if (strcmp(name, opt->name) != 0)
1962 			continue;
1963 		opt->seen = 1;
1964 		if (opt->value == NULL)
1965 			opt->len = len;
1966 		else {
1967 			if (opt->len != len)
1968 				return (EINVAL);
1969 			bcopy(value, opt->value, len);
1970 		}
1971 		return (0);
1972 	}
1973 	return (ENOENT);
1974 }
1975 
1976 int
1977 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
1978 {
1979 	struct vfsopt *opt;
1980 
1981 	TAILQ_FOREACH(opt, opts, link) {
1982 		if (strcmp(name, opt->name) != 0)
1983 			continue;
1984 		opt->seen = 1;
1985 		if (opt->value == NULL)
1986 			opt->len = len;
1987 		else {
1988 			if (opt->len < len)
1989 				return (EINVAL);
1990 			opt->len = len;
1991 			bcopy(value, opt->value, len);
1992 		}
1993 		return (0);
1994 	}
1995 	return (ENOENT);
1996 }
1997 
1998 int
1999 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2000 {
2001 	struct vfsopt *opt;
2002 
2003 	TAILQ_FOREACH(opt, opts, link) {
2004 		if (strcmp(name, opt->name) != 0)
2005 			continue;
2006 		opt->seen = 1;
2007 		if (opt->value == NULL)
2008 			opt->len = strlen(value) + 1;
2009 		else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2010 			return (EINVAL);
2011 		return (0);
2012 	}
2013 	return (ENOENT);
2014 }
2015 
2016 /*
2017  * Find and copy a mount option.
2018  *
2019  * The size of the buffer has to be specified
2020  * in len, if it is not the same length as the
2021  * mount option, EINVAL is returned.
2022  * Returns ENOENT if the option is not found.
2023  */
2024 int
2025 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2026 {
2027 	struct vfsopt *opt;
2028 
2029 	KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2030 
2031 	TAILQ_FOREACH(opt, opts, link) {
2032 		if (strcmp(name, opt->name) == 0) {
2033 			opt->seen = 1;
2034 			if (len != opt->len)
2035 				return (EINVAL);
2036 			bcopy(opt->value, dest, opt->len);
2037 			return (0);
2038 		}
2039 	}
2040 	return (ENOENT);
2041 }
2042 
2043 int
2044 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2045 {
2046 
2047 	/*
2048 	 * Filesystems only fill in part of the structure for updates, we
2049 	 * have to read the entirety first to get all content.
2050 	 */
2051 	memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2052 
2053 	/*
2054 	 * Set these in case the underlying filesystem fails to do so.
2055 	 */
2056 	sbp->f_version = STATFS_VERSION;
2057 	sbp->f_namemax = NAME_MAX;
2058 	sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2059 
2060 	return (mp->mnt_op->vfs_statfs(mp, sbp));
2061 }
2062 
2063 void
2064 vfs_mountedfrom(struct mount *mp, const char *from)
2065 {
2066 
2067 	bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2068 	strlcpy(mp->mnt_stat.f_mntfromname, from,
2069 	    sizeof mp->mnt_stat.f_mntfromname);
2070 }
2071 
2072 /*
2073  * ---------------------------------------------------------------------
2074  * This is the api for building mount args and mounting filesystems from
2075  * inside the kernel.
2076  *
2077  * The API works by accumulation of individual args.  First error is
2078  * latched.
2079  *
2080  * XXX: should be documented in new manpage kernel_mount(9)
2081  */
2082 
2083 /* A memory allocation which must be freed when we are done */
2084 struct mntaarg {
2085 	SLIST_ENTRY(mntaarg)	next;
2086 };
2087 
2088 /* The header for the mount arguments */
2089 struct mntarg {
2090 	struct iovec *v;
2091 	int len;
2092 	int error;
2093 	SLIST_HEAD(, mntaarg)	list;
2094 };
2095 
2096 /*
2097  * Add a boolean argument.
2098  *
2099  * flag is the boolean value.
2100  * name must start with "no".
2101  */
2102 struct mntarg *
2103 mount_argb(struct mntarg *ma, int flag, const char *name)
2104 {
2105 
2106 	KASSERT(name[0] == 'n' && name[1] == 'o',
2107 	    ("mount_argb(...,%s): name must start with 'no'", name));
2108 
2109 	return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2110 }
2111 
2112 /*
2113  * Add an argument printf style
2114  */
2115 struct mntarg *
2116 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2117 {
2118 	va_list ap;
2119 	struct mntaarg *maa;
2120 	struct sbuf *sb;
2121 	int len;
2122 
2123 	if (ma == NULL) {
2124 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2125 		SLIST_INIT(&ma->list);
2126 	}
2127 	if (ma->error)
2128 		return (ma);
2129 
2130 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2131 	    M_MOUNT, M_WAITOK);
2132 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2133 	ma->v[ma->len].iov_len = strlen(name) + 1;
2134 	ma->len++;
2135 
2136 	sb = sbuf_new_auto();
2137 	va_start(ap, fmt);
2138 	sbuf_vprintf(sb, fmt, ap);
2139 	va_end(ap);
2140 	sbuf_finish(sb);
2141 	len = sbuf_len(sb) + 1;
2142 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2143 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2144 	bcopy(sbuf_data(sb), maa + 1, len);
2145 	sbuf_delete(sb);
2146 
2147 	ma->v[ma->len].iov_base = maa + 1;
2148 	ma->v[ma->len].iov_len = len;
2149 	ma->len++;
2150 
2151 	return (ma);
2152 }
2153 
2154 /*
2155  * Add an argument which is a userland string.
2156  */
2157 struct mntarg *
2158 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2159 {
2160 	struct mntaarg *maa;
2161 	char *tbuf;
2162 
2163 	if (val == NULL)
2164 		return (ma);
2165 	if (ma == NULL) {
2166 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2167 		SLIST_INIT(&ma->list);
2168 	}
2169 	if (ma->error)
2170 		return (ma);
2171 	maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2172 	SLIST_INSERT_HEAD(&ma->list, maa, next);
2173 	tbuf = (void *)(maa + 1);
2174 	ma->error = copyinstr(val, tbuf, len, NULL);
2175 	return (mount_arg(ma, name, tbuf, -1));
2176 }
2177 
2178 /*
2179  * Plain argument.
2180  *
2181  * If length is -1, treat value as a C string.
2182  */
2183 struct mntarg *
2184 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2185 {
2186 
2187 	if (ma == NULL) {
2188 		ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2189 		SLIST_INIT(&ma->list);
2190 	}
2191 	if (ma->error)
2192 		return (ma);
2193 
2194 	ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2195 	    M_MOUNT, M_WAITOK);
2196 	ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2197 	ma->v[ma->len].iov_len = strlen(name) + 1;
2198 	ma->len++;
2199 
2200 	ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2201 	if (len < 0)
2202 		ma->v[ma->len].iov_len = strlen(val) + 1;
2203 	else
2204 		ma->v[ma->len].iov_len = len;
2205 	ma->len++;
2206 	return (ma);
2207 }
2208 
2209 /*
2210  * Free a mntarg structure
2211  */
2212 static void
2213 free_mntarg(struct mntarg *ma)
2214 {
2215 	struct mntaarg *maa;
2216 
2217 	while (!SLIST_EMPTY(&ma->list)) {
2218 		maa = SLIST_FIRST(&ma->list);
2219 		SLIST_REMOVE_HEAD(&ma->list, next);
2220 		free(maa, M_MOUNT);
2221 	}
2222 	free(ma->v, M_MOUNT);
2223 	free(ma, M_MOUNT);
2224 }
2225 
2226 /*
2227  * Mount a filesystem
2228  */
2229 int
2230 kernel_mount(struct mntarg *ma, uint64_t flags)
2231 {
2232 	struct uio auio;
2233 	int error;
2234 
2235 	KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2236 	KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2237 	KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2238 
2239 	auio.uio_iov = ma->v;
2240 	auio.uio_iovcnt = ma->len;
2241 	auio.uio_segflg = UIO_SYSSPACE;
2242 
2243 	error = ma->error;
2244 	if (!error)
2245 		error = vfs_donmount(curthread, flags, &auio);
2246 	free_mntarg(ma);
2247 	return (error);
2248 }
2249 
2250 /*
2251  * A printflike function to mount a filesystem.
2252  */
2253 int
2254 kernel_vmount(int flags, ...)
2255 {
2256 	struct mntarg *ma = NULL;
2257 	va_list ap;
2258 	const char *cp;
2259 	const void *vp;
2260 	int error;
2261 
2262 	va_start(ap, flags);
2263 	for (;;) {
2264 		cp = va_arg(ap, const char *);
2265 		if (cp == NULL)
2266 			break;
2267 		vp = va_arg(ap, const void *);
2268 		ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2269 	}
2270 	va_end(ap);
2271 
2272 	error = kernel_mount(ma, flags);
2273 	return (error);
2274 }
2275 
2276 void
2277 vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp)
2278 {
2279 
2280 	bcopy(oexp, exp, sizeof(*oexp));
2281 	exp->ex_numsecflavors = 0;
2282 }
2283