xref: /netbsd-src/sys/kern/vfs_mount.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: vfs_mount.c,v 1.67 2017/08/21 09:00:21 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997-2011 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1989, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  * (c) UNIX System Laboratories, Inc.
37  * All or some portions of this file are derived from material licensed
38  * to the University of California by American Telephone and Telegraph
39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
40  * the permission of UNIX System Laboratories, Inc.
41  *
42  * Redistribution and use in source and binary forms, with or without
43  * modification, are permitted provided that the following conditions
44  * are met:
45  * 1. Redistributions of source code must retain the above copyright
46  *    notice, this list of conditions and the following disclaimer.
47  * 2. Redistributions in binary form must reproduce the above copyright
48  *    notice, this list of conditions and the following disclaimer in the
49  *    documentation and/or other materials provided with the distribution.
50  * 3. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
67  */
68 
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: vfs_mount.c,v 1.67 2017/08/21 09:00:21 hannken Exp $");
71 
72 #include <sys/param.h>
73 #include <sys/kernel.h>
74 
75 #include <sys/atomic.h>
76 #include <sys/buf.h>
77 #include <sys/conf.h>
78 #include <sys/fcntl.h>
79 #include <sys/filedesc.h>
80 #include <sys/device.h>
81 #include <sys/kauth.h>
82 #include <sys/kmem.h>
83 #include <sys/module.h>
84 #include <sys/mount.h>
85 #include <sys/fstrans.h>
86 #include <sys/namei.h>
87 #include <sys/extattr.h>
88 #include <sys/syscallargs.h>
89 #include <sys/sysctl.h>
90 #include <sys/systm.h>
91 #include <sys/vfs_syscalls.h>
92 #include <sys/vnode_impl.h>
93 
94 #include <miscfs/genfs/genfs.h>
95 #include <miscfs/specfs/specdev.h>
96 
97 enum mountlist_type {
98 	ME_MOUNT,
99 	ME_MARKER
100 };
101 struct mountlist_entry {
102 	TAILQ_ENTRY(mountlist_entry) me_list;	/* Mount list. */
103 	struct mount *me_mount;			/* Actual mount if ME_MOUNT,
104 						   current mount else. */
105 	enum mountlist_type me_type;		/* Mount or marker. */
106 };
107 struct mount_iterator {
108 	struct mountlist_entry mi_entry;
109 };
110 
111 static struct vnode *vfs_vnode_iterator_next1(struct vnode_iterator *,
112     bool (*)(void *, struct vnode *), void *, bool);
113 
114 /* Root filesystem. */
115 vnode_t *			rootvnode;
116 
117 /* Mounted filesystem list. */
118 static TAILQ_HEAD(mountlist, mountlist_entry) mountlist;
119 static kmutex_t			mountlist_lock;
120 int vnode_offset_next_by_lru	/* XXX: ugly hack for pstat.c */
121     = offsetof(vnode_impl_t, vi_lrulist.tqe_next);
122 
123 kmutex_t			mntvnode_lock;
124 kmutex_t			vfs_list_lock;
125 
126 static specificdata_domain_t	mount_specificdata_domain;
127 static kmutex_t			mntid_lock;
128 
129 static kmutex_t			mountgen_lock;
130 static uint64_t			mountgen;
131 
132 void
133 vfs_mount_sysinit(void)
134 {
135 
136 	TAILQ_INIT(&mountlist);
137 	mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE);
138 	mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE);
139 	mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE);
140 
141 	mount_specificdata_domain = specificdata_domain_create();
142 	mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE);
143 	mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE);
144 	mountgen = 0;
145 }
146 
147 struct mount *
148 vfs_mountalloc(struct vfsops *vfsops, vnode_t *vp)
149 {
150 	struct mount *mp;
151 	int error __diagused;
152 	extern struct vfsops dead_vfsops;
153 
154 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
155 	mp->mnt_op = vfsops;
156 	mp->mnt_refcnt = 1;
157 	TAILQ_INIT(&mp->mnt_vnodelist);
158 	mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE);
159 	mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE);
160 	mp->mnt_vnodecovered = vp;
161 	mount_initspecific(mp);
162 	if (vfsops != &dead_vfsops) {
163 		error = fstrans_mount(mp);
164 		KASSERT(error == 0);
165 	}
166 
167 	mutex_enter(&mountgen_lock);
168 	mp->mnt_gen = mountgen++;
169 	mutex_exit(&mountgen_lock);
170 
171 	return mp;
172 }
173 
174 /*
175  * vfs_rootmountalloc: lookup a filesystem type, and if found allocate and
176  * initialize a mount structure for it.
177  *
178  * Devname is usually updated by mount(8) after booting.
179  */
180 int
181 vfs_rootmountalloc(const char *fstypename, const char *devname,
182     struct mount **mpp)
183 {
184 	struct vfsops *vfsp = NULL;
185 	struct mount *mp;
186 	int error __diagused;
187 
188 	mutex_enter(&vfs_list_lock);
189 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
190 		if (!strncmp(vfsp->vfs_name, fstypename,
191 		    sizeof(mp->mnt_stat.f_fstypename)))
192 			break;
193 	if (vfsp == NULL) {
194 		mutex_exit(&vfs_list_lock);
195 		return (ENODEV);
196 	}
197 	vfsp->vfs_refcount++;
198 	mutex_exit(&vfs_list_lock);
199 
200 	if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL)
201 		return ENOMEM;
202 	error = vfs_busy(mp);
203 	KASSERT(error == 0);
204 	mp->mnt_flag = MNT_RDONLY;
205 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
206 	    sizeof(mp->mnt_stat.f_fstypename));
207 	mp->mnt_stat.f_mntonname[0] = '/';
208 	mp->mnt_stat.f_mntonname[1] = '\0';
209 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
210 	    '\0';
211 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
212 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
213 	*mpp = mp;
214 	return 0;
215 }
216 
217 /*
218  * vfs_getnewfsid: get a new unique fsid.
219  */
220 void
221 vfs_getnewfsid(struct mount *mp)
222 {
223 	static u_short xxxfs_mntid;
224 	fsid_t tfsid;
225 	int mtype;
226 
227 	mutex_enter(&mntid_lock);
228 	mtype = makefstype(mp->mnt_op->vfs_name);
229 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
230 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
231 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
232 	if (xxxfs_mntid == 0)
233 		++xxxfs_mntid;
234 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
235 	tfsid.__fsid_val[1] = mtype;
236 	while (vfs_getvfs(&tfsid)) {
237 		tfsid.__fsid_val[0]++;
238 		xxxfs_mntid++;
239 	}
240 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
241 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
242 	mutex_exit(&mntid_lock);
243 }
244 
245 /*
246  * Lookup a mount point by filesystem identifier.
247  *
248  * XXX Needs to add a reference to the mount point.
249  */
250 struct mount *
251 vfs_getvfs(fsid_t *fsid)
252 {
253 	mount_iterator_t *iter;
254 	struct mount *mp;
255 
256 	mountlist_iterator_init(&iter);
257 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
258 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
259 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
260 			mountlist_iterator_destroy(iter);
261 			return mp;
262 		}
263 	}
264 	mountlist_iterator_destroy(iter);
265 	return NULL;
266 }
267 
268 /*
269  * Take a reference to a mount structure.
270  */
271 void
272 vfs_ref(struct mount *mp)
273 {
274 
275 	KASSERT(mp->mnt_refcnt > 0 || mutex_owned(&mountlist_lock));
276 
277 	atomic_inc_uint(&mp->mnt_refcnt);
278 }
279 
280 /*
281  * Drop a reference to a mount structure, freeing if the last reference.
282  */
283 void
284 vfs_rele(struct mount *mp)
285 {
286 
287 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
288 		return;
289 	}
290 
291 	/*
292 	 * Nothing else has visibility of the mount: we can now
293 	 * free the data structures.
294 	 */
295 	KASSERT(mp->mnt_refcnt == 0);
296 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
297 	mutex_destroy(&mp->mnt_updating);
298 	mutex_destroy(&mp->mnt_renamelock);
299 	if (mp->mnt_op != NULL) {
300 		vfs_delref(mp->mnt_op);
301 	}
302 	kmem_free(mp, sizeof(*mp));
303 }
304 
305 /*
306  * Mark a mount point as busy, and gain a new reference to it.  Used to
307  * prevent the file system from being unmounted during critical sections.
308  *
309  * vfs_busy can be called multiple times and by multiple threads
310  * and must be accompanied by the same number of vfs_unbusy calls.
311  *
312  * => The caller must hold a pre-existing reference to the mount.
313  * => Will fail if the file system is being unmounted, or is unmounted.
314  */
315 static inline int
316 _vfs_busy(struct mount *mp, bool wait)
317 {
318 
319 	KASSERT(mp->mnt_refcnt > 0);
320 
321 	if (wait) {
322 		fstrans_start(mp);
323 	} else {
324 		if (fstrans_start_nowait(mp))
325 			return EBUSY;
326 	}
327 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
328 		fstrans_done(mp);
329 		return ENOENT;
330 	}
331 	vfs_ref(mp);
332 	return 0;
333 }
334 
335 int
336 vfs_busy(struct mount *mp)
337 {
338 
339 	return _vfs_busy(mp, true);
340 }
341 
342 int
343 vfs_trybusy(struct mount *mp)
344 {
345 
346 	return _vfs_busy(mp, false);
347 }
348 
349 /*
350  * Unbusy a busy filesystem.
351  *
352  * Every successful vfs_busy() call must be undone by a vfs_unbusy() call.
353  */
354 void
355 vfs_unbusy(struct mount *mp)
356 {
357 
358 	KASSERT(mp->mnt_refcnt > 0);
359 
360 	fstrans_done(mp);
361 	vfs_rele(mp);
362 }
363 
364 struct vnode_iterator {
365 	vnode_impl_t vi_vnode;
366 };
367 
368 void
369 vfs_vnode_iterator_init(struct mount *mp, struct vnode_iterator **vnip)
370 {
371 	vnode_t *vp;
372 	vnode_impl_t *vip;
373 
374 	vp = vnalloc_marker(mp);
375 	vip = VNODE_TO_VIMPL(vp);
376 
377 	mutex_enter(&mntvnode_lock);
378 	TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vip, vi_mntvnodes);
379 	vp->v_usecount = 1;
380 	mutex_exit(&mntvnode_lock);
381 
382 	*vnip = (struct vnode_iterator *)vip;
383 }
384 
385 void
386 vfs_vnode_iterator_destroy(struct vnode_iterator *vni)
387 {
388 	vnode_impl_t *mvip = &vni->vi_vnode;
389 	vnode_t *mvp = VIMPL_TO_VNODE(mvip);
390 
391 	mutex_enter(&mntvnode_lock);
392 	KASSERT(vnis_marker(mvp));
393 	if (mvp->v_usecount != 0) {
394 		TAILQ_REMOVE(&mvp->v_mount->mnt_vnodelist, mvip, vi_mntvnodes);
395 		mvp->v_usecount = 0;
396 	}
397 	mutex_exit(&mntvnode_lock);
398 	vnfree_marker(mvp);
399 }
400 
401 static struct vnode *
402 vfs_vnode_iterator_next1(struct vnode_iterator *vni,
403     bool (*f)(void *, struct vnode *), void *cl, bool do_wait)
404 {
405 	vnode_impl_t *mvip = &vni->vi_vnode;
406 	struct mount *mp = VIMPL_TO_VNODE(mvip)->v_mount;
407 	vnode_t *vp;
408 	vnode_impl_t *vip;
409 	int error;
410 
411 	KASSERT(vnis_marker(VIMPL_TO_VNODE(mvip)));
412 
413 	do {
414 		mutex_enter(&mntvnode_lock);
415 		vip = TAILQ_NEXT(mvip, vi_mntvnodes);
416 		TAILQ_REMOVE(&mp->mnt_vnodelist, mvip, vi_mntvnodes);
417 		VIMPL_TO_VNODE(mvip)->v_usecount = 0;
418 again:
419 		vp = VIMPL_TO_VNODE(vip);
420 		if (vp == NULL) {
421 	       		mutex_exit(&mntvnode_lock);
422 	       		return NULL;
423 		}
424 		mutex_enter(vp->v_interlock);
425 		if (vnis_marker(vp) ||
426 		    vdead_check(vp, (do_wait ? 0 : VDEAD_NOWAIT)) ||
427 		    (f && !(*f)(cl, vp))) {
428 			mutex_exit(vp->v_interlock);
429 			vip = TAILQ_NEXT(vip, vi_mntvnodes);
430 			goto again;
431 		}
432 
433 		TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vip, mvip, vi_mntvnodes);
434 		VIMPL_TO_VNODE(mvip)->v_usecount = 1;
435 		mutex_exit(&mntvnode_lock);
436 		error = vcache_vget(vp);
437 		KASSERT(error == 0 || error == ENOENT);
438 	} while (error != 0);
439 
440 	return vp;
441 }
442 
443 struct vnode *
444 vfs_vnode_iterator_next(struct vnode_iterator *vni,
445     bool (*f)(void *, struct vnode *), void *cl)
446 {
447 
448 	return vfs_vnode_iterator_next1(vni, f, cl, false);
449 }
450 
451 /*
452  * Move a vnode from one mount queue to another.
453  */
454 void
455 vfs_insmntque(vnode_t *vp, struct mount *mp)
456 {
457 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
458 	struct mount *omp;
459 
460 	KASSERT(mp == NULL || (mp->mnt_iflag & IMNT_UNMOUNT) == 0 ||
461 	    vp->v_tag == VT_VFS);
462 
463 	mutex_enter(&mntvnode_lock);
464 	/*
465 	 * Delete from old mount point vnode list, if on one.
466 	 */
467 	if ((omp = vp->v_mount) != NULL)
468 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vip, vi_mntvnodes);
469 	/*
470 	 * Insert into list of vnodes for the new mount point, if
471 	 * available.  The caller must take a reference on the mount
472 	 * structure and donate to the vnode.
473 	 */
474 	if ((vp->v_mount = mp) != NULL)
475 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vip, vi_mntvnodes);
476 	mutex_exit(&mntvnode_lock);
477 
478 	if (omp != NULL) {
479 		/* Release reference to old mount. */
480 		vfs_rele(omp);
481 	}
482 }
483 
484 /*
485  * Remove any vnodes in the vnode table belonging to mount point mp.
486  *
487  * If FORCECLOSE is not specified, there should not be any active ones,
488  * return error if any are found (nb: this is a user error, not a
489  * system error). If FORCECLOSE is specified, detach any active vnodes
490  * that are found.
491  *
492  * If WRITECLOSE is set, only flush out regular file vnodes open for
493  * writing.
494  *
495  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
496  */
497 #ifdef DEBUG
498 int busyprt = 0;	/* print out busy vnodes */
499 struct ctldebug debug1 = { "busyprt", &busyprt };
500 #endif
501 
502 static vnode_t *
503 vflushnext(struct vnode_iterator *marker, int *when)
504 {
505 	if (hardclock_ticks > *when) {
506 		yield();
507 		*when = hardclock_ticks + hz / 10;
508 	}
509 	return vfs_vnode_iterator_next1(marker, NULL, NULL, true);
510 }
511 
512 /*
513  * Flush one vnode.  Referenced on entry, unreferenced on return.
514  */
515 static int
516 vflush_one(vnode_t *vp, vnode_t *skipvp, int flags)
517 {
518 	int error;
519 	struct vattr vattr;
520 
521 	if (vp == skipvp ||
522 	    ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM))) {
523 		vrele(vp);
524 		return 0;
525 	}
526 	/*
527 	 * If WRITECLOSE is set, only flush out regular file
528 	 * vnodes open for writing or open and unlinked.
529 	 */
530 	if ((flags & WRITECLOSE)) {
531 		if (vp->v_type != VREG) {
532 			vrele(vp);
533 			return 0;
534 		}
535 		error = vn_lock(vp, LK_EXCLUSIVE);
536 		if (error) {
537 			KASSERT(error == ENOENT);
538 			vrele(vp);
539 			return 0;
540 		}
541 		error = VOP_FSYNC(vp, curlwp->l_cred, FSYNC_WAIT, 0, 0);
542 		if (error == 0)
543 			error = VOP_GETATTR(vp, &vattr, curlwp->l_cred);
544 		VOP_UNLOCK(vp);
545 		if (error) {
546 			vrele(vp);
547 			return error;
548 		}
549 		if (vp->v_writecount == 0 && vattr.va_nlink > 0) {
550 			vrele(vp);
551 			return 0;
552 		}
553 	}
554 	/*
555 	 * First try to recycle the vnode.
556 	 */
557 	if (vrecycle(vp))
558 		return 0;
559 	/*
560 	 * If FORCECLOSE is set, forcibly close the vnode.
561 	 * For block or character devices, revert to an
562 	 * anonymous device.  For all other files, just
563 	 * kill them.
564 	 */
565 	if (flags & FORCECLOSE) {
566 		if (vp->v_usecount > 1 &&
567 		    (vp->v_type == VBLK || vp->v_type == VCHR))
568 			vcache_make_anon(vp);
569 		else
570 			vgone(vp);
571 		return 0;
572 	}
573 	vrele(vp);
574 	return EBUSY;
575 }
576 
577 int
578 vflush(struct mount *mp, vnode_t *skipvp, int flags)
579 {
580 	vnode_t *vp;
581 	struct vnode_iterator *marker;
582 	int busy, error, when, retries = 2;
583 
584 	do {
585 		busy = error = when = 0;
586 
587 		/*
588 		 * First, flush out any vnode references from the
589 		 * deferred vrele list.
590 		 */
591 		vrele_flush(mp);
592 
593 		vfs_vnode_iterator_init(mp, &marker);
594 
595 		while ((vp = vflushnext(marker, &when)) != NULL) {
596 			error = vflush_one(vp, skipvp, flags);
597 			if (error == EBUSY) {
598 				error = 0;
599 				busy++;
600 #ifdef DEBUG
601 				if (busyprt && retries == 0)
602 					vprint("vflush: busy vnode", vp);
603 #endif
604 			} else if (error != 0) {
605 				break;
606 			}
607 		}
608 
609 		vfs_vnode_iterator_destroy(marker);
610 	} while (error == 0 && busy > 0 && retries-- > 0);
611 
612 	if (error)
613 		return error;
614 	if (busy)
615 		return EBUSY;
616 	return 0;
617 }
618 
619 /*
620  * Mount a file system.
621  */
622 
623 /*
624  * Scan all active processes to see if any of them have a current or root
625  * directory onto which the new filesystem has just been  mounted. If so,
626  * replace them with the new mount point.
627  */
628 static void
629 mount_checkdirs(vnode_t *olddp)
630 {
631 	vnode_t *newdp, *rele1, *rele2;
632 	struct cwdinfo *cwdi;
633 	struct proc *p;
634 	bool retry;
635 
636 	if (olddp->v_usecount == 1) {
637 		return;
638 	}
639 	if (VFS_ROOT(olddp->v_mountedhere, &newdp))
640 		panic("mount: lost mount");
641 
642 	do {
643 		retry = false;
644 		mutex_enter(proc_lock);
645 		PROCLIST_FOREACH(p, &allproc) {
646 			if ((cwdi = p->p_cwdi) == NULL)
647 				continue;
648 			/*
649 			 * Cannot change to the old directory any more,
650 			 * so even if we see a stale value it is not a
651 			 * problem.
652 			 */
653 			if (cwdi->cwdi_cdir != olddp &&
654 			    cwdi->cwdi_rdir != olddp)
655 				continue;
656 			retry = true;
657 			rele1 = NULL;
658 			rele2 = NULL;
659 			atomic_inc_uint(&cwdi->cwdi_refcnt);
660 			mutex_exit(proc_lock);
661 			rw_enter(&cwdi->cwdi_lock, RW_WRITER);
662 			if (cwdi->cwdi_cdir == olddp) {
663 				rele1 = cwdi->cwdi_cdir;
664 				vref(newdp);
665 				cwdi->cwdi_cdir = newdp;
666 			}
667 			if (cwdi->cwdi_rdir == olddp) {
668 				rele2 = cwdi->cwdi_rdir;
669 				vref(newdp);
670 				cwdi->cwdi_rdir = newdp;
671 			}
672 			rw_exit(&cwdi->cwdi_lock);
673 			cwdfree(cwdi);
674 			if (rele1 != NULL)
675 				vrele(rele1);
676 			if (rele2 != NULL)
677 				vrele(rele2);
678 			mutex_enter(proc_lock);
679 			break;
680 		}
681 		mutex_exit(proc_lock);
682 	} while (retry);
683 
684 	if (rootvnode == olddp) {
685 		vrele(rootvnode);
686 		vref(newdp);
687 		rootvnode = newdp;
688 	}
689 	vput(newdp);
690 }
691 
692 /*
693  * Start extended attributes
694  */
695 static int
696 start_extattr(struct mount *mp)
697 {
698 	int error;
699 
700 	error = VFS_EXTATTRCTL(mp, EXTATTR_CMD_START, NULL, 0, NULL);
701 	if (error)
702 		printf("%s: failed to start extattr: error = %d\n",
703 		       mp->mnt_stat.f_mntonname, error);
704 
705 	return error;
706 }
707 
708 int
709 mount_domount(struct lwp *l, vnode_t **vpp, struct vfsops *vfsops,
710     const char *path, int flags, void *data, size_t *data_len)
711 {
712 	vnode_t *vp = *vpp;
713 	struct mount *mp;
714 	struct pathbuf *pb;
715 	struct nameidata nd;
716 	int error;
717 
718 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
719 	    KAUTH_REQ_SYSTEM_MOUNT_NEW, vp, KAUTH_ARG(flags), data);
720 	if (error) {
721 		vfs_delref(vfsops);
722 		return error;
723 	}
724 
725 	/* Cannot make a non-dir a mount-point (from here anyway). */
726 	if (vp->v_type != VDIR) {
727 		vfs_delref(vfsops);
728 		return ENOTDIR;
729 	}
730 
731 	if (flags & MNT_EXPORTED) {
732 		vfs_delref(vfsops);
733 		return EINVAL;
734 	}
735 
736 	if ((mp = vfs_mountalloc(vfsops, vp)) == NULL) {
737 		vfs_delref(vfsops);
738 		return ENOMEM;
739 	}
740 
741 	mp->mnt_stat.f_owner = kauth_cred_geteuid(l->l_cred);
742 
743 	/*
744 	 * The underlying file system may refuse the mount for
745 	 * various reasons.  Allow the user to force it to happen.
746 	 *
747 	 * Set the mount level flags.
748 	 */
749 	mp->mnt_flag = flags & (MNT_BASIC_FLAGS | MNT_FORCE | MNT_IGNORE);
750 
751 	mutex_enter(&mp->mnt_updating);
752 	error = VFS_MOUNT(mp, path, data, data_len);
753 	mp->mnt_flag &= ~MNT_OP_FLAGS;
754 
755 	if (error != 0)
756 		goto err_unmounted;
757 
758 	/*
759 	 * Validate and prepare the mount point.
760 	 */
761 	error = pathbuf_copyin(path, &pb);
762 	if (error != 0) {
763 		goto err_mounted;
764 	}
765 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
766 	error = namei(&nd);
767 	pathbuf_destroy(pb);
768 	if (error != 0) {
769 		goto err_mounted;
770 	}
771 	if (nd.ni_vp != vp) {
772 		vput(nd.ni_vp);
773 		error = EINVAL;
774 		goto err_mounted;
775 	}
776 	if (vp->v_mountedhere != NULL) {
777 		vput(nd.ni_vp);
778 		error = EBUSY;
779 		goto err_mounted;
780 	}
781 	error = vinvalbuf(vp, V_SAVE, l->l_cred, l, 0, 0);
782 	if (error != 0) {
783 		vput(nd.ni_vp);
784 		goto err_mounted;
785 	}
786 
787 	/*
788 	 * Put the new filesystem on the mount list after root.
789 	 */
790 	cache_purge(vp);
791 	mp->mnt_iflag &= ~IMNT_WANTRDWR;
792 
793 	mountlist_append(mp);
794 	if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
795 		vfs_syncer_add_to_worklist(mp);
796 	vp->v_mountedhere = mp;
797 	vput(nd.ni_vp);
798 
799 	mount_checkdirs(vp);
800 	mutex_exit(&mp->mnt_updating);
801 
802 	/* Hold an additional reference to the mount across VFS_START(). */
803 	vfs_ref(mp);
804 	(void) VFS_STATVFS(mp, &mp->mnt_stat);
805 	error = VFS_START(mp, 0);
806        if (error) {
807 		vrele(vp);
808 	} else if (flags & MNT_EXTATTR) {
809 		(void)start_extattr(mp);
810 	}
811 	/* Drop reference held for VFS_START(). */
812 	vfs_rele(mp);
813 	*vpp = NULL;
814 	return error;
815 
816 err_mounted:
817 	if (VFS_UNMOUNT(mp, MNT_FORCE) != 0)
818 		panic("Unmounting fresh file system failed");
819 
820 err_unmounted:
821 	vp->v_mountedhere = NULL;
822 	mutex_exit(&mp->mnt_updating);
823 	fstrans_unmount(mp);
824 	vfs_rele(mp);
825 
826 	return error;
827 }
828 
829 /*
830  * Do the actual file system unmount.  File system is assumed to have
831  * been locked by the caller.
832  *
833  * => Caller hold reference to the mount, explicitly for dounmount().
834  */
835 int
836 dounmount(struct mount *mp, int flags, struct lwp *l)
837 {
838 	mount_iterator_t *iter;
839 	struct mount *cmp;
840 	vnode_t *coveredvp;
841 	int error, async, used_syncer, used_extattr;
842 
843 #if NVERIEXEC > 0
844 	error = veriexec_unmountchk(mp);
845 	if (error)
846 		return (error);
847 #endif /* NVERIEXEC > 0 */
848 
849 	/*
850 	 * No unmount below layered mounts.
851 	 */
852 	mountlist_iterator_init(&iter);
853 	while ((cmp = mountlist_iterator_next(iter)) != NULL) {
854 		if (cmp->mnt_lower == mp) {
855 			mountlist_iterator_destroy(iter);
856 			return EBUSY;
857 		}
858 	}
859 	mountlist_iterator_destroy(iter);
860 
861 	error = vfs_suspend(mp, 0);
862 	if (error) {
863 		return error;
864 	}
865 
866 	KASSERT((mp->mnt_iflag & IMNT_GONE) == 0);
867 
868 	used_syncer = (mp->mnt_iflag & IMNT_ONWORKLIST) != 0;
869 	used_extattr = mp->mnt_flag & MNT_EXTATTR;
870 
871 	mp->mnt_iflag |= IMNT_UNMOUNT;
872 	mutex_enter(&mp->mnt_updating);
873 	async = mp->mnt_flag & MNT_ASYNC;
874 	mp->mnt_flag &= ~MNT_ASYNC;
875 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
876 	if (used_syncer)
877 		vfs_syncer_remove_from_worklist(mp);
878 	error = 0;
879 	if (((mp->mnt_flag & MNT_RDONLY) == 0) && ((flags & MNT_FORCE) == 0)) {
880 		error = VFS_SYNC(mp, MNT_WAIT, l->l_cred);
881 	}
882 	if (error == 0 || (flags & MNT_FORCE)) {
883 		error = VFS_UNMOUNT(mp, flags);
884 	}
885 	if (error) {
886 		mp->mnt_iflag &= ~IMNT_UNMOUNT;
887 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
888 			vfs_syncer_add_to_worklist(mp);
889 		mp->mnt_flag |= async;
890 		mutex_exit(&mp->mnt_updating);
891 		vfs_resume(mp);
892 		if (used_extattr) {
893 			if (start_extattr(mp) != 0)
894 				mp->mnt_flag &= ~MNT_EXTATTR;
895 			else
896 				mp->mnt_flag |= MNT_EXTATTR;
897 		}
898 		return (error);
899 	}
900 	mutex_exit(&mp->mnt_updating);
901 
902 	/*
903 	 * mark filesystem as gone to prevent further umounts
904 	 * after mnt_umounting lock is gone, this also prevents
905 	 * vfs_busy() from succeeding.
906 	 */
907 	mp->mnt_iflag |= IMNT_GONE;
908 	vfs_resume(mp);
909 
910 	if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
911 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY);
912 		coveredvp->v_mountedhere = NULL;
913 		VOP_UNLOCK(coveredvp);
914 	}
915 	mountlist_remove(mp);
916 	if (TAILQ_FIRST(&mp->mnt_vnodelist) != NULL)
917 		panic("unmount: dangling vnode");
918 	vfs_hooks_unmount(mp);
919 
920 	fstrans_unmount(mp);
921 	vfs_rele(mp);	/* reference from mount() */
922 	if (coveredvp != NULLVP) {
923 		vrele(coveredvp);
924 	}
925 	return (0);
926 }
927 
928 /*
929  * Unmount all file systems.
930  * We traverse the list in reverse order under the assumption that doing so
931  * will avoid needing to worry about dependencies.
932  */
933 bool
934 vfs_unmountall(struct lwp *l)
935 {
936 
937 	printf("unmounting file systems...\n");
938 	return vfs_unmountall1(l, true, true);
939 }
940 
941 static void
942 vfs_unmount_print(struct mount *mp, const char *pfx)
943 {
944 
945 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
946 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
947 	    mp->mnt_stat.f_fstypename);
948 }
949 
950 /*
951  * Return the mount with the highest generation less than "gen".
952  */
953 static struct mount *
954 vfs_unmount_next(uint64_t gen)
955 {
956 	mount_iterator_t *iter;
957 	struct mount *mp, *nmp;
958 
959 	nmp = NULL;
960 
961 	mountlist_iterator_init(&iter);
962 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
963 		if ((nmp == NULL || mp->mnt_gen > nmp->mnt_gen) &&
964 		    mp->mnt_gen < gen) {
965 			if (nmp != NULL)
966 				vfs_rele(nmp);
967 			nmp = mp;
968 			vfs_ref(nmp);
969 		}
970 	}
971 	mountlist_iterator_destroy(iter);
972 
973 	return nmp;
974 }
975 
976 bool
977 vfs_unmount_forceone(struct lwp *l)
978 {
979 	struct mount *mp;
980 	int error;
981 
982 	mp = vfs_unmount_next(mountgen);
983 	if (mp == NULL) {
984 		return false;
985 	}
986 
987 #ifdef DEBUG
988 	printf("forcefully unmounting %s (%s)...\n",
989 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
990 #endif
991 	if ((error = dounmount(mp, MNT_FORCE, l)) == 0) {
992 		vfs_unmount_print(mp, "forcefully ");
993 		return true;
994 	} else {
995 		vfs_rele(mp);
996 	}
997 
998 #ifdef DEBUG
999 	printf("forceful unmount of %s failed with error %d\n",
1000 	    mp->mnt_stat.f_mntonname, error);
1001 #endif
1002 
1003 	return false;
1004 }
1005 
1006 bool
1007 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
1008 {
1009 	struct mount *mp;
1010 	bool any_error = false, progress = false;
1011 	uint64_t gen;
1012 	int error;
1013 
1014 	gen = mountgen;
1015 	for (;;) {
1016 		mp = vfs_unmount_next(gen);
1017 		if (mp == NULL)
1018 			break;
1019 		gen = mp->mnt_gen;
1020 
1021 #ifdef DEBUG
1022 		printf("unmounting %p %s (%s)...\n",
1023 		    (void *)mp, mp->mnt_stat.f_mntonname,
1024 		    mp->mnt_stat.f_mntfromname);
1025 #endif
1026 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
1027 			vfs_unmount_print(mp, "");
1028 			progress = true;
1029 		} else {
1030 			vfs_rele(mp);
1031 			if (verbose) {
1032 				printf("unmount of %s failed with error %d\n",
1033 				    mp->mnt_stat.f_mntonname, error);
1034 			}
1035 			any_error = true;
1036 		}
1037 	}
1038 	if (verbose) {
1039 		printf("unmounting done\n");
1040 	}
1041 	if (any_error && verbose) {
1042 		printf("WARNING: some file systems would not unmount\n");
1043 	}
1044 	return progress;
1045 }
1046 
1047 void
1048 vfs_sync_all(struct lwp *l)
1049 {
1050 	printf("syncing disks... ");
1051 
1052 	/* remove user processes from run queue */
1053 	suspendsched();
1054 	(void)spl0();
1055 
1056 	/* avoid coming back this way again if we panic. */
1057 	doing_shutdown = 1;
1058 
1059 	do_sys_sync(l);
1060 
1061 	/* Wait for sync to finish. */
1062 	if (buf_syncwait() != 0) {
1063 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
1064 		Debugger();
1065 #endif
1066 		printf("giving up\n");
1067 		return;
1068 	} else
1069 		printf("done\n");
1070 }
1071 
1072 /*
1073  * Sync and unmount file systems before shutting down.
1074  */
1075 void
1076 vfs_shutdown(void)
1077 {
1078 	lwp_t *l = curlwp;
1079 
1080 	vfs_sync_all(l);
1081 
1082 	/*
1083 	 * If we have paniced - do not make the situation potentially
1084 	 * worse by unmounting the file systems.
1085 	 */
1086 	if (panicstr != NULL) {
1087 		return;
1088 	}
1089 
1090 	/* Unmount file systems. */
1091 	vfs_unmountall(l);
1092 }
1093 
1094 /*
1095  * Print a list of supported file system types (used by vfs_mountroot)
1096  */
1097 static void
1098 vfs_print_fstypes(void)
1099 {
1100 	struct vfsops *v;
1101 	int cnt = 0;
1102 
1103 	mutex_enter(&vfs_list_lock);
1104 	LIST_FOREACH(v, &vfs_list, vfs_list)
1105 		++cnt;
1106 	mutex_exit(&vfs_list_lock);
1107 
1108 	if (cnt == 0) {
1109 		printf("WARNING: No file system modules have been loaded.\n");
1110 		return;
1111 	}
1112 
1113 	printf("Supported file systems:");
1114 	mutex_enter(&vfs_list_lock);
1115 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1116 		printf(" %s", v->vfs_name);
1117 	}
1118 	mutex_exit(&vfs_list_lock);
1119 	printf("\n");
1120 }
1121 
1122 /*
1123  * Mount the root file system.  If the operator didn't specify a
1124  * file system to use, try all possible file systems until one
1125  * succeeds.
1126  */
1127 int
1128 vfs_mountroot(void)
1129 {
1130 	struct vfsops *v;
1131 	int error = ENODEV;
1132 
1133 	if (root_device == NULL)
1134 		panic("vfs_mountroot: root device unknown");
1135 
1136 	switch (device_class(root_device)) {
1137 	case DV_IFNET:
1138 		if (rootdev != NODEV)
1139 			panic("vfs_mountroot: rootdev set for DV_IFNET "
1140 			    "(0x%llx -> %llu,%llu)",
1141 			    (unsigned long long)rootdev,
1142 			    (unsigned long long)major(rootdev),
1143 			    (unsigned long long)minor(rootdev));
1144 		break;
1145 
1146 	case DV_DISK:
1147 		if (rootdev == NODEV)
1148 			panic("vfs_mountroot: rootdev not set for DV_DISK");
1149 	        if (bdevvp(rootdev, &rootvp))
1150 	                panic("vfs_mountroot: can't get vnode for rootdev");
1151 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
1152 		if (error) {
1153 			printf("vfs_mountroot: can't open root device\n");
1154 			return (error);
1155 		}
1156 		break;
1157 
1158 	case DV_VIRTUAL:
1159 		break;
1160 
1161 	default:
1162 		printf("%s: inappropriate for root file system\n",
1163 		    device_xname(root_device));
1164 		return (ENODEV);
1165 	}
1166 
1167 	/*
1168 	 * If user specified a root fs type, use it.  Make sure the
1169 	 * specified type exists and has a mount_root()
1170 	 */
1171 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
1172 		v = vfs_getopsbyname(rootfstype);
1173 		error = EFTYPE;
1174 		if (v != NULL) {
1175 			if (v->vfs_mountroot != NULL) {
1176 				error = (v->vfs_mountroot)();
1177 			}
1178 			v->vfs_refcount--;
1179 		}
1180 		goto done;
1181 	}
1182 
1183 	/*
1184 	 * Try each file system currently configured into the kernel.
1185 	 */
1186 	mutex_enter(&vfs_list_lock);
1187 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1188 		if (v->vfs_mountroot == NULL)
1189 			continue;
1190 #ifdef DEBUG
1191 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
1192 #endif
1193 		v->vfs_refcount++;
1194 		mutex_exit(&vfs_list_lock);
1195 		error = (*v->vfs_mountroot)();
1196 		mutex_enter(&vfs_list_lock);
1197 		v->vfs_refcount--;
1198 		if (!error) {
1199 			aprint_normal("root file system type: %s\n",
1200 			    v->vfs_name);
1201 			break;
1202 		}
1203 	}
1204 	mutex_exit(&vfs_list_lock);
1205 
1206 	if (v == NULL) {
1207 		vfs_print_fstypes();
1208 		printf("no file system for %s", device_xname(root_device));
1209 		if (device_class(root_device) == DV_DISK)
1210 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
1211 		printf("\n");
1212 		error = EFTYPE;
1213 	}
1214 
1215 done:
1216 	if (error && device_class(root_device) == DV_DISK) {
1217 		VOP_CLOSE(rootvp, FREAD, FSCRED);
1218 		vrele(rootvp);
1219 	}
1220 	if (error == 0) {
1221 		mount_iterator_t *iter;
1222 		struct mount *mp;
1223 		extern struct cwdinfo cwdi0;
1224 
1225 		mountlist_iterator_init(&iter);
1226 		mp = mountlist_iterator_next(iter);
1227 		KASSERT(mp != NULL);
1228 		mountlist_iterator_destroy(iter);
1229 
1230 		mp->mnt_flag |= MNT_ROOTFS;
1231 		mp->mnt_op->vfs_refcount++;
1232 
1233 		/*
1234 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
1235 		 * reference it.
1236 		 */
1237 		error = VFS_ROOT(mp, &rootvnode);
1238 		if (error)
1239 			panic("cannot find root vnode, error=%d", error);
1240 		cwdi0.cwdi_cdir = rootvnode;
1241 		vref(cwdi0.cwdi_cdir);
1242 		VOP_UNLOCK(rootvnode);
1243 		cwdi0.cwdi_rdir = NULL;
1244 
1245 		/*
1246 		 * Now that root is mounted, we can fixup initproc's CWD
1247 		 * info.  All other processes are kthreads, which merely
1248 		 * share proc0's CWD info.
1249 		 */
1250 		initproc->p_cwdi->cwdi_cdir = rootvnode;
1251 		vref(initproc->p_cwdi->cwdi_cdir);
1252 		initproc->p_cwdi->cwdi_rdir = NULL;
1253 		/*
1254 		 * Enable loading of modules from the filesystem
1255 		 */
1256 		module_load_vfs_init();
1257 
1258 	}
1259 	return (error);
1260 }
1261 
1262 /*
1263  * mount_specific_key_create --
1264  *	Create a key for subsystem mount-specific data.
1265  */
1266 int
1267 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1268 {
1269 
1270 	return specificdata_key_create(mount_specificdata_domain, keyp, dtor);
1271 }
1272 
1273 /*
1274  * mount_specific_key_delete --
1275  *	Delete a key for subsystem mount-specific data.
1276  */
1277 void
1278 mount_specific_key_delete(specificdata_key_t key)
1279 {
1280 
1281 	specificdata_key_delete(mount_specificdata_domain, key);
1282 }
1283 
1284 /*
1285  * mount_initspecific --
1286  *	Initialize a mount's specificdata container.
1287  */
1288 void
1289 mount_initspecific(struct mount *mp)
1290 {
1291 	int error __diagused;
1292 
1293 	error = specificdata_init(mount_specificdata_domain,
1294 				  &mp->mnt_specdataref);
1295 	KASSERT(error == 0);
1296 }
1297 
1298 /*
1299  * mount_finispecific --
1300  *	Finalize a mount's specificdata container.
1301  */
1302 void
1303 mount_finispecific(struct mount *mp)
1304 {
1305 
1306 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
1307 }
1308 
1309 /*
1310  * mount_getspecific --
1311  *	Return mount-specific data corresponding to the specified key.
1312  */
1313 void *
1314 mount_getspecific(struct mount *mp, specificdata_key_t key)
1315 {
1316 
1317 	return specificdata_getspecific(mount_specificdata_domain,
1318 					 &mp->mnt_specdataref, key);
1319 }
1320 
1321 /*
1322  * mount_setspecific --
1323  *	Set mount-specific data corresponding to the specified key.
1324  */
1325 void
1326 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
1327 {
1328 
1329 	specificdata_setspecific(mount_specificdata_domain,
1330 				 &mp->mnt_specdataref, key, data);
1331 }
1332 
1333 /*
1334  * Check to see if a filesystem is mounted on a block device.
1335  */
1336 int
1337 vfs_mountedon(vnode_t *vp)
1338 {
1339 	vnode_t *vq;
1340 	int error = 0;
1341 
1342 	if (vp->v_type != VBLK)
1343 		return ENOTBLK;
1344 	if (spec_node_getmountedfs(vp) != NULL)
1345 		return EBUSY;
1346 	if (spec_node_lookup_by_dev(vp->v_type, vp->v_rdev, &vq) == 0) {
1347 		if (spec_node_getmountedfs(vq) != NULL)
1348 			error = EBUSY;
1349 		vrele(vq);
1350 	}
1351 
1352 	return error;
1353 }
1354 
1355 /*
1356  * Check if a device pointed to by vp is mounted.
1357  *
1358  * Returns:
1359  *   EINVAL	if it's not a disk
1360  *   EBUSY	if it's a disk and mounted
1361  *   0		if it's a disk and not mounted
1362  */
1363 int
1364 rawdev_mounted(vnode_t *vp, vnode_t **bvpp)
1365 {
1366 	vnode_t *bvp;
1367 	dev_t dev;
1368 	int d_type;
1369 
1370 	bvp = NULL;
1371 	d_type = D_OTHER;
1372 
1373 	if (iskmemvp(vp))
1374 		return EINVAL;
1375 
1376 	switch (vp->v_type) {
1377 	case VCHR: {
1378 		const struct cdevsw *cdev;
1379 
1380 		dev = vp->v_rdev;
1381 		cdev = cdevsw_lookup(dev);
1382 		if (cdev != NULL) {
1383 			dev_t blkdev;
1384 
1385 			blkdev = devsw_chr2blk(dev);
1386 			if (blkdev != NODEV) {
1387 				if (vfinddev(blkdev, VBLK, &bvp) != 0) {
1388 					d_type = (cdev->d_flag & D_TYPEMASK);
1389 					/* XXX: what if bvp disappears? */
1390 					vrele(bvp);
1391 				}
1392 			}
1393 		}
1394 
1395 		break;
1396 		}
1397 
1398 	case VBLK: {
1399 		const struct bdevsw *bdev;
1400 
1401 		dev = vp->v_rdev;
1402 		bdev = bdevsw_lookup(dev);
1403 		if (bdev != NULL)
1404 			d_type = (bdev->d_flag & D_TYPEMASK);
1405 
1406 		bvp = vp;
1407 
1408 		break;
1409 		}
1410 
1411 	default:
1412 		break;
1413 	}
1414 
1415 	if (d_type != D_DISK)
1416 		return EINVAL;
1417 
1418 	if (bvpp != NULL)
1419 		*bvpp = bvp;
1420 
1421 	/*
1422 	 * XXX: This is bogus. We should be failing the request
1423 	 * XXX: not only if this specific slice is mounted, but
1424 	 * XXX: if it's on a disk with any other mounted slice.
1425 	 */
1426 	if (vfs_mountedon(bvp))
1427 		return EBUSY;
1428 
1429 	return 0;
1430 }
1431 
1432 /*
1433  * Make a 'unique' number from a mount type name.
1434  */
1435 long
1436 makefstype(const char *type)
1437 {
1438 	long rv;
1439 
1440 	for (rv = 0; *type; type++) {
1441 		rv <<= 2;
1442 		rv ^= *type;
1443 	}
1444 	return rv;
1445 }
1446 
1447 static struct mountlist_entry *
1448 mountlist_alloc(enum mountlist_type type, struct mount *mp)
1449 {
1450 	struct mountlist_entry *me;
1451 
1452 	me = kmem_zalloc(sizeof(*me), KM_SLEEP);
1453 	me->me_mount = mp;
1454 	me->me_type = type;
1455 
1456 	return me;
1457 }
1458 
1459 static void
1460 mountlist_free(struct mountlist_entry *me)
1461 {
1462 
1463 	kmem_free(me, sizeof(*me));
1464 }
1465 
1466 void
1467 mountlist_iterator_init(mount_iterator_t **mip)
1468 {
1469 	struct mountlist_entry *me;
1470 
1471 	me = mountlist_alloc(ME_MARKER, NULL);
1472 	mutex_enter(&mountlist_lock);
1473 	TAILQ_INSERT_HEAD(&mountlist, me, me_list);
1474 	mutex_exit(&mountlist_lock);
1475 	*mip = (mount_iterator_t *)me;
1476 }
1477 
1478 void
1479 mountlist_iterator_destroy(mount_iterator_t *mi)
1480 {
1481 	struct mountlist_entry *marker = &mi->mi_entry;
1482 
1483 	if (marker->me_mount != NULL)
1484 		vfs_unbusy(marker->me_mount);
1485 
1486 	mutex_enter(&mountlist_lock);
1487 	TAILQ_REMOVE(&mountlist, marker, me_list);
1488 	mutex_exit(&mountlist_lock);
1489 
1490 	mountlist_free(marker);
1491 
1492 }
1493 
1494 /*
1495  * Return the next mount or NULL for this iterator.
1496  * Mark it busy on success.
1497  */
1498 static inline struct mount *
1499 _mountlist_iterator_next(mount_iterator_t *mi, bool wait)
1500 {
1501 	struct mountlist_entry *me, *marker = &mi->mi_entry;
1502 	struct mount *mp;
1503 	int error;
1504 
1505 	if (marker->me_mount != NULL) {
1506 		vfs_unbusy(marker->me_mount);
1507 		marker->me_mount = NULL;
1508 	}
1509 
1510 	mutex_enter(&mountlist_lock);
1511 	for (;;) {
1512 		KASSERT(marker->me_type == ME_MARKER);
1513 
1514 		me = TAILQ_NEXT(marker, me_list);
1515 		if (me == NULL) {
1516 			/* End of list: keep marker and return. */
1517 			mutex_exit(&mountlist_lock);
1518 			return NULL;
1519 		}
1520 		TAILQ_REMOVE(&mountlist, marker, me_list);
1521 		TAILQ_INSERT_AFTER(&mountlist, me, marker, me_list);
1522 
1523 		/* Skip other markers. */
1524 		if (me->me_type != ME_MOUNT)
1525 			continue;
1526 
1527 		/* Take an initial reference for vfs_busy() below. */
1528 		mp = me->me_mount;
1529 		KASSERT(mp != NULL);
1530 		vfs_ref(mp);
1531 		mutex_exit(&mountlist_lock);
1532 
1533 		/* Try to mark this mount busy and return on success. */
1534 		if (wait)
1535 			error = vfs_busy(mp);
1536 		else
1537 			error = vfs_trybusy(mp);
1538 		if (error == 0) {
1539 			vfs_rele(mp);
1540 			marker->me_mount = mp;
1541 			return mp;
1542 		}
1543 		vfs_rele(mp);
1544 		mutex_enter(&mountlist_lock);
1545 	}
1546 }
1547 
1548 struct mount *
1549 mountlist_iterator_next(mount_iterator_t *mi)
1550 {
1551 
1552 	return _mountlist_iterator_next(mi, true);
1553 }
1554 
1555 struct mount *
1556 mountlist_iterator_trynext(mount_iterator_t *mi)
1557 {
1558 
1559 	return _mountlist_iterator_next(mi, false);
1560 }
1561 
1562 /*
1563  * Attach new mount to the end of the mount list.
1564  */
1565 void
1566 mountlist_append(struct mount *mp)
1567 {
1568 	struct mountlist_entry *me;
1569 
1570 	me = mountlist_alloc(ME_MOUNT, mp);
1571 	mutex_enter(&mountlist_lock);
1572 	TAILQ_INSERT_TAIL(&mountlist, me, me_list);
1573 	mutex_exit(&mountlist_lock);
1574 }
1575 
1576 /*
1577  * Remove mount from mount list.
1578  */void
1579 mountlist_remove(struct mount *mp)
1580 {
1581 	struct mountlist_entry *me;
1582 
1583 	mutex_enter(&mountlist_lock);
1584 	TAILQ_FOREACH(me, &mountlist, me_list)
1585 		if (me->me_type == ME_MOUNT && me->me_mount == mp)
1586 			break;
1587 	KASSERT(me != NULL);
1588 	TAILQ_REMOVE(&mountlist, me, me_list);
1589 	mutex_exit(&mountlist_lock);
1590 	mountlist_free(me);
1591 }
1592 
1593 /*
1594  * Unlocked variant to traverse the mountlist.
1595  * To be used from DDB only.
1596  */
1597 struct mount *
1598 _mountlist_next(struct mount *mp)
1599 {
1600 	struct mountlist_entry *me;
1601 
1602 	if (mp == NULL) {
1603 		me = TAILQ_FIRST(&mountlist);
1604 	} else {
1605 		TAILQ_FOREACH(me, &mountlist, me_list)
1606 			if (me->me_type == ME_MOUNT && me->me_mount == mp)
1607 				break;
1608 		if (me != NULL)
1609 			me = TAILQ_NEXT(me, me_list);
1610 	}
1611 
1612 	while (me != NULL && me->me_type != ME_MOUNT)
1613 		me = TAILQ_NEXT(me, me_list);
1614 
1615 	return (me ? me->me_mount : NULL);
1616 }
1617