xref: /netbsd-src/sys/kern/vfs_mount.c (revision aef5eb5f59cdfe8314f1b5f78ac04eb144e44010)
1 /*	$NetBSD: vfs_mount.c,v 1.97 2022/09/13 09:35:31 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997-2020 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 1.97 2022/09/13 09:35:31 riastradh 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 #include <uvm/uvm_swap.h>
98 
99 enum mountlist_type {
100 	ME_MOUNT,
101 	ME_MARKER
102 };
103 struct mountlist_entry {
104 	TAILQ_ENTRY(mountlist_entry) me_list;	/* Mount list. */
105 	struct mount *me_mount;			/* Actual mount if ME_MOUNT,
106 						   current mount else. */
107 	enum mountlist_type me_type;		/* Mount or marker. */
108 };
109 struct mount_iterator {
110 	struct mountlist_entry mi_entry;
111 };
112 
113 static struct vnode *vfs_vnode_iterator_next1(struct vnode_iterator *,
114     bool (*)(void *, struct vnode *), void *, bool);
115 
116 /* Root filesystem. */
117 vnode_t *			rootvnode;
118 
119 /* Mounted filesystem list. */
120 static TAILQ_HEAD(mountlist, mountlist_entry) mountlist;
121 static kmutex_t			mountlist_lock __cacheline_aligned;
122 int vnode_offset_next_by_lru	/* XXX: ugly hack for pstat.c */
123     = offsetof(vnode_impl_t, vi_lrulist.tqe_next);
124 
125 kmutex_t			vfs_list_lock __cacheline_aligned;
126 
127 static specificdata_domain_t	mount_specificdata_domain;
128 static kmutex_t			mntid_lock;
129 
130 static kmutex_t			mountgen_lock __cacheline_aligned;
131 static uint64_t			mountgen;
132 
133 void
134 vfs_mount_sysinit(void)
135 {
136 
137 	TAILQ_INIT(&mountlist);
138 	mutex_init(&mountlist_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 
153 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
154 	mp->mnt_op = vfsops;
155 	mp->mnt_refcnt = 1;
156 	TAILQ_INIT(&mp->mnt_vnodelist);
157 	mp->mnt_renamelock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
158 	mp->mnt_vnodelock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
159 	mp->mnt_updating = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
160 	mp->mnt_vnodecovered = vp;
161 	mount_initspecific(mp);
162 
163 	error = fstrans_mount(mp);
164 	KASSERT(error == 0);
165 
166 	mutex_enter(&mountgen_lock);
167 	mp->mnt_gen = mountgen++;
168 	mutex_exit(&mountgen_lock);
169 
170 	return mp;
171 }
172 
173 /*
174  * vfs_rootmountalloc: lookup a filesystem type, and if found allocate and
175  * initialize a mount structure for it.
176  *
177  * Devname is usually updated by mount(8) after booting.
178  */
179 int
180 vfs_rootmountalloc(const char *fstypename, const char *devname,
181     struct mount **mpp)
182 {
183 	struct vfsops *vfsp = NULL;
184 	struct mount *mp;
185 	int error __diagused;
186 
187 	mutex_enter(&vfs_list_lock);
188 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
189 		if (!strncmp(vfsp->vfs_name, fstypename,
190 		    sizeof(mp->mnt_stat.f_fstypename)))
191 			break;
192 	if (vfsp == NULL) {
193 		mutex_exit(&vfs_list_lock);
194 		return (ENODEV);
195 	}
196 	vfsp->vfs_refcount++;
197 	mutex_exit(&vfs_list_lock);
198 
199 	if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL)
200 		return ENOMEM;
201 	error = vfs_busy(mp);
202 	KASSERT(error == 0);
203 	mp->mnt_flag = MNT_RDONLY;
204 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
205 	    sizeof(mp->mnt_stat.f_fstypename));
206 	mp->mnt_stat.f_mntonname[0] = '/';
207 	mp->mnt_stat.f_mntonname[1] = '\0';
208 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
209 	    '\0';
210 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
211 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
212 	*mpp = mp;
213 	return 0;
214 }
215 
216 /*
217  * vfs_getnewfsid: get a new unique fsid.
218  */
219 void
220 vfs_getnewfsid(struct mount *mp)
221 {
222 	static u_short xxxfs_mntid;
223 	struct mountlist_entry *me;
224 	fsid_t tfsid;
225 	int mtype;
226 
227 	mutex_enter(&mntid_lock);
228 	if (xxxfs_mntid == 0)
229 		++xxxfs_mntid;
230 	mtype = makefstype(mp->mnt_op->vfs_name);
231 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
232 	tfsid.__fsid_val[1] = mtype;
233 	/* Always increment to not return the same fsid to parallel mounts. */
234 	xxxfs_mntid++;
235 
236 	/*
237 	 * Directly walk mountlist to prevent deadlock through
238 	 * mountlist_iterator_next() -> vfs_busy().
239 	 */
240 	mutex_enter(&mountlist_lock);
241 	for (me = TAILQ_FIRST(&mountlist); me != TAILQ_END(&mountlist); ) {
242 		if (me->me_type == ME_MOUNT &&
243 		    me->me_mount->mnt_stat.f_fsidx.__fsid_val[0] ==
244 		    tfsid.__fsid_val[0] &&
245 		    me->me_mount->mnt_stat.f_fsidx.__fsid_val[1] ==
246 		    tfsid.__fsid_val[1]) {
247 			tfsid.__fsid_val[0]++;
248 			xxxfs_mntid++;
249 			me = TAILQ_FIRST(&mountlist);
250 		} else {
251 			me = TAILQ_NEXT(me, me_list);
252 		}
253 	}
254 	mutex_exit(&mountlist_lock);
255 
256 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
257 	mp->mnt_stat.f_fsidx.__fsid_val[1] = tfsid.__fsid_val[1];
258 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
259 	mutex_exit(&mntid_lock);
260 }
261 
262 /*
263  * Lookup a mount point by filesystem identifier.
264  *
265  * XXX Needs to add a reference to the mount point.
266  */
267 struct mount *
268 vfs_getvfs(fsid_t *fsid)
269 {
270 	mount_iterator_t *iter;
271 	struct mount *mp;
272 
273 	mountlist_iterator_init(&iter);
274 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
275 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
276 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
277 			mountlist_iterator_destroy(iter);
278 			return mp;
279 		}
280 	}
281 	mountlist_iterator_destroy(iter);
282 	return NULL;
283 }
284 
285 /*
286  * Take a reference to a mount structure.
287  */
288 void
289 vfs_ref(struct mount *mp)
290 {
291 
292 	KASSERT(mp->mnt_refcnt > 0 || mutex_owned(&mountlist_lock));
293 
294 	atomic_inc_uint(&mp->mnt_refcnt);
295 }
296 
297 /*
298  * Drop a reference to a mount structure, freeing if the last reference.
299  */
300 void
301 vfs_rele(struct mount *mp)
302 {
303 
304 #ifndef __HAVE_ATOMIC_AS_MEMBAR
305 	membar_release();
306 #endif
307 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
308 		return;
309 	}
310 #ifndef __HAVE_ATOMIC_AS_MEMBAR
311 	membar_acquire();
312 #endif
313 
314 	/*
315 	 * Nothing else has visibility of the mount: we can now
316 	 * free the data structures.
317 	 */
318 	KASSERT(mp->mnt_refcnt == 0);
319 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
320 	mutex_obj_free(mp->mnt_updating);
321 	mutex_obj_free(mp->mnt_renamelock);
322 	mutex_obj_free(mp->mnt_vnodelock);
323 	if (mp->mnt_op != NULL) {
324 		vfs_delref(mp->mnt_op);
325 	}
326 	fstrans_unmount(mp);
327 	/*
328 	 * Final free of mp gets done from fstrans_mount_dtor().
329 	 *
330 	 * Prevents this memory to be reused as a mount before
331 	 * fstrans releases all references to it.
332 	 */
333 }
334 
335 /*
336  * Mark a mount point as busy, and gain a new reference to it.  Used to
337  * prevent the file system from being unmounted during critical sections.
338  *
339  * vfs_busy can be called multiple times and by multiple threads
340  * and must be accompanied by the same number of vfs_unbusy calls.
341  *
342  * => The caller must hold a pre-existing reference to the mount.
343  * => Will fail if the file system is being unmounted, or is unmounted.
344  */
345 static inline int
346 _vfs_busy(struct mount *mp, bool wait)
347 {
348 
349 	KASSERT(mp->mnt_refcnt > 0);
350 
351 	if (wait) {
352 		fstrans_start(mp);
353 	} else {
354 		if (fstrans_start_nowait(mp))
355 			return EBUSY;
356 	}
357 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
358 		fstrans_done(mp);
359 		return ENOENT;
360 	}
361 	vfs_ref(mp);
362 	return 0;
363 }
364 
365 int
366 vfs_busy(struct mount *mp)
367 {
368 
369 	return _vfs_busy(mp, true);
370 }
371 
372 int
373 vfs_trybusy(struct mount *mp)
374 {
375 
376 	return _vfs_busy(mp, false);
377 }
378 
379 /*
380  * Unbusy a busy filesystem.
381  *
382  * Every successful vfs_busy() call must be undone by a vfs_unbusy() call.
383  */
384 void
385 vfs_unbusy(struct mount *mp)
386 {
387 
388 	KASSERT(mp->mnt_refcnt > 0);
389 
390 	fstrans_done(mp);
391 	vfs_rele(mp);
392 }
393 
394 struct vnode_iterator {
395 	vnode_impl_t vi_vnode;
396 };
397 
398 void
399 vfs_vnode_iterator_init(struct mount *mp, struct vnode_iterator **vnip)
400 {
401 	vnode_t *vp;
402 	vnode_impl_t *vip;
403 
404 	vp = vnalloc_marker(mp);
405 	vip = VNODE_TO_VIMPL(vp);
406 
407 	mutex_enter(mp->mnt_vnodelock);
408 	TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vip, vi_mntvnodes);
409 	vp->v_usecount = 1;
410 	mutex_exit(mp->mnt_vnodelock);
411 
412 	*vnip = (struct vnode_iterator *)vip;
413 }
414 
415 void
416 vfs_vnode_iterator_destroy(struct vnode_iterator *vni)
417 {
418 	vnode_impl_t *mvip = &vni->vi_vnode;
419 	vnode_t *mvp = VIMPL_TO_VNODE(mvip);
420 	kmutex_t *lock;
421 
422 	KASSERT(vnis_marker(mvp));
423 	if (vrefcnt(mvp) != 0) {
424 		lock = mvp->v_mount->mnt_vnodelock;
425 		mutex_enter(lock);
426 		TAILQ_REMOVE(&mvp->v_mount->mnt_vnodelist, mvip, vi_mntvnodes);
427 		mvp->v_usecount = 0;
428 		mutex_exit(lock);
429 	}
430 	vnfree_marker(mvp);
431 }
432 
433 static struct vnode *
434 vfs_vnode_iterator_next1(struct vnode_iterator *vni,
435     bool (*f)(void *, struct vnode *), void *cl, bool do_wait)
436 {
437 	vnode_impl_t *mvip = &vni->vi_vnode;
438 	struct mount *mp = VIMPL_TO_VNODE(mvip)->v_mount;
439 	vnode_t *vp;
440 	vnode_impl_t *vip;
441 	kmutex_t *lock;
442 	int error;
443 
444 	KASSERT(vnis_marker(VIMPL_TO_VNODE(mvip)));
445 
446 	lock = mp->mnt_vnodelock;
447 	do {
448 		mutex_enter(lock);
449 		vip = TAILQ_NEXT(mvip, vi_mntvnodes);
450 		TAILQ_REMOVE(&mp->mnt_vnodelist, mvip, vi_mntvnodes);
451 		VIMPL_TO_VNODE(mvip)->v_usecount = 0;
452 again:
453 		if (vip == NULL) {
454 			mutex_exit(lock);
455 	       		return NULL;
456 		}
457 		vp = VIMPL_TO_VNODE(vip);
458 		KASSERT(vp != NULL);
459 		mutex_enter(vp->v_interlock);
460 		if (vnis_marker(vp) ||
461 		    vdead_check(vp, (do_wait ? 0 : VDEAD_NOWAIT)) ||
462 		    (f && !(*f)(cl, vp))) {
463 			mutex_exit(vp->v_interlock);
464 			vip = TAILQ_NEXT(vip, vi_mntvnodes);
465 			goto again;
466 		}
467 
468 		TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vip, mvip, vi_mntvnodes);
469 		VIMPL_TO_VNODE(mvip)->v_usecount = 1;
470 		mutex_exit(lock);
471 		error = vcache_vget(vp);
472 		KASSERT(error == 0 || error == ENOENT);
473 	} while (error != 0);
474 
475 	return vp;
476 }
477 
478 struct vnode *
479 vfs_vnode_iterator_next(struct vnode_iterator *vni,
480     bool (*f)(void *, struct vnode *), void *cl)
481 {
482 
483 	return vfs_vnode_iterator_next1(vni, f, cl, false);
484 }
485 
486 /*
487  * Move a vnode from one mount queue to another.
488  */
489 void
490 vfs_insmntque(vnode_t *vp, struct mount *mp)
491 {
492 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
493 	struct mount *omp;
494 	kmutex_t *lock;
495 
496 	KASSERT(mp == NULL || (mp->mnt_iflag & IMNT_UNMOUNT) == 0 ||
497 	    vp->v_tag == VT_VFS);
498 
499 	/*
500 	 * Delete from old mount point vnode list, if on one.
501 	 */
502 	if ((omp = vp->v_mount) != NULL) {
503 		lock = omp->mnt_vnodelock;
504 		mutex_enter(lock);
505 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vip, vi_mntvnodes);
506 		mutex_exit(lock);
507 	}
508 
509 	/*
510 	 * Insert into list of vnodes for the new mount point, if
511 	 * available.  The caller must take a reference on the mount
512 	 * structure and donate to the vnode.
513 	 */
514 	if ((vp->v_mount = mp) != NULL) {
515 		lock = mp->mnt_vnodelock;
516 		mutex_enter(lock);
517 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vip, vi_mntvnodes);
518 		mutex_exit(lock);
519 	}
520 
521 	if (omp != NULL) {
522 		/* Release reference to old mount. */
523 		vfs_rele(omp);
524 	}
525 }
526 
527 /*
528  * Remove any vnodes in the vnode table belonging to mount point mp.
529  *
530  * If FORCECLOSE is not specified, there should not be any active ones,
531  * return error if any are found (nb: this is a user error, not a
532  * system error). If FORCECLOSE is specified, detach any active vnodes
533  * that are found.
534  *
535  * If WRITECLOSE is set, only flush out regular file vnodes open for
536  * writing.
537  *
538  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
539  */
540 #ifdef DEBUG
541 int busyprt = 0;	/* print out busy vnodes */
542 struct ctldebug debug1 = { "busyprt", &busyprt };
543 #endif
544 
545 static vnode_t *
546 vflushnext(struct vnode_iterator *marker, int *when)
547 {
548 	if (getticks() > *when) {
549 		yield();
550 		*when = getticks() + hz / 10;
551 	}
552 	preempt_point();
553 	return vfs_vnode_iterator_next1(marker, NULL, NULL, true);
554 }
555 
556 /*
557  * Flush one vnode.  Referenced on entry, unreferenced on return.
558  */
559 static int
560 vflush_one(vnode_t *vp, vnode_t *skipvp, int flags)
561 {
562 	int error;
563 	struct vattr vattr;
564 
565 	if (vp == skipvp ||
566 	    ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM))) {
567 		vrele(vp);
568 		return 0;
569 	}
570 	/*
571 	 * If WRITECLOSE is set, only flush out regular file
572 	 * vnodes open for writing or open and unlinked.
573 	 */
574 	if ((flags & WRITECLOSE)) {
575 		if (vp->v_type != VREG) {
576 			vrele(vp);
577 			return 0;
578 		}
579 		error = vn_lock(vp, LK_EXCLUSIVE);
580 		if (error) {
581 			KASSERT(error == ENOENT);
582 			vrele(vp);
583 			return 0;
584 		}
585 		error = VOP_FSYNC(vp, curlwp->l_cred, FSYNC_WAIT, 0, 0);
586 		if (error == 0)
587 			error = VOP_GETATTR(vp, &vattr, curlwp->l_cred);
588 		VOP_UNLOCK(vp);
589 		if (error) {
590 			vrele(vp);
591 			return error;
592 		}
593 		if (vp->v_writecount == 0 && vattr.va_nlink > 0) {
594 			vrele(vp);
595 			return 0;
596 		}
597 	}
598 	/*
599 	 * First try to recycle the vnode.
600 	 */
601 	if (vrecycle(vp))
602 		return 0;
603 	/*
604 	 * If FORCECLOSE is set, forcibly close the vnode.
605 	 * For block or character devices, revert to an
606 	 * anonymous device.  For all other files, just
607 	 * kill them.
608 	 */
609 	if (flags & FORCECLOSE) {
610 		if (vrefcnt(vp) > 1 &&
611 		    (vp->v_type == VBLK || vp->v_type == VCHR))
612 			vcache_make_anon(vp);
613 		else
614 			vgone(vp);
615 		return 0;
616 	}
617 	vrele(vp);
618 	return EBUSY;
619 }
620 
621 int
622 vflush(struct mount *mp, vnode_t *skipvp, int flags)
623 {
624 	vnode_t *vp;
625 	struct vnode_iterator *marker;
626 	int busy, error, when, retries = 2;
627 
628 	do {
629 		busy = error = when = 0;
630 
631 		/*
632 		 * First, flush out any vnode references from the
633 		 * deferred vrele list.
634 		 */
635 		vrele_flush(mp);
636 
637 		vfs_vnode_iterator_init(mp, &marker);
638 
639 		while ((vp = vflushnext(marker, &when)) != NULL) {
640 			error = vflush_one(vp, skipvp, flags);
641 			if (error == EBUSY) {
642 				error = 0;
643 				busy++;
644 #ifdef DEBUG
645 				if (busyprt && retries == 0)
646 					vprint("vflush: busy vnode", vp);
647 #endif
648 			} else if (error != 0) {
649 				break;
650 			}
651 		}
652 
653 		vfs_vnode_iterator_destroy(marker);
654 	} while (error == 0 && busy > 0 && retries-- > 0);
655 
656 	if (error)
657 		return error;
658 	if (busy)
659 		return EBUSY;
660 	return 0;
661 }
662 
663 /*
664  * Mount a file system.
665  */
666 
667 /*
668  * Scan all active processes to see if any of them have a current or root
669  * directory onto which the new filesystem has just been  mounted. If so,
670  * replace them with the new mount point.
671  */
672 static void
673 mount_checkdirs(vnode_t *olddp)
674 {
675 	vnode_t *newdp, *rele1, *rele2;
676 	struct cwdinfo *cwdi;
677 	struct proc *p;
678 	bool retry;
679 
680 	if (vrefcnt(olddp) == 1) {
681 		return;
682 	}
683 	if (VFS_ROOT(olddp->v_mountedhere, LK_EXCLUSIVE, &newdp))
684 		panic("mount: lost mount");
685 
686 	do {
687 		retry = false;
688 		mutex_enter(&proc_lock);
689 		PROCLIST_FOREACH(p, &allproc) {
690 			if ((cwdi = p->p_cwdi) == NULL)
691 				continue;
692 			/*
693 			 * Cannot change to the old directory any more,
694 			 * so even if we see a stale value it is not a
695 			 * problem.
696 			 */
697 			if (cwdi->cwdi_cdir != olddp &&
698 			    cwdi->cwdi_rdir != olddp)
699 				continue;
700 			retry = true;
701 			rele1 = NULL;
702 			rele2 = NULL;
703 			atomic_inc_uint(&cwdi->cwdi_refcnt);
704 			mutex_exit(&proc_lock);
705 			rw_enter(&cwdi->cwdi_lock, RW_WRITER);
706 			if (cwdi->cwdi_cdir == olddp) {
707 				rele1 = cwdi->cwdi_cdir;
708 				vref(newdp);
709 				cwdi->cwdi_cdir = newdp;
710 			}
711 			if (cwdi->cwdi_rdir == olddp) {
712 				rele2 = cwdi->cwdi_rdir;
713 				vref(newdp);
714 				cwdi->cwdi_rdir = newdp;
715 			}
716 			rw_exit(&cwdi->cwdi_lock);
717 			cwdfree(cwdi);
718 			if (rele1 != NULL)
719 				vrele(rele1);
720 			if (rele2 != NULL)
721 				vrele(rele2);
722 			mutex_enter(&proc_lock);
723 			break;
724 		}
725 		mutex_exit(&proc_lock);
726 	} while (retry);
727 
728 	if (rootvnode == olddp) {
729 		vrele(rootvnode);
730 		vref(newdp);
731 		rootvnode = newdp;
732 	}
733 	vput(newdp);
734 }
735 
736 /*
737  * Start extended attributes
738  */
739 static int
740 start_extattr(struct mount *mp)
741 {
742 	int error;
743 
744 	error = VFS_EXTATTRCTL(mp, EXTATTR_CMD_START, NULL, 0, NULL);
745 	if (error)
746 		printf("%s: failed to start extattr: error = %d\n",
747 		       mp->mnt_stat.f_mntonname, error);
748 
749 	return error;
750 }
751 
752 int
753 mount_domount(struct lwp *l, vnode_t **vpp, struct vfsops *vfsops,
754     const char *path, int flags, void *data, size_t *data_len)
755 {
756 	vnode_t *vp = *vpp;
757 	struct mount *mp;
758 	struct pathbuf *pb;
759 	struct nameidata nd;
760 	int error, error2;
761 
762 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
763 	    KAUTH_REQ_SYSTEM_MOUNT_NEW, vp, KAUTH_ARG(flags), data);
764 	if (error) {
765 		vfs_delref(vfsops);
766 		return error;
767 	}
768 
769 	/* Cannot make a non-dir a mount-point (from here anyway). */
770 	if (vp->v_type != VDIR) {
771 		vfs_delref(vfsops);
772 		return ENOTDIR;
773 	}
774 
775 	if (flags & MNT_EXPORTED) {
776 		vfs_delref(vfsops);
777 		return EINVAL;
778 	}
779 
780 	if ((mp = vfs_mountalloc(vfsops, vp)) == NULL) {
781 		vfs_delref(vfsops);
782 		return ENOMEM;
783 	}
784 
785 	mp->mnt_stat.f_owner = kauth_cred_geteuid(l->l_cred);
786 
787 	/*
788 	 * The underlying file system may refuse the mount for
789 	 * various reasons.  Allow the user to force it to happen.
790 	 *
791 	 * Set the mount level flags.
792 	 */
793 	mp->mnt_flag = flags & (MNT_BASIC_FLAGS | MNT_FORCE | MNT_IGNORE);
794 
795 	error = VFS_MOUNT(mp, path, data, data_len);
796 	mp->mnt_flag &= ~MNT_OP_FLAGS;
797 
798 	if (error != 0) {
799 		vfs_rele(mp);
800 		return error;
801 	}
802 
803 	/* Suspend new file system before taking mnt_updating. */
804 	do {
805 		error2 = vfs_suspend(mp, 0);
806 	} while (error2 == EINTR || error2 == ERESTART);
807 	KASSERT(error2 == 0 || error2 == EOPNOTSUPP);
808 	mutex_enter(mp->mnt_updating);
809 
810 	/*
811 	 * Validate and prepare the mount point.
812 	 */
813 	error = pathbuf_copyin(path, &pb);
814 	if (error != 0) {
815 		goto err_mounted;
816 	}
817 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
818 	error = namei(&nd);
819 	pathbuf_destroy(pb);
820 	if (error != 0) {
821 		goto err_mounted;
822 	}
823 	if (nd.ni_vp != vp) {
824 		vput(nd.ni_vp);
825 		error = EINVAL;
826 		goto err_mounted;
827 	}
828 	if (vp->v_mountedhere != NULL) {
829 		vput(nd.ni_vp);
830 		error = EBUSY;
831 		goto err_mounted;
832 	}
833 	error = vinvalbuf(vp, V_SAVE, l->l_cred, l, 0, 0);
834 	if (error != 0) {
835 		vput(nd.ni_vp);
836 		goto err_mounted;
837 	}
838 
839 	/*
840 	 * Put the new filesystem on the mount list after root.
841 	 */
842 	cache_purge(vp);
843 	mp->mnt_iflag &= ~IMNT_WANTRDWR;
844 
845 	mountlist_append(mp);
846 	if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
847 		vfs_syncer_add_to_worklist(mp);
848 	vp->v_mountedhere = mp;
849 	vput(nd.ni_vp);
850 
851 	mount_checkdirs(vp);
852 	mutex_exit(mp->mnt_updating);
853 	if (error2 == 0)
854 		vfs_resume(mp);
855 
856 	/* Hold an additional reference to the mount across VFS_START(). */
857 	vfs_ref(mp);
858 	(void) VFS_STATVFS(mp, &mp->mnt_stat);
859 	error = VFS_START(mp, 0);
860 	if (error) {
861 		vrele(vp);
862 	} else if (flags & MNT_EXTATTR) {
863 		if (start_extattr(mp) != 0)
864 			mp->mnt_flag &= ~MNT_EXTATTR;
865 	}
866 	/* Drop reference held for VFS_START(). */
867 	vfs_rele(mp);
868 	*vpp = NULL;
869 	return error;
870 
871 err_mounted:
872 	if (VFS_UNMOUNT(mp, MNT_FORCE) != 0)
873 		panic("Unmounting fresh file system failed");
874 	mutex_exit(mp->mnt_updating);
875 	if (error2 == 0)
876 		vfs_resume(mp);
877 	vfs_rele(mp);
878 
879 	return error;
880 }
881 
882 /*
883  * Do the actual file system unmount.  File system is assumed to have
884  * been locked by the caller.
885  *
886  * => Caller hold reference to the mount, explicitly for dounmount().
887  */
888 int
889 dounmount(struct mount *mp, int flags, struct lwp *l)
890 {
891 	vnode_t *coveredvp;
892 	int error, async, used_syncer, used_extattr;
893 	const bool was_suspended = fstrans_is_owner(mp);
894 
895 #if NVERIEXEC > 0
896 	error = veriexec_unmountchk(mp);
897 	if (error)
898 		return (error);
899 #endif /* NVERIEXEC > 0 */
900 
901 	if (!was_suspended) {
902 		error = vfs_suspend(mp, 0);
903 		if (error) {
904 			return error;
905 		}
906 	}
907 
908 	KASSERT((mp->mnt_iflag & IMNT_GONE) == 0);
909 
910 	used_syncer = (mp->mnt_iflag & IMNT_ONWORKLIST) != 0;
911 	used_extattr = mp->mnt_flag & MNT_EXTATTR;
912 
913 	mp->mnt_iflag |= IMNT_UNMOUNT;
914 	mutex_enter(mp->mnt_updating);
915 	async = mp->mnt_flag & MNT_ASYNC;
916 	mp->mnt_flag &= ~MNT_ASYNC;
917 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
918 	if (used_syncer)
919 		vfs_syncer_remove_from_worklist(mp);
920 	error = 0;
921 	if (((mp->mnt_flag & MNT_RDONLY) == 0) && ((flags & MNT_FORCE) == 0)) {
922 		error = VFS_SYNC(mp, MNT_WAIT, l->l_cred);
923 	}
924 	if (error == 0 || (flags & MNT_FORCE)) {
925 		error = VFS_UNMOUNT(mp, flags);
926 	}
927 	if (error) {
928 		mp->mnt_iflag &= ~IMNT_UNMOUNT;
929 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
930 			vfs_syncer_add_to_worklist(mp);
931 		mp->mnt_flag |= async;
932 		mutex_exit(mp->mnt_updating);
933 		if (!was_suspended)
934 			vfs_resume(mp);
935 		if (used_extattr) {
936 			if (start_extattr(mp) != 0)
937 				mp->mnt_flag &= ~MNT_EXTATTR;
938 			else
939 				mp->mnt_flag |= MNT_EXTATTR;
940 		}
941 		return (error);
942 	}
943 	mutex_exit(mp->mnt_updating);
944 
945 	/*
946 	 * mark filesystem as gone to prevent further umounts
947 	 * after mnt_umounting lock is gone, this also prevents
948 	 * vfs_busy() from succeeding.
949 	 */
950 	mp->mnt_iflag |= IMNT_GONE;
951 	if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
952 		coveredvp->v_mountedhere = NULL;
953 	}
954 	if (!was_suspended)
955 		vfs_resume(mp);
956 
957 	mountlist_remove(mp);
958 	if (TAILQ_FIRST(&mp->mnt_vnodelist) != NULL)
959 		panic("unmount: dangling vnode");
960 	vfs_hooks_unmount(mp);
961 
962 	vfs_rele(mp);	/* reference from mount() */
963 	if (coveredvp != NULLVP) {
964 		vrele(coveredvp);
965 	}
966 	return (0);
967 }
968 
969 /*
970  * Unmount all file systems.
971  * We traverse the list in reverse order under the assumption that doing so
972  * will avoid needing to worry about dependencies.
973  */
974 bool
975 vfs_unmountall(struct lwp *l)
976 {
977 
978 	printf("unmounting file systems...\n");
979 	return vfs_unmountall1(l, true, true);
980 }
981 
982 static void
983 vfs_unmount_print(struct mount *mp, const char *pfx)
984 {
985 
986 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
987 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
988 	    mp->mnt_stat.f_fstypename);
989 }
990 
991 /*
992  * Return the mount with the highest generation less than "gen".
993  */
994 static struct mount *
995 vfs_unmount_next(uint64_t gen)
996 {
997 	mount_iterator_t *iter;
998 	struct mount *mp, *nmp;
999 
1000 	nmp = NULL;
1001 
1002 	mountlist_iterator_init(&iter);
1003 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
1004 		if ((nmp == NULL || mp->mnt_gen > nmp->mnt_gen) &&
1005 		    mp->mnt_gen < gen) {
1006 			if (nmp != NULL)
1007 				vfs_rele(nmp);
1008 			nmp = mp;
1009 			vfs_ref(nmp);
1010 		}
1011 	}
1012 	mountlist_iterator_destroy(iter);
1013 
1014 	return nmp;
1015 }
1016 
1017 bool
1018 vfs_unmount_forceone(struct lwp *l)
1019 {
1020 	struct mount *mp;
1021 	int error;
1022 
1023 	mp = vfs_unmount_next(mountgen);
1024 	if (mp == NULL) {
1025 		return false;
1026 	}
1027 
1028 #ifdef DEBUG
1029 	printf("forcefully unmounting %s (%s)...\n",
1030 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
1031 #endif
1032 	if ((error = dounmount(mp, MNT_FORCE, l)) == 0) {
1033 		vfs_unmount_print(mp, "forcefully ");
1034 		return true;
1035 	} else {
1036 		vfs_rele(mp);
1037 	}
1038 
1039 #ifdef DEBUG
1040 	printf("forceful unmount of %s failed with error %d\n",
1041 	    mp->mnt_stat.f_mntonname, error);
1042 #endif
1043 
1044 	return false;
1045 }
1046 
1047 bool
1048 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
1049 {
1050 	struct mount *mp;
1051 	mount_iterator_t *iter;
1052 	bool any_error = false, progress = false;
1053 	uint64_t gen;
1054 	int error;
1055 
1056 	gen = mountgen;
1057 	for (;;) {
1058 		mp = vfs_unmount_next(gen);
1059 		if (mp == NULL)
1060 			break;
1061 		gen = mp->mnt_gen;
1062 
1063 #ifdef DEBUG
1064 		printf("unmounting %p %s (%s)...\n",
1065 		    (void *)mp, mp->mnt_stat.f_mntonname,
1066 		    mp->mnt_stat.f_mntfromname);
1067 #endif
1068 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
1069 			vfs_unmount_print(mp, "");
1070 			progress = true;
1071 		} else {
1072 			vfs_rele(mp);
1073 			if (verbose) {
1074 				printf("unmount of %s failed with error %d\n",
1075 				    mp->mnt_stat.f_mntonname, error);
1076 			}
1077 			any_error = true;
1078 		}
1079 	}
1080 	if (verbose) {
1081 		printf("unmounting done\n");
1082 	}
1083 	if (any_error && verbose) {
1084 		printf("WARNING: some file systems would not unmount\n");
1085 	}
1086 	/* If the mountlist is empty it is time to remove swap. */
1087 	mountlist_iterator_init(&iter);
1088 	if (mountlist_iterator_next(iter) == NULL) {
1089 		uvm_swap_shutdown(l);
1090 	}
1091 	mountlist_iterator_destroy(iter);
1092 
1093 	return progress;
1094 }
1095 
1096 void
1097 vfs_sync_all(struct lwp *l)
1098 {
1099 	printf("syncing disks... ");
1100 
1101 	/* remove user processes from run queue */
1102 	suspendsched();
1103 	(void)spl0();
1104 
1105 	/* avoid coming back this way again if we panic. */
1106 	doing_shutdown = 1;
1107 
1108 	do_sys_sync(l);
1109 
1110 	/* Wait for sync to finish. */
1111 	if (vfs_syncwait() != 0) {
1112 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
1113 		Debugger();
1114 #endif
1115 		printf("giving up\n");
1116 		return;
1117 	} else
1118 		printf("done\n");
1119 }
1120 
1121 /*
1122  * Sync and unmount file systems before shutting down.
1123  */
1124 void
1125 vfs_shutdown(void)
1126 {
1127 	lwp_t *l = curlwp;
1128 
1129 	vfs_sync_all(l);
1130 
1131 	/*
1132 	 * If we have panicked - do not make the situation potentially
1133 	 * worse by unmounting the file systems.
1134 	 */
1135 	if (panicstr != NULL) {
1136 		return;
1137 	}
1138 
1139 	/* Unmount file systems. */
1140 	vfs_unmountall(l);
1141 }
1142 
1143 /*
1144  * Print a list of supported file system types (used by vfs_mountroot)
1145  */
1146 static void
1147 vfs_print_fstypes(void)
1148 {
1149 	struct vfsops *v;
1150 	int cnt = 0;
1151 
1152 	mutex_enter(&vfs_list_lock);
1153 	LIST_FOREACH(v, &vfs_list, vfs_list)
1154 		++cnt;
1155 	mutex_exit(&vfs_list_lock);
1156 
1157 	if (cnt == 0) {
1158 		printf("WARNING: No file system modules have been loaded.\n");
1159 		return;
1160 	}
1161 
1162 	printf("Supported file systems:");
1163 	mutex_enter(&vfs_list_lock);
1164 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1165 		printf(" %s", v->vfs_name);
1166 	}
1167 	mutex_exit(&vfs_list_lock);
1168 	printf("\n");
1169 }
1170 
1171 /*
1172  * Mount the root file system.  If the operator didn't specify a
1173  * file system to use, try all possible file systems until one
1174  * succeeds.
1175  */
1176 int
1177 vfs_mountroot(void)
1178 {
1179 	struct vfsops *v;
1180 	int error = ENODEV;
1181 
1182 	if (root_device == NULL)
1183 		panic("vfs_mountroot: root device unknown");
1184 
1185 	switch (device_class(root_device)) {
1186 	case DV_IFNET:
1187 		if (rootdev != NODEV)
1188 			panic("vfs_mountroot: rootdev set for DV_IFNET "
1189 			    "(0x%llx -> %llu,%llu)",
1190 			    (unsigned long long)rootdev,
1191 			    (unsigned long long)major(rootdev),
1192 			    (unsigned long long)minor(rootdev));
1193 		break;
1194 
1195 	case DV_DISK:
1196 		if (rootdev == NODEV)
1197 			panic("vfs_mountroot: rootdev not set for DV_DISK");
1198 	        if (bdevvp(rootdev, &rootvp))
1199 	                panic("vfs_mountroot: can't get vnode for rootdev");
1200 		vn_lock(rootvp, LK_EXCLUSIVE | LK_RETRY);
1201 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
1202 		VOP_UNLOCK(rootvp);
1203 		if (error) {
1204 			printf("vfs_mountroot: can't open root device\n");
1205 			return (error);
1206 		}
1207 		break;
1208 
1209 	case DV_VIRTUAL:
1210 		break;
1211 
1212 	default:
1213 		printf("%s: inappropriate for root file system\n",
1214 		    device_xname(root_device));
1215 		return (ENODEV);
1216 	}
1217 
1218 	/*
1219 	 * If user specified a root fs type, use it.  Make sure the
1220 	 * specified type exists and has a mount_root()
1221 	 */
1222 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
1223 		v = vfs_getopsbyname(rootfstype);
1224 		error = EFTYPE;
1225 		if (v != NULL) {
1226 			if (v->vfs_mountroot != NULL) {
1227 				error = (v->vfs_mountroot)();
1228 			}
1229 			v->vfs_refcount--;
1230 		}
1231 		goto done;
1232 	}
1233 
1234 	/*
1235 	 * Try each file system currently configured into the kernel.
1236 	 */
1237 	mutex_enter(&vfs_list_lock);
1238 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1239 		if (v->vfs_mountroot == NULL)
1240 			continue;
1241 #ifdef DEBUG
1242 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
1243 #endif
1244 		v->vfs_refcount++;
1245 		mutex_exit(&vfs_list_lock);
1246 		error = (*v->vfs_mountroot)();
1247 		mutex_enter(&vfs_list_lock);
1248 		v->vfs_refcount--;
1249 		if (!error) {
1250 			aprint_normal("root file system type: %s\n",
1251 			    v->vfs_name);
1252 			break;
1253 		}
1254 	}
1255 	mutex_exit(&vfs_list_lock);
1256 
1257 	if (v == NULL) {
1258 		vfs_print_fstypes();
1259 		printf("no file system for %s", device_xname(root_device));
1260 		if (device_class(root_device) == DV_DISK)
1261 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
1262 		printf("\n");
1263 		error = EFTYPE;
1264 	}
1265 
1266 done:
1267 	if (error && device_class(root_device) == DV_DISK) {
1268 		vn_lock(rootvp, LK_EXCLUSIVE | LK_RETRY);
1269 		VOP_CLOSE(rootvp, FREAD, FSCRED);
1270 		VOP_UNLOCK(rootvp);
1271 		vrele(rootvp);
1272 	}
1273 	if (error == 0) {
1274 		mount_iterator_t *iter;
1275 		struct mount *mp;
1276 		extern struct cwdinfo cwdi0;
1277 
1278 		mountlist_iterator_init(&iter);
1279 		mp = mountlist_iterator_next(iter);
1280 		KASSERT(mp != NULL);
1281 		mountlist_iterator_destroy(iter);
1282 
1283 		mp->mnt_flag |= MNT_ROOTFS;
1284 		mp->mnt_op->vfs_refcount++;
1285 
1286 		/*
1287 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
1288 		 * reference it, and donate it the reference grabbed
1289 		 * with VFS_ROOT().
1290 		 */
1291 		error = VFS_ROOT(mp, LK_NONE, &rootvnode);
1292 		if (error)
1293 			panic("cannot find root vnode, error=%d", error);
1294 		cwdi0.cwdi_cdir = rootvnode;
1295 		cwdi0.cwdi_rdir = NULL;
1296 
1297 		/*
1298 		 * Now that root is mounted, we can fixup initproc's CWD
1299 		 * info.  All other processes are kthreads, which merely
1300 		 * share proc0's CWD info.
1301 		 */
1302 		initproc->p_cwdi->cwdi_cdir = rootvnode;
1303 		vref(initproc->p_cwdi->cwdi_cdir);
1304 		initproc->p_cwdi->cwdi_rdir = NULL;
1305 		/*
1306 		 * Enable loading of modules from the filesystem
1307 		 */
1308 		module_load_vfs_init();
1309 
1310 	}
1311 	return (error);
1312 }
1313 
1314 /*
1315  * mount_specific_key_create --
1316  *	Create a key for subsystem mount-specific data.
1317  */
1318 int
1319 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1320 {
1321 
1322 	return specificdata_key_create(mount_specificdata_domain, keyp, dtor);
1323 }
1324 
1325 /*
1326  * mount_specific_key_delete --
1327  *	Delete a key for subsystem mount-specific data.
1328  */
1329 void
1330 mount_specific_key_delete(specificdata_key_t key)
1331 {
1332 
1333 	specificdata_key_delete(mount_specificdata_domain, key);
1334 }
1335 
1336 /*
1337  * mount_initspecific --
1338  *	Initialize a mount's specificdata container.
1339  */
1340 void
1341 mount_initspecific(struct mount *mp)
1342 {
1343 	int error __diagused;
1344 
1345 	error = specificdata_init(mount_specificdata_domain,
1346 				  &mp->mnt_specdataref);
1347 	KASSERT(error == 0);
1348 }
1349 
1350 /*
1351  * mount_finispecific --
1352  *	Finalize a mount's specificdata container.
1353  */
1354 void
1355 mount_finispecific(struct mount *mp)
1356 {
1357 
1358 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
1359 }
1360 
1361 /*
1362  * mount_getspecific --
1363  *	Return mount-specific data corresponding to the specified key.
1364  */
1365 void *
1366 mount_getspecific(struct mount *mp, specificdata_key_t key)
1367 {
1368 
1369 	return specificdata_getspecific(mount_specificdata_domain,
1370 					 &mp->mnt_specdataref, key);
1371 }
1372 
1373 /*
1374  * mount_setspecific --
1375  *	Set mount-specific data corresponding to the specified key.
1376  */
1377 void
1378 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
1379 {
1380 
1381 	specificdata_setspecific(mount_specificdata_domain,
1382 				 &mp->mnt_specdataref, key, data);
1383 }
1384 
1385 /*
1386  * Check to see if a filesystem is mounted on a block device.
1387  */
1388 int
1389 vfs_mountedon(vnode_t *vp)
1390 {
1391 	vnode_t *vq;
1392 	int error = 0;
1393 
1394 	if (vp->v_type != VBLK)
1395 		return ENOTBLK;
1396 	if (spec_node_getmountedfs(vp) != NULL)
1397 		return EBUSY;
1398 	if (spec_node_lookup_by_dev(vp->v_type, vp->v_rdev, VDEAD_NOWAIT, &vq)
1399 	    == 0) {
1400 		if (spec_node_getmountedfs(vq) != NULL)
1401 			error = EBUSY;
1402 		vrele(vq);
1403 	}
1404 
1405 	return error;
1406 }
1407 
1408 /*
1409  * Check if a device pointed to by vp is mounted.
1410  *
1411  * Returns:
1412  *   EINVAL	if it's not a disk
1413  *   EBUSY	if it's a disk and mounted
1414  *   0		if it's a disk and not mounted
1415  */
1416 int
1417 rawdev_mounted(vnode_t *vp, vnode_t **bvpp)
1418 {
1419 	vnode_t *bvp;
1420 	dev_t dev;
1421 	int d_type;
1422 
1423 	bvp = NULL;
1424 	d_type = D_OTHER;
1425 
1426 	if (iskmemvp(vp))
1427 		return EINVAL;
1428 
1429 	switch (vp->v_type) {
1430 	case VCHR: {
1431 		const struct cdevsw *cdev;
1432 
1433 		dev = vp->v_rdev;
1434 		cdev = cdevsw_lookup(dev);
1435 		if (cdev != NULL) {
1436 			dev_t blkdev;
1437 
1438 			blkdev = devsw_chr2blk(dev);
1439 			if (blkdev != NODEV) {
1440 				if (vfinddev(blkdev, VBLK, &bvp) != 0) {
1441 					d_type = (cdev->d_flag & D_TYPEMASK);
1442 					/* XXX: what if bvp disappears? */
1443 					vrele(bvp);
1444 				}
1445 			}
1446 		}
1447 
1448 		break;
1449 		}
1450 
1451 	case VBLK: {
1452 		const struct bdevsw *bdev;
1453 
1454 		dev = vp->v_rdev;
1455 		bdev = bdevsw_lookup(dev);
1456 		if (bdev != NULL)
1457 			d_type = (bdev->d_flag & D_TYPEMASK);
1458 
1459 		bvp = vp;
1460 
1461 		break;
1462 		}
1463 
1464 	default:
1465 		break;
1466 	}
1467 
1468 	if (d_type != D_DISK)
1469 		return EINVAL;
1470 
1471 	if (bvpp != NULL)
1472 		*bvpp = bvp;
1473 
1474 	/*
1475 	 * XXX: This is bogus. We should be failing the request
1476 	 * XXX: not only if this specific slice is mounted, but
1477 	 * XXX: if it's on a disk with any other mounted slice.
1478 	 */
1479 	if (vfs_mountedon(bvp))
1480 		return EBUSY;
1481 
1482 	return 0;
1483 }
1484 
1485 /*
1486  * Make a 'unique' number from a mount type name.
1487  */
1488 long
1489 makefstype(const char *type)
1490 {
1491 	long rv;
1492 
1493 	for (rv = 0; *type; type++) {
1494 		rv <<= 2;
1495 		rv ^= *type;
1496 	}
1497 	return rv;
1498 }
1499 
1500 static struct mountlist_entry *
1501 mountlist_alloc(enum mountlist_type type, struct mount *mp)
1502 {
1503 	struct mountlist_entry *me;
1504 
1505 	me = kmem_zalloc(sizeof(*me), KM_SLEEP);
1506 	me->me_mount = mp;
1507 	me->me_type = type;
1508 
1509 	return me;
1510 }
1511 
1512 static void
1513 mountlist_free(struct mountlist_entry *me)
1514 {
1515 
1516 	kmem_free(me, sizeof(*me));
1517 }
1518 
1519 void
1520 mountlist_iterator_init(mount_iterator_t **mip)
1521 {
1522 	struct mountlist_entry *me;
1523 
1524 	me = mountlist_alloc(ME_MARKER, NULL);
1525 	mutex_enter(&mountlist_lock);
1526 	TAILQ_INSERT_HEAD(&mountlist, me, me_list);
1527 	mutex_exit(&mountlist_lock);
1528 	*mip = (mount_iterator_t *)me;
1529 }
1530 
1531 void
1532 mountlist_iterator_destroy(mount_iterator_t *mi)
1533 {
1534 	struct mountlist_entry *marker = &mi->mi_entry;
1535 
1536 	if (marker->me_mount != NULL)
1537 		vfs_unbusy(marker->me_mount);
1538 
1539 	mutex_enter(&mountlist_lock);
1540 	TAILQ_REMOVE(&mountlist, marker, me_list);
1541 	mutex_exit(&mountlist_lock);
1542 
1543 	mountlist_free(marker);
1544 
1545 }
1546 
1547 /*
1548  * Return the next mount or NULL for this iterator.
1549  * Mark it busy on success.
1550  */
1551 static inline struct mount *
1552 _mountlist_iterator_next(mount_iterator_t *mi, bool wait)
1553 {
1554 	struct mountlist_entry *me, *marker = &mi->mi_entry;
1555 	struct mount *mp;
1556 	int error;
1557 
1558 	if (marker->me_mount != NULL) {
1559 		vfs_unbusy(marker->me_mount);
1560 		marker->me_mount = NULL;
1561 	}
1562 
1563 	mutex_enter(&mountlist_lock);
1564 	for (;;) {
1565 		KASSERT(marker->me_type == ME_MARKER);
1566 
1567 		me = TAILQ_NEXT(marker, me_list);
1568 		if (me == NULL) {
1569 			/* End of list: keep marker and return. */
1570 			mutex_exit(&mountlist_lock);
1571 			return NULL;
1572 		}
1573 		TAILQ_REMOVE(&mountlist, marker, me_list);
1574 		TAILQ_INSERT_AFTER(&mountlist, me, marker, me_list);
1575 
1576 		/* Skip other markers. */
1577 		if (me->me_type != ME_MOUNT)
1578 			continue;
1579 
1580 		/* Take an initial reference for vfs_busy() below. */
1581 		mp = me->me_mount;
1582 		KASSERT(mp != NULL);
1583 		vfs_ref(mp);
1584 		mutex_exit(&mountlist_lock);
1585 
1586 		/* Try to mark this mount busy and return on success. */
1587 		if (wait)
1588 			error = vfs_busy(mp);
1589 		else
1590 			error = vfs_trybusy(mp);
1591 		if (error == 0) {
1592 			vfs_rele(mp);
1593 			marker->me_mount = mp;
1594 			return mp;
1595 		}
1596 		vfs_rele(mp);
1597 		mutex_enter(&mountlist_lock);
1598 	}
1599 }
1600 
1601 struct mount *
1602 mountlist_iterator_next(mount_iterator_t *mi)
1603 {
1604 
1605 	return _mountlist_iterator_next(mi, true);
1606 }
1607 
1608 struct mount *
1609 mountlist_iterator_trynext(mount_iterator_t *mi)
1610 {
1611 
1612 	return _mountlist_iterator_next(mi, false);
1613 }
1614 
1615 /*
1616  * Attach new mount to the end of the mount list.
1617  */
1618 void
1619 mountlist_append(struct mount *mp)
1620 {
1621 	struct mountlist_entry *me;
1622 
1623 	me = mountlist_alloc(ME_MOUNT, mp);
1624 	mutex_enter(&mountlist_lock);
1625 	TAILQ_INSERT_TAIL(&mountlist, me, me_list);
1626 	mutex_exit(&mountlist_lock);
1627 }
1628 
1629 /*
1630  * Remove mount from mount list.
1631  */void
1632 mountlist_remove(struct mount *mp)
1633 {
1634 	struct mountlist_entry *me;
1635 
1636 	mutex_enter(&mountlist_lock);
1637 	TAILQ_FOREACH(me, &mountlist, me_list)
1638 		if (me->me_type == ME_MOUNT && me->me_mount == mp)
1639 			break;
1640 	KASSERT(me != NULL);
1641 	TAILQ_REMOVE(&mountlist, me, me_list);
1642 	mutex_exit(&mountlist_lock);
1643 	mountlist_free(me);
1644 }
1645 
1646 /*
1647  * Unlocked variant to traverse the mountlist.
1648  * To be used from DDB only.
1649  */
1650 struct mount *
1651 _mountlist_next(struct mount *mp)
1652 {
1653 	struct mountlist_entry *me;
1654 
1655 	if (mp == NULL) {
1656 		me = TAILQ_FIRST(&mountlist);
1657 	} else {
1658 		TAILQ_FOREACH(me, &mountlist, me_list)
1659 			if (me->me_type == ME_MOUNT && me->me_mount == mp)
1660 				break;
1661 		if (me != NULL)
1662 			me = TAILQ_NEXT(me, me_list);
1663 	}
1664 
1665 	while (me != NULL && me->me_type != ME_MOUNT)
1666 		me = TAILQ_NEXT(me, me_list);
1667 
1668 	return (me ? me->me_mount : NULL);
1669 }
1670