xref: /netbsd-src/sys/kern/vfs_mount.c (revision 2718af68c3efc72c9769069b5c7f9ed36f6b9def)
1 /*	$NetBSD: vfs_mount.c,v 1.93 2022/04/09 23:38:33 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 1.93 2022/04/09 23:38:33 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 	fsid_t tfsid;
224 	int mtype;
225 
226 	mutex_enter(&mntid_lock);
227 	mtype = makefstype(mp->mnt_op->vfs_name);
228 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
229 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
230 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
231 	if (xxxfs_mntid == 0)
232 		++xxxfs_mntid;
233 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
234 	tfsid.__fsid_val[1] = mtype;
235 	while (vfs_getvfs(&tfsid)) {
236 		tfsid.__fsid_val[0]++;
237 		xxxfs_mntid++;
238 	}
239 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
240 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
241 	mutex_exit(&mntid_lock);
242 }
243 
244 /*
245  * Lookup a mount point by filesystem identifier.
246  *
247  * XXX Needs to add a reference to the mount point.
248  */
249 struct mount *
250 vfs_getvfs(fsid_t *fsid)
251 {
252 	mount_iterator_t *iter;
253 	struct mount *mp;
254 
255 	mountlist_iterator_init(&iter);
256 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
257 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
258 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
259 			mountlist_iterator_destroy(iter);
260 			return mp;
261 		}
262 	}
263 	mountlist_iterator_destroy(iter);
264 	return NULL;
265 }
266 
267 /*
268  * Take a reference to a mount structure.
269  */
270 void
271 vfs_ref(struct mount *mp)
272 {
273 
274 	KASSERT(mp->mnt_refcnt > 0 || mutex_owned(&mountlist_lock));
275 
276 	atomic_inc_uint(&mp->mnt_refcnt);
277 }
278 
279 /*
280  * Drop a reference to a mount structure, freeing if the last reference.
281  */
282 void
283 vfs_rele(struct mount *mp)
284 {
285 
286 #ifndef __HAVE_ATOMIC_AS_MEMBAR
287 	membar_release();
288 #endif
289 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
290 		return;
291 	}
292 #ifndef __HAVE_ATOMIC_AS_MEMBAR
293 	membar_acquire();
294 #endif
295 
296 	/*
297 	 * Nothing else has visibility of the mount: we can now
298 	 * free the data structures.
299 	 */
300 	KASSERT(mp->mnt_refcnt == 0);
301 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
302 	mutex_obj_free(mp->mnt_updating);
303 	mutex_obj_free(mp->mnt_renamelock);
304 	mutex_obj_free(mp->mnt_vnodelock);
305 	if (mp->mnt_op != NULL) {
306 		vfs_delref(mp->mnt_op);
307 	}
308 	fstrans_unmount(mp);
309 	/*
310 	 * Final free of mp gets done from fstrans_mount_dtor().
311 	 *
312 	 * Prevents this memory to be reused as a mount before
313 	 * fstrans releases all references to it.
314 	 */
315 }
316 
317 /*
318  * Mark a mount point as busy, and gain a new reference to it.  Used to
319  * prevent the file system from being unmounted during critical sections.
320  *
321  * vfs_busy can be called multiple times and by multiple threads
322  * and must be accompanied by the same number of vfs_unbusy calls.
323  *
324  * => The caller must hold a pre-existing reference to the mount.
325  * => Will fail if the file system is being unmounted, or is unmounted.
326  */
327 static inline int
328 _vfs_busy(struct mount *mp, bool wait)
329 {
330 
331 	KASSERT(mp->mnt_refcnt > 0);
332 
333 	if (wait) {
334 		fstrans_start(mp);
335 	} else {
336 		if (fstrans_start_nowait(mp))
337 			return EBUSY;
338 	}
339 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
340 		fstrans_done(mp);
341 		return ENOENT;
342 	}
343 	vfs_ref(mp);
344 	return 0;
345 }
346 
347 int
348 vfs_busy(struct mount *mp)
349 {
350 
351 	return _vfs_busy(mp, true);
352 }
353 
354 int
355 vfs_trybusy(struct mount *mp)
356 {
357 
358 	return _vfs_busy(mp, false);
359 }
360 
361 /*
362  * Unbusy a busy filesystem.
363  *
364  * Every successful vfs_busy() call must be undone by a vfs_unbusy() call.
365  */
366 void
367 vfs_unbusy(struct mount *mp)
368 {
369 
370 	KASSERT(mp->mnt_refcnt > 0);
371 
372 	fstrans_done(mp);
373 	vfs_rele(mp);
374 }
375 
376 struct vnode_iterator {
377 	vnode_impl_t vi_vnode;
378 };
379 
380 void
381 vfs_vnode_iterator_init(struct mount *mp, struct vnode_iterator **vnip)
382 {
383 	vnode_t *vp;
384 	vnode_impl_t *vip;
385 
386 	vp = vnalloc_marker(mp);
387 	vip = VNODE_TO_VIMPL(vp);
388 
389 	mutex_enter(mp->mnt_vnodelock);
390 	TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vip, vi_mntvnodes);
391 	vp->v_usecount = 1;
392 	mutex_exit(mp->mnt_vnodelock);
393 
394 	*vnip = (struct vnode_iterator *)vip;
395 }
396 
397 void
398 vfs_vnode_iterator_destroy(struct vnode_iterator *vni)
399 {
400 	vnode_impl_t *mvip = &vni->vi_vnode;
401 	vnode_t *mvp = VIMPL_TO_VNODE(mvip);
402 	kmutex_t *lock;
403 
404 	KASSERT(vnis_marker(mvp));
405 	if (vrefcnt(mvp) != 0) {
406 		lock = mvp->v_mount->mnt_vnodelock;
407 		mutex_enter(lock);
408 		TAILQ_REMOVE(&mvp->v_mount->mnt_vnodelist, mvip, vi_mntvnodes);
409 		mvp->v_usecount = 0;
410 		mutex_exit(lock);
411 	}
412 	vnfree_marker(mvp);
413 }
414 
415 static struct vnode *
416 vfs_vnode_iterator_next1(struct vnode_iterator *vni,
417     bool (*f)(void *, struct vnode *), void *cl, bool do_wait)
418 {
419 	vnode_impl_t *mvip = &vni->vi_vnode;
420 	struct mount *mp = VIMPL_TO_VNODE(mvip)->v_mount;
421 	vnode_t *vp;
422 	vnode_impl_t *vip;
423 	kmutex_t *lock;
424 	int error;
425 
426 	KASSERT(vnis_marker(VIMPL_TO_VNODE(mvip)));
427 
428 	lock = mp->mnt_vnodelock;
429 	do {
430 		mutex_enter(lock);
431 		vip = TAILQ_NEXT(mvip, vi_mntvnodes);
432 		TAILQ_REMOVE(&mp->mnt_vnodelist, mvip, vi_mntvnodes);
433 		VIMPL_TO_VNODE(mvip)->v_usecount = 0;
434 again:
435 		if (vip == NULL) {
436 			mutex_exit(lock);
437 	       		return NULL;
438 		}
439 		vp = VIMPL_TO_VNODE(vip);
440 		KASSERT(vp != NULL);
441 		mutex_enter(vp->v_interlock);
442 		if (vnis_marker(vp) ||
443 		    vdead_check(vp, (do_wait ? 0 : VDEAD_NOWAIT)) ||
444 		    (f && !(*f)(cl, vp))) {
445 			mutex_exit(vp->v_interlock);
446 			vip = TAILQ_NEXT(vip, vi_mntvnodes);
447 			goto again;
448 		}
449 
450 		TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vip, mvip, vi_mntvnodes);
451 		VIMPL_TO_VNODE(mvip)->v_usecount = 1;
452 		mutex_exit(lock);
453 		error = vcache_vget(vp);
454 		KASSERT(error == 0 || error == ENOENT);
455 	} while (error != 0);
456 
457 	return vp;
458 }
459 
460 struct vnode *
461 vfs_vnode_iterator_next(struct vnode_iterator *vni,
462     bool (*f)(void *, struct vnode *), void *cl)
463 {
464 
465 	return vfs_vnode_iterator_next1(vni, f, cl, false);
466 }
467 
468 /*
469  * Move a vnode from one mount queue to another.
470  */
471 void
472 vfs_insmntque(vnode_t *vp, struct mount *mp)
473 {
474 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
475 	struct mount *omp;
476 	kmutex_t *lock;
477 
478 	KASSERT(mp == NULL || (mp->mnt_iflag & IMNT_UNMOUNT) == 0 ||
479 	    vp->v_tag == VT_VFS);
480 
481 	/*
482 	 * Delete from old mount point vnode list, if on one.
483 	 */
484 	if ((omp = vp->v_mount) != NULL) {
485 		lock = omp->mnt_vnodelock;
486 		mutex_enter(lock);
487 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vip, vi_mntvnodes);
488 		mutex_exit(lock);
489 	}
490 
491 	/*
492 	 * Insert into list of vnodes for the new mount point, if
493 	 * available.  The caller must take a reference on the mount
494 	 * structure and donate to the vnode.
495 	 */
496 	if ((vp->v_mount = mp) != NULL) {
497 		lock = mp->mnt_vnodelock;
498 		mutex_enter(lock);
499 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vip, vi_mntvnodes);
500 		mutex_exit(lock);
501 	}
502 
503 	if (omp != NULL) {
504 		/* Release reference to old mount. */
505 		vfs_rele(omp);
506 	}
507 }
508 
509 /*
510  * Remove any vnodes in the vnode table belonging to mount point mp.
511  *
512  * If FORCECLOSE is not specified, there should not be any active ones,
513  * return error if any are found (nb: this is a user error, not a
514  * system error). If FORCECLOSE is specified, detach any active vnodes
515  * that are found.
516  *
517  * If WRITECLOSE is set, only flush out regular file vnodes open for
518  * writing.
519  *
520  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
521  */
522 #ifdef DEBUG
523 int busyprt = 0;	/* print out busy vnodes */
524 struct ctldebug debug1 = { "busyprt", &busyprt };
525 #endif
526 
527 static vnode_t *
528 vflushnext(struct vnode_iterator *marker, int *when)
529 {
530 	if (getticks() > *when) {
531 		yield();
532 		*when = getticks() + hz / 10;
533 	}
534 	return vfs_vnode_iterator_next1(marker, NULL, NULL, true);
535 }
536 
537 /*
538  * Flush one vnode.  Referenced on entry, unreferenced on return.
539  */
540 static int
541 vflush_one(vnode_t *vp, vnode_t *skipvp, int flags)
542 {
543 	int error;
544 	struct vattr vattr;
545 
546 	if (vp == skipvp ||
547 	    ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM))) {
548 		vrele(vp);
549 		return 0;
550 	}
551 	/*
552 	 * If WRITECLOSE is set, only flush out regular file
553 	 * vnodes open for writing or open and unlinked.
554 	 */
555 	if ((flags & WRITECLOSE)) {
556 		if (vp->v_type != VREG) {
557 			vrele(vp);
558 			return 0;
559 		}
560 		error = vn_lock(vp, LK_EXCLUSIVE);
561 		if (error) {
562 			KASSERT(error == ENOENT);
563 			vrele(vp);
564 			return 0;
565 		}
566 		error = VOP_FSYNC(vp, curlwp->l_cred, FSYNC_WAIT, 0, 0);
567 		if (error == 0)
568 			error = VOP_GETATTR(vp, &vattr, curlwp->l_cred);
569 		VOP_UNLOCK(vp);
570 		if (error) {
571 			vrele(vp);
572 			return error;
573 		}
574 		if (vp->v_writecount == 0 && vattr.va_nlink > 0) {
575 			vrele(vp);
576 			return 0;
577 		}
578 	}
579 	/*
580 	 * First try to recycle the vnode.
581 	 */
582 	if (vrecycle(vp))
583 		return 0;
584 	/*
585 	 * If FORCECLOSE is set, forcibly close the vnode.
586 	 * For block or character devices, revert to an
587 	 * anonymous device.  For all other files, just
588 	 * kill them.
589 	 */
590 	if (flags & FORCECLOSE) {
591 		if (vrefcnt(vp) > 1 &&
592 		    (vp->v_type == VBLK || vp->v_type == VCHR))
593 			vcache_make_anon(vp);
594 		else
595 			vgone(vp);
596 		return 0;
597 	}
598 	vrele(vp);
599 	return EBUSY;
600 }
601 
602 int
603 vflush(struct mount *mp, vnode_t *skipvp, int flags)
604 {
605 	vnode_t *vp;
606 	struct vnode_iterator *marker;
607 	int busy, error, when, retries = 2;
608 
609 	do {
610 		busy = error = when = 0;
611 
612 		/*
613 		 * First, flush out any vnode references from the
614 		 * deferred vrele list.
615 		 */
616 		vrele_flush(mp);
617 
618 		vfs_vnode_iterator_init(mp, &marker);
619 
620 		while ((vp = vflushnext(marker, &when)) != NULL) {
621 			error = vflush_one(vp, skipvp, flags);
622 			if (error == EBUSY) {
623 				error = 0;
624 				busy++;
625 #ifdef DEBUG
626 				if (busyprt && retries == 0)
627 					vprint("vflush: busy vnode", vp);
628 #endif
629 			} else if (error != 0) {
630 				break;
631 			}
632 		}
633 
634 		vfs_vnode_iterator_destroy(marker);
635 	} while (error == 0 && busy > 0 && retries-- > 0);
636 
637 	if (error)
638 		return error;
639 	if (busy)
640 		return EBUSY;
641 	return 0;
642 }
643 
644 /*
645  * Mount a file system.
646  */
647 
648 /*
649  * Scan all active processes to see if any of them have a current or root
650  * directory onto which the new filesystem has just been  mounted. If so,
651  * replace them with the new mount point.
652  */
653 static void
654 mount_checkdirs(vnode_t *olddp)
655 {
656 	vnode_t *newdp, *rele1, *rele2;
657 	struct cwdinfo *cwdi;
658 	struct proc *p;
659 	bool retry;
660 
661 	if (vrefcnt(olddp) == 1) {
662 		return;
663 	}
664 	if (VFS_ROOT(olddp->v_mountedhere, LK_EXCLUSIVE, &newdp))
665 		panic("mount: lost mount");
666 
667 	do {
668 		retry = false;
669 		mutex_enter(&proc_lock);
670 		PROCLIST_FOREACH(p, &allproc) {
671 			if ((cwdi = p->p_cwdi) == NULL)
672 				continue;
673 			/*
674 			 * Cannot change to the old directory any more,
675 			 * so even if we see a stale value it is not a
676 			 * problem.
677 			 */
678 			if (cwdi->cwdi_cdir != olddp &&
679 			    cwdi->cwdi_rdir != olddp)
680 				continue;
681 			retry = true;
682 			rele1 = NULL;
683 			rele2 = NULL;
684 			atomic_inc_uint(&cwdi->cwdi_refcnt);
685 			mutex_exit(&proc_lock);
686 			rw_enter(&cwdi->cwdi_lock, RW_WRITER);
687 			if (cwdi->cwdi_cdir == olddp) {
688 				rele1 = cwdi->cwdi_cdir;
689 				vref(newdp);
690 				cwdi->cwdi_cdir = newdp;
691 			}
692 			if (cwdi->cwdi_rdir == olddp) {
693 				rele2 = cwdi->cwdi_rdir;
694 				vref(newdp);
695 				cwdi->cwdi_rdir = newdp;
696 			}
697 			rw_exit(&cwdi->cwdi_lock);
698 			cwdfree(cwdi);
699 			if (rele1 != NULL)
700 				vrele(rele1);
701 			if (rele2 != NULL)
702 				vrele(rele2);
703 			mutex_enter(&proc_lock);
704 			break;
705 		}
706 		mutex_exit(&proc_lock);
707 	} while (retry);
708 
709 	if (rootvnode == olddp) {
710 		vrele(rootvnode);
711 		vref(newdp);
712 		rootvnode = newdp;
713 	}
714 	vput(newdp);
715 }
716 
717 /*
718  * Start extended attributes
719  */
720 static int
721 start_extattr(struct mount *mp)
722 {
723 	int error;
724 
725 	error = VFS_EXTATTRCTL(mp, EXTATTR_CMD_START, NULL, 0, NULL);
726 	if (error)
727 		printf("%s: failed to start extattr: error = %d\n",
728 		       mp->mnt_stat.f_mntonname, error);
729 
730 	return error;
731 }
732 
733 int
734 mount_domount(struct lwp *l, vnode_t **vpp, struct vfsops *vfsops,
735     const char *path, int flags, void *data, size_t *data_len)
736 {
737 	vnode_t *vp = *vpp;
738 	struct mount *mp;
739 	struct pathbuf *pb;
740 	struct nameidata nd;
741 	int error, error2;
742 
743 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
744 	    KAUTH_REQ_SYSTEM_MOUNT_NEW, vp, KAUTH_ARG(flags), data);
745 	if (error) {
746 		vfs_delref(vfsops);
747 		return error;
748 	}
749 
750 	/* Cannot make a non-dir a mount-point (from here anyway). */
751 	if (vp->v_type != VDIR) {
752 		vfs_delref(vfsops);
753 		return ENOTDIR;
754 	}
755 
756 	if (flags & MNT_EXPORTED) {
757 		vfs_delref(vfsops);
758 		return EINVAL;
759 	}
760 
761 	if ((mp = vfs_mountalloc(vfsops, vp)) == NULL) {
762 		vfs_delref(vfsops);
763 		return ENOMEM;
764 	}
765 
766 	mp->mnt_stat.f_owner = kauth_cred_geteuid(l->l_cred);
767 
768 	/*
769 	 * The underlying file system may refuse the mount for
770 	 * various reasons.  Allow the user to force it to happen.
771 	 *
772 	 * Set the mount level flags.
773 	 */
774 	mp->mnt_flag = flags & (MNT_BASIC_FLAGS | MNT_FORCE | MNT_IGNORE);
775 
776 	mutex_enter(mp->mnt_updating);
777 	error = VFS_MOUNT(mp, path, data, data_len);
778 	mp->mnt_flag &= ~MNT_OP_FLAGS;
779 
780 	if (error != 0)
781 		goto err_unmounted;
782 
783 	/*
784 	 * Validate and prepare the mount point.
785 	 */
786 	error = pathbuf_copyin(path, &pb);
787 	if (error != 0) {
788 		goto err_mounted;
789 	}
790 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
791 	error = namei(&nd);
792 	pathbuf_destroy(pb);
793 	if (error != 0) {
794 		goto err_mounted;
795 	}
796 	if (nd.ni_vp != vp) {
797 		vput(nd.ni_vp);
798 		error = EINVAL;
799 		goto err_mounted;
800 	}
801 	if (vp->v_mountedhere != NULL) {
802 		vput(nd.ni_vp);
803 		error = EBUSY;
804 		goto err_mounted;
805 	}
806 	error = vinvalbuf(vp, V_SAVE, l->l_cred, l, 0, 0);
807 	if (error != 0) {
808 		vput(nd.ni_vp);
809 		goto err_mounted;
810 	}
811 
812 	/*
813 	 * Put the new filesystem on the mount list after root.
814 	 */
815 	cache_purge(vp);
816 	mp->mnt_iflag &= ~IMNT_WANTRDWR;
817 
818 	mountlist_append(mp);
819 	if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
820 		vfs_syncer_add_to_worklist(mp);
821 	vp->v_mountedhere = mp;
822 	vput(nd.ni_vp);
823 
824 	mount_checkdirs(vp);
825 	mutex_exit(mp->mnt_updating);
826 
827 	/* Hold an additional reference to the mount across VFS_START(). */
828 	vfs_ref(mp);
829 	(void) VFS_STATVFS(mp, &mp->mnt_stat);
830 	error = VFS_START(mp, 0);
831 	if (error) {
832 		vrele(vp);
833 	} else if (flags & MNT_EXTATTR) {
834 		if (start_extattr(mp) != 0)
835 			mp->mnt_flag &= ~MNT_EXTATTR;
836 	}
837 	/* Drop reference held for VFS_START(). */
838 	vfs_rele(mp);
839 	*vpp = NULL;
840 	return error;
841 
842 err_mounted:
843 	do {
844 		error2 = vfs_suspend(mp, 0);
845 	} while (error2 == EINTR || error2 == ERESTART);
846 	KASSERT(error2 == 0 || error2 == EOPNOTSUPP);
847 
848 	if (VFS_UNMOUNT(mp, MNT_FORCE) != 0)
849 		panic("Unmounting fresh file system failed");
850 
851 	if (error2 == 0)
852 		vfs_resume(mp);
853 
854 err_unmounted:
855 	mutex_exit(mp->mnt_updating);
856 	vfs_rele(mp);
857 
858 	return error;
859 }
860 
861 /*
862  * Do the actual file system unmount.  File system is assumed to have
863  * been locked by the caller.
864  *
865  * => Caller hold reference to the mount, explicitly for dounmount().
866  */
867 int
868 dounmount(struct mount *mp, int flags, struct lwp *l)
869 {
870 	vnode_t *coveredvp;
871 	int error, async, used_syncer, used_extattr;
872 	const bool was_suspended = fstrans_is_owner(mp);
873 
874 #if NVERIEXEC > 0
875 	error = veriexec_unmountchk(mp);
876 	if (error)
877 		return (error);
878 #endif /* NVERIEXEC > 0 */
879 
880 	if (!was_suspended) {
881 		error = vfs_suspend(mp, 0);
882 		if (error) {
883 			return error;
884 		}
885 	}
886 
887 	KASSERT((mp->mnt_iflag & IMNT_GONE) == 0);
888 
889 	used_syncer = (mp->mnt_iflag & IMNT_ONWORKLIST) != 0;
890 	used_extattr = mp->mnt_flag & MNT_EXTATTR;
891 
892 	mp->mnt_iflag |= IMNT_UNMOUNT;
893 	mutex_enter(mp->mnt_updating);
894 	async = mp->mnt_flag & MNT_ASYNC;
895 	mp->mnt_flag &= ~MNT_ASYNC;
896 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
897 	if (used_syncer)
898 		vfs_syncer_remove_from_worklist(mp);
899 	error = 0;
900 	if (((mp->mnt_flag & MNT_RDONLY) == 0) && ((flags & MNT_FORCE) == 0)) {
901 		error = VFS_SYNC(mp, MNT_WAIT, l->l_cred);
902 	}
903 	if (error == 0 || (flags & MNT_FORCE)) {
904 		error = VFS_UNMOUNT(mp, flags);
905 	}
906 	if (error) {
907 		mp->mnt_iflag &= ~IMNT_UNMOUNT;
908 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
909 			vfs_syncer_add_to_worklist(mp);
910 		mp->mnt_flag |= async;
911 		mutex_exit(mp->mnt_updating);
912 		if (!was_suspended)
913 			vfs_resume(mp);
914 		if (used_extattr) {
915 			if (start_extattr(mp) != 0)
916 				mp->mnt_flag &= ~MNT_EXTATTR;
917 			else
918 				mp->mnt_flag |= MNT_EXTATTR;
919 		}
920 		return (error);
921 	}
922 	mutex_exit(mp->mnt_updating);
923 
924 	/*
925 	 * mark filesystem as gone to prevent further umounts
926 	 * after mnt_umounting lock is gone, this also prevents
927 	 * vfs_busy() from succeeding.
928 	 */
929 	mp->mnt_iflag |= IMNT_GONE;
930 	if (!was_suspended)
931 		vfs_resume(mp);
932 
933 	if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
934 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY);
935 		coveredvp->v_mountedhere = NULL;
936 		VOP_UNLOCK(coveredvp);
937 	}
938 	mountlist_remove(mp);
939 	if (TAILQ_FIRST(&mp->mnt_vnodelist) != NULL)
940 		panic("unmount: dangling vnode");
941 	vfs_hooks_unmount(mp);
942 
943 	vfs_rele(mp);	/* reference from mount() */
944 	if (coveredvp != NULLVP) {
945 		vrele(coveredvp);
946 	}
947 	return (0);
948 }
949 
950 /*
951  * Unmount all file systems.
952  * We traverse the list in reverse order under the assumption that doing so
953  * will avoid needing to worry about dependencies.
954  */
955 bool
956 vfs_unmountall(struct lwp *l)
957 {
958 
959 	printf("unmounting file systems...\n");
960 	return vfs_unmountall1(l, true, true);
961 }
962 
963 static void
964 vfs_unmount_print(struct mount *mp, const char *pfx)
965 {
966 
967 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
968 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
969 	    mp->mnt_stat.f_fstypename);
970 }
971 
972 /*
973  * Return the mount with the highest generation less than "gen".
974  */
975 static struct mount *
976 vfs_unmount_next(uint64_t gen)
977 {
978 	mount_iterator_t *iter;
979 	struct mount *mp, *nmp;
980 
981 	nmp = NULL;
982 
983 	mountlist_iterator_init(&iter);
984 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
985 		if ((nmp == NULL || mp->mnt_gen > nmp->mnt_gen) &&
986 		    mp->mnt_gen < gen) {
987 			if (nmp != NULL)
988 				vfs_rele(nmp);
989 			nmp = mp;
990 			vfs_ref(nmp);
991 		}
992 	}
993 	mountlist_iterator_destroy(iter);
994 
995 	return nmp;
996 }
997 
998 bool
999 vfs_unmount_forceone(struct lwp *l)
1000 {
1001 	struct mount *mp;
1002 	int error;
1003 
1004 	mp = vfs_unmount_next(mountgen);
1005 	if (mp == NULL) {
1006 		return false;
1007 	}
1008 
1009 #ifdef DEBUG
1010 	printf("forcefully unmounting %s (%s)...\n",
1011 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
1012 #endif
1013 	if ((error = dounmount(mp, MNT_FORCE, l)) == 0) {
1014 		vfs_unmount_print(mp, "forcefully ");
1015 		return true;
1016 	} else {
1017 		vfs_rele(mp);
1018 	}
1019 
1020 #ifdef DEBUG
1021 	printf("forceful unmount of %s failed with error %d\n",
1022 	    mp->mnt_stat.f_mntonname, error);
1023 #endif
1024 
1025 	return false;
1026 }
1027 
1028 bool
1029 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
1030 {
1031 	struct mount *mp;
1032 	mount_iterator_t *iter;
1033 	bool any_error = false, progress = false;
1034 	uint64_t gen;
1035 	int error;
1036 
1037 	gen = mountgen;
1038 	for (;;) {
1039 		mp = vfs_unmount_next(gen);
1040 		if (mp == NULL)
1041 			break;
1042 		gen = mp->mnt_gen;
1043 
1044 #ifdef DEBUG
1045 		printf("unmounting %p %s (%s)...\n",
1046 		    (void *)mp, mp->mnt_stat.f_mntonname,
1047 		    mp->mnt_stat.f_mntfromname);
1048 #endif
1049 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
1050 			vfs_unmount_print(mp, "");
1051 			progress = true;
1052 		} else {
1053 			vfs_rele(mp);
1054 			if (verbose) {
1055 				printf("unmount of %s failed with error %d\n",
1056 				    mp->mnt_stat.f_mntonname, error);
1057 			}
1058 			any_error = true;
1059 		}
1060 	}
1061 	if (verbose) {
1062 		printf("unmounting done\n");
1063 	}
1064 	if (any_error && verbose) {
1065 		printf("WARNING: some file systems would not unmount\n");
1066 	}
1067 	/* If the mountlist is empty it is time to remove swap. */
1068 	mountlist_iterator_init(&iter);
1069 	if (mountlist_iterator_next(iter) == NULL) {
1070 		uvm_swap_shutdown(l);
1071 	}
1072 	mountlist_iterator_destroy(iter);
1073 
1074 	return progress;
1075 }
1076 
1077 void
1078 vfs_sync_all(struct lwp *l)
1079 {
1080 	printf("syncing disks... ");
1081 
1082 	/* remove user processes from run queue */
1083 	suspendsched();
1084 	(void)spl0();
1085 
1086 	/* avoid coming back this way again if we panic. */
1087 	doing_shutdown = 1;
1088 
1089 	do_sys_sync(l);
1090 
1091 	/* Wait for sync to finish. */
1092 	if (vfs_syncwait() != 0) {
1093 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
1094 		Debugger();
1095 #endif
1096 		printf("giving up\n");
1097 		return;
1098 	} else
1099 		printf("done\n");
1100 }
1101 
1102 /*
1103  * Sync and unmount file systems before shutting down.
1104  */
1105 void
1106 vfs_shutdown(void)
1107 {
1108 	lwp_t *l = curlwp;
1109 
1110 	vfs_sync_all(l);
1111 
1112 	/*
1113 	 * If we have panicked - do not make the situation potentially
1114 	 * worse by unmounting the file systems.
1115 	 */
1116 	if (panicstr != NULL) {
1117 		return;
1118 	}
1119 
1120 	/* Unmount file systems. */
1121 	vfs_unmountall(l);
1122 }
1123 
1124 /*
1125  * Print a list of supported file system types (used by vfs_mountroot)
1126  */
1127 static void
1128 vfs_print_fstypes(void)
1129 {
1130 	struct vfsops *v;
1131 	int cnt = 0;
1132 
1133 	mutex_enter(&vfs_list_lock);
1134 	LIST_FOREACH(v, &vfs_list, vfs_list)
1135 		++cnt;
1136 	mutex_exit(&vfs_list_lock);
1137 
1138 	if (cnt == 0) {
1139 		printf("WARNING: No file system modules have been loaded.\n");
1140 		return;
1141 	}
1142 
1143 	printf("Supported file systems:");
1144 	mutex_enter(&vfs_list_lock);
1145 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1146 		printf(" %s", v->vfs_name);
1147 	}
1148 	mutex_exit(&vfs_list_lock);
1149 	printf("\n");
1150 }
1151 
1152 /*
1153  * Mount the root file system.  If the operator didn't specify a
1154  * file system to use, try all possible file systems until one
1155  * succeeds.
1156  */
1157 int
1158 vfs_mountroot(void)
1159 {
1160 	struct vfsops *v;
1161 	int error = ENODEV;
1162 
1163 	if (root_device == NULL)
1164 		panic("vfs_mountroot: root device unknown");
1165 
1166 	switch (device_class(root_device)) {
1167 	case DV_IFNET:
1168 		if (rootdev != NODEV)
1169 			panic("vfs_mountroot: rootdev set for DV_IFNET "
1170 			    "(0x%llx -> %llu,%llu)",
1171 			    (unsigned long long)rootdev,
1172 			    (unsigned long long)major(rootdev),
1173 			    (unsigned long long)minor(rootdev));
1174 		break;
1175 
1176 	case DV_DISK:
1177 		if (rootdev == NODEV)
1178 			panic("vfs_mountroot: rootdev not set for DV_DISK");
1179 	        if (bdevvp(rootdev, &rootvp))
1180 	                panic("vfs_mountroot: can't get vnode for rootdev");
1181 		vn_lock(rootvp, LK_EXCLUSIVE | LK_RETRY);
1182 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
1183 		VOP_UNLOCK(rootvp);
1184 		if (error) {
1185 			printf("vfs_mountroot: can't open root device\n");
1186 			return (error);
1187 		}
1188 		break;
1189 
1190 	case DV_VIRTUAL:
1191 		break;
1192 
1193 	default:
1194 		printf("%s: inappropriate for root file system\n",
1195 		    device_xname(root_device));
1196 		return (ENODEV);
1197 	}
1198 
1199 	/*
1200 	 * If user specified a root fs type, use it.  Make sure the
1201 	 * specified type exists and has a mount_root()
1202 	 */
1203 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
1204 		v = vfs_getopsbyname(rootfstype);
1205 		error = EFTYPE;
1206 		if (v != NULL) {
1207 			if (v->vfs_mountroot != NULL) {
1208 				error = (v->vfs_mountroot)();
1209 			}
1210 			v->vfs_refcount--;
1211 		}
1212 		goto done;
1213 	}
1214 
1215 	/*
1216 	 * Try each file system currently configured into the kernel.
1217 	 */
1218 	mutex_enter(&vfs_list_lock);
1219 	LIST_FOREACH(v, &vfs_list, vfs_list) {
1220 		if (v->vfs_mountroot == NULL)
1221 			continue;
1222 #ifdef DEBUG
1223 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
1224 #endif
1225 		v->vfs_refcount++;
1226 		mutex_exit(&vfs_list_lock);
1227 		error = (*v->vfs_mountroot)();
1228 		mutex_enter(&vfs_list_lock);
1229 		v->vfs_refcount--;
1230 		if (!error) {
1231 			aprint_normal("root file system type: %s\n",
1232 			    v->vfs_name);
1233 			break;
1234 		}
1235 	}
1236 	mutex_exit(&vfs_list_lock);
1237 
1238 	if (v == NULL) {
1239 		vfs_print_fstypes();
1240 		printf("no file system for %s", device_xname(root_device));
1241 		if (device_class(root_device) == DV_DISK)
1242 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
1243 		printf("\n");
1244 		error = EFTYPE;
1245 	}
1246 
1247 done:
1248 	if (error && device_class(root_device) == DV_DISK) {
1249 		vn_lock(rootvp, LK_EXCLUSIVE | LK_RETRY);
1250 		VOP_CLOSE(rootvp, FREAD, FSCRED);
1251 		VOP_UNLOCK(rootvp);
1252 		vrele(rootvp);
1253 	}
1254 	if (error == 0) {
1255 		mount_iterator_t *iter;
1256 		struct mount *mp;
1257 		extern struct cwdinfo cwdi0;
1258 
1259 		mountlist_iterator_init(&iter);
1260 		mp = mountlist_iterator_next(iter);
1261 		KASSERT(mp != NULL);
1262 		mountlist_iterator_destroy(iter);
1263 
1264 		mp->mnt_flag |= MNT_ROOTFS;
1265 		mp->mnt_op->vfs_refcount++;
1266 
1267 		/*
1268 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
1269 		 * reference it, and donate it the reference grabbed
1270 		 * with VFS_ROOT().
1271 		 */
1272 		error = VFS_ROOT(mp, LK_NONE, &rootvnode);
1273 		if (error)
1274 			panic("cannot find root vnode, error=%d", error);
1275 		cwdi0.cwdi_cdir = rootvnode;
1276 		cwdi0.cwdi_rdir = NULL;
1277 
1278 		/*
1279 		 * Now that root is mounted, we can fixup initproc's CWD
1280 		 * info.  All other processes are kthreads, which merely
1281 		 * share proc0's CWD info.
1282 		 */
1283 		initproc->p_cwdi->cwdi_cdir = rootvnode;
1284 		vref(initproc->p_cwdi->cwdi_cdir);
1285 		initproc->p_cwdi->cwdi_rdir = NULL;
1286 		/*
1287 		 * Enable loading of modules from the filesystem
1288 		 */
1289 		module_load_vfs_init();
1290 
1291 	}
1292 	return (error);
1293 }
1294 
1295 /*
1296  * mount_specific_key_create --
1297  *	Create a key for subsystem mount-specific data.
1298  */
1299 int
1300 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
1301 {
1302 
1303 	return specificdata_key_create(mount_specificdata_domain, keyp, dtor);
1304 }
1305 
1306 /*
1307  * mount_specific_key_delete --
1308  *	Delete a key for subsystem mount-specific data.
1309  */
1310 void
1311 mount_specific_key_delete(specificdata_key_t key)
1312 {
1313 
1314 	specificdata_key_delete(mount_specificdata_domain, key);
1315 }
1316 
1317 /*
1318  * mount_initspecific --
1319  *	Initialize a mount's specificdata container.
1320  */
1321 void
1322 mount_initspecific(struct mount *mp)
1323 {
1324 	int error __diagused;
1325 
1326 	error = specificdata_init(mount_specificdata_domain,
1327 				  &mp->mnt_specdataref);
1328 	KASSERT(error == 0);
1329 }
1330 
1331 /*
1332  * mount_finispecific --
1333  *	Finalize a mount's specificdata container.
1334  */
1335 void
1336 mount_finispecific(struct mount *mp)
1337 {
1338 
1339 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
1340 }
1341 
1342 /*
1343  * mount_getspecific --
1344  *	Return mount-specific data corresponding to the specified key.
1345  */
1346 void *
1347 mount_getspecific(struct mount *mp, specificdata_key_t key)
1348 {
1349 
1350 	return specificdata_getspecific(mount_specificdata_domain,
1351 					 &mp->mnt_specdataref, key);
1352 }
1353 
1354 /*
1355  * mount_setspecific --
1356  *	Set mount-specific data corresponding to the specified key.
1357  */
1358 void
1359 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
1360 {
1361 
1362 	specificdata_setspecific(mount_specificdata_domain,
1363 				 &mp->mnt_specdataref, key, data);
1364 }
1365 
1366 /*
1367  * Check to see if a filesystem is mounted on a block device.
1368  */
1369 int
1370 vfs_mountedon(vnode_t *vp)
1371 {
1372 	vnode_t *vq;
1373 	int error = 0;
1374 
1375 	if (vp->v_type != VBLK)
1376 		return ENOTBLK;
1377 	if (spec_node_getmountedfs(vp) != NULL)
1378 		return EBUSY;
1379 	if (spec_node_lookup_by_dev(vp->v_type, vp->v_rdev, VDEAD_NOWAIT, &vq)
1380 	    == 0) {
1381 		if (spec_node_getmountedfs(vq) != NULL)
1382 			error = EBUSY;
1383 		vrele(vq);
1384 	}
1385 
1386 	return error;
1387 }
1388 
1389 /*
1390  * Check if a device pointed to by vp is mounted.
1391  *
1392  * Returns:
1393  *   EINVAL	if it's not a disk
1394  *   EBUSY	if it's a disk and mounted
1395  *   0		if it's a disk and not mounted
1396  */
1397 int
1398 rawdev_mounted(vnode_t *vp, vnode_t **bvpp)
1399 {
1400 	vnode_t *bvp;
1401 	dev_t dev;
1402 	int d_type;
1403 
1404 	bvp = NULL;
1405 	d_type = D_OTHER;
1406 
1407 	if (iskmemvp(vp))
1408 		return EINVAL;
1409 
1410 	switch (vp->v_type) {
1411 	case VCHR: {
1412 		const struct cdevsw *cdev;
1413 
1414 		dev = vp->v_rdev;
1415 		cdev = cdevsw_lookup(dev);
1416 		if (cdev != NULL) {
1417 			dev_t blkdev;
1418 
1419 			blkdev = devsw_chr2blk(dev);
1420 			if (blkdev != NODEV) {
1421 				if (vfinddev(blkdev, VBLK, &bvp) != 0) {
1422 					d_type = (cdev->d_flag & D_TYPEMASK);
1423 					/* XXX: what if bvp disappears? */
1424 					vrele(bvp);
1425 				}
1426 			}
1427 		}
1428 
1429 		break;
1430 		}
1431 
1432 	case VBLK: {
1433 		const struct bdevsw *bdev;
1434 
1435 		dev = vp->v_rdev;
1436 		bdev = bdevsw_lookup(dev);
1437 		if (bdev != NULL)
1438 			d_type = (bdev->d_flag & D_TYPEMASK);
1439 
1440 		bvp = vp;
1441 
1442 		break;
1443 		}
1444 
1445 	default:
1446 		break;
1447 	}
1448 
1449 	if (d_type != D_DISK)
1450 		return EINVAL;
1451 
1452 	if (bvpp != NULL)
1453 		*bvpp = bvp;
1454 
1455 	/*
1456 	 * XXX: This is bogus. We should be failing the request
1457 	 * XXX: not only if this specific slice is mounted, but
1458 	 * XXX: if it's on a disk with any other mounted slice.
1459 	 */
1460 	if (vfs_mountedon(bvp))
1461 		return EBUSY;
1462 
1463 	return 0;
1464 }
1465 
1466 /*
1467  * Make a 'unique' number from a mount type name.
1468  */
1469 long
1470 makefstype(const char *type)
1471 {
1472 	long rv;
1473 
1474 	for (rv = 0; *type; type++) {
1475 		rv <<= 2;
1476 		rv ^= *type;
1477 	}
1478 	return rv;
1479 }
1480 
1481 static struct mountlist_entry *
1482 mountlist_alloc(enum mountlist_type type, struct mount *mp)
1483 {
1484 	struct mountlist_entry *me;
1485 
1486 	me = kmem_zalloc(sizeof(*me), KM_SLEEP);
1487 	me->me_mount = mp;
1488 	me->me_type = type;
1489 
1490 	return me;
1491 }
1492 
1493 static void
1494 mountlist_free(struct mountlist_entry *me)
1495 {
1496 
1497 	kmem_free(me, sizeof(*me));
1498 }
1499 
1500 void
1501 mountlist_iterator_init(mount_iterator_t **mip)
1502 {
1503 	struct mountlist_entry *me;
1504 
1505 	me = mountlist_alloc(ME_MARKER, NULL);
1506 	mutex_enter(&mountlist_lock);
1507 	TAILQ_INSERT_HEAD(&mountlist, me, me_list);
1508 	mutex_exit(&mountlist_lock);
1509 	*mip = (mount_iterator_t *)me;
1510 }
1511 
1512 void
1513 mountlist_iterator_destroy(mount_iterator_t *mi)
1514 {
1515 	struct mountlist_entry *marker = &mi->mi_entry;
1516 
1517 	if (marker->me_mount != NULL)
1518 		vfs_unbusy(marker->me_mount);
1519 
1520 	mutex_enter(&mountlist_lock);
1521 	TAILQ_REMOVE(&mountlist, marker, me_list);
1522 	mutex_exit(&mountlist_lock);
1523 
1524 	mountlist_free(marker);
1525 
1526 }
1527 
1528 /*
1529  * Return the next mount or NULL for this iterator.
1530  * Mark it busy on success.
1531  */
1532 static inline struct mount *
1533 _mountlist_iterator_next(mount_iterator_t *mi, bool wait)
1534 {
1535 	struct mountlist_entry *me, *marker = &mi->mi_entry;
1536 	struct mount *mp;
1537 	int error;
1538 
1539 	if (marker->me_mount != NULL) {
1540 		vfs_unbusy(marker->me_mount);
1541 		marker->me_mount = NULL;
1542 	}
1543 
1544 	mutex_enter(&mountlist_lock);
1545 	for (;;) {
1546 		KASSERT(marker->me_type == ME_MARKER);
1547 
1548 		me = TAILQ_NEXT(marker, me_list);
1549 		if (me == NULL) {
1550 			/* End of list: keep marker and return. */
1551 			mutex_exit(&mountlist_lock);
1552 			return NULL;
1553 		}
1554 		TAILQ_REMOVE(&mountlist, marker, me_list);
1555 		TAILQ_INSERT_AFTER(&mountlist, me, marker, me_list);
1556 
1557 		/* Skip other markers. */
1558 		if (me->me_type != ME_MOUNT)
1559 			continue;
1560 
1561 		/* Take an initial reference for vfs_busy() below. */
1562 		mp = me->me_mount;
1563 		KASSERT(mp != NULL);
1564 		vfs_ref(mp);
1565 		mutex_exit(&mountlist_lock);
1566 
1567 		/* Try to mark this mount busy and return on success. */
1568 		if (wait)
1569 			error = vfs_busy(mp);
1570 		else
1571 			error = vfs_trybusy(mp);
1572 		if (error == 0) {
1573 			vfs_rele(mp);
1574 			marker->me_mount = mp;
1575 			return mp;
1576 		}
1577 		vfs_rele(mp);
1578 		mutex_enter(&mountlist_lock);
1579 	}
1580 }
1581 
1582 struct mount *
1583 mountlist_iterator_next(mount_iterator_t *mi)
1584 {
1585 
1586 	return _mountlist_iterator_next(mi, true);
1587 }
1588 
1589 struct mount *
1590 mountlist_iterator_trynext(mount_iterator_t *mi)
1591 {
1592 
1593 	return _mountlist_iterator_next(mi, false);
1594 }
1595 
1596 /*
1597  * Attach new mount to the end of the mount list.
1598  */
1599 void
1600 mountlist_append(struct mount *mp)
1601 {
1602 	struct mountlist_entry *me;
1603 
1604 	me = mountlist_alloc(ME_MOUNT, mp);
1605 	mutex_enter(&mountlist_lock);
1606 	TAILQ_INSERT_TAIL(&mountlist, me, me_list);
1607 	mutex_exit(&mountlist_lock);
1608 }
1609 
1610 /*
1611  * Remove mount from mount list.
1612  */void
1613 mountlist_remove(struct mount *mp)
1614 {
1615 	struct mountlist_entry *me;
1616 
1617 	mutex_enter(&mountlist_lock);
1618 	TAILQ_FOREACH(me, &mountlist, me_list)
1619 		if (me->me_type == ME_MOUNT && me->me_mount == mp)
1620 			break;
1621 	KASSERT(me != NULL);
1622 	TAILQ_REMOVE(&mountlist, me, me_list);
1623 	mutex_exit(&mountlist_lock);
1624 	mountlist_free(me);
1625 }
1626 
1627 /*
1628  * Unlocked variant to traverse the mountlist.
1629  * To be used from DDB only.
1630  */
1631 struct mount *
1632 _mountlist_next(struct mount *mp)
1633 {
1634 	struct mountlist_entry *me;
1635 
1636 	if (mp == NULL) {
1637 		me = TAILQ_FIRST(&mountlist);
1638 	} else {
1639 		TAILQ_FOREACH(me, &mountlist, me_list)
1640 			if (me->me_type == ME_MOUNT && me->me_mount == mp)
1641 				break;
1642 		if (me != NULL)
1643 			me = TAILQ_NEXT(me, me_list);
1644 	}
1645 
1646 	while (me != NULL && me->me_type != ME_MOUNT)
1647 		me = TAILQ_NEXT(me, me_list);
1648 
1649 	return (me ? me->me_mount : NULL);
1650 }
1651