xref: /netbsd-src/sys/kern/vfs_subr.c (revision b8c616269f5ebf18ab2e35cb8099d683130a177c)
1 /*	$NetBSD: vfs_subr.c,v 1.186 2003/02/01 06:23:45 thorpej Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998 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.
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  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1989, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  * (c) UNIX System Laboratories, Inc.
44  * All or some portions of this file are derived from material licensed
45  * to the University of California by American Telephone and Telegraph
46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47  * the permission of UNIX System Laboratories, Inc.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  * 3. All advertising materials mentioning features or use of this software
58  *    must display the following acknowledgement:
59  *	This product includes software developed by the University of
60  *	California, Berkeley and its contributors.
61  * 4. Neither the name of the University nor the names of its contributors
62  *    may be used to endorse or promote products derived from this software
63  *    without specific prior written permission.
64  *
65  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
66  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
67  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
68  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
69  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
70  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
71  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
72  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
73  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
74  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
75  * SUCH DAMAGE.
76  *
77  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
78  */
79 
80 /*
81  * External virtual filesystem routines
82  */
83 
84 #include <sys/cdefs.h>
85 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.186 2003/02/01 06:23:45 thorpej Exp $");
86 
87 #include "opt_ddb.h"
88 #include "opt_compat_netbsd.h"
89 #include "opt_compat_43.h"
90 
91 #include <sys/param.h>
92 #include <sys/systm.h>
93 #include <sys/proc.h>
94 #include <sys/kernel.h>
95 #include <sys/mount.h>
96 #include <sys/time.h>
97 #include <sys/event.h>
98 #include <sys/fcntl.h>
99 #include <sys/vnode.h>
100 #include <sys/stat.h>
101 #include <sys/namei.h>
102 #include <sys/ucred.h>
103 #include <sys/buf.h>
104 #include <sys/errno.h>
105 #include <sys/malloc.h>
106 #include <sys/domain.h>
107 #include <sys/mbuf.h>
108 #include <sys/sa.h>
109 #include <sys/syscallargs.h>
110 #include <sys/device.h>
111 #include <sys/dirent.h>
112 
113 #include <miscfs/specfs/specdev.h>
114 #include <miscfs/genfs/genfs.h>
115 #include <miscfs/syncfs/syncfs.h>
116 
117 #include <uvm/uvm.h>
118 #include <uvm/uvm_ddb.h>
119 
120 #include <sys/sysctl.h>
121 
122 enum vtype iftovt_tab[16] = {
123 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
124 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
125 };
126 const int	vttoif_tab[9] = {
127 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
128 	S_IFSOCK, S_IFIFO, S_IFMT,
129 };
130 
131 int doforce = 1;		/* 1 => permit forcible unmounting */
132 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
133 
134 extern int dovfsusermount;	/* 1 => permit any user to mount filesystems */
135 
136 /*
137  * Insq/Remq for the vnode usage lists.
138  */
139 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
140 #define	bufremvn(bp) {							\
141 	LIST_REMOVE(bp, b_vnbufs);					\
142 	(bp)->b_vnbufs.le_next = NOLIST;				\
143 }
144 /* TAILQ_HEAD(freelst, vnode) vnode_free_list =	vnode free list (in vnode.h) */
145 struct freelst vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
146 struct freelst vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
147 
148 struct mntlist mountlist =			/* mounted filesystem list */
149     CIRCLEQ_HEAD_INITIALIZER(mountlist);
150 struct vfs_list_head vfs_list =			/* vfs list */
151     LIST_HEAD_INITIALIZER(vfs_list);
152 
153 struct nfs_public nfs_pub;			/* publicly exported FS */
154 
155 struct simplelock mountlist_slock = SIMPLELOCK_INITIALIZER;
156 static struct simplelock mntid_slock = SIMPLELOCK_INITIALIZER;
157 struct simplelock mntvnode_slock = SIMPLELOCK_INITIALIZER;
158 struct simplelock vnode_free_list_slock = SIMPLELOCK_INITIALIZER;
159 struct simplelock spechash_slock = SIMPLELOCK_INITIALIZER;
160 
161 /*
162  * These define the root filesystem and device.
163  */
164 struct mount *rootfs;
165 struct vnode *rootvnode;
166 struct device *root_device;			/* root device */
167 
168 struct pool vnode_pool;				/* memory pool for vnodes */
169 
170 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
171 
172 /*
173  * Local declarations.
174  */
175 void insmntque __P((struct vnode *, struct mount *));
176 int getdevvp __P((dev_t, struct vnode **, enum vtype));
177 void vgoneall __P((struct vnode *));
178 
179 void vclean(struct vnode *, int, struct proc *);
180 
181 static int vfs_hang_addrlist __P((struct mount *, struct netexport *,
182 				  struct export_args *));
183 static int vfs_free_netcred __P((struct radix_node *, void *));
184 static void vfs_free_addrlist __P((struct netexport *));
185 
186 #ifdef DEBUG
187 void printlockedvnodes __P((void));
188 #endif
189 
190 /*
191  * Initialize the vnode management data structures.
192  */
193 void
194 vntblinit()
195 {
196 
197 	pool_init(&vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodepl",
198 	    &pool_allocator_nointr);
199 
200 	/*
201 	 * Initialize the filesystem syncer.
202 	 */
203 	vn_initialize_syncerd();
204 }
205 
206 /*
207  * Mark a mount point as busy. Used to synchronize access and to delay
208  * unmounting. Interlock is not released on failure.
209  */
210 int
211 vfs_busy(mp, flags, interlkp)
212 	struct mount *mp;
213 	int flags;
214 	struct simplelock *interlkp;
215 {
216 	int lkflags;
217 
218 	while (mp->mnt_flag & MNT_UNMOUNT) {
219 		int gone;
220 
221 		if (flags & LK_NOWAIT)
222 			return (ENOENT);
223 		if ((flags & LK_RECURSEFAIL) && mp->mnt_unmounter != NULL
224 		    && mp->mnt_unmounter == curproc)
225 			return (EDEADLK);
226 		if (interlkp)
227 			simple_unlock(interlkp);
228 		/*
229 		 * Since all busy locks are shared except the exclusive
230 		 * lock granted when unmounting, the only place that a
231 		 * wakeup needs to be done is at the release of the
232 		 * exclusive lock at the end of dounmount.
233 		 *
234 		 * XXX MP: add spinlock protecting mnt_wcnt here once you
235 		 * can atomically unlock-and-sleep.
236 		 */
237 		mp->mnt_wcnt++;
238 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
239 		mp->mnt_wcnt--;
240 		gone = mp->mnt_flag & MNT_GONE;
241 
242 		if (mp->mnt_wcnt == 0)
243 			wakeup(&mp->mnt_wcnt);
244 		if (interlkp)
245 			simple_lock(interlkp);
246 		if (gone)
247 			return (ENOENT);
248 	}
249 	lkflags = LK_SHARED;
250 	if (interlkp)
251 		lkflags |= LK_INTERLOCK;
252 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp))
253 		panic("vfs_busy: unexpected lock failure");
254 	return (0);
255 }
256 
257 /*
258  * Free a busy filesystem.
259  */
260 void
261 vfs_unbusy(mp)
262 	struct mount *mp;
263 {
264 
265 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
266 }
267 
268 /*
269  * Lookup a filesystem type, and if found allocate and initialize
270  * a mount structure for it.
271  *
272  * Devname is usually updated by mount(8) after booting.
273  */
274 int
275 vfs_rootmountalloc(fstypename, devname, mpp)
276 	char *fstypename;
277 	char *devname;
278 	struct mount **mpp;
279 {
280 	struct vfsops *vfsp = NULL;
281 	struct mount *mp;
282 
283 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
284 		if (!strncmp(vfsp->vfs_name, fstypename, MFSNAMELEN))
285 			break;
286 
287 	if (vfsp == NULL)
288 		return (ENODEV);
289 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
290 	memset((char *)mp, 0, (u_long)sizeof(struct mount));
291 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
292 	(void)vfs_busy(mp, LK_NOWAIT, 0);
293 	LIST_INIT(&mp->mnt_vnodelist);
294 	mp->mnt_op = vfsp;
295 	mp->mnt_flag = MNT_RDONLY;
296 	mp->mnt_vnodecovered = NULLVP;
297 	vfsp->vfs_refcount++;
298 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN);
299 	mp->mnt_stat.f_mntonname[0] = '/';
300 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
301 	*mpp = mp;
302 	return (0);
303 }
304 
305 /*
306  * Lookup a mount point by filesystem identifier.
307  */
308 struct mount *
309 vfs_getvfs(fsid)
310 	fsid_t *fsid;
311 {
312 	struct mount *mp;
313 
314 	simple_lock(&mountlist_slock);
315 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
316 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
317 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
318 			simple_unlock(&mountlist_slock);
319 			return (mp);
320 		}
321 	}
322 	simple_unlock(&mountlist_slock);
323 	return ((struct mount *)0);
324 }
325 
326 /*
327  * Get a new unique fsid
328  */
329 void
330 vfs_getnewfsid(mp)
331 	struct mount *mp;
332 {
333 	static u_short xxxfs_mntid;
334 	fsid_t tfsid;
335 	int mtype;
336 
337 	simple_lock(&mntid_slock);
338 	mtype = makefstype(mp->mnt_op->vfs_name);
339 	mp->mnt_stat.f_fsid.val[0] = makedev(mtype, 0);
340 	mp->mnt_stat.f_fsid.val[1] = mtype;
341 	if (xxxfs_mntid == 0)
342 		++xxxfs_mntid;
343 	tfsid.val[0] = makedev(mtype & 0xff, xxxfs_mntid);
344 	tfsid.val[1] = mtype;
345 	if (!CIRCLEQ_EMPTY(&mountlist)) {
346 		while (vfs_getvfs(&tfsid)) {
347 			tfsid.val[0]++;
348 			xxxfs_mntid++;
349 		}
350 	}
351 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
352 	simple_unlock(&mntid_slock);
353 }
354 
355 /*
356  * Make a 'unique' number from a mount type name.
357  */
358 long
359 makefstype(type)
360 	const char *type;
361 {
362 	long rv;
363 
364 	for (rv = 0; *type; type++) {
365 		rv <<= 2;
366 		rv ^= *type;
367 	}
368 	return rv;
369 }
370 
371 
372 /*
373  * Set vnode attributes to VNOVAL
374  */
375 void
376 vattr_null(vap)
377 	struct vattr *vap;
378 {
379 
380 	vap->va_type = VNON;
381 
382 	/*
383 	 * Assign individually so that it is safe even if size and
384 	 * sign of each member are varied.
385 	 */
386 	vap->va_mode = VNOVAL;
387 	vap->va_nlink = VNOVAL;
388 	vap->va_uid = VNOVAL;
389 	vap->va_gid = VNOVAL;
390 	vap->va_fsid = VNOVAL;
391 	vap->va_fileid = VNOVAL;
392 	vap->va_size = VNOVAL;
393 	vap->va_blocksize = VNOVAL;
394 	vap->va_atime.tv_sec =
395 	    vap->va_mtime.tv_sec =
396 	    vap->va_ctime.tv_sec = VNOVAL;
397 	vap->va_atime.tv_nsec =
398 	    vap->va_mtime.tv_nsec =
399 	    vap->va_ctime.tv_nsec = VNOVAL;
400 	vap->va_gen = VNOVAL;
401 	vap->va_flags = VNOVAL;
402 	vap->va_rdev = VNOVAL;
403 	vap->va_bytes = VNOVAL;
404 	vap->va_vaflags = 0;
405 }
406 
407 /*
408  * Routines having to do with the management of the vnode table.
409  */
410 extern int (**dead_vnodeop_p) __P((void *));
411 long numvnodes;
412 
413 /*
414  * Return the next vnode from the free list.
415  */
416 int
417 getnewvnode(tag, mp, vops, vpp)
418 	enum vtagtype tag;
419 	struct mount *mp;
420 	int (**vops) __P((void *));
421 	struct vnode **vpp;
422 {
423 	extern struct uvm_pagerops uvm_vnodeops;
424 	struct uvm_object *uobj;
425 	struct proc *p = curproc;	/* XXX */
426 	struct freelst *listhd;
427 	static int toggle;
428 	struct vnode *vp;
429 	int error = 0, tryalloc;
430 
431  try_again:
432 	if (mp) {
433 		/*
434 		 * Mark filesystem busy while we're creating a vnode.
435 		 * If unmount is in progress, this will wait; if the
436 		 * unmount succeeds (only if umount -f), this will
437 		 * return an error.  If the unmount fails, we'll keep
438 		 * going afterwards.
439 		 * (This puts the per-mount vnode list logically under
440 		 * the protection of the vfs_busy lock).
441 		 */
442 		error = vfs_busy(mp, LK_RECURSEFAIL, 0);
443 		if (error && error != EDEADLK)
444 			return error;
445 	}
446 
447 	/*
448 	 * We must choose whether to allocate a new vnode or recycle an
449 	 * existing one. The criterion for allocating a new one is that
450 	 * the total number of vnodes is less than the number desired or
451 	 * there are no vnodes on either free list. Generally we only
452 	 * want to recycle vnodes that have no buffers associated with
453 	 * them, so we look first on the vnode_free_list. If it is empty,
454 	 * we next consider vnodes with referencing buffers on the
455 	 * vnode_hold_list. The toggle ensures that half the time we
456 	 * will use a buffer from the vnode_hold_list, and half the time
457 	 * we will allocate a new one unless the list has grown to twice
458 	 * the desired size. We are reticent to recycle vnodes from the
459 	 * vnode_hold_list because we will lose the identity of all its
460 	 * referencing buffers.
461 	 */
462 
463 	vp = NULL;
464 
465 	simple_lock(&vnode_free_list_slock);
466 
467 	toggle ^= 1;
468 	if (numvnodes > 2 * desiredvnodes)
469 		toggle = 0;
470 
471 	tryalloc = numvnodes < desiredvnodes ||
472 	    (TAILQ_FIRST(&vnode_free_list) == NULL &&
473 	     (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle));
474 
475 	if (tryalloc &&
476 	    (vp = pool_get(&vnode_pool, PR_NOWAIT)) != NULL) {
477 		simple_unlock(&vnode_free_list_slock);
478 		memset(vp, 0, sizeof(*vp));
479 		simple_lock_init(&vp->v_interlock);
480 		uobj = &vp->v_uobj;
481 		uobj->pgops = &uvm_vnodeops;
482 		uobj->uo_npages = 0;
483 		TAILQ_INIT(&uobj->memq);
484 		numvnodes++;
485 	} else {
486 		if ((vp = TAILQ_FIRST(listhd = &vnode_free_list)) == NULL)
487 			vp = TAILQ_FIRST(listhd = &vnode_hold_list);
488 		for (; vp != NULL; vp = TAILQ_NEXT(vp, v_freelist)) {
489 			if (simple_lock_try(&vp->v_interlock)) {
490 				if ((vp->v_flag & VLAYER) == 0) {
491 					break;
492 				}
493 				if (VOP_ISLOCKED(vp) == 0)
494 					break;
495 				else
496 					simple_unlock(&vp->v_interlock);
497 			}
498 		}
499 		/*
500 		 * Unless this is a bad time of the month, at most
501 		 * the first NCPUS items on the free list are
502 		 * locked, so this is close enough to being empty.
503 		 */
504 		if (vp == NULLVP) {
505 			simple_unlock(&vnode_free_list_slock);
506 			if (mp && error != EDEADLK)
507 				vfs_unbusy(mp);
508 			if (tryalloc) {
509 				printf("WARNING: unable to allocate new "
510 				    "vnode, retrying...\n");
511 				(void) tsleep(&lbolt, PRIBIO, "newvn", hz);
512 				goto try_again;
513 			}
514 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
515 			*vpp = 0;
516 			return (ENFILE);
517 		}
518 		if (vp->v_usecount)
519 			panic("free vnode isn't, vp %p", vp);
520 		TAILQ_REMOVE(listhd, vp, v_freelist);
521 		/* see comment on why 0xdeadb is set at end of vgone (below) */
522 		vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
523 		simple_unlock(&vnode_free_list_slock);
524 		vp->v_lease = NULL;
525 
526 		if (vp->v_type != VBAD)
527 			vgonel(vp, p);
528 		else
529 			simple_unlock(&vp->v_interlock);
530 #ifdef DIAGNOSTIC
531 		if (vp->v_data || vp->v_uobj.uo_npages ||
532 		    TAILQ_FIRST(&vp->v_uobj.memq))
533 			panic("cleaned vnode isn't, vp %p", vp);
534 		if (vp->v_numoutput)
535 			panic("clean vnode has pending I/O's, vp %p", vp);
536 #endif
537 		KASSERT((vp->v_flag & VONWORKLST) == 0);
538 		vp->v_flag = 0;
539 		vp->v_socket = NULL;
540 #ifdef VERIFIED_EXEC
541 		vp->fp_status = FINGERPRINT_INVALID;
542 #endif
543 	}
544 	vp->v_type = VNON;
545 	vp->v_vnlock = &vp->v_lock;
546 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
547 	cache_purge(vp);
548 	vp->v_tag = tag;
549 	vp->v_op = vops;
550 	insmntque(vp, mp);
551 	*vpp = vp;
552 	vp->v_usecount = 1;
553 	vp->v_data = 0;
554 	simple_lock_init(&vp->v_uobj.vmobjlock);
555 
556 	/*
557 	 * initialize uvm_object within vnode.
558 	 */
559 
560 	uobj = &vp->v_uobj;
561 	KASSERT(uobj->pgops == &uvm_vnodeops);
562 	KASSERT(uobj->uo_npages == 0);
563 	KASSERT(TAILQ_FIRST(&uobj->memq) == NULL);
564 	vp->v_size = VSIZENOTSET;
565 
566 	if (mp && error != EDEADLK)
567 		vfs_unbusy(mp);
568 	return (0);
569 }
570 
571 /*
572  * This is really just the reverse of getnewvnode(). Needed for
573  * VFS_VGET functions who may need to push back a vnode in case
574  * of a locking race.
575  */
576 void
577 ungetnewvnode(vp)
578 	struct vnode *vp;
579 {
580 #ifdef DIAGNOSTIC
581 	if (vp->v_usecount != 1)
582 		panic("ungetnewvnode: busy vnode");
583 #endif
584 	vp->v_usecount--;
585 	insmntque(vp, NULL);
586 	vp->v_type = VBAD;
587 
588 	simple_lock(&vp->v_interlock);
589 	/*
590 	 * Insert at head of LRU list
591 	 */
592 	simple_lock(&vnode_free_list_slock);
593 	if (vp->v_holdcnt > 0)
594 		TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist);
595 	else
596 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
597 	simple_unlock(&vnode_free_list_slock);
598 	simple_unlock(&vp->v_interlock);
599 }
600 
601 /*
602  * Move a vnode from one mount queue to another.
603  */
604 void
605 insmntque(vp, mp)
606 	struct vnode *vp;
607 	struct mount *mp;
608 {
609 
610 #ifdef DIAGNOSTIC
611 	if ((mp != NULL) &&
612 	    (mp->mnt_flag & MNT_UNMOUNT) &&
613 	    !(mp->mnt_flag & MNT_SOFTDEP) &&
614 	    vp->v_tag != VT_VFS) {
615 		panic("insmntque into dying filesystem");
616 	}
617 #endif
618 
619 	simple_lock(&mntvnode_slock);
620 	/*
621 	 * Delete from old mount point vnode list, if on one.
622 	 */
623 	if (vp->v_mount != NULL)
624 		LIST_REMOVE(vp, v_mntvnodes);
625 	/*
626 	 * Insert into list of vnodes for the new mount point, if available.
627 	 */
628 	if ((vp->v_mount = mp) != NULL)
629 		LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
630 	simple_unlock(&mntvnode_slock);
631 }
632 
633 /*
634  * Update outstanding I/O count and do wakeup if requested.
635  */
636 void
637 vwakeup(bp)
638 	struct buf *bp;
639 {
640 	struct vnode *vp;
641 
642 	if ((vp = bp->b_vp) != NULL) {
643 		if (--vp->v_numoutput < 0)
644 			panic("vwakeup: neg numoutput, vp %p", vp);
645 		if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
646 			vp->v_flag &= ~VBWAIT;
647 			wakeup((caddr_t)&vp->v_numoutput);
648 		}
649 	}
650 }
651 
652 /*
653  * Flush out and invalidate all buffers associated with a vnode.
654  * Called with the underlying vnode locked, which should prevent new dirty
655  * buffers from being queued.
656  */
657 int
658 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
659 	struct vnode *vp;
660 	int flags;
661 	struct ucred *cred;
662 	struct proc *p;
663 	int slpflag, slptimeo;
664 {
665 	struct buf *bp, *nbp;
666 	int s, error;
667 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
668 		(flags & V_SAVE ? PGO_CLEANIT : 0);
669 
670 	/* XXXUBC this doesn't look at flags or slp* */
671 	simple_lock(&vp->v_interlock);
672 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
673 	if (error) {
674 		return error;
675 	}
676 
677 	if (flags & V_SAVE) {
678 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p);
679 		if (error)
680 		        return (error);
681 #ifdef DIAGNOSTIC
682 		s = splbio();
683 		if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd))
684 		        panic("vinvalbuf: dirty bufs, vp %p", vp);
685 		splx(s);
686 #endif
687 	}
688 
689 	s = splbio();
690 
691 restart:
692 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
693 		nbp = LIST_NEXT(bp, b_vnbufs);
694 		if (bp->b_flags & B_BUSY) {
695 			bp->b_flags |= B_WANTED;
696 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
697 			    "vinvalbuf", slptimeo);
698 			if (error) {
699 				splx(s);
700 				return (error);
701 			}
702 			goto restart;
703 		}
704 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
705 		brelse(bp);
706 	}
707 
708 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
709 		nbp = LIST_NEXT(bp, b_vnbufs);
710 		if (bp->b_flags & B_BUSY) {
711 			bp->b_flags |= B_WANTED;
712 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
713 			    "vinvalbuf", slptimeo);
714 			if (error) {
715 				splx(s);
716 				return (error);
717 			}
718 			goto restart;
719 		}
720 		/*
721 		 * XXX Since there are no node locks for NFS, I believe
722 		 * there is a slight chance that a delayed write will
723 		 * occur while sleeping just above, so check for it.
724 		 */
725 		if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
726 #ifdef DEBUG
727 			printf("buffer still DELWRI\n");
728 #endif
729 			bp->b_flags |= B_BUSY | B_VFLUSH;
730 			VOP_BWRITE(bp);
731 			goto restart;
732 		}
733 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
734 		brelse(bp);
735 	}
736 
737 #ifdef DIAGNOSTIC
738 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
739 		panic("vinvalbuf: flush failed, vp %p", vp);
740 #endif
741 
742 	splx(s);
743 
744 	return (0);
745 }
746 
747 /*
748  * Destroy any in core blocks past the truncation length.
749  * Called with the underlying vnode locked, which should prevent new dirty
750  * buffers from being queued.
751  */
752 int
753 vtruncbuf(vp, lbn, slpflag, slptimeo)
754 	struct vnode *vp;
755 	daddr_t lbn;
756 	int slpflag, slptimeo;
757 {
758 	struct buf *bp, *nbp;
759 	int s, error;
760 	voff_t off;
761 
762 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
763 	simple_lock(&vp->v_interlock);
764 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
765 	if (error) {
766 		return error;
767 	}
768 
769 	s = splbio();
770 
771 restart:
772 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
773 		nbp = LIST_NEXT(bp, b_vnbufs);
774 		if (bp->b_lblkno < lbn)
775 			continue;
776 		if (bp->b_flags & B_BUSY) {
777 			bp->b_flags |= B_WANTED;
778 			error = tsleep(bp, slpflag | (PRIBIO + 1),
779 			    "vtruncbuf", slptimeo);
780 			if (error) {
781 				splx(s);
782 				return (error);
783 			}
784 			goto restart;
785 		}
786 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
787 		brelse(bp);
788 	}
789 
790 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
791 		nbp = LIST_NEXT(bp, b_vnbufs);
792 		if (bp->b_lblkno < lbn)
793 			continue;
794 		if (bp->b_flags & B_BUSY) {
795 			bp->b_flags |= B_WANTED;
796 			error = tsleep(bp, slpflag | (PRIBIO + 1),
797 			    "vtruncbuf", slptimeo);
798 			if (error) {
799 				splx(s);
800 				return (error);
801 			}
802 			goto restart;
803 		}
804 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
805 		brelse(bp);
806 	}
807 
808 	splx(s);
809 
810 	return (0);
811 }
812 
813 void
814 vflushbuf(vp, sync)
815 	struct vnode *vp;
816 	int sync;
817 {
818 	struct buf *bp, *nbp;
819 	int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0);
820 	int s;
821 
822 	simple_lock(&vp->v_interlock);
823 	(void) VOP_PUTPAGES(vp, 0, 0, flags);
824 
825 loop:
826 	s = splbio();
827 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
828 		nbp = LIST_NEXT(bp, b_vnbufs);
829 		if ((bp->b_flags & B_BUSY))
830 			continue;
831 		if ((bp->b_flags & B_DELWRI) == 0)
832 			panic("vflushbuf: not dirty, bp %p", bp);
833 		bp->b_flags |= B_BUSY | B_VFLUSH;
834 		splx(s);
835 		/*
836 		 * Wait for I/O associated with indirect blocks to complete,
837 		 * since there is no way to quickly wait for them below.
838 		 */
839 		if (bp->b_vp == vp || sync == 0)
840 			(void) bawrite(bp);
841 		else
842 			(void) bwrite(bp);
843 		goto loop;
844 	}
845 	if (sync == 0) {
846 		splx(s);
847 		return;
848 	}
849 	while (vp->v_numoutput) {
850 		vp->v_flag |= VBWAIT;
851 		tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0);
852 	}
853 	splx(s);
854 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
855 		vprint("vflushbuf: dirty", vp);
856 		goto loop;
857 	}
858 }
859 
860 /*
861  * Associate a buffer with a vnode.
862  */
863 void
864 bgetvp(vp, bp)
865 	struct vnode *vp;
866 	struct buf *bp;
867 {
868 	int s;
869 
870 	if (bp->b_vp)
871 		panic("bgetvp: not free, bp %p", bp);
872 	VHOLD(vp);
873 	s = splbio();
874 	bp->b_vp = vp;
875 	if (vp->v_type == VBLK || vp->v_type == VCHR)
876 		bp->b_dev = vp->v_rdev;
877 	else
878 		bp->b_dev = NODEV;
879 	/*
880 	 * Insert onto list for new vnode.
881 	 */
882 	bufinsvn(bp, &vp->v_cleanblkhd);
883 	splx(s);
884 }
885 
886 /*
887  * Disassociate a buffer from a vnode.
888  */
889 void
890 brelvp(bp)
891 	struct buf *bp;
892 {
893 	struct vnode *vp;
894 	int s;
895 
896 	if (bp->b_vp == NULL)
897 		panic("brelvp: vp NULL, bp %p", bp);
898 
899 	s = splbio();
900 	vp = bp->b_vp;
901 	/*
902 	 * Delete from old vnode list, if on one.
903 	 */
904 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
905 		bufremvn(bp);
906 
907 	if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_flag & VONWORKLST) &&
908 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
909 		vp->v_flag &= ~VONWORKLST;
910 		LIST_REMOVE(vp, v_synclist);
911 	}
912 
913 	bp->b_vp = NULL;
914 	HOLDRELE(vp);
915 	splx(s);
916 }
917 
918 /*
919  * Reassign a buffer from one vnode to another.
920  * Used to assign file specific control information
921  * (indirect blocks) to the vnode to which they belong.
922  *
923  * This function must be called at splbio().
924  */
925 void
926 reassignbuf(bp, newvp)
927 	struct buf *bp;
928 	struct vnode *newvp;
929 {
930 	struct buflists *listheadp;
931 	int delay;
932 
933 	/*
934 	 * Delete from old vnode list, if on one.
935 	 */
936 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
937 		bufremvn(bp);
938 	/*
939 	 * If dirty, put on list of dirty buffers;
940 	 * otherwise insert onto list of clean buffers.
941 	 */
942 	if ((bp->b_flags & B_DELWRI) == 0) {
943 		listheadp = &newvp->v_cleanblkhd;
944 		if (TAILQ_EMPTY(&newvp->v_uobj.memq) &&
945 		    (newvp->v_flag & VONWORKLST) &&
946 		    LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
947 			newvp->v_flag &= ~VONWORKLST;
948 			LIST_REMOVE(newvp, v_synclist);
949 		}
950 	} else {
951 		listheadp = &newvp->v_dirtyblkhd;
952 		if ((newvp->v_flag & VONWORKLST) == 0) {
953 			switch (newvp->v_type) {
954 			case VDIR:
955 				delay = dirdelay;
956 				break;
957 			case VBLK:
958 				if (newvp->v_specmountpoint != NULL) {
959 					delay = metadelay;
960 					break;
961 				}
962 				/* fall through */
963 			default:
964 				delay = filedelay;
965 				break;
966 			}
967 			if (!newvp->v_mount ||
968 			    (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0)
969 				vn_syncer_add_to_worklist(newvp, delay);
970 		}
971 	}
972 	bufinsvn(bp, listheadp);
973 }
974 
975 /*
976  * Create a vnode for a block device.
977  * Used for root filesystem and swap areas.
978  * Also used for memory file system special devices.
979  */
980 int
981 bdevvp(dev, vpp)
982 	dev_t dev;
983 	struct vnode **vpp;
984 {
985 
986 	return (getdevvp(dev, vpp, VBLK));
987 }
988 
989 /*
990  * Create a vnode for a character device.
991  * Used for kernfs and some console handling.
992  */
993 int
994 cdevvp(dev, vpp)
995 	dev_t dev;
996 	struct vnode **vpp;
997 {
998 
999 	return (getdevvp(dev, vpp, VCHR));
1000 }
1001 
1002 /*
1003  * Create a vnode for a device.
1004  * Used by bdevvp (block device) for root file system etc.,
1005  * and by cdevvp (character device) for console and kernfs.
1006  */
1007 int
1008 getdevvp(dev, vpp, type)
1009 	dev_t dev;
1010 	struct vnode **vpp;
1011 	enum vtype type;
1012 {
1013 	struct vnode *vp;
1014 	struct vnode *nvp;
1015 	int error;
1016 
1017 	if (dev == NODEV) {
1018 		*vpp = NULLVP;
1019 		return (0);
1020 	}
1021 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
1022 	if (error) {
1023 		*vpp = NULLVP;
1024 		return (error);
1025 	}
1026 	vp = nvp;
1027 	vp->v_type = type;
1028 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
1029 		vput(vp);
1030 		vp = nvp;
1031 	}
1032 	*vpp = vp;
1033 	return (0);
1034 }
1035 
1036 /*
1037  * Check to see if the new vnode represents a special device
1038  * for which we already have a vnode (either because of
1039  * bdevvp() or because of a different vnode representing
1040  * the same block device). If such an alias exists, deallocate
1041  * the existing contents and return the aliased vnode. The
1042  * caller is responsible for filling it with its new contents.
1043  */
1044 struct vnode *
1045 checkalias(nvp, nvp_rdev, mp)
1046 	struct vnode *nvp;
1047 	dev_t nvp_rdev;
1048 	struct mount *mp;
1049 {
1050 	struct proc *p = curproc;       /* XXX */
1051 	struct vnode *vp;
1052 	struct vnode **vpp;
1053 
1054 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1055 		return (NULLVP);
1056 
1057 	vpp = &speclisth[SPECHASH(nvp_rdev)];
1058 loop:
1059 	simple_lock(&spechash_slock);
1060 	for (vp = *vpp; vp; vp = vp->v_specnext) {
1061 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1062 			continue;
1063 		/*
1064 		 * Alias, but not in use, so flush it out.
1065 		 */
1066 		simple_lock(&vp->v_interlock);
1067 		if (vp->v_usecount == 0) {
1068 			simple_unlock(&spechash_slock);
1069 			vgonel(vp, p);
1070 			goto loop;
1071 		}
1072 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
1073 			simple_unlock(&spechash_slock);
1074 			goto loop;
1075 		}
1076 		break;
1077 	}
1078 	if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) {
1079 		MALLOC(nvp->v_specinfo, struct specinfo *,
1080 			sizeof(struct specinfo), M_VNODE, M_NOWAIT);
1081 		/* XXX Erg. */
1082 		if (nvp->v_specinfo == NULL) {
1083 			simple_unlock(&spechash_slock);
1084 			uvm_wait("checkalias");
1085 			goto loop;
1086 		}
1087 
1088 		nvp->v_rdev = nvp_rdev;
1089 		nvp->v_hashchain = vpp;
1090 		nvp->v_specnext = *vpp;
1091 		nvp->v_specmountpoint = NULL;
1092 		simple_unlock(&spechash_slock);
1093 		nvp->v_speclockf = NULL;
1094 		*vpp = nvp;
1095 		if (vp != NULLVP) {
1096 			nvp->v_flag |= VALIASED;
1097 			vp->v_flag |= VALIASED;
1098 			vput(vp);
1099 		}
1100 		return (NULLVP);
1101 	}
1102 	simple_unlock(&spechash_slock);
1103 	VOP_UNLOCK(vp, 0);
1104 	simple_lock(&vp->v_interlock);
1105 	vclean(vp, 0, p);
1106 	vp->v_op = nvp->v_op;
1107 	vp->v_tag = nvp->v_tag;
1108 	vp->v_vnlock = &vp->v_lock;
1109 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
1110 	nvp->v_type = VNON;
1111 	insmntque(vp, mp);
1112 	return (vp);
1113 }
1114 
1115 /*
1116  * Grab a particular vnode from the free list, increment its
1117  * reference count and lock it. If the vnode lock bit is set the
1118  * vnode is being eliminated in vgone. In that case, we can not
1119  * grab the vnode, so the process is awakened when the transition is
1120  * completed, and an error returned to indicate that the vnode is no
1121  * longer usable (possibly having been changed to a new file system type).
1122  */
1123 int
1124 vget(vp, flags)
1125 	struct vnode *vp;
1126 	int flags;
1127 {
1128 	int error;
1129 
1130 	/*
1131 	 * If the vnode is in the process of being cleaned out for
1132 	 * another use, we wait for the cleaning to finish and then
1133 	 * return failure. Cleaning is determined by checking that
1134 	 * the VXLOCK flag is set.
1135 	 */
1136 
1137 	if ((flags & LK_INTERLOCK) == 0)
1138 		simple_lock(&vp->v_interlock);
1139 	if (vp->v_flag & VXLOCK) {
1140 		if (flags & LK_NOWAIT) {
1141 			simple_unlock(&vp->v_interlock);
1142 			return EBUSY;
1143 		}
1144 		vp->v_flag |= VXWANT;
1145 		ltsleep(vp, PINOD|PNORELOCK, "vget", 0, &vp->v_interlock);
1146 		return (ENOENT);
1147 	}
1148 	if (vp->v_usecount == 0) {
1149 		simple_lock(&vnode_free_list_slock);
1150 		if (vp->v_holdcnt > 0)
1151 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1152 		else
1153 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1154 		simple_unlock(&vnode_free_list_slock);
1155 	}
1156 	vp->v_usecount++;
1157 #ifdef DIAGNOSTIC
1158 	if (vp->v_usecount == 0) {
1159 		vprint("vget", vp);
1160 		panic("vget: usecount overflow, vp %p", vp);
1161 	}
1162 #endif
1163 	if (flags & LK_TYPE_MASK) {
1164 		if ((error = vn_lock(vp, flags | LK_INTERLOCK))) {
1165 			/*
1166 			 * must expand vrele here because we do not want
1167 			 * to call VOP_INACTIVE if the reference count
1168 			 * drops back to zero since it was never really
1169 			 * active. We must remove it from the free list
1170 			 * before sleeping so that multiple processes do
1171 			 * not try to recycle it.
1172 			 */
1173 			simple_lock(&vp->v_interlock);
1174 			vp->v_usecount--;
1175 			if (vp->v_usecount > 0) {
1176 				simple_unlock(&vp->v_interlock);
1177 				return (error);
1178 			}
1179 			/*
1180 			 * insert at tail of LRU list
1181 			 */
1182 			simple_lock(&vnode_free_list_slock);
1183 			if (vp->v_holdcnt > 0)
1184 				TAILQ_INSERT_TAIL(&vnode_hold_list, vp,
1185 				    v_freelist);
1186 			else
1187 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
1188 				    v_freelist);
1189 			simple_unlock(&vnode_free_list_slock);
1190 			simple_unlock(&vp->v_interlock);
1191 		}
1192 		return (error);
1193 	}
1194 	simple_unlock(&vp->v_interlock);
1195 	return (0);
1196 }
1197 
1198 /*
1199  * vput(), just unlock and vrele()
1200  */
1201 void
1202 vput(vp)
1203 	struct vnode *vp;
1204 {
1205 	struct proc *p = curproc;	/* XXX */
1206 
1207 #ifdef DIAGNOSTIC
1208 	if (vp == NULL)
1209 		panic("vput: null vp");
1210 #endif
1211 	simple_lock(&vp->v_interlock);
1212 	vp->v_usecount--;
1213 	if (vp->v_usecount > 0) {
1214 		simple_unlock(&vp->v_interlock);
1215 		VOP_UNLOCK(vp, 0);
1216 		return;
1217 	}
1218 #ifdef DIAGNOSTIC
1219 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1220 		vprint("vput: bad ref count", vp);
1221 		panic("vput: ref cnt");
1222 	}
1223 #endif
1224 	/*
1225 	 * Insert at tail of LRU list.
1226 	 */
1227 	simple_lock(&vnode_free_list_slock);
1228 	if (vp->v_holdcnt > 0)
1229 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1230 	else
1231 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1232 	simple_unlock(&vnode_free_list_slock);
1233 	if (vp->v_flag & VEXECMAP) {
1234 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1235 		uvmexp.filepages += vp->v_uobj.uo_npages;
1236 	}
1237 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1238 	simple_unlock(&vp->v_interlock);
1239 	VOP_INACTIVE(vp, p);
1240 }
1241 
1242 /*
1243  * Vnode release.
1244  * If count drops to zero, call inactive routine and return to freelist.
1245  */
1246 void
1247 vrele(vp)
1248 	struct vnode *vp;
1249 {
1250 	struct proc *p = curproc;	/* XXX */
1251 
1252 #ifdef DIAGNOSTIC
1253 	if (vp == NULL)
1254 		panic("vrele: null vp");
1255 #endif
1256 	simple_lock(&vp->v_interlock);
1257 	vp->v_usecount--;
1258 	if (vp->v_usecount > 0) {
1259 		simple_unlock(&vp->v_interlock);
1260 		return;
1261 	}
1262 #ifdef DIAGNOSTIC
1263 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1264 		vprint("vrele: bad ref count", vp);
1265 		panic("vrele: ref cnt vp %p", vp);
1266 	}
1267 #endif
1268 	/*
1269 	 * Insert at tail of LRU list.
1270 	 */
1271 	simple_lock(&vnode_free_list_slock);
1272 	if (vp->v_holdcnt > 0)
1273 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1274 	else
1275 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1276 	simple_unlock(&vnode_free_list_slock);
1277 	if (vp->v_flag & VEXECMAP) {
1278 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1279 		uvmexp.filepages += vp->v_uobj.uo_npages;
1280 	}
1281 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1282 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0)
1283 		VOP_INACTIVE(vp, p);
1284 }
1285 
1286 #ifdef DIAGNOSTIC
1287 /*
1288  * Page or buffer structure gets a reference.
1289  */
1290 void
1291 vhold(vp)
1292 	struct vnode *vp;
1293 {
1294 
1295 	/*
1296 	 * If it is on the freelist and the hold count is currently
1297 	 * zero, move it to the hold list. The test of the back
1298 	 * pointer and the use reference count of zero is because
1299 	 * it will be removed from a free list by getnewvnode,
1300 	 * but will not have its reference count incremented until
1301 	 * after calling vgone. If the reference count were
1302 	 * incremented first, vgone would (incorrectly) try to
1303 	 * close the previous instance of the underlying object.
1304 	 * So, the back pointer is explicitly set to `0xdeadb' in
1305 	 * getnewvnode after removing it from a freelist to ensure
1306 	 * that we do not try to move it here.
1307 	 */
1308   	simple_lock(&vp->v_interlock);
1309 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1310 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1311 		simple_lock(&vnode_free_list_slock);
1312 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1313 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1314 		simple_unlock(&vnode_free_list_slock);
1315 	}
1316 	vp->v_holdcnt++;
1317 	simple_unlock(&vp->v_interlock);
1318 }
1319 
1320 /*
1321  * Page or buffer structure frees a reference.
1322  */
1323 void
1324 holdrele(vp)
1325 	struct vnode *vp;
1326 {
1327 
1328 	simple_lock(&vp->v_interlock);
1329 	if (vp->v_holdcnt <= 0)
1330 		panic("holdrele: holdcnt vp %p", vp);
1331 	vp->v_holdcnt--;
1332 
1333 	/*
1334 	 * If it is on the holdlist and the hold count drops to
1335 	 * zero, move it to the free list. The test of the back
1336 	 * pointer and the use reference count of zero is because
1337 	 * it will be removed from a free list by getnewvnode,
1338 	 * but will not have its reference count incremented until
1339 	 * after calling vgone. If the reference count were
1340 	 * incremented first, vgone would (incorrectly) try to
1341 	 * close the previous instance of the underlying object.
1342 	 * So, the back pointer is explicitly set to `0xdeadb' in
1343 	 * getnewvnode after removing it from a freelist to ensure
1344 	 * that we do not try to move it here.
1345 	 */
1346 
1347 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1348 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1349 		simple_lock(&vnode_free_list_slock);
1350 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1351 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1352 		simple_unlock(&vnode_free_list_slock);
1353 	}
1354 	simple_unlock(&vp->v_interlock);
1355 }
1356 
1357 /*
1358  * Vnode reference.
1359  */
1360 void
1361 vref(vp)
1362 	struct vnode *vp;
1363 {
1364 
1365 	simple_lock(&vp->v_interlock);
1366 	if (vp->v_usecount <= 0)
1367 		panic("vref used where vget required, vp %p", vp);
1368 	vp->v_usecount++;
1369 #ifdef DIAGNOSTIC
1370 	if (vp->v_usecount == 0) {
1371 		vprint("vref", vp);
1372 		panic("vref: usecount overflow, vp %p", vp);
1373 	}
1374 #endif
1375 	simple_unlock(&vp->v_interlock);
1376 }
1377 #endif /* DIAGNOSTIC */
1378 
1379 /*
1380  * Remove any vnodes in the vnode table belonging to mount point mp.
1381  *
1382  * If FORCECLOSE is not specified, there should not be any active ones,
1383  * return error if any are found (nb: this is a user error, not a
1384  * system error). If FORCECLOSE is specified, detach any active vnodes
1385  * that are found.
1386  *
1387  * If WRITECLOSE is set, only flush out regular file vnodes open for
1388  * writing.
1389  *
1390  * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped.
1391  */
1392 #ifdef DEBUG
1393 int busyprt = 0;	/* print out busy vnodes */
1394 struct ctldebug debug1 = { "busyprt", &busyprt };
1395 #endif
1396 
1397 int
1398 vflush(mp, skipvp, flags)
1399 	struct mount *mp;
1400 	struct vnode *skipvp;
1401 	int flags;
1402 {
1403 	struct proc *p = curproc;	/* XXX */
1404 	struct vnode *vp, *nvp;
1405 	int busy = 0;
1406 
1407 	simple_lock(&mntvnode_slock);
1408 loop:
1409 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1410 		if (vp->v_mount != mp)
1411 			goto loop;
1412 		nvp = LIST_NEXT(vp, v_mntvnodes);
1413 		/*
1414 		 * Skip over a selected vnode.
1415 		 */
1416 		if (vp == skipvp)
1417 			continue;
1418 		simple_lock(&vp->v_interlock);
1419 		/*
1420 		 * Skip over a vnodes marked VSYSTEM.
1421 		 */
1422 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1423 			simple_unlock(&vp->v_interlock);
1424 			continue;
1425 		}
1426 		/*
1427 		 * If WRITECLOSE is set, only flush out regular file
1428 		 * vnodes open for writing.
1429 		 */
1430 		if ((flags & WRITECLOSE) &&
1431 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1432 			simple_unlock(&vp->v_interlock);
1433 			continue;
1434 		}
1435 		/*
1436 		 * With v_usecount == 0, all we need to do is clear
1437 		 * out the vnode data structures and we are done.
1438 		 */
1439 		if (vp->v_usecount == 0) {
1440 			simple_unlock(&mntvnode_slock);
1441 			vgonel(vp, p);
1442 			simple_lock(&mntvnode_slock);
1443 			continue;
1444 		}
1445 		/*
1446 		 * If FORCECLOSE is set, forcibly close the vnode.
1447 		 * For block or character devices, revert to an
1448 		 * anonymous device. For all other files, just kill them.
1449 		 */
1450 		if (flags & FORCECLOSE) {
1451 			simple_unlock(&mntvnode_slock);
1452 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1453 				vgonel(vp, p);
1454 			} else {
1455 				vclean(vp, 0, p);
1456 				vp->v_op = spec_vnodeop_p;
1457 				insmntque(vp, (struct mount *)0);
1458 			}
1459 			simple_lock(&mntvnode_slock);
1460 			continue;
1461 		}
1462 #ifdef DEBUG
1463 		if (busyprt)
1464 			vprint("vflush: busy vnode", vp);
1465 #endif
1466 		simple_unlock(&vp->v_interlock);
1467 		busy++;
1468 	}
1469 	simple_unlock(&mntvnode_slock);
1470 	if (busy)
1471 		return (EBUSY);
1472 	return (0);
1473 }
1474 
1475 /*
1476  * Disassociate the underlying file system from a vnode.
1477  */
1478 void
1479 vclean(vp, flags, p)
1480 	struct vnode *vp;
1481 	int flags;
1482 	struct proc *p;
1483 {
1484 	int active;
1485 
1486 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1487 
1488 	/*
1489 	 * Check to see if the vnode is in use.
1490 	 * If so we have to reference it before we clean it out
1491 	 * so that its count cannot fall to zero and generate a
1492 	 * race against ourselves to recycle it.
1493 	 */
1494 
1495 	if ((active = vp->v_usecount) != 0) {
1496 		vp->v_usecount++;
1497 #ifdef DIAGNOSTIC
1498 		if (vp->v_usecount == 0) {
1499 			vprint("vclean", vp);
1500 			panic("vclean: usecount overflow");
1501 		}
1502 #endif
1503 	}
1504 
1505 	/*
1506 	 * Prevent the vnode from being recycled or
1507 	 * brought into use while we clean it out.
1508 	 */
1509 	if (vp->v_flag & VXLOCK)
1510 		panic("vclean: deadlock, vp %p", vp);
1511 	vp->v_flag |= VXLOCK;
1512 	if (vp->v_flag & VEXECMAP) {
1513 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1514 		uvmexp.filepages += vp->v_uobj.uo_npages;
1515 	}
1516 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1517 
1518 	/*
1519 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1520 	 * have the object locked while it cleans it out. The VOP_LOCK
1521 	 * ensures that the VOP_INACTIVE routine is done with its work.
1522 	 * For active vnodes, it ensures that no other activity can
1523 	 * occur while the underlying object is being cleaned out.
1524 	 */
1525 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK);
1526 
1527 	/*
1528 	 * Clean out any cached data associated with the vnode.
1529 	 */
1530 	if (flags & DOCLOSE) {
1531 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1532 		KASSERT((vp->v_flag & VONWORKLST) == 0);
1533 	}
1534 	LOCK_ASSERT(!simple_lock_held(&vp->v_interlock));
1535 
1536 	/*
1537 	 * If purging an active vnode, it must be closed and
1538 	 * deactivated before being reclaimed. Note that the
1539 	 * VOP_INACTIVE will unlock the vnode.
1540 	 */
1541 	if (active) {
1542 		if (flags & DOCLOSE)
1543 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL);
1544 		VOP_INACTIVE(vp, p);
1545 	} else {
1546 		/*
1547 		 * Any other processes trying to obtain this lock must first
1548 		 * wait for VXLOCK to clear, then call the new lock operation.
1549 		 */
1550 		VOP_UNLOCK(vp, 0);
1551 	}
1552 	/*
1553 	 * Reclaim the vnode.
1554 	 */
1555 	if (VOP_RECLAIM(vp, p))
1556 		panic("vclean: cannot reclaim, vp %p", vp);
1557 	if (active) {
1558 		/*
1559 		 * Inline copy of vrele() since VOP_INACTIVE
1560 		 * has already been called.
1561 		 */
1562 		simple_lock(&vp->v_interlock);
1563 		if (--vp->v_usecount <= 0) {
1564 #ifdef DIAGNOSTIC
1565 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1566 				vprint("vclean: bad ref count", vp);
1567 				panic("vclean: ref cnt");
1568 			}
1569 #endif
1570 			/*
1571 			 * Insert at tail of LRU list.
1572 			 */
1573 
1574 			simple_unlock(&vp->v_interlock);
1575 			simple_lock(&vnode_free_list_slock);
1576 #ifdef DIAGNOSTIC
1577 			if (vp->v_holdcnt > 0)
1578 				panic("vclean: not clean, vp %p", vp);
1579 #endif
1580 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1581 			simple_unlock(&vnode_free_list_slock);
1582 		} else
1583 			simple_unlock(&vp->v_interlock);
1584 	}
1585 
1586 	KASSERT(vp->v_uobj.uo_npages == 0);
1587 	cache_purge(vp);
1588 
1589 	/*
1590 	 * Done with purge, notify sleepers of the grim news.
1591 	 */
1592 	vp->v_op = dead_vnodeop_p;
1593 	vp->v_tag = VT_NON;
1594 	simple_lock(&vp->v_interlock);
1595 	VN_KNOTE(vp, NOTE_REVOKE);	/* FreeBSD has this in vn_pollgone() */
1596 	vp->v_flag &= ~VXLOCK;
1597 	if (vp->v_flag & VXWANT) {
1598 		vp->v_flag &= ~VXWANT;
1599 		simple_unlock(&vp->v_interlock);
1600 		wakeup((caddr_t)vp);
1601 	} else
1602 		simple_unlock(&vp->v_interlock);
1603 }
1604 
1605 /*
1606  * Recycle an unused vnode to the front of the free list.
1607  * Release the passed interlock if the vnode will be recycled.
1608  */
1609 int
1610 vrecycle(vp, inter_lkp, p)
1611 	struct vnode *vp;
1612 	struct simplelock *inter_lkp;
1613 	struct proc *p;
1614 {
1615 
1616 	simple_lock(&vp->v_interlock);
1617 	if (vp->v_usecount == 0) {
1618 		if (inter_lkp)
1619 			simple_unlock(inter_lkp);
1620 		vgonel(vp, p);
1621 		return (1);
1622 	}
1623 	simple_unlock(&vp->v_interlock);
1624 	return (0);
1625 }
1626 
1627 /*
1628  * Eliminate all activity associated with a vnode
1629  * in preparation for reuse.
1630  */
1631 void
1632 vgone(vp)
1633 	struct vnode *vp;
1634 {
1635 	struct proc *p = curproc;	/* XXX */
1636 
1637 	simple_lock(&vp->v_interlock);
1638 	vgonel(vp, p);
1639 }
1640 
1641 /*
1642  * vgone, with the vp interlock held.
1643  */
1644 void
1645 vgonel(vp, p)
1646 	struct vnode *vp;
1647 	struct proc *p;
1648 {
1649 	struct vnode *vq;
1650 	struct vnode *vx;
1651 
1652 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1653 
1654 	/*
1655 	 * If a vgone (or vclean) is already in progress,
1656 	 * wait until it is done and return.
1657 	 */
1658 
1659 	if (vp->v_flag & VXLOCK) {
1660 		vp->v_flag |= VXWANT;
1661 		ltsleep(vp, PINOD | PNORELOCK, "vgone", 0, &vp->v_interlock);
1662 		return;
1663 	}
1664 
1665 	/*
1666 	 * Clean out the filesystem specific data.
1667 	 */
1668 
1669 	vclean(vp, DOCLOSE, p);
1670 	KASSERT((vp->v_flag & VONWORKLST) == 0);
1671 
1672 	/*
1673 	 * Delete from old mount point vnode list, if on one.
1674 	 */
1675 
1676 	if (vp->v_mount != NULL)
1677 		insmntque(vp, (struct mount *)0);
1678 
1679 	/*
1680 	 * If special device, remove it from special device alias list.
1681 	 * if it is on one.
1682 	 */
1683 
1684 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1685 		simple_lock(&spechash_slock);
1686 		if (vp->v_hashchain != NULL) {
1687 			if (*vp->v_hashchain == vp) {
1688 				*vp->v_hashchain = vp->v_specnext;
1689 			} else {
1690 				for (vq = *vp->v_hashchain; vq;
1691 							vq = vq->v_specnext) {
1692 					if (vq->v_specnext != vp)
1693 						continue;
1694 					vq->v_specnext = vp->v_specnext;
1695 					break;
1696 				}
1697 				if (vq == NULL)
1698 					panic("missing bdev");
1699 			}
1700 			if (vp->v_flag & VALIASED) {
1701 				vx = NULL;
1702 				for (vq = *vp->v_hashchain; vq;
1703 							vq = vq->v_specnext) {
1704 					if (vq->v_rdev != vp->v_rdev ||
1705 					    vq->v_type != vp->v_type)
1706 						continue;
1707 					if (vx)
1708 						break;
1709 					vx = vq;
1710 				}
1711 				if (vx == NULL)
1712 					panic("missing alias");
1713 				if (vq == NULL)
1714 					vx->v_flag &= ~VALIASED;
1715 				vp->v_flag &= ~VALIASED;
1716 			}
1717 		}
1718 		simple_unlock(&spechash_slock);
1719 		FREE(vp->v_specinfo, M_VNODE);
1720 		vp->v_specinfo = NULL;
1721 	}
1722 
1723 	/*
1724 	 * If it is on the freelist and not already at the head,
1725 	 * move it to the head of the list. The test of the back
1726 	 * pointer and the reference count of zero is because
1727 	 * it will be removed from the free list by getnewvnode,
1728 	 * but will not have its reference count incremented until
1729 	 * after calling vgone. If the reference count were
1730 	 * incremented first, vgone would (incorrectly) try to
1731 	 * close the previous instance of the underlying object.
1732 	 * So, the back pointer is explicitly set to `0xdeadb' in
1733 	 * getnewvnode after removing it from the freelist to ensure
1734 	 * that we do not try to move it here.
1735 	 */
1736 
1737 	if (vp->v_usecount == 0) {
1738 		simple_lock(&vnode_free_list_slock);
1739 		if (vp->v_holdcnt > 0)
1740 			panic("vgonel: not clean, vp %p", vp);
1741 		if (vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb &&
1742 		    TAILQ_FIRST(&vnode_free_list) != vp) {
1743 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1744 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1745 		}
1746 		simple_unlock(&vnode_free_list_slock);
1747 	}
1748 	vp->v_type = VBAD;
1749 }
1750 
1751 /*
1752  * Lookup a vnode by device number.
1753  */
1754 int
1755 vfinddev(dev, type, vpp)
1756 	dev_t dev;
1757 	enum vtype type;
1758 	struct vnode **vpp;
1759 {
1760 	struct vnode *vp;
1761 	int rc = 0;
1762 
1763 	simple_lock(&spechash_slock);
1764 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1765 		if (dev != vp->v_rdev || type != vp->v_type)
1766 			continue;
1767 		*vpp = vp;
1768 		rc = 1;
1769 		break;
1770 	}
1771 	simple_unlock(&spechash_slock);
1772 	return (rc);
1773 }
1774 
1775 /*
1776  * Revoke all the vnodes corresponding to the specified minor number
1777  * range (endpoints inclusive) of the specified major.
1778  */
1779 void
1780 vdevgone(maj, minl, minh, type)
1781 	int maj, minl, minh;
1782 	enum vtype type;
1783 {
1784 	struct vnode *vp;
1785 	int mn;
1786 
1787 	for (mn = minl; mn <= minh; mn++)
1788 		if (vfinddev(makedev(maj, mn), type, &vp))
1789 			VOP_REVOKE(vp, REVOKEALL);
1790 }
1791 
1792 /*
1793  * Calculate the total number of references to a special device.
1794  */
1795 int
1796 vcount(vp)
1797 	struct vnode *vp;
1798 {
1799 	struct vnode *vq, *vnext;
1800 	int count;
1801 
1802 loop:
1803 	if ((vp->v_flag & VALIASED) == 0)
1804 		return (vp->v_usecount);
1805 	simple_lock(&spechash_slock);
1806 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1807 		vnext = vq->v_specnext;
1808 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1809 			continue;
1810 		/*
1811 		 * Alias, but not in use, so flush it out.
1812 		 */
1813 		if (vq->v_usecount == 0 && vq != vp &&
1814 		    (vq->v_flag & VXLOCK) == 0) {
1815 			simple_unlock(&spechash_slock);
1816 			vgone(vq);
1817 			goto loop;
1818 		}
1819 		count += vq->v_usecount;
1820 	}
1821 	simple_unlock(&spechash_slock);
1822 	return (count);
1823 }
1824 
1825 /*
1826  * Print out a description of a vnode.
1827  */
1828 static const char * const typename[] =
1829    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1830 
1831 void
1832 vprint(label, vp)
1833 	char *label;
1834 	struct vnode *vp;
1835 {
1836 	char buf[96];
1837 
1838 	if (label != NULL)
1839 		printf("%s: ", label);
1840 	printf("tag %d type %s, usecount %d, writecount %ld, refcount %ld,",
1841 	    vp->v_tag, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1842 	    vp->v_holdcnt);
1843 	buf[0] = '\0';
1844 	if (vp->v_flag & VROOT)
1845 		strcat(buf, "|VROOT");
1846 	if (vp->v_flag & VTEXT)
1847 		strcat(buf, "|VTEXT");
1848 	if (vp->v_flag & VEXECMAP)
1849 		strcat(buf, "|VEXECMAP");
1850 	if (vp->v_flag & VSYSTEM)
1851 		strcat(buf, "|VSYSTEM");
1852 	if (vp->v_flag & VXLOCK)
1853 		strcat(buf, "|VXLOCK");
1854 	if (vp->v_flag & VXWANT)
1855 		strcat(buf, "|VXWANT");
1856 	if (vp->v_flag & VBWAIT)
1857 		strcat(buf, "|VBWAIT");
1858 	if (vp->v_flag & VALIASED)
1859 		strcat(buf, "|VALIASED");
1860 	if (buf[0] != '\0')
1861 		printf(" flags (%s)", &buf[1]);
1862 	if (vp->v_data == NULL) {
1863 		printf("\n");
1864 	} else {
1865 		printf("\n\t");
1866 		VOP_PRINT(vp);
1867 	}
1868 }
1869 
1870 #ifdef DEBUG
1871 /*
1872  * List all of the locked vnodes in the system.
1873  * Called when debugging the kernel.
1874  */
1875 void
1876 printlockedvnodes()
1877 {
1878 	struct mount *mp, *nmp;
1879 	struct vnode *vp;
1880 
1881 	printf("Locked vnodes\n");
1882 	simple_lock(&mountlist_slock);
1883 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
1884 	     mp = nmp) {
1885 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
1886 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
1887 			continue;
1888 		}
1889 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1890 			if (VOP_ISLOCKED(vp))
1891 				vprint(NULL, vp);
1892 		}
1893 		simple_lock(&mountlist_slock);
1894 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
1895 		vfs_unbusy(mp);
1896 	}
1897 	simple_unlock(&mountlist_slock);
1898 }
1899 #endif
1900 
1901 /*
1902  * Top level filesystem related information gathering.
1903  */
1904 int
1905 vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
1906 	int *name;
1907 	u_int namelen;
1908 	void *oldp;
1909 	size_t *oldlenp;
1910 	void *newp;
1911 	size_t newlen;
1912 	struct proc *p;
1913 {
1914 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1915 	struct vfsconf vfc;
1916 	extern const char * const mountcompatnames[];
1917 	extern int nmountcompatnames;
1918 #endif
1919 	struct vfsops *vfsp;
1920 
1921 	/* all sysctl names at this level are at least name and field */
1922 	if (namelen < 2)
1923 		return (ENOTDIR);		/* overloaded */
1924 
1925 	/* Not generic: goes to file system. */
1926 	if (name[0] != VFS_GENERIC) {
1927 		static const struct ctlname vfsnames[VFS_MAXID+1]=CTL_VFS_NAMES;
1928 		const char *vfsname;
1929 
1930 		if (name[0] < 0 || name[0] > VFS_MAXID
1931 		    || (vfsname = vfsnames[name[0]].ctl_name) == NULL)
1932 			return (EOPNOTSUPP);
1933 
1934 		vfsp = vfs_getopsbyname(vfsname);
1935 		if (vfsp == NULL || vfsp->vfs_sysctl == NULL)
1936 			return (EOPNOTSUPP);
1937 		return ((*vfsp->vfs_sysctl)(&name[1], namelen - 1,
1938 		    oldp, oldlenp, newp, newlen, p));
1939 	}
1940 
1941 	/* The rest are generic vfs sysctls. */
1942 	switch (name[1]) {
1943 	case VFS_USERMOUNT:
1944 		return sysctl_int(oldp, oldlenp, newp, newlen, &dovfsusermount);
1945 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1946 	case VFS_MAXTYPENUM:
1947 		/*
1948 		 * Provided for 4.4BSD-Lite2 compatibility.
1949 		 */
1950 		return (sysctl_rdint(oldp, oldlenp, newp, nmountcompatnames));
1951 	case VFS_CONF:
1952 		/*
1953 		 * Special: a node, next is a file system name.
1954 		 * Provided for 4.4BSD-Lite2 compatibility.
1955 		 */
1956 		if (namelen < 3)
1957 			return (ENOTDIR);	/* overloaded */
1958 		if (name[2] >= nmountcompatnames || name[2] < 0 ||
1959 		    mountcompatnames[name[2]] == NULL)
1960 			return (EOPNOTSUPP);
1961 		vfsp = vfs_getopsbyname(mountcompatnames[name[2]]);
1962 		if (vfsp == NULL)
1963 			return (EOPNOTSUPP);
1964 		vfc.vfc_vfsops = vfsp;
1965 		strncpy(vfc.vfc_name, vfsp->vfs_name, MFSNAMELEN);
1966 		vfc.vfc_typenum = name[2];
1967 		vfc.vfc_refcount = vfsp->vfs_refcount;
1968 		vfc.vfc_flags = 0;
1969 		vfc.vfc_mountroot = vfsp->vfs_mountroot;
1970 		vfc.vfc_next = NULL;
1971 		return (sysctl_rdstruct(oldp, oldlenp, newp, &vfc,
1972 		    sizeof(struct vfsconf)));
1973 #endif
1974 	default:
1975 		break;
1976 	}
1977 	return (EOPNOTSUPP);
1978 }
1979 
1980 int kinfo_vdebug = 1;
1981 int kinfo_vgetfailed;
1982 #define KINFO_VNODESLOP	10
1983 /*
1984  * Dump vnode list (via sysctl).
1985  * Copyout address of vnode followed by vnode.
1986  */
1987 /* ARGSUSED */
1988 int
1989 sysctl_vnode(where, sizep, p)
1990 	char *where;
1991 	size_t *sizep;
1992 	struct proc *p;
1993 {
1994 	struct mount *mp, *nmp;
1995 	struct vnode *nvp, *vp;
1996 	char *bp = where, *savebp;
1997 	char *ewhere;
1998 	int error;
1999 
2000 #define VPTRSZ	sizeof(struct vnode *)
2001 #define VNODESZ	sizeof(struct vnode)
2002 	if (where == NULL) {
2003 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
2004 		return (0);
2005 	}
2006 	ewhere = where + *sizep;
2007 
2008 	simple_lock(&mountlist_slock);
2009 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2010 	     mp = nmp) {
2011 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
2012 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
2013 			continue;
2014 		}
2015 		savebp = bp;
2016 again:
2017 		simple_lock(&mntvnode_slock);
2018 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2019 		     vp != NULL;
2020 		     vp = nvp) {
2021 			/*
2022 			 * Check that the vp is still associated with
2023 			 * this filesystem.  RACE: could have been
2024 			 * recycled onto the same filesystem.
2025 			 */
2026 			if (vp->v_mount != mp) {
2027 				simple_unlock(&mntvnode_slock);
2028 				if (kinfo_vdebug)
2029 					printf("kinfo: vp changed\n");
2030 				bp = savebp;
2031 				goto again;
2032 			}
2033 			nvp = LIST_NEXT(vp, v_mntvnodes);
2034 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2035 				simple_unlock(&mntvnode_slock);
2036 				*sizep = bp - where;
2037 				return (ENOMEM);
2038 			}
2039 			simple_unlock(&mntvnode_slock);
2040 			if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
2041 			   (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
2042 				return (error);
2043 			bp += VPTRSZ + VNODESZ;
2044 			simple_lock(&mntvnode_slock);
2045 		}
2046 		simple_unlock(&mntvnode_slock);
2047 		simple_lock(&mountlist_slock);
2048 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
2049 		vfs_unbusy(mp);
2050 	}
2051 	simple_unlock(&mountlist_slock);
2052 
2053 	*sizep = bp - where;
2054 	return (0);
2055 }
2056 
2057 /*
2058  * Check to see if a filesystem is mounted on a block device.
2059  */
2060 int
2061 vfs_mountedon(vp)
2062 	struct vnode *vp;
2063 {
2064 	struct vnode *vq;
2065 	int error = 0;
2066 
2067 	if (vp->v_specmountpoint != NULL)
2068 		return (EBUSY);
2069 	if (vp->v_flag & VALIASED) {
2070 		simple_lock(&spechash_slock);
2071 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2072 			if (vq->v_rdev != vp->v_rdev ||
2073 			    vq->v_type != vp->v_type)
2074 				continue;
2075 			if (vq->v_specmountpoint != NULL) {
2076 				error = EBUSY;
2077 				break;
2078 			}
2079 		}
2080 		simple_unlock(&spechash_slock);
2081 	}
2082 	return (error);
2083 }
2084 
2085 /*
2086  * Build hash lists of net addresses and hang them off the mount point.
2087  * Called by ufs_mount() to set up the lists of export addresses.
2088  */
2089 static int
2090 vfs_hang_addrlist(mp, nep, argp)
2091 	struct mount *mp;
2092 	struct netexport *nep;
2093 	struct export_args *argp;
2094 {
2095 	struct netcred *np, *enp;
2096 	struct radix_node_head *rnh;
2097 	int i;
2098 	struct radix_node *rn;
2099 	struct sockaddr *saddr, *smask = 0;
2100 	struct domain *dom;
2101 	int error;
2102 
2103 	if (argp->ex_addrlen == 0) {
2104 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2105 			return (EPERM);
2106 		np = &nep->ne_defexported;
2107 		np->netc_exflags = argp->ex_flags;
2108 		crcvt(&np->netc_anon, &argp->ex_anon);
2109 		np->netc_anon.cr_ref = 1;
2110 		mp->mnt_flag |= MNT_DEFEXPORTED;
2111 		return (0);
2112 	}
2113 
2114 	if (argp->ex_addrlen > MLEN)
2115 		return (EINVAL);
2116 
2117 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2118 	np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK);
2119 	memset((caddr_t)np, 0, i);
2120 	saddr = (struct sockaddr *)(np + 1);
2121 	error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen);
2122 	if (error)
2123 		goto out;
2124 	if (saddr->sa_len > argp->ex_addrlen)
2125 		saddr->sa_len = argp->ex_addrlen;
2126 	if (argp->ex_masklen) {
2127 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2128 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2129 		if (error)
2130 			goto out;
2131 		if (smask->sa_len > argp->ex_masklen)
2132 			smask->sa_len = argp->ex_masklen;
2133 	}
2134 	i = saddr->sa_family;
2135 	if ((rnh = nep->ne_rtable[i]) == 0) {
2136 		/*
2137 		 * Seems silly to initialize every AF when most are not
2138 		 * used, do so on demand here
2139 		 */
2140 		for (dom = domains; dom; dom = dom->dom_next)
2141 			if (dom->dom_family == i && dom->dom_rtattach) {
2142 				dom->dom_rtattach((void **)&nep->ne_rtable[i],
2143 					dom->dom_rtoffset);
2144 				break;
2145 			}
2146 		if ((rnh = nep->ne_rtable[i]) == 0) {
2147 			error = ENOBUFS;
2148 			goto out;
2149 		}
2150 	}
2151 	rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh,
2152 		np->netc_rnodes);
2153 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
2154 		if (rn == 0) {
2155 			enp = (struct netcred *)(*rnh->rnh_lookup)(saddr,
2156 				smask, rnh);
2157 			if (enp == 0) {
2158 				error = EPERM;
2159 				goto out;
2160 			}
2161 		} else
2162 			enp = (struct netcred *)rn;
2163 
2164 		if (enp->netc_exflags != argp->ex_flags ||
2165 		    enp->netc_anon.cr_uid != argp->ex_anon.cr_uid ||
2166 		    enp->netc_anon.cr_gid != argp->ex_anon.cr_gid ||
2167 		    enp->netc_anon.cr_ngroups !=
2168 		    			(uint32_t) argp->ex_anon.cr_ngroups ||
2169 		    memcmp(&enp->netc_anon.cr_groups, &argp->ex_anon.cr_groups,
2170 			enp->netc_anon.cr_ngroups))
2171 				error = EPERM;
2172 		else
2173 			error = 0;
2174 		goto out;
2175 	}
2176 	np->netc_exflags = argp->ex_flags;
2177 	crcvt(&np->netc_anon, &argp->ex_anon);
2178 	np->netc_anon.cr_ref = 1;
2179 	return (0);
2180 out:
2181 	free(np, M_NETADDR);
2182 	return (error);
2183 }
2184 
2185 /* ARGSUSED */
2186 static int
2187 vfs_free_netcred(rn, w)
2188 	struct radix_node *rn;
2189 	void *w;
2190 {
2191 	struct radix_node_head *rnh = (struct radix_node_head *)w;
2192 
2193 	(*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh);
2194 	free((caddr_t)rn, M_NETADDR);
2195 	return (0);
2196 }
2197 
2198 /*
2199  * Free the net address hash lists that are hanging off the mount points.
2200  */
2201 static void
2202 vfs_free_addrlist(nep)
2203 	struct netexport *nep;
2204 {
2205 	int i;
2206 	struct radix_node_head *rnh;
2207 
2208 	for (i = 0; i <= AF_MAX; i++)
2209 		if ((rnh = nep->ne_rtable[i]) != NULL) {
2210 			(*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh);
2211 			free((caddr_t)rnh, M_RTABLE);
2212 			nep->ne_rtable[i] = 0;
2213 		}
2214 }
2215 
2216 int
2217 vfs_export(mp, nep, argp)
2218 	struct mount *mp;
2219 	struct netexport *nep;
2220 	struct export_args *argp;
2221 {
2222 	int error;
2223 
2224 	if (argp->ex_flags & MNT_DELEXPORT) {
2225 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2226 			vfs_setpublicfs(NULL, NULL, NULL);
2227 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2228 		}
2229 		vfs_free_addrlist(nep);
2230 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2231 	}
2232 	if (argp->ex_flags & MNT_EXPORTED) {
2233 		if (argp->ex_flags & MNT_EXPUBLIC) {
2234 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2235 				return (error);
2236 			mp->mnt_flag |= MNT_EXPUBLIC;
2237 		}
2238 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
2239 			return (error);
2240 		mp->mnt_flag |= MNT_EXPORTED;
2241 	}
2242 	return (0);
2243 }
2244 
2245 /*
2246  * Set the publicly exported filesystem (WebNFS). Currently, only
2247  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2248  */
2249 int
2250 vfs_setpublicfs(mp, nep, argp)
2251 	struct mount *mp;
2252 	struct netexport *nep;
2253 	struct export_args *argp;
2254 {
2255 	int error;
2256 	struct vnode *rvp;
2257 	char *cp;
2258 
2259 	/*
2260 	 * mp == NULL -> invalidate the current info, the FS is
2261 	 * no longer exported. May be called from either vfs_export
2262 	 * or unmount, so check if it hasn't already been done.
2263 	 */
2264 	if (mp == NULL) {
2265 		if (nfs_pub.np_valid) {
2266 			nfs_pub.np_valid = 0;
2267 			if (nfs_pub.np_index != NULL) {
2268 				FREE(nfs_pub.np_index, M_TEMP);
2269 				nfs_pub.np_index = NULL;
2270 			}
2271 		}
2272 		return (0);
2273 	}
2274 
2275 	/*
2276 	 * Only one allowed at a time.
2277 	 */
2278 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2279 		return (EBUSY);
2280 
2281 	/*
2282 	 * Get real filehandle for root of exported FS.
2283 	 */
2284 	memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle));
2285 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2286 
2287 	if ((error = VFS_ROOT(mp, &rvp)))
2288 		return (error);
2289 
2290 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2291 		return (error);
2292 
2293 	vput(rvp);
2294 
2295 	/*
2296 	 * If an indexfile was specified, pull it in.
2297 	 */
2298 	if (argp->ex_indexfile != NULL) {
2299 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2300 		    M_WAITOK);
2301 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2302 		    MAXNAMLEN, (size_t *)0);
2303 		if (!error) {
2304 			/*
2305 			 * Check for illegal filenames.
2306 			 */
2307 			for (cp = nfs_pub.np_index; *cp; cp++) {
2308 				if (*cp == '/') {
2309 					error = EINVAL;
2310 					break;
2311 				}
2312 			}
2313 		}
2314 		if (error) {
2315 			FREE(nfs_pub.np_index, M_TEMP);
2316 			return (error);
2317 		}
2318 	}
2319 
2320 	nfs_pub.np_mount = mp;
2321 	nfs_pub.np_valid = 1;
2322 	return (0);
2323 }
2324 
2325 struct netcred *
2326 vfs_export_lookup(mp, nep, nam)
2327 	struct mount *mp;
2328 	struct netexport *nep;
2329 	struct mbuf *nam;
2330 {
2331 	struct netcred *np;
2332 	struct radix_node_head *rnh;
2333 	struct sockaddr *saddr;
2334 
2335 	np = NULL;
2336 	if (mp->mnt_flag & MNT_EXPORTED) {
2337 		/*
2338 		 * Lookup in the export list first.
2339 		 */
2340 		if (nam != NULL) {
2341 			saddr = mtod(nam, struct sockaddr *);
2342 			rnh = nep->ne_rtable[saddr->sa_family];
2343 			if (rnh != NULL) {
2344 				np = (struct netcred *)
2345 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2346 							      rnh);
2347 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2348 					np = NULL;
2349 			}
2350 		}
2351 		/*
2352 		 * If no address match, use the default if it exists.
2353 		 */
2354 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2355 			np = &nep->ne_defexported;
2356 	}
2357 	return (np);
2358 }
2359 
2360 /*
2361  * Do the usual access checking.
2362  * file_mode, uid and gid are from the vnode in question,
2363  * while acc_mode and cred are from the VOP_ACCESS parameter list
2364  */
2365 int
2366 vaccess(type, file_mode, uid, gid, acc_mode, cred)
2367 	enum vtype type;
2368 	mode_t file_mode;
2369 	uid_t uid;
2370 	gid_t gid;
2371 	mode_t acc_mode;
2372 	struct ucred *cred;
2373 {
2374 	mode_t mask;
2375 
2376 	/*
2377 	 * Super-user always gets read/write access, but execute access depends
2378 	 * on at least one execute bit being set.
2379 	 */
2380 	if (cred->cr_uid == 0) {
2381 		if ((acc_mode & VEXEC) && type != VDIR &&
2382 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
2383 			return (EACCES);
2384 		return (0);
2385 	}
2386 
2387 	mask = 0;
2388 
2389 	/* Otherwise, check the owner. */
2390 	if (cred->cr_uid == uid) {
2391 		if (acc_mode & VEXEC)
2392 			mask |= S_IXUSR;
2393 		if (acc_mode & VREAD)
2394 			mask |= S_IRUSR;
2395 		if (acc_mode & VWRITE)
2396 			mask |= S_IWUSR;
2397 		return ((file_mode & mask) == mask ? 0 : EACCES);
2398 	}
2399 
2400 	/* Otherwise, check the groups. */
2401 	if (cred->cr_gid == gid || groupmember(gid, cred)) {
2402 		if (acc_mode & VEXEC)
2403 			mask |= S_IXGRP;
2404 		if (acc_mode & VREAD)
2405 			mask |= S_IRGRP;
2406 		if (acc_mode & VWRITE)
2407 			mask |= S_IWGRP;
2408 		return ((file_mode & mask) == mask ? 0 : EACCES);
2409 	}
2410 
2411 	/* Otherwise, check everyone else. */
2412 	if (acc_mode & VEXEC)
2413 		mask |= S_IXOTH;
2414 	if (acc_mode & VREAD)
2415 		mask |= S_IROTH;
2416 	if (acc_mode & VWRITE)
2417 		mask |= S_IWOTH;
2418 	return ((file_mode & mask) == mask ? 0 : EACCES);
2419 }
2420 
2421 /*
2422  * Unmount all file systems.
2423  * We traverse the list in reverse order under the assumption that doing so
2424  * will avoid needing to worry about dependencies.
2425  */
2426 void
2427 vfs_unmountall(p)
2428 	struct proc *p;
2429 {
2430 	struct mount *mp, *nmp;
2431 	int allerror, error;
2432 
2433 	for (allerror = 0,
2434 	     mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2435 		nmp = mp->mnt_list.cqe_prev;
2436 #ifdef DEBUG
2437 		printf("unmounting %s (%s)...\n",
2438 		    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
2439 #endif
2440 		/*
2441 		 * XXX Freeze syncer.  Must do this before locking the
2442 		 * mount point.  See dounmount() for details.
2443 		 */
2444 		lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL);
2445 		if (vfs_busy(mp, 0, 0)) {
2446 			lockmgr(&syncer_lock, LK_RELEASE, NULL);
2447 			continue;
2448 		}
2449 		if ((error = dounmount(mp, MNT_FORCE, p)) != 0) {
2450 			printf("unmount of %s failed with error %d\n",
2451 			    mp->mnt_stat.f_mntonname, error);
2452 			allerror = 1;
2453 		}
2454 	}
2455 	if (allerror)
2456 		printf("WARNING: some file systems would not unmount\n");
2457 }
2458 
2459 /*
2460  * Sync and unmount file systems before shutting down.
2461  */
2462 void
2463 vfs_shutdown()
2464 {
2465 	struct buf *bp;
2466 	int iter, nbusy, nbusy_prev = 0, dcount, s;
2467 	struct lwp *l = curlwp;
2468 	struct proc *p;
2469 
2470 	/* XXX we're certainly not running in proc0's context! */
2471 	if (l == NULL || (p = l->l_proc) == NULL)
2472 		p = &proc0;
2473 
2474 	printf("syncing disks... ");
2475 
2476 	/* remove user process from run queue */
2477 	suspendsched();
2478 	(void) spl0();
2479 
2480 	/* avoid coming back this way again if we panic. */
2481 	doing_shutdown = 1;
2482 
2483 	sys_sync(l, NULL, NULL);
2484 
2485 	/* Wait for sync to finish. */
2486 	dcount = 10000;
2487 	for (iter = 0; iter < 20;) {
2488 		nbusy = 0;
2489 		for (bp = &buf[nbuf]; --bp >= buf; ) {
2490 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2491 				nbusy++;
2492 			/*
2493 			 * With soft updates, some buffers that are
2494 			 * written will be remarked as dirty until other
2495 			 * buffers are written.
2496 			 */
2497 			if (bp->b_vp && bp->b_vp->v_mount
2498 			    && (bp->b_vp->v_mount->mnt_flag & MNT_SOFTDEP)
2499 			    && (bp->b_flags & B_DELWRI)) {
2500 				s = splbio();
2501 				bremfree(bp);
2502 				bp->b_flags |= B_BUSY;
2503 				splx(s);
2504 				nbusy++;
2505 				bawrite(bp);
2506 				if (dcount-- <= 0) {
2507 					printf("softdep ");
2508 					goto fail;
2509 				}
2510 			}
2511 		}
2512 		if (nbusy == 0)
2513 			break;
2514 		if (nbusy_prev == 0)
2515 			nbusy_prev = nbusy;
2516 		printf("%d ", nbusy);
2517 		tsleep(&nbusy, PRIBIO, "bflush",
2518 		    (iter == 0) ? 1 : hz / 25 * iter);
2519 		if (nbusy >= nbusy_prev) /* we didn't flush anything */
2520 			iter++;
2521 		else
2522 			nbusy_prev = nbusy;
2523 	}
2524 	if (nbusy) {
2525 fail:
2526 #if defined(DEBUG) || defined(DEBUG_HALT_BUSY)
2527 		printf("giving up\nPrinting vnodes for busy buffers\n");
2528 		for (bp = &buf[nbuf]; --bp >= buf; )
2529 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2530 				vprint(NULL, bp->b_vp);
2531 
2532 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2533 		Debugger();
2534 #endif
2535 
2536 #else  /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2537 		printf("giving up\n");
2538 #endif /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2539 		return;
2540 	} else
2541 		printf("done\n");
2542 
2543 	/*
2544 	 * If we've panic'd, don't make the situation potentially
2545 	 * worse by unmounting the file systems.
2546 	 */
2547 	if (panicstr != NULL)
2548 		return;
2549 
2550 	/* Release inodes held by texts before update. */
2551 #ifdef notdef
2552 	vnshutdown();
2553 #endif
2554 	/* Unmount file systems. */
2555 	vfs_unmountall(p);
2556 }
2557 
2558 /*
2559  * Mount the root file system.  If the operator didn't specify a
2560  * file system to use, try all possible file systems until one
2561  * succeeds.
2562  */
2563 int
2564 vfs_mountroot()
2565 {
2566 	struct vfsops *v;
2567 
2568 	if (root_device == NULL)
2569 		panic("vfs_mountroot: root device unknown");
2570 
2571 	switch (root_device->dv_class) {
2572 	case DV_IFNET:
2573 		if (rootdev != NODEV)
2574 			panic("vfs_mountroot: rootdev set for DV_IFNET "
2575 			    "(0x%08x -> %d,%d)", rootdev,
2576 			    major(rootdev), minor(rootdev));
2577 		break;
2578 
2579 	case DV_DISK:
2580 		if (rootdev == NODEV)
2581 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2582 		break;
2583 
2584 	default:
2585 		printf("%s: inappropriate for root file system\n",
2586 		    root_device->dv_xname);
2587 		return (ENODEV);
2588 	}
2589 
2590 	/*
2591 	 * If user specified a file system, use it.
2592 	 */
2593 	if (mountroot != NULL)
2594 		return ((*mountroot)());
2595 
2596 	/*
2597 	 * Try each file system currently configured into the kernel.
2598 	 */
2599 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2600 		if (v->vfs_mountroot == NULL)
2601 			continue;
2602 #ifdef DEBUG
2603 		printf("mountroot: trying %s...\n", v->vfs_name);
2604 #endif
2605 		if ((*v->vfs_mountroot)() == 0) {
2606 			printf("root file system type: %s\n", v->vfs_name);
2607 			break;
2608 		}
2609 	}
2610 
2611 	if (v == NULL) {
2612 		printf("no file system for %s", root_device->dv_xname);
2613 		if (root_device->dv_class == DV_DISK)
2614 			printf(" (dev 0x%x)", rootdev);
2615 		printf("\n");
2616 		return (EFTYPE);
2617 	}
2618 	return (0);
2619 }
2620 
2621 /*
2622  * Given a file system name, look up the vfsops for that
2623  * file system, or return NULL if file system isn't present
2624  * in the kernel.
2625  */
2626 struct vfsops *
2627 vfs_getopsbyname(name)
2628 	const char *name;
2629 {
2630 	struct vfsops *v;
2631 
2632 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2633 		if (strcmp(v->vfs_name, name) == 0)
2634 			break;
2635 	}
2636 
2637 	return (v);
2638 }
2639 
2640 /*
2641  * Establish a file system and initialize it.
2642  */
2643 int
2644 vfs_attach(vfs)
2645 	struct vfsops *vfs;
2646 {
2647 	struct vfsops *v;
2648 	int error = 0;
2649 
2650 
2651 	/*
2652 	 * Make sure this file system doesn't already exist.
2653 	 */
2654 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2655 		if (strcmp(vfs->vfs_name, v->vfs_name) == 0) {
2656 			error = EEXIST;
2657 			goto out;
2658 		}
2659 	}
2660 
2661 	/*
2662 	 * Initialize the vnode operations for this file system.
2663 	 */
2664 	vfs_opv_init(vfs->vfs_opv_descs);
2665 
2666 	/*
2667 	 * Now initialize the file system itself.
2668 	 */
2669 	(*vfs->vfs_init)();
2670 
2671 	/*
2672 	 * ...and link it into the kernel's list.
2673 	 */
2674 	LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list);
2675 
2676 	/*
2677 	 * Sanity: make sure the reference count is 0.
2678 	 */
2679 	vfs->vfs_refcount = 0;
2680 
2681  out:
2682 	return (error);
2683 }
2684 
2685 /*
2686  * Remove a file system from the kernel.
2687  */
2688 int
2689 vfs_detach(vfs)
2690 	struct vfsops *vfs;
2691 {
2692 	struct vfsops *v;
2693 
2694 	/*
2695 	 * Make sure no one is using the filesystem.
2696 	 */
2697 	if (vfs->vfs_refcount != 0)
2698 		return (EBUSY);
2699 
2700 	/*
2701 	 * ...and remove it from the kernel's list.
2702 	 */
2703 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2704 		if (v == vfs) {
2705 			LIST_REMOVE(v, vfs_list);
2706 			break;
2707 		}
2708 	}
2709 
2710 	if (v == NULL)
2711 		return (ESRCH);
2712 
2713 	/*
2714 	 * Now run the file system-specific cleanups.
2715 	 */
2716 	(*vfs->vfs_done)();
2717 
2718 	/*
2719 	 * Free the vnode operations vector.
2720 	 */
2721 	vfs_opv_free(vfs->vfs_opv_descs);
2722 	return (0);
2723 }
2724 
2725 void
2726 vfs_reinit(void)
2727 {
2728 	struct vfsops *vfs;
2729 
2730 	LIST_FOREACH(vfs, &vfs_list, vfs_list) {
2731 		if (vfs->vfs_reinit) {
2732 			(*vfs->vfs_reinit)();
2733 		}
2734 	}
2735 }
2736 
2737 #ifdef DDB
2738 const char buf_flagbits[] =
2739 	"\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI"
2740 	"\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE"
2741 	"\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED"
2742 	"\32XXX\33VFLUSH";
2743 
2744 void
2745 vfs_buf_print(bp, full, pr)
2746 	struct buf *bp;
2747 	int full;
2748 	void (*pr) __P((const char *, ...));
2749 {
2750 	char buf[1024];
2751 
2752 	(*pr)("  vp %p lblkno 0x%x blkno 0x%x dev 0x%x\n",
2753 		  bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev);
2754 
2755 	bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf));
2756 	(*pr)("  error %d flags 0x%s\n", bp->b_error, buf);
2757 
2758 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
2759 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
2760 	(*pr)("  data %p saveaddr %p dep %p\n",
2761 		  bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep));
2762 	(*pr)("  iodone %p\n", bp->b_iodone);
2763 }
2764 
2765 
2766 const char vnode_flagbits[] =
2767 	"\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\5EXECMAP"
2768 	"\11XLOCK\12XWANT\13BWAIT\14ALIASED"
2769 	"\15DIROP\16LAYER\17ONWORKLIST\20DIRTY";
2770 
2771 const char *vnode_types[] = {
2772 	"VNON",
2773 	"VREG",
2774 	"VDIR",
2775 	"VBLK",
2776 	"VCHR",
2777 	"VLNK",
2778 	"VSOCK",
2779 	"VFIFO",
2780 	"VBAD",
2781 };
2782 
2783 const char *vnode_tags[] = {
2784 	"VT_NON",
2785 	"VT_UFS",
2786 	"VT_NFS",
2787 	"VT_MFS",
2788 	"VT_MSDOSFS",
2789 	"VT_LFS",
2790 	"VT_LOFS",
2791 	"VT_FDESC",
2792 	"VT_PORTAL",
2793 	"VT_NULL",
2794 	"VT_UMAP",
2795 	"VT_KERNFS",
2796 	"VT_PROCFS",
2797 	"VT_AFS",
2798 	"VT_ISOFS",
2799 	"VT_UNION",
2800 	"VT_ADOSFS",
2801 	"VT_EXT2FS",
2802 	"VT_CODA",
2803 	"VT_FILECORE",
2804 	"VT_NTFS",
2805 	"VT_VFS",
2806 	"VT_OVERLAY"
2807 };
2808 
2809 void
2810 vfs_vnode_print(vp, full, pr)
2811 	struct vnode *vp;
2812 	int full;
2813 	void (*pr) __P((const char *, ...));
2814 {
2815 	char buf[256];
2816 	const char *vtype, *vtag;
2817 
2818 	uvm_object_printit(&vp->v_uobj, full, pr);
2819 	bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf));
2820 	(*pr)("\nVNODE flags %s\n", buf);
2821 	(*pr)("mp %p numoutput %d size 0x%llx\n",
2822 	      vp->v_mount, vp->v_numoutput, vp->v_size);
2823 
2824 	(*pr)("data %p usecount %d writecount %ld holdcnt %ld numoutput %d\n",
2825 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2826 	      vp->v_holdcnt, vp->v_numoutput);
2827 
2828 	vtype = (vp->v_type >= 0 &&
2829 		 vp->v_type < sizeof(vnode_types) / sizeof(vnode_types[0])) ?
2830 		vnode_types[vp->v_type] : "UNKNOWN";
2831 	vtag = (vp->v_tag >= 0 &&
2832 		vp->v_tag < sizeof(vnode_tags) / sizeof(vnode_tags[0])) ?
2833 		vnode_tags[vp->v_tag] : "UNKNOWN";
2834 
2835 	(*pr)("type %s(%d) tag %s(%d) id 0x%lx mount %p typedata %p\n",
2836 	      vtype, vp->v_type, vtag, vp->v_tag,
2837 	      vp->v_id, vp->v_mount, vp->v_mountedhere);
2838 
2839 	if (full) {
2840 		struct buf *bp;
2841 
2842 		(*pr)("clean bufs:\n");
2843 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2844 			(*pr)(" bp %p\n", bp);
2845 			vfs_buf_print(bp, full, pr);
2846 		}
2847 
2848 		(*pr)("dirty bufs:\n");
2849 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2850 			(*pr)(" bp %p\n", bp);
2851 			vfs_buf_print(bp, full, pr);
2852 		}
2853 	}
2854 }
2855 #endif
2856