xref: /netbsd-src/sys/kern/vfs_subr.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: vfs_subr.c,v 1.156 2001/08/03 06:00:13 jdolecek 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 "opt_ddb.h"
85 #include "opt_compat_netbsd.h"
86 #include "opt_compat_43.h"
87 
88 #include <sys/param.h>
89 #include <sys/systm.h>
90 #include <sys/proc.h>
91 #include <sys/kernel.h>
92 #include <sys/mount.h>
93 #include <sys/time.h>
94 #include <sys/fcntl.h>
95 #include <sys/vnode.h>
96 #include <sys/stat.h>
97 #include <sys/namei.h>
98 #include <sys/ucred.h>
99 #include <sys/buf.h>
100 #include <sys/errno.h>
101 #include <sys/malloc.h>
102 #include <sys/domain.h>
103 #include <sys/mbuf.h>
104 #include <sys/syscallargs.h>
105 #include <sys/device.h>
106 #include <sys/dirent.h>
107 
108 #include <miscfs/specfs/specdev.h>
109 #include <miscfs/genfs/genfs.h>
110 #include <miscfs/syncfs/syncfs.h>
111 
112 #include <uvm/uvm.h>
113 #include <uvm/uvm_ddb.h>
114 
115 #include <sys/sysctl.h>
116 
117 enum vtype iftovt_tab[16] = {
118 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
119 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
120 };
121 const int	vttoif_tab[9] = {
122 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
123 	S_IFSOCK, S_IFIFO, S_IFMT,
124 };
125 
126 int doforce = 1;		/* 1 => permit forcible unmounting */
127 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
128 
129 extern int dovfsusermount;	/* 1 => permit any user to mount filesystems */
130 
131 /*
132  * Insq/Remq for the vnode usage lists.
133  */
134 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
135 #define	bufremvn(bp) {							\
136 	LIST_REMOVE(bp, b_vnbufs);					\
137 	(bp)->b_vnbufs.le_next = NOLIST;				\
138 }
139 /* TAILQ_HEAD(freelst, vnode) vnode_free_list =	vnode free list (in vnode.h) */
140 struct freelst vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
141 struct freelst vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
142 
143 struct mntlist mountlist =			/* mounted filesystem list */
144     CIRCLEQ_HEAD_INITIALIZER(mountlist);
145 struct vfs_list_head vfs_list =			/* vfs list */
146     LIST_HEAD_INITIALIZER(vfs_list);
147 
148 struct nfs_public nfs_pub;			/* publicly exported FS */
149 
150 struct simplelock mountlist_slock = SIMPLELOCK_INITIALIZER;
151 static struct simplelock mntid_slock = SIMPLELOCK_INITIALIZER;
152 struct simplelock mntvnode_slock = SIMPLELOCK_INITIALIZER;
153 struct simplelock vnode_free_list_slock = SIMPLELOCK_INITIALIZER;
154 struct simplelock spechash_slock = SIMPLELOCK_INITIALIZER;
155 
156 /*
157  * These define the root filesystem and device.
158  */
159 struct mount *rootfs;
160 struct vnode *rootvnode;
161 struct device *root_device;			/* root device */
162 
163 struct pool vnode_pool;				/* memory pool for vnodes */
164 
165 /*
166  * Local declarations.
167  */
168 void insmntque __P((struct vnode *, struct mount *));
169 int getdevvp __P((dev_t, struct vnode **, enum vtype));
170 void vgoneall __P((struct vnode *));
171 
172 static int vfs_hang_addrlist __P((struct mount *, struct netexport *,
173 				  struct export_args *));
174 static int vfs_free_netcred __P((struct radix_node *, void *));
175 static void vfs_free_addrlist __P((struct netexport *));
176 
177 #ifdef DEBUG
178 void printlockedvnodes __P((void));
179 #endif
180 
181 /*
182  * Initialize the vnode management data structures.
183  */
184 void
185 vntblinit()
186 {
187 
188 	pool_init(&vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodepl",
189 	    0, pool_page_alloc_nointr, pool_page_free_nointr, M_VNODE);
190 
191 	/*
192 	 * Initialize the filesystem syncer.
193 	 */
194 	vn_initialize_syncerd();
195 }
196 
197 /*
198  * Mark a mount point as busy. Used to synchronize access and to delay
199  * unmounting. Interlock is not released on failure.
200  */
201 int
202 vfs_busy(mp, flags, interlkp)
203 	struct mount *mp;
204 	int flags;
205 	struct simplelock *interlkp;
206 {
207 	int lkflags;
208 
209 	while (mp->mnt_flag & MNT_UNMOUNT) {
210 		int gone;
211 
212 		if (flags & LK_NOWAIT)
213 			return (ENOENT);
214 		if ((flags & LK_RECURSEFAIL) && mp->mnt_unmounter != NULL
215 		    && mp->mnt_unmounter == curproc)
216 			return (EDEADLK);
217 		if (interlkp)
218 			simple_unlock(interlkp);
219 		/*
220 		 * Since all busy locks are shared except the exclusive
221 		 * lock granted when unmounting, the only place that a
222 		 * wakeup needs to be done is at the release of the
223 		 * exclusive lock at the end of dounmount.
224 		 *
225 		 * XXX MP: add spinlock protecting mnt_wcnt here once you
226 		 * can atomically unlock-and-sleep.
227 		 */
228 		mp->mnt_wcnt++;
229 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
230 		mp->mnt_wcnt--;
231 		gone = mp->mnt_flag & MNT_GONE;
232 
233 		if (mp->mnt_wcnt == 0)
234 			wakeup(&mp->mnt_wcnt);
235 		if (interlkp)
236 			simple_lock(interlkp);
237 		if (gone)
238 			return (ENOENT);
239 	}
240 	lkflags = LK_SHARED;
241 	if (interlkp)
242 		lkflags |= LK_INTERLOCK;
243 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp))
244 		panic("vfs_busy: unexpected lock failure");
245 	return (0);
246 }
247 
248 /*
249  * Free a busy filesystem.
250  */
251 void
252 vfs_unbusy(mp)
253 	struct mount *mp;
254 {
255 
256 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
257 }
258 
259 /*
260  * Lookup a filesystem type, and if found allocate and initialize
261  * a mount structure for it.
262  *
263  * Devname is usually updated by mount(8) after booting.
264  */
265 int
266 vfs_rootmountalloc(fstypename, devname, mpp)
267 	char *fstypename;
268 	char *devname;
269 	struct mount **mpp;
270 {
271 	struct vfsops *vfsp = NULL;
272 	struct mount *mp;
273 
274 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
275 		if (!strncmp(vfsp->vfs_name, fstypename, MFSNAMELEN))
276 			break;
277 
278 	if (vfsp == NULL)
279 		return (ENODEV);
280 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
281 	memset((char *)mp, 0, (u_long)sizeof(struct mount));
282 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
283 	(void)vfs_busy(mp, LK_NOWAIT, 0);
284 	LIST_INIT(&mp->mnt_vnodelist);
285 	mp->mnt_op = vfsp;
286 	mp->mnt_flag = MNT_RDONLY;
287 	mp->mnt_vnodecovered = NULLVP;
288 	vfsp->vfs_refcount++;
289 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN);
290 	mp->mnt_stat.f_mntonname[0] = '/';
291 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
292 	*mpp = mp;
293 	return (0);
294 }
295 
296 /*
297  * Lookup a mount point by filesystem identifier.
298  */
299 struct mount *
300 vfs_getvfs(fsid)
301 	fsid_t *fsid;
302 {
303 	struct mount *mp;
304 
305 	simple_lock(&mountlist_slock);
306 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
307 	     mp = mp->mnt_list.cqe_next) {
308 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
309 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
310 			simple_unlock(&mountlist_slock);
311 			return (mp);
312 		}
313 	}
314 	simple_unlock(&mountlist_slock);
315 	return ((struct mount *)0);
316 }
317 
318 /*
319  * Get a new unique fsid
320  */
321 void
322 vfs_getnewfsid(mp)
323 	struct mount *mp;
324 {
325 	static u_short xxxfs_mntid;
326 	fsid_t tfsid;
327 	int mtype;
328 
329 	simple_lock(&mntid_slock);
330 	mtype = makefstype(mp->mnt_op->vfs_name);
331 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
332 	mp->mnt_stat.f_fsid.val[1] = mtype;
333 	if (xxxfs_mntid == 0)
334 		++xxxfs_mntid;
335 	tfsid.val[0] = makedev((nblkdev + mtype) & 0xff, xxxfs_mntid);
336 	tfsid.val[1] = mtype;
337 	if (mountlist.cqh_first != (void *)&mountlist) {
338 		while (vfs_getvfs(&tfsid)) {
339 			tfsid.val[0]++;
340 			xxxfs_mntid++;
341 		}
342 	}
343 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
344 	simple_unlock(&mntid_slock);
345 }
346 
347 /*
348  * Make a 'unique' number from a mount type name.
349  */
350 long
351 makefstype(type)
352 	const char *type;
353 {
354 	long rv;
355 
356 	for (rv = 0; *type; type++) {
357 		rv <<= 2;
358 		rv ^= *type;
359 	}
360 	return rv;
361 }
362 
363 
364 /*
365  * Set vnode attributes to VNOVAL
366  */
367 void
368 vattr_null(vap)
369 	struct vattr *vap;
370 {
371 
372 	vap->va_type = VNON;
373 
374 	/*
375 	 * Assign individually so that it is safe even if size and
376 	 * sign of each member are varied.
377 	 */
378 	vap->va_mode = VNOVAL;
379 	vap->va_nlink = VNOVAL;
380 	vap->va_uid = VNOVAL;
381 	vap->va_gid = VNOVAL;
382 	vap->va_fsid = VNOVAL;
383 	vap->va_fileid = VNOVAL;
384 	vap->va_size = VNOVAL;
385 	vap->va_blocksize = VNOVAL;
386 	vap->va_atime.tv_sec =
387 	    vap->va_mtime.tv_sec =
388 	    vap->va_ctime.tv_sec = VNOVAL;
389 	vap->va_atime.tv_nsec =
390 	    vap->va_mtime.tv_nsec =
391 	    vap->va_ctime.tv_nsec = VNOVAL;
392 	vap->va_gen = VNOVAL;
393 	vap->va_flags = VNOVAL;
394 	vap->va_rdev = VNOVAL;
395 	vap->va_bytes = VNOVAL;
396 	vap->va_vaflags = 0;
397 }
398 
399 /*
400  * Routines having to do with the management of the vnode table.
401  */
402 extern int (**dead_vnodeop_p) __P((void *));
403 long numvnodes;
404 
405 /*
406  * Return the next vnode from the free list.
407  */
408 int
409 getnewvnode(tag, mp, vops, vpp)
410 	enum vtagtype tag;
411 	struct mount *mp;
412 	int (**vops) __P((void *));
413 	struct vnode **vpp;
414 {
415 	extern struct uvm_pagerops uvm_vnodeops;
416 	struct uvm_object *uobj;
417 	struct proc *p = curproc;	/* XXX */
418 	struct freelst *listhd;
419 	static int toggle;
420 	struct vnode *vp;
421 	int error = 0, tryalloc;
422 #ifdef DIAGNOSTIC
423 	int s;
424 #endif
425 	if (mp) {
426 		/*
427 		 * Mark filesystem busy while we're creating a vnode.
428 		 * If unmount is in progress, this will wait; if the
429 		 * unmount succeeds (only if umount -f), this will
430 		 * return an error.  If the unmount fails, we'll keep
431 		 * going afterwards.
432 		 * (This puts the per-mount vnode list logically under
433 		 * the protection of the vfs_busy lock).
434 		 */
435 		error = vfs_busy(mp, LK_RECURSEFAIL, 0);
436 		if (error && error != EDEADLK)
437 			return error;
438 	}
439 
440 	/*
441 	 * We must choose whether to allocate a new vnode or recycle an
442 	 * existing one. The criterion for allocating a new one is that
443 	 * the total number of vnodes is less than the number desired or
444 	 * there are no vnodes on either free list. Generally we only
445 	 * want to recycle vnodes that have no buffers associated with
446 	 * them, so we look first on the vnode_free_list. If it is empty,
447 	 * we next consider vnodes with referencing buffers on the
448 	 * vnode_hold_list. The toggle ensures that half the time we
449 	 * will use a buffer from the vnode_hold_list, and half the time
450 	 * we will allocate a new one unless the list has grown to twice
451 	 * the desired size. We are reticent to recycle vnodes from the
452 	 * vnode_hold_list because we will lose the identity of all its
453 	 * referencing buffers.
454 	 */
455 
456  try_again:
457 	vp = NULL;
458 
459 	simple_lock(&vnode_free_list_slock);
460 
461 	toggle ^= 1;
462 	if (numvnodes > 2 * desiredvnodes)
463 		toggle = 0;
464 
465 	tryalloc = numvnodes < desiredvnodes ||
466 	    (TAILQ_FIRST(listhd = &vnode_free_list) == NULL &&
467 	     (TAILQ_FIRST(listhd = &vnode_hold_list) == NULL || toggle));
468 
469 	if (tryalloc &&
470 	    (vp = pool_get(&vnode_pool, PR_NOWAIT)) != NULL) {
471 		simple_unlock(&vnode_free_list_slock);
472 		memset(vp, 0, sizeof(*vp));
473 		simple_lock_init(&vp->v_interlock);
474 		numvnodes++;
475 	} else {
476 		for (vp = TAILQ_FIRST(listhd); vp != NULLVP;
477 		    vp = TAILQ_NEXT(vp, v_freelist)) {
478 			if (simple_lock_try(&vp->v_interlock)) {
479 				if ((vp->v_flag & VLAYER) == 0)
480 					break;
481 				if (VOP_ISLOCKED(vp) == 0)
482 					break;
483 				else
484 					simple_unlock(&vp->v_interlock);
485 			}
486 		}
487 		/*
488 		 * Unless this is a bad time of the month, at most
489 		 * the first NCPUS items on the free list are
490 		 * locked, so this is close enough to being empty.
491 		 */
492 		if (vp == NULLVP) {
493 			simple_unlock(&vnode_free_list_slock);
494 			if (mp && error != EDEADLK)
495 				vfs_unbusy(mp);
496 			if (tryalloc) {
497 				printf("WARNING: unable to allocate new "
498 				    "vnode, retrying...\n");
499 				(void) tsleep(&lbolt, PRIBIO, "newvn", hz);
500 				goto try_again;
501 			}
502 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
503 			*vpp = 0;
504 			return (ENFILE);
505 		}
506 		if (vp->v_usecount)
507 			panic("free vnode isn't, vp %p", vp);
508 		TAILQ_REMOVE(listhd, vp, v_freelist);
509 		/* see comment on why 0xdeadb is set at end of vgone (below) */
510 		vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
511 		simple_unlock(&vnode_free_list_slock);
512 		vp->v_lease = NULL;
513 		if (vp->v_type != VBAD)
514 			vgonel(vp, p);
515 		else
516 			simple_unlock(&vp->v_interlock);
517 #ifdef DIAGNOSTIC
518 		if (vp->v_data)
519 			panic("cleaned vnode isn't, vp %p", vp);
520 		s = splbio();
521 		if (vp->v_numoutput)
522 			panic("clean vnode has pending I/O's, vp %p", vp);
523 		splx(s);
524 #endif
525 		vp->v_flag = 0;
526 		vp->v_lastr = 0;
527 		vp->v_ralen = 0;
528 		vp->v_maxra = 0;
529 		vp->v_lastw = 0;
530 		vp->v_lasta = 0;
531 		vp->v_cstart = 0;
532 		vp->v_clen = 0;
533 		vp->v_socket = 0;
534 	}
535 	vp->v_type = VNON;
536 	vp->v_vnlock = &vp->v_lock;
537 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
538 	lockinit(&vp->v_glock, PVFS, "glock", 0, 0);
539 	cache_purge(vp);
540 	vp->v_tag = tag;
541 	vp->v_op = vops;
542 	insmntque(vp, mp);
543 	*vpp = vp;
544 	vp->v_usecount = 1;
545 	vp->v_data = 0;
546 	simple_lock_init(&vp->v_uvm.u_obj.vmobjlock);
547 
548 	/*
549 	 * initialize uvm_object within vnode.
550 	 */
551 
552 	uobj = &vp->v_uvm.u_obj;
553 	uobj->pgops = &uvm_vnodeops;
554 	TAILQ_INIT(&uobj->memq);
555 	vp->v_uvm.u_size = VSIZENOTSET;
556 
557 	if (mp && error != EDEADLK)
558 		vfs_unbusy(mp);
559 	return (0);
560 }
561 
562 /*
563  * This is really just the reverse of getnewvnode(). Needed for
564  * VFS_VGET functions who may need to push back a vnode in case
565  * of a locking race.
566  */
567 void
568 ungetnewvnode(vp)
569 	struct vnode *vp;
570 {
571 #ifdef DIAGNOSTIC
572 	if (vp->v_usecount != 1)
573 		panic("ungetnewvnode: busy vnode");
574 #endif
575 	vp->v_usecount--;
576 	insmntque(vp, NULL);
577 	vp->v_type = VBAD;
578 
579 	simple_lock(&vp->v_interlock);
580 	/*
581 	 * Insert at head of LRU list
582 	 */
583 	simple_lock(&vnode_free_list_slock);
584 	if (vp->v_holdcnt > 0)
585 		TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist);
586 	else
587 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
588 	simple_unlock(&vnode_free_list_slock);
589 	simple_unlock(&vp->v_interlock);
590 }
591 
592 /*
593  * Move a vnode from one mount queue to another.
594  */
595 void
596 insmntque(vp, mp)
597 	struct vnode *vp;
598 	struct mount *mp;
599 {
600 
601 #ifdef DIAGNOSTIC
602 	if ((mp != NULL) &&
603 	    (mp->mnt_flag & MNT_UNMOUNT) &&
604 	    !(mp->mnt_flag & MNT_SOFTDEP) &&
605 	    vp->v_tag != VT_VFS) {
606 		panic("insmntque into dying filesystem");
607 	}
608 #endif
609 
610 	simple_lock(&mntvnode_slock);
611 	/*
612 	 * Delete from old mount point vnode list, if on one.
613 	 */
614 	if (vp->v_mount != NULL)
615 		LIST_REMOVE(vp, v_mntvnodes);
616 	/*
617 	 * Insert into list of vnodes for the new mount point, if available.
618 	 */
619 	if ((vp->v_mount = mp) != NULL)
620 		LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
621 	simple_unlock(&mntvnode_slock);
622 }
623 
624 /*
625  * Update outstanding I/O count and do wakeup if requested.
626  */
627 void
628 vwakeup(bp)
629 	struct buf *bp;
630 {
631 	struct vnode *vp;
632 
633 	if ((vp = bp->b_vp) != NULL) {
634 		if (--vp->v_numoutput < 0)
635 			panic("vwakeup: neg numoutput, vp %p", vp);
636 		if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
637 			vp->v_flag &= ~VBWAIT;
638 			wakeup((caddr_t)&vp->v_numoutput);
639 		}
640 	}
641 }
642 
643 /*
644  * Flush out and invalidate all buffers associated with a vnode.
645  * Called with the underlying vnode locked, which should prevent new dirty
646  * buffers from being queued.
647  */
648 int
649 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
650 	struct vnode *vp;
651 	int flags;
652 	struct ucred *cred;
653 	struct proc *p;
654 	int slpflag, slptimeo;
655 {
656 	struct uvm_object *uobj = &vp->v_uvm.u_obj;
657 	struct buf *bp, *nbp;
658 	int s, error, rv;
659 	int flushflags = PGO_ALLPAGES|PGO_FREE|PGO_SYNCIO|
660 		(flags & V_SAVE ? PGO_CLEANIT : 0);
661 
662 	/* XXXUBC this doesn't look at flags or slp* */
663 	if (vp->v_type == VREG) {
664 		simple_lock(&uobj->vmobjlock);
665 		rv = (uobj->pgops->pgo_flush)(uobj, 0, 0, flushflags);
666 		simple_unlock(&uobj->vmobjlock);
667 		if (!rv) {
668 			return EIO;
669 		}
670 	}
671 	if (flags & V_SAVE) {
672 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p);
673 		if (error)
674 		        return (error);
675 #ifdef DIAGNOSTIC
676 		s = splbio();
677 		if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd))
678 		        panic("vinvalbuf: dirty bufs, vp %p", vp);
679 		splx(s);
680 #endif
681 	}
682 
683 	s = splbio();
684 
685 restart:
686 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
687 		nbp = LIST_NEXT(bp, b_vnbufs);
688 		if (bp->b_flags & B_BUSY) {
689 			bp->b_flags |= B_WANTED;
690 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
691 			    "vinvalbuf", slptimeo);
692 			if (error) {
693 				splx(s);
694 				return (error);
695 			}
696 			goto restart;
697 		}
698 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
699 		brelse(bp);
700 	}
701 
702 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
703 		nbp = LIST_NEXT(bp, b_vnbufs);
704 		if (bp->b_flags & B_BUSY) {
705 			bp->b_flags |= B_WANTED;
706 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
707 			    "vinvalbuf", slptimeo);
708 			if (error) {
709 				splx(s);
710 				return (error);
711 			}
712 			goto restart;
713 		}
714 		/*
715 		 * XXX Since there are no node locks for NFS, I believe
716 		 * there is a slight chance that a delayed write will
717 		 * occur while sleeping just above, so check for it.
718 		 */
719 		if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
720 #ifdef DEBUG
721 			printf("buffer still DELWRI\n");
722 #endif
723 			bp->b_flags |= B_BUSY | B_VFLUSH;
724 			VOP_BWRITE(bp);
725 			goto restart;
726 		}
727 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
728 		brelse(bp);
729 	}
730 
731 #ifdef DIAGNOSTIC
732 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
733 		panic("vinvalbuf: flush failed, vp %p", vp);
734 #endif
735 
736 	splx(s);
737 
738 	return (0);
739 }
740 
741 /*
742  * Destroy any in core blocks past the truncation length.
743  * Called with the underlying vnode locked, which should prevent new dirty
744  * buffers from being queued.
745  */
746 int
747 vtruncbuf(vp, lbn, slpflag, slptimeo)
748 	struct vnode *vp;
749 	daddr_t lbn;
750 	int slpflag, slptimeo;
751 {
752 	struct uvm_object *uobj = &vp->v_uvm.u_obj;
753 	struct buf *bp, *nbp;
754 	int s, error, rv;
755 
756 	s = splbio();
757 	if (vp->v_type == VREG) {
758 		simple_lock(&uobj->vmobjlock);
759 		rv = (uobj->pgops->pgo_flush)(uobj,
760 		    round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift), 0,
761 		    PGO_FREE|PGO_SYNCIO);
762 		simple_unlock(&uobj->vmobjlock);
763 		if (!rv) {
764 			splx(s);
765 			return EIO;
766 		}
767 	}
768 
769 restart:
770 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
771 		nbp = LIST_NEXT(bp, b_vnbufs);
772 		if (bp->b_lblkno < lbn)
773 			continue;
774 		if (bp->b_flags & B_BUSY) {
775 			bp->b_flags |= B_WANTED;
776 			error = tsleep(bp, slpflag | (PRIBIO + 1),
777 			    "vtruncbuf", slptimeo);
778 			if (error) {
779 				splx(s);
780 				return (error);
781 			}
782 			goto restart;
783 		}
784 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
785 		brelse(bp);
786 	}
787 
788 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
789 		nbp = LIST_NEXT(bp, b_vnbufs);
790 		if (bp->b_lblkno < lbn)
791 			continue;
792 		if (bp->b_flags & B_BUSY) {
793 			bp->b_flags |= B_WANTED;
794 			error = tsleep(bp, slpflag | (PRIBIO + 1),
795 			    "vtruncbuf", slptimeo);
796 			if (error) {
797 				splx(s);
798 				return (error);
799 			}
800 			goto restart;
801 		}
802 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
803 		brelse(bp);
804 	}
805 
806 	splx(s);
807 
808 	return (0);
809 }
810 
811 void
812 vflushbuf(vp, sync)
813 	struct vnode *vp;
814 	int sync;
815 {
816 	struct uvm_object *uobj = &vp->v_uvm.u_obj;
817 	struct buf *bp, *nbp;
818 	int s;
819 
820 	if (vp->v_type == VREG) {
821 		int flags = PGO_CLEANIT|PGO_ALLPAGES| (sync ? PGO_SYNCIO : 0);
822 
823 		simple_lock(&uobj->vmobjlock);
824 		(uobj->pgops->pgo_flush)(uobj, 0, 0, flags);
825 		simple_unlock(&uobj->vmobjlock);
826 	}
827 
828 loop:
829 	s = splbio();
830 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
831 		nbp = LIST_NEXT(bp, b_vnbufs);
832 		if ((bp->b_flags & B_BUSY))
833 			continue;
834 		if ((bp->b_flags & B_DELWRI) == 0)
835 			panic("vflushbuf: not dirty, bp %p", bp);
836 		bp->b_flags |= B_BUSY | B_VFLUSH;
837 		splx(s);
838 		/*
839 		 * Wait for I/O associated with indirect blocks to complete,
840 		 * since there is no way to quickly wait for them below.
841 		 */
842 		if (bp->b_vp == vp || sync == 0)
843 			(void) bawrite(bp);
844 		else
845 			(void) bwrite(bp);
846 		goto loop;
847 	}
848 	if (sync == 0) {
849 		splx(s);
850 		return;
851 	}
852 	while (vp->v_numoutput) {
853 		vp->v_flag |= VBWAIT;
854 		tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0);
855 	}
856 	splx(s);
857 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
858 		vprint("vflushbuf: dirty", vp);
859 		goto loop;
860 	}
861 }
862 
863 /*
864  * Associate a buffer with a vnode.
865  */
866 void
867 bgetvp(vp, bp)
868 	struct vnode *vp;
869 	struct buf *bp;
870 {
871 	int s;
872 
873 	if (bp->b_vp)
874 		panic("bgetvp: not free, bp %p", bp);
875 	VHOLD(vp);
876 	s = splbio();
877 	bp->b_vp = vp;
878 	if (vp->v_type == VBLK || vp->v_type == VCHR)
879 		bp->b_dev = vp->v_rdev;
880 	else
881 		bp->b_dev = NODEV;
882 	/*
883 	 * Insert onto list for new vnode.
884 	 */
885 	bufinsvn(bp, &vp->v_cleanblkhd);
886 	splx(s);
887 }
888 
889 /*
890  * Disassociate a buffer from a vnode.
891  */
892 void
893 brelvp(bp)
894 	struct buf *bp;
895 {
896 	struct vnode *vp;
897 	int s;
898 
899 	if (bp->b_vp == NULL)
900 		panic("brelvp: vp NULL, bp %p", bp);
901 
902 	s = splbio();
903 	vp = bp->b_vp;
904 	/*
905 	 * Delete from old vnode list, if on one.
906 	 */
907 	if (bp->b_vnbufs.le_next != NOLIST)
908 		bufremvn(bp);
909 
910 	if (vp->v_type != VREG && (vp->v_flag & VONWORKLST) &&
911 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
912 		vp->v_flag &= ~VONWORKLST;
913 		LIST_REMOVE(vp, v_synclist);
914 	}
915 
916 	bp->b_vp = NULL;
917 	HOLDRELE(vp);
918 	splx(s);
919 }
920 
921 /*
922  * Reassign a buffer from one vnode to another.
923  * Used to assign file specific control information
924  * (indirect blocks) to the vnode to which they belong.
925  *
926  * This function must be called at splbio().
927  */
928 void
929 reassignbuf(bp, newvp)
930 	struct buf *bp;
931 	struct vnode *newvp;
932 {
933 	struct buflists *listheadp;
934 	int delay;
935 
936 	/*
937 	 * Delete from old vnode list, if on one.
938 	 */
939 	if (bp->b_vnbufs.le_next != NOLIST)
940 		bufremvn(bp);
941 	/*
942 	 * If dirty, put on list of dirty buffers;
943 	 * otherwise insert onto list of clean buffers.
944 	 */
945 	if ((bp->b_flags & B_DELWRI) == 0) {
946 		listheadp = &newvp->v_cleanblkhd;
947 		if (newvp->v_type != VREG &&
948 		    (newvp->v_flag & VONWORKLST) &&
949 		    LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
950 			newvp->v_flag &= ~VONWORKLST;
951 			LIST_REMOVE(newvp, v_synclist);
952 		}
953 	} else {
954 		listheadp = &newvp->v_dirtyblkhd;
955 		if ((newvp->v_flag & VONWORKLST) == 0) {
956 			switch (newvp->v_type) {
957 			case VDIR:
958 				delay = dirdelay;
959 				break;
960 			case VBLK:
961 				if (newvp->v_specmountpoint != NULL) {
962 					delay = metadelay;
963 					break;
964 				}
965 				/* fall through */
966 			default:
967 				delay = filedelay;
968 				break;
969 			}
970 			if (!newvp->v_mount ||
971 			    (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0)
972 				vn_syncer_add_to_worklist(newvp, delay);
973 		}
974 	}
975 	bufinsvn(bp, listheadp);
976 }
977 
978 /*
979  * Create a vnode for a block device.
980  * Used for root filesystem and swap areas.
981  * Also used for memory file system special devices.
982  */
983 int
984 bdevvp(dev, vpp)
985 	dev_t dev;
986 	struct vnode **vpp;
987 {
988 
989 	return (getdevvp(dev, vpp, VBLK));
990 }
991 
992 /*
993  * Create a vnode for a character device.
994  * Used for kernfs and some console handling.
995  */
996 int
997 cdevvp(dev, vpp)
998 	dev_t dev;
999 	struct vnode **vpp;
1000 {
1001 
1002 	return (getdevvp(dev, vpp, VCHR));
1003 }
1004 
1005 /*
1006  * Create a vnode for a device.
1007  * Used by bdevvp (block device) for root file system etc.,
1008  * and by cdevvp (character device) for console and kernfs.
1009  */
1010 int
1011 getdevvp(dev, vpp, type)
1012 	dev_t dev;
1013 	struct vnode **vpp;
1014 	enum vtype type;
1015 {
1016 	struct vnode *vp;
1017 	struct vnode *nvp;
1018 	int error;
1019 
1020 	if (dev == NODEV) {
1021 		*vpp = NULLVP;
1022 		return (0);
1023 	}
1024 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
1025 	if (error) {
1026 		*vpp = NULLVP;
1027 		return (error);
1028 	}
1029 	vp = nvp;
1030 	vp->v_type = type;
1031 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
1032 		vput(vp);
1033 		vp = nvp;
1034 	}
1035 	*vpp = vp;
1036 	return (0);
1037 }
1038 
1039 /*
1040  * Check to see if the new vnode represents a special device
1041  * for which we already have a vnode (either because of
1042  * bdevvp() or because of a different vnode representing
1043  * the same block device). If such an alias exists, deallocate
1044  * the existing contents and return the aliased vnode. The
1045  * caller is responsible for filling it with its new contents.
1046  */
1047 struct vnode *
1048 checkalias(nvp, nvp_rdev, mp)
1049 	struct vnode *nvp;
1050 	dev_t nvp_rdev;
1051 	struct mount *mp;
1052 {
1053 	struct proc *p = curproc;       /* XXX */
1054 	struct vnode *vp;
1055 	struct vnode **vpp;
1056 
1057 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1058 		return (NULLVP);
1059 
1060 	vpp = &speclisth[SPECHASH(nvp_rdev)];
1061 loop:
1062 	simple_lock(&spechash_slock);
1063 	for (vp = *vpp; vp; vp = vp->v_specnext) {
1064 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1065 			continue;
1066 		/*
1067 		 * Alias, but not in use, so flush it out.
1068 		 */
1069 		simple_lock(&vp->v_interlock);
1070 		if (vp->v_usecount == 0) {
1071 			simple_unlock(&spechash_slock);
1072 			vgonel(vp, p);
1073 			goto loop;
1074 		}
1075 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
1076 			simple_unlock(&spechash_slock);
1077 			goto loop;
1078 		}
1079 		break;
1080 	}
1081 	if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) {
1082 		MALLOC(nvp->v_specinfo, struct specinfo *,
1083 			sizeof(struct specinfo), M_VNODE, M_NOWAIT);
1084 		/* XXX Erg. */
1085 		if (nvp->v_specinfo == NULL) {
1086 			simple_unlock(&spechash_slock);
1087 			uvm_wait("checkalias");
1088 			goto loop;
1089 		}
1090 
1091 		nvp->v_rdev = nvp_rdev;
1092 		nvp->v_hashchain = vpp;
1093 		nvp->v_specnext = *vpp;
1094 		nvp->v_specmountpoint = NULL;
1095 		simple_unlock(&spechash_slock);
1096 		nvp->v_speclockf = NULL;
1097 		*vpp = nvp;
1098 		if (vp != NULLVP) {
1099 			nvp->v_flag |= VALIASED;
1100 			vp->v_flag |= VALIASED;
1101 			vput(vp);
1102 		}
1103 		return (NULLVP);
1104 	}
1105 	simple_unlock(&spechash_slock);
1106 	VOP_UNLOCK(vp, 0);
1107 	simple_lock(&vp->v_interlock);
1108 	vclean(vp, 0, p);
1109 	vp->v_op = nvp->v_op;
1110 	vp->v_tag = nvp->v_tag;
1111 	vp->v_vnlock = &vp->v_lock;
1112 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
1113 	nvp->v_type = VNON;
1114 	insmntque(vp, mp);
1115 	return (vp);
1116 }
1117 
1118 /*
1119  * Grab a particular vnode from the free list, increment its
1120  * reference count and lock it. If the vnode lock bit is set the
1121  * vnode is being eliminated in vgone. In that case, we can not
1122  * grab the vnode, so the process is awakened when the transition is
1123  * completed, and an error returned to indicate that the vnode is no
1124  * longer usable (possibly having been changed to a new file system type).
1125  */
1126 int
1127 vget(vp, flags)
1128 	struct vnode *vp;
1129 	int flags;
1130 {
1131 	int error;
1132 
1133 	/*
1134 	 * If the vnode is in the process of being cleaned out for
1135 	 * another use, we wait for the cleaning to finish and then
1136 	 * return failure. Cleaning is determined by checking that
1137 	 * the VXLOCK flag is set.
1138 	 */
1139 
1140 	if ((flags & LK_INTERLOCK) == 0)
1141 		simple_lock(&vp->v_interlock);
1142 	if (vp->v_flag & VXLOCK) {
1143 		if (flags & LK_NOWAIT) {
1144 			simple_unlock(&vp->v_interlock);
1145 			return EBUSY;
1146 		}
1147 		vp->v_flag |= VXWANT;
1148 		ltsleep((caddr_t)vp, PINOD|PNORELOCK,
1149 		    "vget", 0, &vp->v_interlock);
1150 		return (ENOENT);
1151 	}
1152 	if (vp->v_usecount == 0) {
1153 		simple_lock(&vnode_free_list_slock);
1154 		if (vp->v_holdcnt > 0)
1155 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1156 		else
1157 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1158 		simple_unlock(&vnode_free_list_slock);
1159 	}
1160 	vp->v_usecount++;
1161 #ifdef DIAGNOSTIC
1162 	if (vp->v_usecount == 0) {
1163 		vprint("vget", vp);
1164 		panic("vget: usecount overflow, vp %p", vp);
1165 	}
1166 #endif
1167 	if (flags & LK_TYPE_MASK) {
1168 		if ((error = vn_lock(vp, flags | LK_INTERLOCK))) {
1169 			/*
1170 			 * must expand vrele here because we do not want
1171 			 * to call VOP_INACTIVE if the reference count
1172 			 * drops back to zero since it was never really
1173 			 * active. We must remove it from the free list
1174 			 * before sleeping so that multiple processes do
1175 			 * not try to recycle it.
1176 			 */
1177 			simple_lock(&vp->v_interlock);
1178 			vp->v_usecount--;
1179 			if (vp->v_usecount > 0) {
1180 				simple_unlock(&vp->v_interlock);
1181 				return (error);
1182 			}
1183 			/*
1184 			 * insert at tail of LRU list
1185 			 */
1186 			simple_lock(&vnode_free_list_slock);
1187 			if (vp->v_holdcnt > 0)
1188 				TAILQ_INSERT_TAIL(&vnode_hold_list, vp,
1189 				    v_freelist);
1190 			else
1191 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
1192 				    v_freelist);
1193 			simple_unlock(&vnode_free_list_slock);
1194 			simple_unlock(&vp->v_interlock);
1195 		}
1196 		return (error);
1197 	}
1198 	simple_unlock(&vp->v_interlock);
1199 	return (0);
1200 }
1201 
1202 /*
1203  * vput(), just unlock and vrele()
1204  */
1205 void
1206 vput(vp)
1207 	struct vnode *vp;
1208 {
1209 	struct proc *p = curproc;	/* XXX */
1210 
1211 #ifdef DIAGNOSTIC
1212 	if (vp == NULL)
1213 		panic("vput: null vp");
1214 #endif
1215 	simple_lock(&vp->v_interlock);
1216 	vp->v_usecount--;
1217 	if (vp->v_usecount > 0) {
1218 		simple_unlock(&vp->v_interlock);
1219 		VOP_UNLOCK(vp, 0);
1220 		return;
1221 	}
1222 #ifdef DIAGNOSTIC
1223 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1224 		vprint("vput: bad ref count", vp);
1225 		panic("vput: ref cnt");
1226 	}
1227 #endif
1228 	/*
1229 	 * Insert at tail of LRU list.
1230 	 */
1231 	simple_lock(&vnode_free_list_slock);
1232 	if (vp->v_holdcnt > 0)
1233 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1234 	else
1235 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1236 	simple_unlock(&vnode_free_list_slock);
1237 	if (vp->v_flag & VTEXT) {
1238 		uvmexp.vtextpages -= vp->v_uvm.u_obj.uo_npages;
1239 		uvmexp.vnodepages += vp->v_uvm.u_obj.uo_npages;
1240 	}
1241 	vp->v_flag &= ~VTEXT;
1242 	simple_unlock(&vp->v_interlock);
1243 	VOP_INACTIVE(vp, p);
1244 }
1245 
1246 /*
1247  * Vnode release.
1248  * If count drops to zero, call inactive routine and return to freelist.
1249  */
1250 void
1251 vrele(vp)
1252 	struct vnode *vp;
1253 {
1254 	struct proc *p = curproc;	/* XXX */
1255 
1256 #ifdef DIAGNOSTIC
1257 	if (vp == NULL)
1258 		panic("vrele: null vp");
1259 #endif
1260 	simple_lock(&vp->v_interlock);
1261 	vp->v_usecount--;
1262 	if (vp->v_usecount > 0) {
1263 		simple_unlock(&vp->v_interlock);
1264 		return;
1265 	}
1266 #ifdef DIAGNOSTIC
1267 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1268 		vprint("vrele: bad ref count", vp);
1269 		panic("vrele: ref cnt vp %p", vp);
1270 	}
1271 #endif
1272 	/*
1273 	 * Insert at tail of LRU list.
1274 	 */
1275 	simple_lock(&vnode_free_list_slock);
1276 	if (vp->v_holdcnt > 0)
1277 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1278 	else
1279 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1280 	simple_unlock(&vnode_free_list_slock);
1281 	if (vp->v_flag & VTEXT) {
1282 		uvmexp.vtextpages -= vp->v_uvm.u_obj.uo_npages;
1283 		uvmexp.vnodepages += vp->v_uvm.u_obj.uo_npages;
1284 	}
1285 	vp->v_flag &= ~VTEXT;
1286 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0)
1287 		VOP_INACTIVE(vp, p);
1288 }
1289 
1290 #ifdef DIAGNOSTIC
1291 /*
1292  * Page or buffer structure gets a reference.
1293  */
1294 void
1295 vhold(vp)
1296 	struct vnode *vp;
1297 {
1298 
1299 	/*
1300 	 * If it is on the freelist and the hold count is currently
1301 	 * zero, move it to the hold list. The test of the back
1302 	 * pointer and the use reference count of zero is because
1303 	 * it will be removed from a free list by getnewvnode,
1304 	 * but will not have its reference count incremented until
1305 	 * after calling vgone. If the reference count were
1306 	 * incremented first, vgone would (incorrectly) try to
1307 	 * close the previous instance of the underlying object.
1308 	 * So, the back pointer is explicitly set to `0xdeadb' in
1309 	 * getnewvnode after removing it from a freelist to ensure
1310 	 * that we do not try to move it here.
1311 	 */
1312   	simple_lock(&vp->v_interlock);
1313 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1314 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1315 		simple_lock(&vnode_free_list_slock);
1316 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1317 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1318 		simple_unlock(&vnode_free_list_slock);
1319 	}
1320 	vp->v_holdcnt++;
1321 	simple_unlock(&vp->v_interlock);
1322 }
1323 
1324 /*
1325  * Page or buffer structure frees a reference.
1326  */
1327 void
1328 holdrele(vp)
1329 	struct vnode *vp;
1330 {
1331 
1332 	simple_lock(&vp->v_interlock);
1333 	if (vp->v_holdcnt <= 0)
1334 		panic("holdrele: holdcnt vp %p", vp);
1335 	vp->v_holdcnt--;
1336 
1337 	/*
1338 	 * If it is on the holdlist and the hold count drops to
1339 	 * zero, move it to the free list. The test of the back
1340 	 * pointer and the use reference count of zero is because
1341 	 * it will be removed from a free list by getnewvnode,
1342 	 * but will not have its reference count incremented until
1343 	 * after calling vgone. If the reference count were
1344 	 * incremented first, vgone would (incorrectly) try to
1345 	 * close the previous instance of the underlying object.
1346 	 * So, the back pointer is explicitly set to `0xdeadb' in
1347 	 * getnewvnode after removing it from a freelist to ensure
1348 	 * that we do not try to move it here.
1349 	 */
1350 
1351 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1352 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1353 		simple_lock(&vnode_free_list_slock);
1354 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1355 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1356 		simple_unlock(&vnode_free_list_slock);
1357 	}
1358 	simple_unlock(&vp->v_interlock);
1359 }
1360 
1361 /*
1362  * Vnode reference.
1363  */
1364 void
1365 vref(vp)
1366 	struct vnode *vp;
1367 {
1368 
1369 	simple_lock(&vp->v_interlock);
1370 	if (vp->v_usecount <= 0)
1371 		panic("vref used where vget required, vp %p", vp);
1372 	vp->v_usecount++;
1373 #ifdef DIAGNOSTIC
1374 	if (vp->v_usecount == 0) {
1375 		vprint("vref", vp);
1376 		panic("vref: usecount overflow, vp %p", vp);
1377 	}
1378 #endif
1379 	simple_unlock(&vp->v_interlock);
1380 }
1381 #endif /* DIAGNOSTIC */
1382 
1383 /*
1384  * Remove any vnodes in the vnode table belonging to mount point mp.
1385  *
1386  * If MNT_NOFORCE is specified, there should not be any active ones,
1387  * return error if any are found (nb: this is a user error, not a
1388  * system error). If MNT_FORCE is specified, detach any active vnodes
1389  * that are found.
1390  */
1391 #ifdef DEBUG
1392 int busyprt = 0;	/* print out busy vnodes */
1393 struct ctldebug debug1 = { "busyprt", &busyprt };
1394 #endif
1395 
1396 int
1397 vflush(mp, skipvp, flags)
1398 	struct mount *mp;
1399 	struct vnode *skipvp;
1400 	int flags;
1401 {
1402 	struct proc *p = curproc;	/* XXX */
1403 	struct vnode *vp, *nvp;
1404 	int busy = 0;
1405 
1406 	simple_lock(&mntvnode_slock);
1407 loop:
1408 	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
1409 		if (vp->v_mount != mp)
1410 			goto loop;
1411 		nvp = vp->v_mntvnodes.le_next;
1412 		/*
1413 		 * Skip over a selected vnode.
1414 		 */
1415 		if (vp == skipvp)
1416 			continue;
1417 		simple_lock(&vp->v_interlock);
1418 		/*
1419 		 * Skip over a vnodes marked VSYSTEM.
1420 		 */
1421 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1422 			simple_unlock(&vp->v_interlock);
1423 			continue;
1424 		}
1425 		/*
1426 		 * If WRITECLOSE is set, only flush out regular file
1427 		 * vnodes open for writing.
1428 		 */
1429 		if ((flags & WRITECLOSE) &&
1430 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1431 			simple_unlock(&vp->v_interlock);
1432 			continue;
1433 		}
1434 		/*
1435 		 * With v_usecount == 0, all we need to do is clear
1436 		 * out the vnode data structures and we are done.
1437 		 */
1438 		if (vp->v_usecount == 0) {
1439 			simple_unlock(&mntvnode_slock);
1440 			vgonel(vp, p);
1441 			simple_lock(&mntvnode_slock);
1442 			continue;
1443 		}
1444 		/*
1445 		 * If FORCECLOSE is set, forcibly close the vnode.
1446 		 * For block or character devices, revert to an
1447 		 * anonymous device. For all other files, just kill them.
1448 		 */
1449 		if (flags & FORCECLOSE) {
1450 			simple_unlock(&mntvnode_slock);
1451 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1452 				vgonel(vp, p);
1453 			} else {
1454 				vclean(vp, 0, p);
1455 				vp->v_op = spec_vnodeop_p;
1456 				insmntque(vp, (struct mount *)0);
1457 			}
1458 			simple_lock(&mntvnode_slock);
1459 			continue;
1460 		}
1461 #ifdef DEBUG
1462 		if (busyprt)
1463 			vprint("vflush: busy vnode", vp);
1464 #endif
1465 		simple_unlock(&vp->v_interlock);
1466 		busy++;
1467 	}
1468 	simple_unlock(&mntvnode_slock);
1469 	if (busy)
1470 		return (EBUSY);
1471 	return (0);
1472 }
1473 
1474 /*
1475  * Disassociate the underlying file system from a vnode.
1476  */
1477 void
1478 vclean(vp, flags, p)
1479 	struct vnode *vp;
1480 	int flags;
1481 	struct proc *p;
1482 {
1483 	int active;
1484 
1485 	/*
1486 	 * Check to see if the vnode is in use.
1487 	 * If so we have to reference it before we clean it out
1488 	 * so that its count cannot fall to zero and generate a
1489 	 * race against ourselves to recycle it.
1490 	 */
1491 	if ((active = vp->v_usecount) != 0) {
1492 		/* We have the vnode interlock. */
1493 		vp->v_usecount++;
1494 #ifdef DIAGNOSTIC
1495 		if (vp->v_usecount == 0) {
1496 			vprint("vclean", vp);
1497 			panic("vclean: usecount overflow");
1498 		}
1499 #endif
1500 	}
1501 
1502 	/*
1503 	 * Prevent the vnode from being recycled or
1504 	 * brought into use while we clean it out.
1505 	 */
1506 	if (vp->v_flag & VXLOCK)
1507 		panic("vclean: deadlock, vp %p", vp);
1508 	vp->v_flag |= VXLOCK;
1509 	if (vp->v_flag & VTEXT) {
1510 		uvmexp.vtextpages -= vp->v_uvm.u_obj.uo_npages;
1511 		uvmexp.vnodepages += vp->v_uvm.u_obj.uo_npages;
1512 	}
1513 	vp->v_flag &= ~VTEXT;
1514 
1515 	/*
1516 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1517 	 * have the object locked while it cleans it out. The VOP_LOCK
1518 	 * ensures that the VOP_INACTIVE routine is done with its work.
1519 	 * For active vnodes, it ensures that no other activity can
1520 	 * occur while the underlying object is being cleaned out.
1521 	 */
1522 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK);
1523 
1524 	/*
1525 	 * Clean out any cached data associated with the vnode.
1526 	 */
1527 	if (flags & DOCLOSE)
1528 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1529 
1530 	/*
1531 	 * If purging an active vnode, it must be closed and
1532 	 * deactivated before being reclaimed. Note that the
1533 	 * VOP_INACTIVE will unlock the vnode.
1534 	 */
1535 	if (active) {
1536 		if (flags & DOCLOSE)
1537 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL);
1538 		VOP_INACTIVE(vp, p);
1539 	} else {
1540 		/*
1541 		 * Any other processes trying to obtain this lock must first
1542 		 * wait for VXLOCK to clear, then call the new lock operation.
1543 		 */
1544 		VOP_UNLOCK(vp, 0);
1545 	}
1546 	/*
1547 	 * Reclaim the vnode.
1548 	 */
1549 	if (VOP_RECLAIM(vp, p))
1550 		panic("vclean: cannot reclaim, vp %p", vp);
1551 	if (active) {
1552 		/*
1553 		 * Inline copy of vrele() since VOP_INACTIVE
1554 		 * has already been called.
1555 		 */
1556 		simple_lock(&vp->v_interlock);
1557 		if (--vp->v_usecount <= 0) {
1558 #ifdef DIAGNOSTIC
1559 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1560 				vprint("vclean: bad ref count", vp);
1561 				panic("vclean: ref cnt");
1562 			}
1563 #endif
1564 			/*
1565 			 * Insert at tail of LRU list.
1566 			 */
1567 
1568 			simple_unlock(&vp->v_interlock);
1569 			simple_lock(&vnode_free_list_slock);
1570 #ifdef DIAGNOSTIC
1571 			if (vp->v_vnlock) {
1572 				if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0)
1573 					vprint("vclean: lock not drained", vp);
1574 			}
1575 			if (vp->v_holdcnt > 0)
1576 				panic("vclean: not clean, vp %p", vp);
1577 #endif
1578 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1579 			simple_unlock(&vnode_free_list_slock);
1580 		} else
1581 			simple_unlock(&vp->v_interlock);
1582 	}
1583 
1584 	cache_purge(vp);
1585 
1586 	/*
1587 	 * Done with purge, notify sleepers of the grim news.
1588 	 */
1589 	vp->v_op = dead_vnodeop_p;
1590 	vp->v_tag = VT_NON;
1591 	simple_lock(&vp->v_interlock);
1592 	vp->v_flag &= ~VXLOCK;
1593 	if (vp->v_flag & VXWANT) {
1594 		vp->v_flag &= ~VXWANT;
1595 		simple_unlock(&vp->v_interlock);
1596 		wakeup((caddr_t)vp);
1597 	} else
1598 		simple_unlock(&vp->v_interlock);
1599 }
1600 
1601 /*
1602  * Recycle an unused vnode to the front of the free list.
1603  * Release the passed interlock if the vnode will be recycled.
1604  */
1605 int
1606 vrecycle(vp, inter_lkp, p)
1607 	struct vnode *vp;
1608 	struct simplelock *inter_lkp;
1609 	struct proc *p;
1610 {
1611 
1612 	simple_lock(&vp->v_interlock);
1613 	if (vp->v_usecount == 0) {
1614 		if (inter_lkp)
1615 			simple_unlock(inter_lkp);
1616 		vgonel(vp, p);
1617 		return (1);
1618 	}
1619 	simple_unlock(&vp->v_interlock);
1620 	return (0);
1621 }
1622 
1623 /*
1624  * Eliminate all activity associated with a vnode
1625  * in preparation for reuse.
1626  */
1627 void
1628 vgone(vp)
1629 	struct vnode *vp;
1630 {
1631 	struct proc *p = curproc;	/* XXX */
1632 
1633 	simple_lock(&vp->v_interlock);
1634 	vgonel(vp, p);
1635 }
1636 
1637 /*
1638  * vgone, with the vp interlock held.
1639  */
1640 void
1641 vgonel(vp, p)
1642 	struct vnode *vp;
1643 	struct proc *p;
1644 {
1645 	struct vnode *vq;
1646 	struct vnode *vx;
1647 
1648 	/*
1649 	 * If a vgone (or vclean) is already in progress,
1650 	 * wait until it is done and return.
1651 	 */
1652 	if (vp->v_flag & VXLOCK) {
1653 		vp->v_flag |= VXWANT;
1654 		ltsleep((caddr_t)vp, PINOD | PNORELOCK,
1655 		    "vgone", 0, &vp->v_interlock);
1656 		return;
1657 	}
1658 	/*
1659 	 * Clean out the filesystem specific data.
1660 	 */
1661 	vclean(vp, DOCLOSE, p);
1662 	/*
1663 	 * Delete from old mount point vnode list, if on one.
1664 	 */
1665 	if (vp->v_mount != NULL)
1666 		insmntque(vp, (struct mount *)0);
1667 	/*
1668 	 * If special device, remove it from special device alias list.
1669 	 * if it is on one.
1670 	 */
1671 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1672 		simple_lock(&spechash_slock);
1673 		if (vp->v_hashchain != NULL) {
1674 			if (*vp->v_hashchain == vp) {
1675 				*vp->v_hashchain = vp->v_specnext;
1676 			} else {
1677 				for (vq = *vp->v_hashchain; vq;
1678 							vq = vq->v_specnext) {
1679 					if (vq->v_specnext != vp)
1680 						continue;
1681 					vq->v_specnext = vp->v_specnext;
1682 					break;
1683 				}
1684 				if (vq == NULL)
1685 					panic("missing bdev");
1686 			}
1687 			if (vp->v_flag & VALIASED) {
1688 				vx = NULL;
1689 				for (vq = *vp->v_hashchain; vq;
1690 							vq = vq->v_specnext) {
1691 					if (vq->v_rdev != vp->v_rdev ||
1692 					    vq->v_type != vp->v_type)
1693 						continue;
1694 					if (vx)
1695 						break;
1696 					vx = vq;
1697 				}
1698 				if (vx == NULL)
1699 					panic("missing alias");
1700 				if (vq == NULL)
1701 					vx->v_flag &= ~VALIASED;
1702 				vp->v_flag &= ~VALIASED;
1703 			}
1704 		}
1705 		simple_unlock(&spechash_slock);
1706 		FREE(vp->v_specinfo, M_VNODE);
1707 		vp->v_specinfo = NULL;
1708 	}
1709 	/*
1710 	 * If it is on the freelist and not already at the head,
1711 	 * move it to the head of the list. The test of the back
1712 	 * pointer and the reference count of zero is because
1713 	 * it will be removed from the free list by getnewvnode,
1714 	 * but will not have its reference count incremented until
1715 	 * after calling vgone. If the reference count were
1716 	 * incremented first, vgone would (incorrectly) try to
1717 	 * close the previous instance of the underlying object.
1718 	 * So, the back pointer is explicitly set to `0xdeadb' in
1719 	 * getnewvnode after removing it from the freelist to ensure
1720 	 * that we do not try to move it here.
1721 	 */
1722 	if (vp->v_usecount == 0) {
1723 		simple_lock(&vnode_free_list_slock);
1724 		if (vp->v_holdcnt > 0)
1725 			panic("vgonel: not clean, vp %p", vp);
1726 		if (vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb &&
1727 		    TAILQ_FIRST(&vnode_free_list) != vp) {
1728 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1729 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1730 		}
1731 		simple_unlock(&vnode_free_list_slock);
1732 	}
1733 	vp->v_type = VBAD;
1734 }
1735 
1736 /*
1737  * Lookup a vnode by device number.
1738  */
1739 int
1740 vfinddev(dev, type, vpp)
1741 	dev_t dev;
1742 	enum vtype type;
1743 	struct vnode **vpp;
1744 {
1745 	struct vnode *vp;
1746 	int rc = 0;
1747 
1748 	simple_lock(&spechash_slock);
1749 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1750 		if (dev != vp->v_rdev || type != vp->v_type)
1751 			continue;
1752 		*vpp = vp;
1753 		rc = 1;
1754 		break;
1755 	}
1756 	simple_unlock(&spechash_slock);
1757 	return (rc);
1758 }
1759 
1760 /*
1761  * Revoke all the vnodes corresponding to the specified minor number
1762  * range (endpoints inclusive) of the specified major.
1763  */
1764 void
1765 vdevgone(maj, minl, minh, type)
1766 	int maj, minl, minh;
1767 	enum vtype type;
1768 {
1769 	struct vnode *vp;
1770 	int mn;
1771 
1772 	for (mn = minl; mn <= minh; mn++)
1773 		if (vfinddev(makedev(maj, mn), type, &vp))
1774 			VOP_REVOKE(vp, REVOKEALL);
1775 }
1776 
1777 /*
1778  * Calculate the total number of references to a special device.
1779  */
1780 int
1781 vcount(vp)
1782 	struct vnode *vp;
1783 {
1784 	struct vnode *vq, *vnext;
1785 	int count;
1786 
1787 loop:
1788 	if ((vp->v_flag & VALIASED) == 0)
1789 		return (vp->v_usecount);
1790 	simple_lock(&spechash_slock);
1791 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1792 		vnext = vq->v_specnext;
1793 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1794 			continue;
1795 		/*
1796 		 * Alias, but not in use, so flush it out.
1797 		 */
1798 		if (vq->v_usecount == 0 && vq != vp &&
1799 		    (vq->v_flag & VXLOCK) == 0) {
1800 			simple_unlock(&spechash_slock);
1801 			vgone(vq);
1802 			goto loop;
1803 		}
1804 		count += vq->v_usecount;
1805 	}
1806 	simple_unlock(&spechash_slock);
1807 	return (count);
1808 }
1809 
1810 /*
1811  * Print out a description of a vnode.
1812  */
1813 static const char * const typename[] =
1814    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1815 
1816 void
1817 vprint(label, vp)
1818 	char *label;
1819 	struct vnode *vp;
1820 {
1821 	char buf[64];
1822 
1823 	if (label != NULL)
1824 		printf("%s: ", label);
1825 	printf("tag %d type %s, usecount %d, writecount %ld, refcount %ld,",
1826 	    vp->v_tag, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1827 	    vp->v_holdcnt);
1828 	buf[0] = '\0';
1829 	if (vp->v_flag & VROOT)
1830 		strcat(buf, "|VROOT");
1831 	if (vp->v_flag & VTEXT)
1832 		strcat(buf, "|VTEXT");
1833 	if (vp->v_flag & VSYSTEM)
1834 		strcat(buf, "|VSYSTEM");
1835 	if (vp->v_flag & VXLOCK)
1836 		strcat(buf, "|VXLOCK");
1837 	if (vp->v_flag & VXWANT)
1838 		strcat(buf, "|VXWANT");
1839 	if (vp->v_flag & VBWAIT)
1840 		strcat(buf, "|VBWAIT");
1841 	if (vp->v_flag & VALIASED)
1842 		strcat(buf, "|VALIASED");
1843 	if (buf[0] != '\0')
1844 		printf(" flags (%s)", &buf[1]);
1845 	if (vp->v_data == NULL) {
1846 		printf("\n");
1847 	} else {
1848 		printf("\n\t");
1849 		VOP_PRINT(vp);
1850 	}
1851 }
1852 
1853 #ifdef DEBUG
1854 /*
1855  * List all of the locked vnodes in the system.
1856  * Called when debugging the kernel.
1857  */
1858 void
1859 printlockedvnodes()
1860 {
1861 	struct mount *mp, *nmp;
1862 	struct vnode *vp;
1863 
1864 	printf("Locked vnodes\n");
1865 	simple_lock(&mountlist_slock);
1866 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
1867 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
1868 			nmp = mp->mnt_list.cqe_next;
1869 			continue;
1870 		}
1871 		for (vp = mp->mnt_vnodelist.lh_first;
1872 		     vp != NULL;
1873 		     vp = vp->v_mntvnodes.le_next) {
1874 			if (VOP_ISLOCKED(vp))
1875 				vprint((char *)0, vp);
1876 		}
1877 		simple_lock(&mountlist_slock);
1878 		nmp = mp->mnt_list.cqe_next;
1879 		vfs_unbusy(mp);
1880 	}
1881 	simple_unlock(&mountlist_slock);
1882 }
1883 #endif
1884 
1885 /*
1886  * Top level filesystem related information gathering.
1887  */
1888 int
1889 vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
1890 	int *name;
1891 	u_int namelen;
1892 	void *oldp;
1893 	size_t *oldlenp;
1894 	void *newp;
1895 	size_t newlen;
1896 	struct proc *p;
1897 {
1898 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1899 	struct vfsconf vfc;
1900 	extern const char * const mountcompatnames[];
1901 	extern int nmountcompatnames;
1902 #endif
1903 	struct vfsops *vfsp;
1904 
1905 	/* all sysctl names at this level are at least name and field */
1906 	if (namelen < 2)
1907 		return (ENOTDIR);		/* overloaded */
1908 
1909 	/* Not generic: goes to file system. */
1910 	if (name[0] != VFS_GENERIC) {
1911 		static const struct ctlname vfsnames[VFS_MAXID+1]=CTL_VFS_NAMES;
1912 		const char *vfsname;
1913 
1914 		if (name[0] < 0 || name[0] > VFS_MAXID
1915 		    || (vfsname = vfsnames[name[0]].ctl_name) == NULL)
1916 			return (EOPNOTSUPP);
1917 
1918 		vfsp = vfs_getopsbyname(vfsname);
1919 		if (vfsp == NULL || vfsp->vfs_sysctl == NULL)
1920 			return (EOPNOTSUPP);
1921 		return ((*vfsp->vfs_sysctl)(&name[1], namelen - 1,
1922 		    oldp, oldlenp, newp, newlen, p));
1923 	}
1924 
1925 	/* The rest are generic vfs sysctls. */
1926 	switch (name[1]) {
1927 	case VFS_USERMOUNT:
1928 		return sysctl_int(oldp, oldlenp, newp, newlen, &dovfsusermount);
1929 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1930 	case VFS_MAXTYPENUM:
1931 		/*
1932 		 * Provided for 4.4BSD-Lite2 compatibility.
1933 		 */
1934 		return (sysctl_rdint(oldp, oldlenp, newp, nmountcompatnames));
1935 	case VFS_CONF:
1936 		/*
1937 		 * Special: a node, next is a file system name.
1938 		 * Provided for 4.4BSD-Lite2 compatibility.
1939 		 */
1940 		if (namelen < 3)
1941 			return (ENOTDIR);	/* overloaded */
1942 		if (name[2] >= nmountcompatnames || name[2] < 0 ||
1943 		    mountcompatnames[name[2]] == NULL)
1944 			return (EOPNOTSUPP);
1945 		vfsp = vfs_getopsbyname(mountcompatnames[name[2]]);
1946 		if (vfsp == NULL)
1947 			return (EOPNOTSUPP);
1948 		vfc.vfc_vfsops = vfsp;
1949 		strncpy(vfc.vfc_name, vfsp->vfs_name, MFSNAMELEN);
1950 		vfc.vfc_typenum = name[2];
1951 		vfc.vfc_refcount = vfsp->vfs_refcount;
1952 		vfc.vfc_flags = 0;
1953 		vfc.vfc_mountroot = vfsp->vfs_mountroot;
1954 		vfc.vfc_next = NULL;
1955 		return (sysctl_rdstruct(oldp, oldlenp, newp, &vfc,
1956 		    sizeof(struct vfsconf)));
1957 #endif
1958 	default:
1959 		break;
1960 	}
1961 	return (EOPNOTSUPP);
1962 }
1963 
1964 int kinfo_vdebug = 1;
1965 int kinfo_vgetfailed;
1966 #define KINFO_VNODESLOP	10
1967 /*
1968  * Dump vnode list (via sysctl).
1969  * Copyout address of vnode followed by vnode.
1970  */
1971 /* ARGSUSED */
1972 int
1973 sysctl_vnode(where, sizep, p)
1974 	char *where;
1975 	size_t *sizep;
1976 	struct proc *p;
1977 {
1978 	struct mount *mp, *nmp;
1979 	struct vnode *nvp, *vp;
1980 	char *bp = where, *savebp;
1981 	char *ewhere;
1982 	int error;
1983 
1984 #define VPTRSZ	sizeof(struct vnode *)
1985 #define VNODESZ	sizeof(struct vnode)
1986 	if (where == NULL) {
1987 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
1988 		return (0);
1989 	}
1990 	ewhere = where + *sizep;
1991 
1992 	simple_lock(&mountlist_slock);
1993 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
1994 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
1995 			nmp = mp->mnt_list.cqe_next;
1996 			continue;
1997 		}
1998 		savebp = bp;
1999 again:
2000 		simple_lock(&mntvnode_slock);
2001 		for (vp = mp->mnt_vnodelist.lh_first;
2002 		     vp != NULL;
2003 		     vp = nvp) {
2004 			/*
2005 			 * Check that the vp is still associated with
2006 			 * this filesystem.  RACE: could have been
2007 			 * recycled onto the same filesystem.
2008 			 */
2009 			if (vp->v_mount != mp) {
2010 				simple_unlock(&mntvnode_slock);
2011 				if (kinfo_vdebug)
2012 					printf("kinfo: vp changed\n");
2013 				bp = savebp;
2014 				goto again;
2015 			}
2016 			nvp = vp->v_mntvnodes.le_next;
2017 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2018 				simple_unlock(&mntvnode_slock);
2019 				*sizep = bp - where;
2020 				return (ENOMEM);
2021 			}
2022 			simple_unlock(&mntvnode_slock);
2023 			if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
2024 			   (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
2025 				return (error);
2026 			bp += VPTRSZ + VNODESZ;
2027 			simple_lock(&mntvnode_slock);
2028 		}
2029 		simple_unlock(&mntvnode_slock);
2030 		simple_lock(&mountlist_slock);
2031 		nmp = mp->mnt_list.cqe_next;
2032 		vfs_unbusy(mp);
2033 	}
2034 	simple_unlock(&mountlist_slock);
2035 
2036 	*sizep = bp - where;
2037 	return (0);
2038 }
2039 
2040 /*
2041  * Check to see if a filesystem is mounted on a block device.
2042  */
2043 int
2044 vfs_mountedon(vp)
2045 	struct vnode *vp;
2046 {
2047 	struct vnode *vq;
2048 	int error = 0;
2049 
2050 	if (vp->v_specmountpoint != NULL)
2051 		return (EBUSY);
2052 	if (vp->v_flag & VALIASED) {
2053 		simple_lock(&spechash_slock);
2054 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2055 			if (vq->v_rdev != vp->v_rdev ||
2056 			    vq->v_type != vp->v_type)
2057 				continue;
2058 			if (vq->v_specmountpoint != NULL) {
2059 				error = EBUSY;
2060 				break;
2061 			}
2062 		}
2063 		simple_unlock(&spechash_slock);
2064 	}
2065 	return (error);
2066 }
2067 
2068 /*
2069  * Build hash lists of net addresses and hang them off the mount point.
2070  * Called by ufs_mount() to set up the lists of export addresses.
2071  */
2072 static int
2073 vfs_hang_addrlist(mp, nep, argp)
2074 	struct mount *mp;
2075 	struct netexport *nep;
2076 	struct export_args *argp;
2077 {
2078 	struct netcred *np, *enp;
2079 	struct radix_node_head *rnh;
2080 	int i;
2081 	struct radix_node *rn;
2082 	struct sockaddr *saddr, *smask = 0;
2083 	struct domain *dom;
2084 	int error;
2085 
2086 	if (argp->ex_addrlen == 0) {
2087 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2088 			return (EPERM);
2089 		np = &nep->ne_defexported;
2090 		np->netc_exflags = argp->ex_flags;
2091 		np->netc_anon = argp->ex_anon;
2092 		np->netc_anon.cr_ref = 1;
2093 		mp->mnt_flag |= MNT_DEFEXPORTED;
2094 		return (0);
2095 	}
2096 
2097 	if (argp->ex_addrlen > MLEN)
2098 		return (EINVAL);
2099 
2100 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2101 	np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK);
2102 	memset((caddr_t)np, 0, i);
2103 	saddr = (struct sockaddr *)(np + 1);
2104 	error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen);
2105 	if (error)
2106 		goto out;
2107 	if (saddr->sa_len > argp->ex_addrlen)
2108 		saddr->sa_len = argp->ex_addrlen;
2109 	if (argp->ex_masklen) {
2110 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2111 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2112 		if (error)
2113 			goto out;
2114 		if (smask->sa_len > argp->ex_masklen)
2115 			smask->sa_len = argp->ex_masklen;
2116 	}
2117 	i = saddr->sa_family;
2118 	if ((rnh = nep->ne_rtable[i]) == 0) {
2119 		/*
2120 		 * Seems silly to initialize every AF when most are not
2121 		 * used, do so on demand here
2122 		 */
2123 		for (dom = domains; dom; dom = dom->dom_next)
2124 			if (dom->dom_family == i && dom->dom_rtattach) {
2125 				dom->dom_rtattach((void **)&nep->ne_rtable[i],
2126 					dom->dom_rtoffset);
2127 				break;
2128 			}
2129 		if ((rnh = nep->ne_rtable[i]) == 0) {
2130 			error = ENOBUFS;
2131 			goto out;
2132 		}
2133 	}
2134 	rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh,
2135 		np->netc_rnodes);
2136 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
2137 		if (rn == 0) {
2138 			enp = (struct netcred *)(*rnh->rnh_lookup)(saddr,
2139 				smask, rnh);
2140 			if (enp == 0) {
2141 				error = EPERM;
2142 				goto out;
2143 			}
2144 		} else
2145 			enp = (struct netcred *)rn;
2146 
2147 		if (enp->netc_exflags != argp->ex_flags ||
2148 		    enp->netc_anon.cr_uid != argp->ex_anon.cr_uid ||
2149 		    enp->netc_anon.cr_gid != argp->ex_anon.cr_gid ||
2150 		    enp->netc_anon.cr_ngroups != argp->ex_anon.cr_ngroups ||
2151 		    memcmp(&enp->netc_anon.cr_groups, &argp->ex_anon.cr_groups,
2152 			enp->netc_anon.cr_ngroups))
2153 				error = EPERM;
2154 		else
2155 			error = 0;
2156 		goto out;
2157 	}
2158 	np->netc_exflags = argp->ex_flags;
2159 	np->netc_anon = argp->ex_anon;
2160 	np->netc_anon.cr_ref = 1;
2161 	return (0);
2162 out:
2163 	free(np, M_NETADDR);
2164 	return (error);
2165 }
2166 
2167 /* ARGSUSED */
2168 static int
2169 vfs_free_netcred(rn, w)
2170 	struct radix_node *rn;
2171 	void *w;
2172 {
2173 	struct radix_node_head *rnh = (struct radix_node_head *)w;
2174 
2175 	(*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh);
2176 	free((caddr_t)rn, M_NETADDR);
2177 	return (0);
2178 }
2179 
2180 /*
2181  * Free the net address hash lists that are hanging off the mount points.
2182  */
2183 static void
2184 vfs_free_addrlist(nep)
2185 	struct netexport *nep;
2186 {
2187 	int i;
2188 	struct radix_node_head *rnh;
2189 
2190 	for (i = 0; i <= AF_MAX; i++)
2191 		if ((rnh = nep->ne_rtable[i]) != NULL) {
2192 			(*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh);
2193 			free((caddr_t)rnh, M_RTABLE);
2194 			nep->ne_rtable[i] = 0;
2195 		}
2196 }
2197 
2198 int
2199 vfs_export(mp, nep, argp)
2200 	struct mount *mp;
2201 	struct netexport *nep;
2202 	struct export_args *argp;
2203 {
2204 	int error;
2205 
2206 	if (argp->ex_flags & MNT_DELEXPORT) {
2207 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2208 			vfs_setpublicfs(NULL, NULL, NULL);
2209 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2210 		}
2211 		vfs_free_addrlist(nep);
2212 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2213 	}
2214 	if (argp->ex_flags & MNT_EXPORTED) {
2215 		if (argp->ex_flags & MNT_EXPUBLIC) {
2216 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2217 				return (error);
2218 			mp->mnt_flag |= MNT_EXPUBLIC;
2219 		}
2220 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
2221 			return (error);
2222 		mp->mnt_flag |= MNT_EXPORTED;
2223 	}
2224 	return (0);
2225 }
2226 
2227 /*
2228  * Set the publicly exported filesystem (WebNFS). Currently, only
2229  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2230  */
2231 int
2232 vfs_setpublicfs(mp, nep, argp)
2233 	struct mount *mp;
2234 	struct netexport *nep;
2235 	struct export_args *argp;
2236 {
2237 	int error;
2238 	struct vnode *rvp;
2239 	char *cp;
2240 
2241 	/*
2242 	 * mp == NULL -> invalidate the current info, the FS is
2243 	 * no longer exported. May be called from either vfs_export
2244 	 * or unmount, so check if it hasn't already been done.
2245 	 */
2246 	if (mp == NULL) {
2247 		if (nfs_pub.np_valid) {
2248 			nfs_pub.np_valid = 0;
2249 			if (nfs_pub.np_index != NULL) {
2250 				FREE(nfs_pub.np_index, M_TEMP);
2251 				nfs_pub.np_index = NULL;
2252 			}
2253 		}
2254 		return (0);
2255 	}
2256 
2257 	/*
2258 	 * Only one allowed at a time.
2259 	 */
2260 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2261 		return (EBUSY);
2262 
2263 	/*
2264 	 * Get real filehandle for root of exported FS.
2265 	 */
2266 	memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle));
2267 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2268 
2269 	if ((error = VFS_ROOT(mp, &rvp)))
2270 		return (error);
2271 
2272 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2273 		return (error);
2274 
2275 	vput(rvp);
2276 
2277 	/*
2278 	 * If an indexfile was specified, pull it in.
2279 	 */
2280 	if (argp->ex_indexfile != NULL) {
2281 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2282 		    M_WAITOK);
2283 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2284 		    MAXNAMLEN, (size_t *)0);
2285 		if (!error) {
2286 			/*
2287 			 * Check for illegal filenames.
2288 			 */
2289 			for (cp = nfs_pub.np_index; *cp; cp++) {
2290 				if (*cp == '/') {
2291 					error = EINVAL;
2292 					break;
2293 				}
2294 			}
2295 		}
2296 		if (error) {
2297 			FREE(nfs_pub.np_index, M_TEMP);
2298 			return (error);
2299 		}
2300 	}
2301 
2302 	nfs_pub.np_mount = mp;
2303 	nfs_pub.np_valid = 1;
2304 	return (0);
2305 }
2306 
2307 struct netcred *
2308 vfs_export_lookup(mp, nep, nam)
2309 	struct mount *mp;
2310 	struct netexport *nep;
2311 	struct mbuf *nam;
2312 {
2313 	struct netcred *np;
2314 	struct radix_node_head *rnh;
2315 	struct sockaddr *saddr;
2316 
2317 	np = NULL;
2318 	if (mp->mnt_flag & MNT_EXPORTED) {
2319 		/*
2320 		 * Lookup in the export list first.
2321 		 */
2322 		if (nam != NULL) {
2323 			saddr = mtod(nam, struct sockaddr *);
2324 			rnh = nep->ne_rtable[saddr->sa_family];
2325 			if (rnh != NULL) {
2326 				np = (struct netcred *)
2327 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2328 							      rnh);
2329 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2330 					np = NULL;
2331 			}
2332 		}
2333 		/*
2334 		 * If no address match, use the default if it exists.
2335 		 */
2336 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2337 			np = &nep->ne_defexported;
2338 	}
2339 	return (np);
2340 }
2341 
2342 /*
2343  * Do the usual access checking.
2344  * file_mode, uid and gid are from the vnode in question,
2345  * while acc_mode and cred are from the VOP_ACCESS parameter list
2346  */
2347 int
2348 vaccess(type, file_mode, uid, gid, acc_mode, cred)
2349 	enum vtype type;
2350 	mode_t file_mode;
2351 	uid_t uid;
2352 	gid_t gid;
2353 	mode_t acc_mode;
2354 	struct ucred *cred;
2355 {
2356 	mode_t mask;
2357 
2358 	/*
2359 	 * Super-user always gets read/write access, but execute access depends
2360 	 * on at least one execute bit being set.
2361 	 */
2362 	if (cred->cr_uid == 0) {
2363 		if ((acc_mode & VEXEC) && type != VDIR &&
2364 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
2365 			return (EACCES);
2366 		return (0);
2367 	}
2368 
2369 	mask = 0;
2370 
2371 	/* Otherwise, check the owner. */
2372 	if (cred->cr_uid == uid) {
2373 		if (acc_mode & VEXEC)
2374 			mask |= S_IXUSR;
2375 		if (acc_mode & VREAD)
2376 			mask |= S_IRUSR;
2377 		if (acc_mode & VWRITE)
2378 			mask |= S_IWUSR;
2379 		return ((file_mode & mask) == mask ? 0 : EACCES);
2380 	}
2381 
2382 	/* Otherwise, check the groups. */
2383 	if (cred->cr_gid == gid || groupmember(gid, cred)) {
2384 		if (acc_mode & VEXEC)
2385 			mask |= S_IXGRP;
2386 		if (acc_mode & VREAD)
2387 			mask |= S_IRGRP;
2388 		if (acc_mode & VWRITE)
2389 			mask |= S_IWGRP;
2390 		return ((file_mode & mask) == mask ? 0 : EACCES);
2391 	}
2392 
2393 	/* Otherwise, check everyone else. */
2394 	if (acc_mode & VEXEC)
2395 		mask |= S_IXOTH;
2396 	if (acc_mode & VREAD)
2397 		mask |= S_IROTH;
2398 	if (acc_mode & VWRITE)
2399 		mask |= S_IWOTH;
2400 	return ((file_mode & mask) == mask ? 0 : EACCES);
2401 }
2402 
2403 /*
2404  * Unmount all file systems.
2405  * We traverse the list in reverse order under the assumption that doing so
2406  * will avoid needing to worry about dependencies.
2407  */
2408 void
2409 vfs_unmountall(p)
2410 	struct proc *p;
2411 {
2412 	struct mount *mp, *nmp;
2413 	int allerror, error;
2414 
2415 	for (allerror = 0,
2416 	     mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2417 		nmp = mp->mnt_list.cqe_prev;
2418 #ifdef DEBUG
2419 		printf("unmounting %s (%s)...\n",
2420 		    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
2421 #endif
2422 		/*
2423 		 * XXX Freeze syncer.  Must do this before locking the
2424 		 * mount point.  See dounmount() for details.
2425 		 */
2426 		lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL);
2427 		if (vfs_busy(mp, 0, 0)) {
2428 			lockmgr(&syncer_lock, LK_RELEASE, NULL);
2429 			continue;
2430 		}
2431 		if ((error = dounmount(mp, MNT_FORCE, p)) != 0) {
2432 			printf("unmount of %s failed with error %d\n",
2433 			    mp->mnt_stat.f_mntonname, error);
2434 			allerror = 1;
2435 		}
2436 	}
2437 	if (allerror)
2438 		printf("WARNING: some file systems would not unmount\n");
2439 }
2440 
2441 /*
2442  * Sync and unmount file systems before shutting down.
2443  */
2444 void
2445 vfs_shutdown()
2446 {
2447 	struct buf *bp;
2448 	int iter, nbusy, nbusy_prev = 0, dcount, s;
2449 	struct proc *p = curproc;
2450 
2451 	/* XXX we're certainly not running in proc0's context! */
2452 	if (p == NULL)
2453 		p = &proc0;
2454 
2455 	printf("syncing disks... ");
2456 
2457 	/* remove user process from run queue */
2458 	suspendsched();
2459 	(void) spl0();
2460 
2461 	/* avoid coming back this way again if we panic. */
2462 	doing_shutdown = 1;
2463 
2464 	sys_sync(p, NULL, NULL);
2465 
2466 	/* Wait for sync to finish. */
2467 	dcount = 10000;
2468 	for (iter = 0; iter < 20;) {
2469 		nbusy = 0;
2470 		for (bp = &buf[nbuf]; --bp >= buf; ) {
2471 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2472 				nbusy++;
2473 			/*
2474 			 * With soft updates, some buffers that are
2475 			 * written will be remarked as dirty until other
2476 			 * buffers are written.
2477 			 */
2478 			if (bp->b_vp && bp->b_vp->v_mount
2479 			    && (bp->b_vp->v_mount->mnt_flag & MNT_SOFTDEP)
2480 			    && (bp->b_flags & B_DELWRI)) {
2481 				s = splbio();
2482 				bremfree(bp);
2483 				bp->b_flags |= B_BUSY;
2484 				splx(s);
2485 				nbusy++;
2486 				bawrite(bp);
2487 				if (dcount-- <= 0) {
2488 					printf("softdep ");
2489 					goto fail;
2490 				}
2491 			}
2492 		}
2493 		if (nbusy == 0)
2494 			break;
2495 		if (nbusy_prev == 0)
2496 			nbusy_prev = nbusy;
2497 		printf("%d ", nbusy);
2498 		tsleep(&nbusy, PRIBIO, "bflush",
2499 		    (iter == 0) ? 1 : hz / 25 * iter);
2500 		if (nbusy >= nbusy_prev) /* we didn't flush anything */
2501 			iter++;
2502 		else
2503 			nbusy_prev = nbusy;
2504 	}
2505 	if (nbusy) {
2506 fail:
2507 #if defined(DEBUG) || defined(DEBUG_HALT_BUSY)
2508 		printf("giving up\nPrinting vnodes for busy buffers\n");
2509 		for (bp = &buf[nbuf]; --bp >= buf; )
2510 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2511 				vprint(NULL, bp->b_vp);
2512 
2513 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2514 		Debugger();
2515 #endif
2516 
2517 #else  /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2518 		printf("giving up\n");
2519 #endif /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2520 		return;
2521 	} else
2522 		printf("done\n");
2523 
2524 	/*
2525 	 * If we've panic'd, don't make the situation potentially
2526 	 * worse by unmounting the file systems.
2527 	 */
2528 	if (panicstr != NULL)
2529 		return;
2530 
2531 	/* Release inodes held by texts before update. */
2532 #ifdef notdef
2533 	vnshutdown();
2534 #endif
2535 	/* Unmount file systems. */
2536 	vfs_unmountall(p);
2537 }
2538 
2539 /*
2540  * Mount the root file system.  If the operator didn't specify a
2541  * file system to use, try all possible file systems until one
2542  * succeeds.
2543  */
2544 int
2545 vfs_mountroot()
2546 {
2547 	extern int (*mountroot) __P((void));
2548 	struct vfsops *v;
2549 
2550 	if (root_device == NULL)
2551 		panic("vfs_mountroot: root device unknown");
2552 
2553 	switch (root_device->dv_class) {
2554 	case DV_IFNET:
2555 		if (rootdev != NODEV)
2556 			panic("vfs_mountroot: rootdev set for DV_IFNET");
2557 		break;
2558 
2559 	case DV_DISK:
2560 		if (rootdev == NODEV)
2561 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2562 		break;
2563 
2564 	default:
2565 		printf("%s: inappropriate for root file system\n",
2566 		    root_device->dv_xname);
2567 		return (ENODEV);
2568 	}
2569 
2570 	/*
2571 	 * If user specified a file system, use it.
2572 	 */
2573 	if (mountroot != NULL)
2574 		return ((*mountroot)());
2575 
2576 	/*
2577 	 * Try each file system currently configured into the kernel.
2578 	 */
2579 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2580 		if (v->vfs_mountroot == NULL)
2581 			continue;
2582 #ifdef DEBUG
2583 		printf("mountroot: trying %s...\n", v->vfs_name);
2584 #endif
2585 		if ((*v->vfs_mountroot)() == 0) {
2586 			printf("root file system type: %s\n", v->vfs_name);
2587 			break;
2588 		}
2589 	}
2590 
2591 	if (v == NULL) {
2592 		printf("no file system for %s", root_device->dv_xname);
2593 		if (root_device->dv_class == DV_DISK)
2594 			printf(" (dev 0x%x)", rootdev);
2595 		printf("\n");
2596 		return (EFTYPE);
2597 	}
2598 	return (0);
2599 }
2600 
2601 /*
2602  * Given a file system name, look up the vfsops for that
2603  * file system, or return NULL if file system isn't present
2604  * in the kernel.
2605  */
2606 struct vfsops *
2607 vfs_getopsbyname(name)
2608 	const char *name;
2609 {
2610 	struct vfsops *v;
2611 
2612 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2613 		if (strcmp(v->vfs_name, name) == 0)
2614 			break;
2615 	}
2616 
2617 	return (v);
2618 }
2619 
2620 /*
2621  * Establish a file system and initialize it.
2622  */
2623 int
2624 vfs_attach(vfs)
2625 	struct vfsops *vfs;
2626 {
2627 	struct vfsops *v;
2628 	int error = 0;
2629 
2630 
2631 	/*
2632 	 * Make sure this file system doesn't already exist.
2633 	 */
2634 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2635 		if (strcmp(vfs->vfs_name, v->vfs_name) == 0) {
2636 			error = EEXIST;
2637 			goto out;
2638 		}
2639 	}
2640 
2641 	/*
2642 	 * Initialize the vnode operations for this file system.
2643 	 */
2644 	vfs_opv_init(vfs->vfs_opv_descs);
2645 
2646 	/*
2647 	 * Now initialize the file system itself.
2648 	 */
2649 	(*vfs->vfs_init)();
2650 
2651 	/*
2652 	 * ...and link it into the kernel's list.
2653 	 */
2654 	LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list);
2655 
2656 	/*
2657 	 * Sanity: make sure the reference count is 0.
2658 	 */
2659 	vfs->vfs_refcount = 0;
2660 
2661  out:
2662 	return (error);
2663 }
2664 
2665 /*
2666  * Remove a file system from the kernel.
2667  */
2668 int
2669 vfs_detach(vfs)
2670 	struct vfsops *vfs;
2671 {
2672 	struct vfsops *v;
2673 
2674 	/*
2675 	 * Make sure no one is using the filesystem.
2676 	 */
2677 	if (vfs->vfs_refcount != 0)
2678 		return (EBUSY);
2679 
2680 	/*
2681 	 * ...and remove it from the kernel's list.
2682 	 */
2683 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2684 		if (v == vfs) {
2685 			LIST_REMOVE(v, vfs_list);
2686 			break;
2687 		}
2688 	}
2689 
2690 	if (v == NULL)
2691 		return (ESRCH);
2692 
2693 	/*
2694 	 * Now run the file system-specific cleanups.
2695 	 */
2696 	(*vfs->vfs_done)();
2697 
2698 	/*
2699 	 * Free the vnode operations vector.
2700 	 */
2701 	vfs_opv_free(vfs->vfs_opv_descs);
2702 	return (0);
2703 }
2704 
2705 #ifdef DDB
2706 const char buf_flagbits[] =
2707 	"\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI"
2708 	"\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE"
2709 	"\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED"
2710 	"\32XXX\33VFLUSH";
2711 
2712 void
2713 vfs_buf_print(bp, full, pr)
2714 	struct buf *bp;
2715 	int full;
2716 	void (*pr) __P((const char *, ...));
2717 {
2718 	char buf[1024];
2719 
2720 	(*pr)("  vp %p lblkno 0x%x blkno 0x%x dev 0x%x\n",
2721 		  bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev);
2722 
2723 	bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf));
2724 	(*pr)("  error %d flags 0x%s\n", bp->b_error, buf);
2725 
2726 	(*pr)("  bufsize 0x%x bcount 0x%x resid 0x%x\n",
2727 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
2728 	(*pr)("  data %p saveaddr %p dep %p\n",
2729 		  bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep));
2730 	(*pr)("  iodone %p\n", bp->b_iodone);
2731 }
2732 
2733 
2734 const char vnode_flagbits[] =
2735 	"\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\11XLOCK\12XWANT\13BWAIT\14ALIASED"
2736 	"\15DIROP\16LAYER\17ONWORKLIST\20DIRTY";
2737 
2738 const char *vnode_types[] = {
2739 	"VNON",
2740 	"VREG",
2741 	"VDIR",
2742 	"VBLK",
2743 	"VCHR",
2744 	"VLNK",
2745 	"VSOCK",
2746 	"VFIFO",
2747 	"VBAD",
2748 };
2749 
2750 const char *vnode_tags[] = {
2751 	"VT_NON",
2752 	"VT_UFS",
2753 	"VT_NFS",
2754 	"VT_MFS",
2755 	"VT_MSDOSFS",
2756 	"VT_LFS",
2757 	"VT_LOFS",
2758 	"VT_FDESC",
2759 	"VT_PORTAL",
2760 	"VT_NULL",
2761 	"VT_UMAP",
2762 	"VT_KERNFS",
2763 	"VT_PROCFS",
2764 	"VT_AFS",
2765 	"VT_ISOFS",
2766 	"VT_UNION",
2767 	"VT_ADOSFS",
2768 	"VT_EXT2FS",
2769 	"VT_CODA",
2770 	"VT_FILECORE",
2771 	"VT_NTFS",
2772 	"VT_VFS",
2773 	"VT_OVERLAY"
2774 };
2775 
2776 void
2777 vfs_vnode_print(vp, full, pr)
2778 	struct vnode *vp;
2779 	int full;
2780 	void (*pr) __P((const char *, ...));
2781 {
2782 	char buf[256];
2783 
2784 	const char *vtype, *vtag;
2785 
2786 	uvm_object_printit(&vp->v_uvm.u_obj, full, pr);
2787 	bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf));
2788 	(*pr)("\nVNODE flags %s\n", buf);
2789 	(*pr)("mp %p nio %d size 0x%x rwlock 0x%x glock 0x%x\n",
2790 	      vp->v_mount, vp->v_uvm.u_nio, (int)vp->v_uvm.u_size,
2791 	      vp->v_vnlock ? lockstatus(vp->v_vnlock) : 0x999,
2792 	      lockstatus(&vp->v_glock));
2793 
2794 	(*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n",
2795 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2796 	      vp->v_holdcnt, vp->v_numoutput);
2797 
2798 	vtype = (vp->v_type >= 0 &&
2799 		 vp->v_type < sizeof(vnode_types) / sizeof(vnode_types[0])) ?
2800 		vnode_types[vp->v_type] : "UNKNOWN";
2801 	vtag = (vp->v_tag >= 0 &&
2802 		vp->v_tag < sizeof(vnode_tags) / sizeof(vnode_tags[0])) ?
2803 		vnode_tags[vp->v_tag] : "UNKNOWN";
2804 
2805 	(*pr)("type %s(%d) tag %s(%d) id 0x%x mount %p typedata %p\n",
2806 	      vtype, vp->v_type, vtag, vp->v_tag,
2807 	      vp->v_id, vp->v_mount, vp->v_mountedhere);
2808 	(*pr)("lastr 0x%x lastw 0x%x lasta 0x%x\n",
2809 	      vp->v_lastr, vp->v_lastw, vp->v_lasta);
2810 	(*pr)("cstart 0x%x clen 0x%x ralen 0x%x maxra 0x%x\n",
2811 	      vp->v_cstart, vp->v_clen, vp->v_ralen, vp->v_maxra);
2812 
2813 	if (full) {
2814 		struct buf *bp;
2815 
2816 		(*pr)("clean bufs:\n");
2817 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2818 			(*pr)(" bp %p\n", bp);
2819 			vfs_buf_print(bp, full, pr);
2820 		}
2821 
2822 		(*pr)("dirty bufs:\n");
2823 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2824 			(*pr)(" bp %p\n", bp);
2825 			vfs_buf_print(bp, full, pr);
2826 		}
2827 	}
2828 }
2829 #endif
2830