xref: /netbsd-src/sys/kern/vfs_subr.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: vfs_subr.c,v 1.178 2002/09/06 13:18:43 gehenna 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 1.178 2002/09/06 13:18:43 gehenna 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/fcntl.h>
98 #include <sys/vnode.h>
99 #include <sys/stat.h>
100 #include <sys/namei.h>
101 #include <sys/ucred.h>
102 #include <sys/buf.h>
103 #include <sys/errno.h>
104 #include <sys/malloc.h>
105 #include <sys/domain.h>
106 #include <sys/mbuf.h>
107 #include <sys/syscallargs.h>
108 #include <sys/device.h>
109 #include <sys/dirent.h>
110 
111 #include <miscfs/specfs/specdev.h>
112 #include <miscfs/genfs/genfs.h>
113 #include <miscfs/syncfs/syncfs.h>
114 
115 #include <uvm/uvm.h>
116 #include <uvm/uvm_ddb.h>
117 
118 #include <sys/sysctl.h>
119 
120 enum vtype iftovt_tab[16] = {
121 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
122 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
123 };
124 const int	vttoif_tab[9] = {
125 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
126 	S_IFSOCK, S_IFIFO, S_IFMT,
127 };
128 
129 int doforce = 1;		/* 1 => permit forcible unmounting */
130 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
131 
132 extern int dovfsusermount;	/* 1 => permit any user to mount filesystems */
133 
134 /*
135  * Insq/Remq for the vnode usage lists.
136  */
137 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
138 #define	bufremvn(bp) {							\
139 	LIST_REMOVE(bp, b_vnbufs);					\
140 	(bp)->b_vnbufs.le_next = NOLIST;				\
141 }
142 /* TAILQ_HEAD(freelst, vnode) vnode_free_list =	vnode free list (in vnode.h) */
143 struct freelst vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
144 struct freelst vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
145 
146 struct mntlist mountlist =			/* mounted filesystem list */
147     CIRCLEQ_HEAD_INITIALIZER(mountlist);
148 struct vfs_list_head vfs_list =			/* vfs list */
149     LIST_HEAD_INITIALIZER(vfs_list);
150 
151 struct nfs_public nfs_pub;			/* publicly exported FS */
152 
153 struct simplelock mountlist_slock = SIMPLELOCK_INITIALIZER;
154 static struct simplelock mntid_slock = SIMPLELOCK_INITIALIZER;
155 struct simplelock mntvnode_slock = SIMPLELOCK_INITIALIZER;
156 struct simplelock vnode_free_list_slock = SIMPLELOCK_INITIALIZER;
157 struct simplelock spechash_slock = SIMPLELOCK_INITIALIZER;
158 
159 /*
160  * These define the root filesystem and device.
161  */
162 struct mount *rootfs;
163 struct vnode *rootvnode;
164 struct device *root_device;			/* root device */
165 
166 struct pool vnode_pool;				/* memory pool for vnodes */
167 
168 /*
169  * Local declarations.
170  */
171 void insmntque __P((struct vnode *, struct mount *));
172 int getdevvp __P((dev_t, struct vnode **, enum vtype));
173 void vgoneall __P((struct vnode *));
174 
175 static int vfs_hang_addrlist __P((struct mount *, struct netexport *,
176 				  struct export_args *));
177 static int vfs_free_netcred __P((struct radix_node *, void *));
178 static void vfs_free_addrlist __P((struct netexport *));
179 
180 #ifdef DEBUG
181 void printlockedvnodes __P((void));
182 #endif
183 
184 /*
185  * Initialize the vnode management data structures.
186  */
187 void
188 vntblinit()
189 {
190 
191 	pool_init(&vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodepl",
192 	    &pool_allocator_nointr);
193 
194 	/*
195 	 * Initialize the filesystem syncer.
196 	 */
197 	vn_initialize_syncerd();
198 }
199 
200 /*
201  * Mark a mount point as busy. Used to synchronize access and to delay
202  * unmounting. Interlock is not released on failure.
203  */
204 int
205 vfs_busy(mp, flags, interlkp)
206 	struct mount *mp;
207 	int flags;
208 	struct simplelock *interlkp;
209 {
210 	int lkflags;
211 
212 	while (mp->mnt_flag & MNT_UNMOUNT) {
213 		int gone;
214 
215 		if (flags & LK_NOWAIT)
216 			return (ENOENT);
217 		if ((flags & LK_RECURSEFAIL) && mp->mnt_unmounter != NULL
218 		    && mp->mnt_unmounter == curproc)
219 			return (EDEADLK);
220 		if (interlkp)
221 			simple_unlock(interlkp);
222 		/*
223 		 * Since all busy locks are shared except the exclusive
224 		 * lock granted when unmounting, the only place that a
225 		 * wakeup needs to be done is at the release of the
226 		 * exclusive lock at the end of dounmount.
227 		 *
228 		 * XXX MP: add spinlock protecting mnt_wcnt here once you
229 		 * can atomically unlock-and-sleep.
230 		 */
231 		mp->mnt_wcnt++;
232 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
233 		mp->mnt_wcnt--;
234 		gone = mp->mnt_flag & MNT_GONE;
235 
236 		if (mp->mnt_wcnt == 0)
237 			wakeup(&mp->mnt_wcnt);
238 		if (interlkp)
239 			simple_lock(interlkp);
240 		if (gone)
241 			return (ENOENT);
242 	}
243 	lkflags = LK_SHARED;
244 	if (interlkp)
245 		lkflags |= LK_INTERLOCK;
246 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp))
247 		panic("vfs_busy: unexpected lock failure");
248 	return (0);
249 }
250 
251 /*
252  * Free a busy filesystem.
253  */
254 void
255 vfs_unbusy(mp)
256 	struct mount *mp;
257 {
258 
259 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
260 }
261 
262 /*
263  * Lookup a filesystem type, and if found allocate and initialize
264  * a mount structure for it.
265  *
266  * Devname is usually updated by mount(8) after booting.
267  */
268 int
269 vfs_rootmountalloc(fstypename, devname, mpp)
270 	char *fstypename;
271 	char *devname;
272 	struct mount **mpp;
273 {
274 	struct vfsops *vfsp = NULL;
275 	struct mount *mp;
276 
277 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
278 		if (!strncmp(vfsp->vfs_name, fstypename, MFSNAMELEN))
279 			break;
280 
281 	if (vfsp == NULL)
282 		return (ENODEV);
283 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
284 	memset((char *)mp, 0, (u_long)sizeof(struct mount));
285 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
286 	(void)vfs_busy(mp, LK_NOWAIT, 0);
287 	LIST_INIT(&mp->mnt_vnodelist);
288 	mp->mnt_op = vfsp;
289 	mp->mnt_flag = MNT_RDONLY;
290 	mp->mnt_vnodecovered = NULLVP;
291 	vfsp->vfs_refcount++;
292 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN);
293 	mp->mnt_stat.f_mntonname[0] = '/';
294 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
295 	*mpp = mp;
296 	return (0);
297 }
298 
299 /*
300  * Lookup a mount point by filesystem identifier.
301  */
302 struct mount *
303 vfs_getvfs(fsid)
304 	fsid_t *fsid;
305 {
306 	struct mount *mp;
307 
308 	simple_lock(&mountlist_slock);
309 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
310 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
311 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
312 			simple_unlock(&mountlist_slock);
313 			return (mp);
314 		}
315 	}
316 	simple_unlock(&mountlist_slock);
317 	return ((struct mount *)0);
318 }
319 
320 /*
321  * Get a new unique fsid
322  */
323 void
324 vfs_getnewfsid(mp)
325 	struct mount *mp;
326 {
327 	static u_short xxxfs_mntid;
328 	fsid_t tfsid;
329 	int mtype;
330 
331 	simple_lock(&mntid_slock);
332 	mtype = makefstype(mp->mnt_op->vfs_name);
333 	mp->mnt_stat.f_fsid.val[0] = makedev(mtype, 0);
334 	mp->mnt_stat.f_fsid.val[1] = mtype;
335 	if (xxxfs_mntid == 0)
336 		++xxxfs_mntid;
337 	tfsid.val[0] = makedev(mtype & 0xff, xxxfs_mntid);
338 	tfsid.val[1] = mtype;
339 	if (!CIRCLEQ_EMPTY(&mountlist)) {
340 		while (vfs_getvfs(&tfsid)) {
341 			tfsid.val[0]++;
342 			xxxfs_mntid++;
343 		}
344 	}
345 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
346 	simple_unlock(&mntid_slock);
347 }
348 
349 /*
350  * Make a 'unique' number from a mount type name.
351  */
352 long
353 makefstype(type)
354 	const char *type;
355 {
356 	long rv;
357 
358 	for (rv = 0; *type; type++) {
359 		rv <<= 2;
360 		rv ^= *type;
361 	}
362 	return rv;
363 }
364 
365 
366 /*
367  * Set vnode attributes to VNOVAL
368  */
369 void
370 vattr_null(vap)
371 	struct vattr *vap;
372 {
373 
374 	vap->va_type = VNON;
375 
376 	/*
377 	 * Assign individually so that it is safe even if size and
378 	 * sign of each member are varied.
379 	 */
380 	vap->va_mode = VNOVAL;
381 	vap->va_nlink = VNOVAL;
382 	vap->va_uid = VNOVAL;
383 	vap->va_gid = VNOVAL;
384 	vap->va_fsid = VNOVAL;
385 	vap->va_fileid = VNOVAL;
386 	vap->va_size = VNOVAL;
387 	vap->va_blocksize = VNOVAL;
388 	vap->va_atime.tv_sec =
389 	    vap->va_mtime.tv_sec =
390 	    vap->va_ctime.tv_sec = VNOVAL;
391 	vap->va_atime.tv_nsec =
392 	    vap->va_mtime.tv_nsec =
393 	    vap->va_ctime.tv_nsec = VNOVAL;
394 	vap->va_gen = VNOVAL;
395 	vap->va_flags = VNOVAL;
396 	vap->va_rdev = VNOVAL;
397 	vap->va_bytes = VNOVAL;
398 	vap->va_vaflags = 0;
399 }
400 
401 /*
402  * Routines having to do with the management of the vnode table.
403  */
404 extern int (**dead_vnodeop_p) __P((void *));
405 long numvnodes;
406 
407 /*
408  * Return the next vnode from the free list.
409  */
410 int
411 getnewvnode(tag, mp, vops, vpp)
412 	enum vtagtype tag;
413 	struct mount *mp;
414 	int (**vops) __P((void *));
415 	struct vnode **vpp;
416 {
417 	extern struct uvm_pagerops uvm_vnodeops;
418 	struct uvm_object *uobj;
419 	struct proc *p = curproc;	/* XXX */
420 	struct freelst *listhd;
421 	static int toggle;
422 	struct vnode *vp;
423 	int error = 0, tryalloc;
424 
425  try_again:
426 	if (mp) {
427 		/*
428 		 * Mark filesystem busy while we're creating a vnode.
429 		 * If unmount is in progress, this will wait; if the
430 		 * unmount succeeds (only if umount -f), this will
431 		 * return an error.  If the unmount fails, we'll keep
432 		 * going afterwards.
433 		 * (This puts the per-mount vnode list logically under
434 		 * the protection of the vfs_busy lock).
435 		 */
436 		error = vfs_busy(mp, LK_RECURSEFAIL, 0);
437 		if (error && error != EDEADLK)
438 			return error;
439 	}
440 
441 	/*
442 	 * We must choose whether to allocate a new vnode or recycle an
443 	 * existing one. The criterion for allocating a new one is that
444 	 * the total number of vnodes is less than the number desired or
445 	 * there are no vnodes on either free list. Generally we only
446 	 * want to recycle vnodes that have no buffers associated with
447 	 * them, so we look first on the vnode_free_list. If it is empty,
448 	 * we next consider vnodes with referencing buffers on the
449 	 * vnode_hold_list. The toggle ensures that half the time we
450 	 * will use a buffer from the vnode_hold_list, and half the time
451 	 * we will allocate a new one unless the list has grown to twice
452 	 * the desired size. We are reticent to recycle vnodes from the
453 	 * vnode_hold_list because we will lose the identity of all its
454 	 * referencing buffers.
455 	 */
456 
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(&vnode_free_list) == NULL &&
467 	     (TAILQ_FIRST(&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 		uobj = &vp->v_uobj;
475 		uobj->pgops = &uvm_vnodeops;
476 		uobj->uo_npages = 0;
477 		TAILQ_INIT(&uobj->memq);
478 		numvnodes++;
479 	} else {
480 		if ((vp = TAILQ_FIRST(listhd = &vnode_free_list)) == NULL)
481 			vp = TAILQ_FIRST(listhd = &vnode_hold_list);
482 		for (; vp != NULL; vp = TAILQ_NEXT(vp, v_freelist)) {
483 			if (simple_lock_try(&vp->v_interlock)) {
484 				if ((vp->v_flag & VLAYER) == 0) {
485 					break;
486 				}
487 				if (VOP_ISLOCKED(vp) == 0)
488 					break;
489 				else
490 					simple_unlock(&vp->v_interlock);
491 			}
492 		}
493 		/*
494 		 * Unless this is a bad time of the month, at most
495 		 * the first NCPUS items on the free list are
496 		 * locked, so this is close enough to being empty.
497 		 */
498 		if (vp == NULLVP) {
499 			simple_unlock(&vnode_free_list_slock);
500 			if (mp && error != EDEADLK)
501 				vfs_unbusy(mp);
502 			if (tryalloc) {
503 				printf("WARNING: unable to allocate new "
504 				    "vnode, retrying...\n");
505 				(void) tsleep(&lbolt, PRIBIO, "newvn", hz);
506 				goto try_again;
507 			}
508 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
509 			*vpp = 0;
510 			return (ENFILE);
511 		}
512 		if (vp->v_usecount)
513 			panic("free vnode isn't, vp %p", vp);
514 		TAILQ_REMOVE(listhd, vp, v_freelist);
515 		/* see comment on why 0xdeadb is set at end of vgone (below) */
516 		vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
517 		simple_unlock(&vnode_free_list_slock);
518 		vp->v_lease = NULL;
519 
520 		if (vp->v_type != VBAD)
521 			vgonel(vp, p);
522 		else
523 			simple_unlock(&vp->v_interlock);
524 #ifdef DIAGNOSTIC
525 		if (vp->v_data || vp->v_uobj.uo_npages ||
526 		    TAILQ_FIRST(&vp->v_uobj.memq))
527 			panic("cleaned vnode isn't, vp %p", vp);
528 		if (vp->v_numoutput)
529 			panic("clean vnode has pending I/O's, vp %p", vp);
530 #endif
531 		KASSERT((vp->v_flag & VONWORKLST) == 0);
532 		vp->v_flag = 0;
533 		vp->v_socket = NULL;
534 	}
535 	vp->v_type = VNON;
536 	vp->v_vnlock = &vp->v_lock;
537 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
538 	cache_purge(vp);
539 	vp->v_tag = tag;
540 	vp->v_op = vops;
541 	insmntque(vp, mp);
542 	*vpp = vp;
543 	vp->v_usecount = 1;
544 	vp->v_data = 0;
545 	simple_lock_init(&vp->v_uobj.vmobjlock);
546 
547 	/*
548 	 * initialize uvm_object within vnode.
549 	 */
550 
551 	uobj = &vp->v_uobj;
552 	KASSERT(uobj->pgops == &uvm_vnodeops);
553 	KASSERT(uobj->uo_npages == 0);
554 	KASSERT(TAILQ_FIRST(&uobj->memq) == NULL);
555 	vp->v_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 buf *bp, *nbp;
657 	int s, error;
658 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
659 		(flags & V_SAVE ? PGO_CLEANIT : 0);
660 
661 	/* XXXUBC this doesn't look at flags or slp* */
662 	simple_lock(&vp->v_interlock);
663 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
664 	if (error) {
665 		return error;
666 	}
667 
668 	if (flags & V_SAVE) {
669 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p);
670 		if (error)
671 		        return (error);
672 #ifdef DIAGNOSTIC
673 		s = splbio();
674 		if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd))
675 		        panic("vinvalbuf: dirty bufs, vp %p", vp);
676 		splx(s);
677 #endif
678 	}
679 
680 	s = splbio();
681 
682 restart:
683 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
684 		nbp = LIST_NEXT(bp, b_vnbufs);
685 		if (bp->b_flags & B_BUSY) {
686 			bp->b_flags |= B_WANTED;
687 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
688 			    "vinvalbuf", slptimeo);
689 			if (error) {
690 				splx(s);
691 				return (error);
692 			}
693 			goto restart;
694 		}
695 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
696 		brelse(bp);
697 	}
698 
699 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
700 		nbp = LIST_NEXT(bp, b_vnbufs);
701 		if (bp->b_flags & B_BUSY) {
702 			bp->b_flags |= B_WANTED;
703 			error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1),
704 			    "vinvalbuf", slptimeo);
705 			if (error) {
706 				splx(s);
707 				return (error);
708 			}
709 			goto restart;
710 		}
711 		/*
712 		 * XXX Since there are no node locks for NFS, I believe
713 		 * there is a slight chance that a delayed write will
714 		 * occur while sleeping just above, so check for it.
715 		 */
716 		if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
717 #ifdef DEBUG
718 			printf("buffer still DELWRI\n");
719 #endif
720 			bp->b_flags |= B_BUSY | B_VFLUSH;
721 			VOP_BWRITE(bp);
722 			goto restart;
723 		}
724 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
725 		brelse(bp);
726 	}
727 
728 #ifdef DIAGNOSTIC
729 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
730 		panic("vinvalbuf: flush failed, vp %p", vp);
731 #endif
732 
733 	splx(s);
734 
735 	return (0);
736 }
737 
738 /*
739  * Destroy any in core blocks past the truncation length.
740  * Called with the underlying vnode locked, which should prevent new dirty
741  * buffers from being queued.
742  */
743 int
744 vtruncbuf(vp, lbn, slpflag, slptimeo)
745 	struct vnode *vp;
746 	daddr_t lbn;
747 	int slpflag, slptimeo;
748 {
749 	struct buf *bp, *nbp;
750 	int s, error;
751 	voff_t off;
752 
753 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
754 	simple_lock(&vp->v_interlock);
755 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
756 	if (error) {
757 		return error;
758 	}
759 
760 	s = splbio();
761 
762 restart:
763 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
764 		nbp = LIST_NEXT(bp, b_vnbufs);
765 		if (bp->b_lblkno < lbn)
766 			continue;
767 		if (bp->b_flags & B_BUSY) {
768 			bp->b_flags |= B_WANTED;
769 			error = tsleep(bp, slpflag | (PRIBIO + 1),
770 			    "vtruncbuf", slptimeo);
771 			if (error) {
772 				splx(s);
773 				return (error);
774 			}
775 			goto restart;
776 		}
777 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
778 		brelse(bp);
779 	}
780 
781 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
782 		nbp = LIST_NEXT(bp, b_vnbufs);
783 		if (bp->b_lblkno < lbn)
784 			continue;
785 		if (bp->b_flags & B_BUSY) {
786 			bp->b_flags |= B_WANTED;
787 			error = tsleep(bp, slpflag | (PRIBIO + 1),
788 			    "vtruncbuf", slptimeo);
789 			if (error) {
790 				splx(s);
791 				return (error);
792 			}
793 			goto restart;
794 		}
795 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
796 		brelse(bp);
797 	}
798 
799 	splx(s);
800 
801 	return (0);
802 }
803 
804 void
805 vflushbuf(vp, sync)
806 	struct vnode *vp;
807 	int sync;
808 {
809 	struct buf *bp, *nbp;
810 	int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0);
811 	int s;
812 
813 	simple_lock(&vp->v_interlock);
814 	(void) VOP_PUTPAGES(vp, 0, 0, flags);
815 
816 loop:
817 	s = splbio();
818 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
819 		nbp = LIST_NEXT(bp, b_vnbufs);
820 		if ((bp->b_flags & B_BUSY))
821 			continue;
822 		if ((bp->b_flags & B_DELWRI) == 0)
823 			panic("vflushbuf: not dirty, bp %p", bp);
824 		bp->b_flags |= B_BUSY | B_VFLUSH;
825 		splx(s);
826 		/*
827 		 * Wait for I/O associated with indirect blocks to complete,
828 		 * since there is no way to quickly wait for them below.
829 		 */
830 		if (bp->b_vp == vp || sync == 0)
831 			(void) bawrite(bp);
832 		else
833 			(void) bwrite(bp);
834 		goto loop;
835 	}
836 	if (sync == 0) {
837 		splx(s);
838 		return;
839 	}
840 	while (vp->v_numoutput) {
841 		vp->v_flag |= VBWAIT;
842 		tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0);
843 	}
844 	splx(s);
845 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
846 		vprint("vflushbuf: dirty", vp);
847 		goto loop;
848 	}
849 }
850 
851 /*
852  * Associate a buffer with a vnode.
853  */
854 void
855 bgetvp(vp, bp)
856 	struct vnode *vp;
857 	struct buf *bp;
858 {
859 	int s;
860 
861 	if (bp->b_vp)
862 		panic("bgetvp: not free, bp %p", bp);
863 	VHOLD(vp);
864 	s = splbio();
865 	bp->b_vp = vp;
866 	if (vp->v_type == VBLK || vp->v_type == VCHR)
867 		bp->b_dev = vp->v_rdev;
868 	else
869 		bp->b_dev = NODEV;
870 	/*
871 	 * Insert onto list for new vnode.
872 	 */
873 	bufinsvn(bp, &vp->v_cleanblkhd);
874 	splx(s);
875 }
876 
877 /*
878  * Disassociate a buffer from a vnode.
879  */
880 void
881 brelvp(bp)
882 	struct buf *bp;
883 {
884 	struct vnode *vp;
885 	int s;
886 
887 	if (bp->b_vp == NULL)
888 		panic("brelvp: vp NULL, bp %p", bp);
889 
890 	s = splbio();
891 	vp = bp->b_vp;
892 	/*
893 	 * Delete from old vnode list, if on one.
894 	 */
895 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
896 		bufremvn(bp);
897 
898 	if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_flag & VONWORKLST) &&
899 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
900 		vp->v_flag &= ~VONWORKLST;
901 		LIST_REMOVE(vp, v_synclist);
902 	}
903 
904 	bp->b_vp = NULL;
905 	HOLDRELE(vp);
906 	splx(s);
907 }
908 
909 /*
910  * Reassign a buffer from one vnode to another.
911  * Used to assign file specific control information
912  * (indirect blocks) to the vnode to which they belong.
913  *
914  * This function must be called at splbio().
915  */
916 void
917 reassignbuf(bp, newvp)
918 	struct buf *bp;
919 	struct vnode *newvp;
920 {
921 	struct buflists *listheadp;
922 	int delay;
923 
924 	/*
925 	 * Delete from old vnode list, if on one.
926 	 */
927 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
928 		bufremvn(bp);
929 	/*
930 	 * If dirty, put on list of dirty buffers;
931 	 * otherwise insert onto list of clean buffers.
932 	 */
933 	if ((bp->b_flags & B_DELWRI) == 0) {
934 		listheadp = &newvp->v_cleanblkhd;
935 		if (TAILQ_EMPTY(&newvp->v_uobj.memq) &&
936 		    (newvp->v_flag & VONWORKLST) &&
937 		    LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
938 			newvp->v_flag &= ~VONWORKLST;
939 			LIST_REMOVE(newvp, v_synclist);
940 		}
941 	} else {
942 		listheadp = &newvp->v_dirtyblkhd;
943 		if ((newvp->v_flag & VONWORKLST) == 0) {
944 			switch (newvp->v_type) {
945 			case VDIR:
946 				delay = dirdelay;
947 				break;
948 			case VBLK:
949 				if (newvp->v_specmountpoint != NULL) {
950 					delay = metadelay;
951 					break;
952 				}
953 				/* fall through */
954 			default:
955 				delay = filedelay;
956 				break;
957 			}
958 			if (!newvp->v_mount ||
959 			    (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0)
960 				vn_syncer_add_to_worklist(newvp, delay);
961 		}
962 	}
963 	bufinsvn(bp, listheadp);
964 }
965 
966 /*
967  * Create a vnode for a block device.
968  * Used for root filesystem and swap areas.
969  * Also used for memory file system special devices.
970  */
971 int
972 bdevvp(dev, vpp)
973 	dev_t dev;
974 	struct vnode **vpp;
975 {
976 
977 	return (getdevvp(dev, vpp, VBLK));
978 }
979 
980 /*
981  * Create a vnode for a character device.
982  * Used for kernfs and some console handling.
983  */
984 int
985 cdevvp(dev, vpp)
986 	dev_t dev;
987 	struct vnode **vpp;
988 {
989 
990 	return (getdevvp(dev, vpp, VCHR));
991 }
992 
993 /*
994  * Create a vnode for a device.
995  * Used by bdevvp (block device) for root file system etc.,
996  * and by cdevvp (character device) for console and kernfs.
997  */
998 int
999 getdevvp(dev, vpp, type)
1000 	dev_t dev;
1001 	struct vnode **vpp;
1002 	enum vtype type;
1003 {
1004 	struct vnode *vp;
1005 	struct vnode *nvp;
1006 	int error;
1007 
1008 	if (dev == NODEV) {
1009 		*vpp = NULLVP;
1010 		return (0);
1011 	}
1012 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
1013 	if (error) {
1014 		*vpp = NULLVP;
1015 		return (error);
1016 	}
1017 	vp = nvp;
1018 	vp->v_type = type;
1019 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
1020 		vput(vp);
1021 		vp = nvp;
1022 	}
1023 	*vpp = vp;
1024 	return (0);
1025 }
1026 
1027 /*
1028  * Check to see if the new vnode represents a special device
1029  * for which we already have a vnode (either because of
1030  * bdevvp() or because of a different vnode representing
1031  * the same block device). If such an alias exists, deallocate
1032  * the existing contents and return the aliased vnode. The
1033  * caller is responsible for filling it with its new contents.
1034  */
1035 struct vnode *
1036 checkalias(nvp, nvp_rdev, mp)
1037 	struct vnode *nvp;
1038 	dev_t nvp_rdev;
1039 	struct mount *mp;
1040 {
1041 	struct proc *p = curproc;       /* XXX */
1042 	struct vnode *vp;
1043 	struct vnode **vpp;
1044 
1045 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1046 		return (NULLVP);
1047 
1048 	vpp = &speclisth[SPECHASH(nvp_rdev)];
1049 loop:
1050 	simple_lock(&spechash_slock);
1051 	for (vp = *vpp; vp; vp = vp->v_specnext) {
1052 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1053 			continue;
1054 		/*
1055 		 * Alias, but not in use, so flush it out.
1056 		 */
1057 		simple_lock(&vp->v_interlock);
1058 		if (vp->v_usecount == 0) {
1059 			simple_unlock(&spechash_slock);
1060 			vgonel(vp, p);
1061 			goto loop;
1062 		}
1063 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
1064 			simple_unlock(&spechash_slock);
1065 			goto loop;
1066 		}
1067 		break;
1068 	}
1069 	if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) {
1070 		MALLOC(nvp->v_specinfo, struct specinfo *,
1071 			sizeof(struct specinfo), M_VNODE, M_NOWAIT);
1072 		/* XXX Erg. */
1073 		if (nvp->v_specinfo == NULL) {
1074 			simple_unlock(&spechash_slock);
1075 			uvm_wait("checkalias");
1076 			goto loop;
1077 		}
1078 
1079 		nvp->v_rdev = nvp_rdev;
1080 		nvp->v_hashchain = vpp;
1081 		nvp->v_specnext = *vpp;
1082 		nvp->v_specmountpoint = NULL;
1083 		simple_unlock(&spechash_slock);
1084 		nvp->v_speclockf = NULL;
1085 		*vpp = nvp;
1086 		if (vp != NULLVP) {
1087 			nvp->v_flag |= VALIASED;
1088 			vp->v_flag |= VALIASED;
1089 			vput(vp);
1090 		}
1091 		return (NULLVP);
1092 	}
1093 	simple_unlock(&spechash_slock);
1094 	VOP_UNLOCK(vp, 0);
1095 	simple_lock(&vp->v_interlock);
1096 	vclean(vp, 0, p);
1097 	vp->v_op = nvp->v_op;
1098 	vp->v_tag = nvp->v_tag;
1099 	vp->v_vnlock = &vp->v_lock;
1100 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
1101 	nvp->v_type = VNON;
1102 	insmntque(vp, mp);
1103 	return (vp);
1104 }
1105 
1106 /*
1107  * Grab a particular vnode from the free list, increment its
1108  * reference count and lock it. If the vnode lock bit is set the
1109  * vnode is being eliminated in vgone. In that case, we can not
1110  * grab the vnode, so the process is awakened when the transition is
1111  * completed, and an error returned to indicate that the vnode is no
1112  * longer usable (possibly having been changed to a new file system type).
1113  */
1114 int
1115 vget(vp, flags)
1116 	struct vnode *vp;
1117 	int flags;
1118 {
1119 	int error;
1120 
1121 	/*
1122 	 * If the vnode is in the process of being cleaned out for
1123 	 * another use, we wait for the cleaning to finish and then
1124 	 * return failure. Cleaning is determined by checking that
1125 	 * the VXLOCK flag is set.
1126 	 */
1127 
1128 	if ((flags & LK_INTERLOCK) == 0)
1129 		simple_lock(&vp->v_interlock);
1130 	if (vp->v_flag & VXLOCK) {
1131 		if (flags & LK_NOWAIT) {
1132 			simple_unlock(&vp->v_interlock);
1133 			return EBUSY;
1134 		}
1135 		vp->v_flag |= VXWANT;
1136 		ltsleep(vp, PINOD|PNORELOCK, "vget", 0, &vp->v_interlock);
1137 		return (ENOENT);
1138 	}
1139 	if (vp->v_usecount == 0) {
1140 		simple_lock(&vnode_free_list_slock);
1141 		if (vp->v_holdcnt > 0)
1142 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1143 		else
1144 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1145 		simple_unlock(&vnode_free_list_slock);
1146 	}
1147 	vp->v_usecount++;
1148 #ifdef DIAGNOSTIC
1149 	if (vp->v_usecount == 0) {
1150 		vprint("vget", vp);
1151 		panic("vget: usecount overflow, vp %p", vp);
1152 	}
1153 #endif
1154 	if (flags & LK_TYPE_MASK) {
1155 		if ((error = vn_lock(vp, flags | LK_INTERLOCK))) {
1156 			/*
1157 			 * must expand vrele here because we do not want
1158 			 * to call VOP_INACTIVE if the reference count
1159 			 * drops back to zero since it was never really
1160 			 * active. We must remove it from the free list
1161 			 * before sleeping so that multiple processes do
1162 			 * not try to recycle it.
1163 			 */
1164 			simple_lock(&vp->v_interlock);
1165 			vp->v_usecount--;
1166 			if (vp->v_usecount > 0) {
1167 				simple_unlock(&vp->v_interlock);
1168 				return (error);
1169 			}
1170 			/*
1171 			 * insert at tail of LRU list
1172 			 */
1173 			simple_lock(&vnode_free_list_slock);
1174 			if (vp->v_holdcnt > 0)
1175 				TAILQ_INSERT_TAIL(&vnode_hold_list, vp,
1176 				    v_freelist);
1177 			else
1178 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
1179 				    v_freelist);
1180 			simple_unlock(&vnode_free_list_slock);
1181 			simple_unlock(&vp->v_interlock);
1182 		}
1183 		return (error);
1184 	}
1185 	simple_unlock(&vp->v_interlock);
1186 	return (0);
1187 }
1188 
1189 /*
1190  * vput(), just unlock and vrele()
1191  */
1192 void
1193 vput(vp)
1194 	struct vnode *vp;
1195 {
1196 	struct proc *p = curproc;	/* XXX */
1197 
1198 #ifdef DIAGNOSTIC
1199 	if (vp == NULL)
1200 		panic("vput: null vp");
1201 #endif
1202 	simple_lock(&vp->v_interlock);
1203 	vp->v_usecount--;
1204 	if (vp->v_usecount > 0) {
1205 		simple_unlock(&vp->v_interlock);
1206 		VOP_UNLOCK(vp, 0);
1207 		return;
1208 	}
1209 #ifdef DIAGNOSTIC
1210 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1211 		vprint("vput: bad ref count", vp);
1212 		panic("vput: ref cnt");
1213 	}
1214 #endif
1215 	/*
1216 	 * Insert at tail of LRU list.
1217 	 */
1218 	simple_lock(&vnode_free_list_slock);
1219 	if (vp->v_holdcnt > 0)
1220 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1221 	else
1222 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1223 	simple_unlock(&vnode_free_list_slock);
1224 	if (vp->v_flag & VEXECMAP) {
1225 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1226 		uvmexp.filepages += vp->v_uobj.uo_npages;
1227 	}
1228 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1229 	simple_unlock(&vp->v_interlock);
1230 	VOP_INACTIVE(vp, p);
1231 }
1232 
1233 /*
1234  * Vnode release.
1235  * If count drops to zero, call inactive routine and return to freelist.
1236  */
1237 void
1238 vrele(vp)
1239 	struct vnode *vp;
1240 {
1241 	struct proc *p = curproc;	/* XXX */
1242 
1243 #ifdef DIAGNOSTIC
1244 	if (vp == NULL)
1245 		panic("vrele: null vp");
1246 #endif
1247 	simple_lock(&vp->v_interlock);
1248 	vp->v_usecount--;
1249 	if (vp->v_usecount > 0) {
1250 		simple_unlock(&vp->v_interlock);
1251 		return;
1252 	}
1253 #ifdef DIAGNOSTIC
1254 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1255 		vprint("vrele: bad ref count", vp);
1256 		panic("vrele: ref cnt vp %p", vp);
1257 	}
1258 #endif
1259 	/*
1260 	 * Insert at tail of LRU list.
1261 	 */
1262 	simple_lock(&vnode_free_list_slock);
1263 	if (vp->v_holdcnt > 0)
1264 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1265 	else
1266 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1267 	simple_unlock(&vnode_free_list_slock);
1268 	if (vp->v_flag & VEXECMAP) {
1269 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1270 		uvmexp.filepages += vp->v_uobj.uo_npages;
1271 	}
1272 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1273 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0)
1274 		VOP_INACTIVE(vp, p);
1275 }
1276 
1277 #ifdef DIAGNOSTIC
1278 /*
1279  * Page or buffer structure gets a reference.
1280  */
1281 void
1282 vhold(vp)
1283 	struct vnode *vp;
1284 {
1285 
1286 	/*
1287 	 * If it is on the freelist and the hold count is currently
1288 	 * zero, move it to the hold list. The test of the back
1289 	 * pointer and the use reference count of zero is because
1290 	 * it will be removed from a free list by getnewvnode,
1291 	 * but will not have its reference count incremented until
1292 	 * after calling vgone. If the reference count were
1293 	 * incremented first, vgone would (incorrectly) try to
1294 	 * close the previous instance of the underlying object.
1295 	 * So, the back pointer is explicitly set to `0xdeadb' in
1296 	 * getnewvnode after removing it from a freelist to ensure
1297 	 * that we do not try to move it here.
1298 	 */
1299   	simple_lock(&vp->v_interlock);
1300 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1301 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1302 		simple_lock(&vnode_free_list_slock);
1303 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1304 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1305 		simple_unlock(&vnode_free_list_slock);
1306 	}
1307 	vp->v_holdcnt++;
1308 	simple_unlock(&vp->v_interlock);
1309 }
1310 
1311 /*
1312  * Page or buffer structure frees a reference.
1313  */
1314 void
1315 holdrele(vp)
1316 	struct vnode *vp;
1317 {
1318 
1319 	simple_lock(&vp->v_interlock);
1320 	if (vp->v_holdcnt <= 0)
1321 		panic("holdrele: holdcnt vp %p", vp);
1322 	vp->v_holdcnt--;
1323 
1324 	/*
1325 	 * If it is on the holdlist and the hold count drops to
1326 	 * zero, move it to the free list. The test of the back
1327 	 * pointer and the use reference count of zero is because
1328 	 * it will be removed from a free list by getnewvnode,
1329 	 * but will not have its reference count incremented until
1330 	 * after calling vgone. If the reference count were
1331 	 * incremented first, vgone would (incorrectly) try to
1332 	 * close the previous instance of the underlying object.
1333 	 * So, the back pointer is explicitly set to `0xdeadb' in
1334 	 * getnewvnode after removing it from a freelist to ensure
1335 	 * that we do not try to move it here.
1336 	 */
1337 
1338 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1339 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1340 		simple_lock(&vnode_free_list_slock);
1341 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1342 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1343 		simple_unlock(&vnode_free_list_slock);
1344 	}
1345 	simple_unlock(&vp->v_interlock);
1346 }
1347 
1348 /*
1349  * Vnode reference.
1350  */
1351 void
1352 vref(vp)
1353 	struct vnode *vp;
1354 {
1355 
1356 	simple_lock(&vp->v_interlock);
1357 	if (vp->v_usecount <= 0)
1358 		panic("vref used where vget required, vp %p", vp);
1359 	vp->v_usecount++;
1360 #ifdef DIAGNOSTIC
1361 	if (vp->v_usecount == 0) {
1362 		vprint("vref", vp);
1363 		panic("vref: usecount overflow, vp %p", vp);
1364 	}
1365 #endif
1366 	simple_unlock(&vp->v_interlock);
1367 }
1368 #endif /* DIAGNOSTIC */
1369 
1370 /*
1371  * Remove any vnodes in the vnode table belonging to mount point mp.
1372  *
1373  * If MNT_NOFORCE is specified, there should not be any active ones,
1374  * return error if any are found (nb: this is a user error, not a
1375  * system error). If MNT_FORCE is specified, detach any active vnodes
1376  * that are found.
1377  */
1378 #ifdef DEBUG
1379 int busyprt = 0;	/* print out busy vnodes */
1380 struct ctldebug debug1 = { "busyprt", &busyprt };
1381 #endif
1382 
1383 int
1384 vflush(mp, skipvp, flags)
1385 	struct mount *mp;
1386 	struct vnode *skipvp;
1387 	int flags;
1388 {
1389 	struct proc *p = curproc;	/* XXX */
1390 	struct vnode *vp, *nvp;
1391 	int busy = 0;
1392 
1393 	simple_lock(&mntvnode_slock);
1394 loop:
1395 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1396 		if (vp->v_mount != mp)
1397 			goto loop;
1398 		nvp = LIST_NEXT(vp, v_mntvnodes);
1399 		/*
1400 		 * Skip over a selected vnode.
1401 		 */
1402 		if (vp == skipvp)
1403 			continue;
1404 		simple_lock(&vp->v_interlock);
1405 		/*
1406 		 * Skip over a vnodes marked VSYSTEM.
1407 		 */
1408 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1409 			simple_unlock(&vp->v_interlock);
1410 			continue;
1411 		}
1412 		/*
1413 		 * If WRITECLOSE is set, only flush out regular file
1414 		 * vnodes open for writing.
1415 		 */
1416 		if ((flags & WRITECLOSE) &&
1417 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1418 			simple_unlock(&vp->v_interlock);
1419 			continue;
1420 		}
1421 		/*
1422 		 * With v_usecount == 0, all we need to do is clear
1423 		 * out the vnode data structures and we are done.
1424 		 */
1425 		if (vp->v_usecount == 0) {
1426 			simple_unlock(&mntvnode_slock);
1427 			vgonel(vp, p);
1428 			simple_lock(&mntvnode_slock);
1429 			continue;
1430 		}
1431 		/*
1432 		 * If FORCECLOSE is set, forcibly close the vnode.
1433 		 * For block or character devices, revert to an
1434 		 * anonymous device. For all other files, just kill them.
1435 		 */
1436 		if (flags & FORCECLOSE) {
1437 			simple_unlock(&mntvnode_slock);
1438 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1439 				vgonel(vp, p);
1440 			} else {
1441 				vclean(vp, 0, p);
1442 				vp->v_op = spec_vnodeop_p;
1443 				insmntque(vp, (struct mount *)0);
1444 			}
1445 			simple_lock(&mntvnode_slock);
1446 			continue;
1447 		}
1448 #ifdef DEBUG
1449 		if (busyprt)
1450 			vprint("vflush: busy vnode", vp);
1451 #endif
1452 		simple_unlock(&vp->v_interlock);
1453 		busy++;
1454 	}
1455 	simple_unlock(&mntvnode_slock);
1456 	if (busy)
1457 		return (EBUSY);
1458 	return (0);
1459 }
1460 
1461 /*
1462  * Disassociate the underlying file system from a vnode.
1463  */
1464 void
1465 vclean(vp, flags, p)
1466 	struct vnode *vp;
1467 	int flags;
1468 	struct proc *p;
1469 {
1470 	int active;
1471 
1472 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1473 
1474 	/*
1475 	 * Check to see if the vnode is in use.
1476 	 * If so we have to reference it before we clean it out
1477 	 * so that its count cannot fall to zero and generate a
1478 	 * race against ourselves to recycle it.
1479 	 */
1480 
1481 	if ((active = vp->v_usecount) != 0) {
1482 		vp->v_usecount++;
1483 #ifdef DIAGNOSTIC
1484 		if (vp->v_usecount == 0) {
1485 			vprint("vclean", vp);
1486 			panic("vclean: usecount overflow");
1487 		}
1488 #endif
1489 	}
1490 
1491 	/*
1492 	 * Prevent the vnode from being recycled or
1493 	 * brought into use while we clean it out.
1494 	 */
1495 	if (vp->v_flag & VXLOCK)
1496 		panic("vclean: deadlock, vp %p", vp);
1497 	vp->v_flag |= VXLOCK;
1498 	if (vp->v_flag & VEXECMAP) {
1499 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1500 		uvmexp.filepages += vp->v_uobj.uo_npages;
1501 	}
1502 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1503 
1504 	/*
1505 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1506 	 * have the object locked while it cleans it out. The VOP_LOCK
1507 	 * ensures that the VOP_INACTIVE routine is done with its work.
1508 	 * For active vnodes, it ensures that no other activity can
1509 	 * occur while the underlying object is being cleaned out.
1510 	 */
1511 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK);
1512 
1513 	/*
1514 	 * Clean out any cached data associated with the vnode.
1515 	 */
1516 	if (flags & DOCLOSE) {
1517 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1518 		KASSERT((vp->v_flag & VONWORKLST) == 0);
1519 	}
1520 	LOCK_ASSERT(!simple_lock_held(&vp->v_interlock));
1521 
1522 	/*
1523 	 * If purging an active vnode, it must be closed and
1524 	 * deactivated before being reclaimed. Note that the
1525 	 * VOP_INACTIVE will unlock the vnode.
1526 	 */
1527 	if (active) {
1528 		if (flags & DOCLOSE)
1529 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL);
1530 		VOP_INACTIVE(vp, p);
1531 	} else {
1532 		/*
1533 		 * Any other processes trying to obtain this lock must first
1534 		 * wait for VXLOCK to clear, then call the new lock operation.
1535 		 */
1536 		VOP_UNLOCK(vp, 0);
1537 	}
1538 	/*
1539 	 * Reclaim the vnode.
1540 	 */
1541 	if (VOP_RECLAIM(vp, p))
1542 		panic("vclean: cannot reclaim, vp %p", vp);
1543 	if (active) {
1544 		/*
1545 		 * Inline copy of vrele() since VOP_INACTIVE
1546 		 * has already been called.
1547 		 */
1548 		simple_lock(&vp->v_interlock);
1549 		if (--vp->v_usecount <= 0) {
1550 #ifdef DIAGNOSTIC
1551 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1552 				vprint("vclean: bad ref count", vp);
1553 				panic("vclean: ref cnt");
1554 			}
1555 #endif
1556 			/*
1557 			 * Insert at tail of LRU list.
1558 			 */
1559 
1560 			simple_unlock(&vp->v_interlock);
1561 			simple_lock(&vnode_free_list_slock);
1562 #ifdef DIAGNOSTIC
1563 			if (vp->v_holdcnt > 0)
1564 				panic("vclean: not clean, vp %p", vp);
1565 #endif
1566 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1567 			simple_unlock(&vnode_free_list_slock);
1568 		} else
1569 			simple_unlock(&vp->v_interlock);
1570 	}
1571 
1572 	KASSERT(vp->v_uobj.uo_npages == 0);
1573 	cache_purge(vp);
1574 
1575 	/*
1576 	 * Done with purge, notify sleepers of the grim news.
1577 	 */
1578 	vp->v_op = dead_vnodeop_p;
1579 	vp->v_tag = VT_NON;
1580 	simple_lock(&vp->v_interlock);
1581 	vp->v_flag &= ~VXLOCK;
1582 	if (vp->v_flag & VXWANT) {
1583 		vp->v_flag &= ~VXWANT;
1584 		simple_unlock(&vp->v_interlock);
1585 		wakeup((caddr_t)vp);
1586 	} else
1587 		simple_unlock(&vp->v_interlock);
1588 }
1589 
1590 /*
1591  * Recycle an unused vnode to the front of the free list.
1592  * Release the passed interlock if the vnode will be recycled.
1593  */
1594 int
1595 vrecycle(vp, inter_lkp, p)
1596 	struct vnode *vp;
1597 	struct simplelock *inter_lkp;
1598 	struct proc *p;
1599 {
1600 
1601 	simple_lock(&vp->v_interlock);
1602 	if (vp->v_usecount == 0) {
1603 		if (inter_lkp)
1604 			simple_unlock(inter_lkp);
1605 		vgonel(vp, p);
1606 		return (1);
1607 	}
1608 	simple_unlock(&vp->v_interlock);
1609 	return (0);
1610 }
1611 
1612 /*
1613  * Eliminate all activity associated with a vnode
1614  * in preparation for reuse.
1615  */
1616 void
1617 vgone(vp)
1618 	struct vnode *vp;
1619 {
1620 	struct proc *p = curproc;	/* XXX */
1621 
1622 	simple_lock(&vp->v_interlock);
1623 	vgonel(vp, p);
1624 }
1625 
1626 /*
1627  * vgone, with the vp interlock held.
1628  */
1629 void
1630 vgonel(vp, p)
1631 	struct vnode *vp;
1632 	struct proc *p;
1633 {
1634 	struct vnode *vq;
1635 	struct vnode *vx;
1636 
1637 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1638 
1639 	/*
1640 	 * If a vgone (or vclean) is already in progress,
1641 	 * wait until it is done and return.
1642 	 */
1643 
1644 	if (vp->v_flag & VXLOCK) {
1645 		vp->v_flag |= VXWANT;
1646 		ltsleep(vp, PINOD | PNORELOCK, "vgone", 0, &vp->v_interlock);
1647 		return;
1648 	}
1649 
1650 	/*
1651 	 * Clean out the filesystem specific data.
1652 	 */
1653 
1654 	vclean(vp, DOCLOSE, p);
1655 	KASSERT((vp->v_flag & VONWORKLST) == 0);
1656 
1657 	/*
1658 	 * Delete from old mount point vnode list, if on one.
1659 	 */
1660 
1661 	if (vp->v_mount != NULL)
1662 		insmntque(vp, (struct mount *)0);
1663 
1664 	/*
1665 	 * If special device, remove it from special device alias list.
1666 	 * if it is on one.
1667 	 */
1668 
1669 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1670 		simple_lock(&spechash_slock);
1671 		if (vp->v_hashchain != NULL) {
1672 			if (*vp->v_hashchain == vp) {
1673 				*vp->v_hashchain = vp->v_specnext;
1674 			} else {
1675 				for (vq = *vp->v_hashchain; vq;
1676 							vq = vq->v_specnext) {
1677 					if (vq->v_specnext != vp)
1678 						continue;
1679 					vq->v_specnext = vp->v_specnext;
1680 					break;
1681 				}
1682 				if (vq == NULL)
1683 					panic("missing bdev");
1684 			}
1685 			if (vp->v_flag & VALIASED) {
1686 				vx = NULL;
1687 				for (vq = *vp->v_hashchain; vq;
1688 							vq = vq->v_specnext) {
1689 					if (vq->v_rdev != vp->v_rdev ||
1690 					    vq->v_type != vp->v_type)
1691 						continue;
1692 					if (vx)
1693 						break;
1694 					vx = vq;
1695 				}
1696 				if (vx == NULL)
1697 					panic("missing alias");
1698 				if (vq == NULL)
1699 					vx->v_flag &= ~VALIASED;
1700 				vp->v_flag &= ~VALIASED;
1701 			}
1702 		}
1703 		simple_unlock(&spechash_slock);
1704 		FREE(vp->v_specinfo, M_VNODE);
1705 		vp->v_specinfo = NULL;
1706 	}
1707 
1708 	/*
1709 	 * If it is on the freelist and not already at the head,
1710 	 * move it to the head of the list. The test of the back
1711 	 * pointer and the reference count of zero is because
1712 	 * it will be removed from the free list by getnewvnode,
1713 	 * but will not have its reference count incremented until
1714 	 * after calling vgone. If the reference count were
1715 	 * incremented first, vgone would (incorrectly) try to
1716 	 * close the previous instance of the underlying object.
1717 	 * So, the back pointer is explicitly set to `0xdeadb' in
1718 	 * getnewvnode after removing it from the freelist to ensure
1719 	 * that we do not try to move it here.
1720 	 */
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[96];
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 & VEXECMAP)
1834 		strcat(buf, "|VEXECMAP");
1835 	if (vp->v_flag & VSYSTEM)
1836 		strcat(buf, "|VSYSTEM");
1837 	if (vp->v_flag & VXLOCK)
1838 		strcat(buf, "|VXLOCK");
1839 	if (vp->v_flag & VXWANT)
1840 		strcat(buf, "|VXWANT");
1841 	if (vp->v_flag & VBWAIT)
1842 		strcat(buf, "|VBWAIT");
1843 	if (vp->v_flag & VALIASED)
1844 		strcat(buf, "|VALIASED");
1845 	if (buf[0] != '\0')
1846 		printf(" flags (%s)", &buf[1]);
1847 	if (vp->v_data == NULL) {
1848 		printf("\n");
1849 	} else {
1850 		printf("\n\t");
1851 		VOP_PRINT(vp);
1852 	}
1853 }
1854 
1855 #ifdef DEBUG
1856 /*
1857  * List all of the locked vnodes in the system.
1858  * Called when debugging the kernel.
1859  */
1860 void
1861 printlockedvnodes()
1862 {
1863 	struct mount *mp, *nmp;
1864 	struct vnode *vp;
1865 
1866 	printf("Locked vnodes\n");
1867 	simple_lock(&mountlist_slock);
1868 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
1869 	     mp = nmp) {
1870 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
1871 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
1872 			continue;
1873 		}
1874 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1875 			if (VOP_ISLOCKED(vp))
1876 				vprint(NULL, vp);
1877 		}
1878 		simple_lock(&mountlist_slock);
1879 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
1880 		vfs_unbusy(mp);
1881 	}
1882 	simple_unlock(&mountlist_slock);
1883 }
1884 #endif
1885 
1886 /*
1887  * Top level filesystem related information gathering.
1888  */
1889 int
1890 vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
1891 	int *name;
1892 	u_int namelen;
1893 	void *oldp;
1894 	size_t *oldlenp;
1895 	void *newp;
1896 	size_t newlen;
1897 	struct proc *p;
1898 {
1899 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1900 	struct vfsconf vfc;
1901 	extern const char * const mountcompatnames[];
1902 	extern int nmountcompatnames;
1903 #endif
1904 	struct vfsops *vfsp;
1905 
1906 	/* all sysctl names at this level are at least name and field */
1907 	if (namelen < 2)
1908 		return (ENOTDIR);		/* overloaded */
1909 
1910 	/* Not generic: goes to file system. */
1911 	if (name[0] != VFS_GENERIC) {
1912 		static const struct ctlname vfsnames[VFS_MAXID+1]=CTL_VFS_NAMES;
1913 		const char *vfsname;
1914 
1915 		if (name[0] < 0 || name[0] > VFS_MAXID
1916 		    || (vfsname = vfsnames[name[0]].ctl_name) == NULL)
1917 			return (EOPNOTSUPP);
1918 
1919 		vfsp = vfs_getopsbyname(vfsname);
1920 		if (vfsp == NULL || vfsp->vfs_sysctl == NULL)
1921 			return (EOPNOTSUPP);
1922 		return ((*vfsp->vfs_sysctl)(&name[1], namelen - 1,
1923 		    oldp, oldlenp, newp, newlen, p));
1924 	}
1925 
1926 	/* The rest are generic vfs sysctls. */
1927 	switch (name[1]) {
1928 	case VFS_USERMOUNT:
1929 		return sysctl_int(oldp, oldlenp, newp, newlen, &dovfsusermount);
1930 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1931 	case VFS_MAXTYPENUM:
1932 		/*
1933 		 * Provided for 4.4BSD-Lite2 compatibility.
1934 		 */
1935 		return (sysctl_rdint(oldp, oldlenp, newp, nmountcompatnames));
1936 	case VFS_CONF:
1937 		/*
1938 		 * Special: a node, next is a file system name.
1939 		 * Provided for 4.4BSD-Lite2 compatibility.
1940 		 */
1941 		if (namelen < 3)
1942 			return (ENOTDIR);	/* overloaded */
1943 		if (name[2] >= nmountcompatnames || name[2] < 0 ||
1944 		    mountcompatnames[name[2]] == NULL)
1945 			return (EOPNOTSUPP);
1946 		vfsp = vfs_getopsbyname(mountcompatnames[name[2]]);
1947 		if (vfsp == NULL)
1948 			return (EOPNOTSUPP);
1949 		vfc.vfc_vfsops = vfsp;
1950 		strncpy(vfc.vfc_name, vfsp->vfs_name, MFSNAMELEN);
1951 		vfc.vfc_typenum = name[2];
1952 		vfc.vfc_refcount = vfsp->vfs_refcount;
1953 		vfc.vfc_flags = 0;
1954 		vfc.vfc_mountroot = vfsp->vfs_mountroot;
1955 		vfc.vfc_next = NULL;
1956 		return (sysctl_rdstruct(oldp, oldlenp, newp, &vfc,
1957 		    sizeof(struct vfsconf)));
1958 #endif
1959 	default:
1960 		break;
1961 	}
1962 	return (EOPNOTSUPP);
1963 }
1964 
1965 int kinfo_vdebug = 1;
1966 int kinfo_vgetfailed;
1967 #define KINFO_VNODESLOP	10
1968 /*
1969  * Dump vnode list (via sysctl).
1970  * Copyout address of vnode followed by vnode.
1971  */
1972 /* ARGSUSED */
1973 int
1974 sysctl_vnode(where, sizep, p)
1975 	char *where;
1976 	size_t *sizep;
1977 	struct proc *p;
1978 {
1979 	struct mount *mp, *nmp;
1980 	struct vnode *nvp, *vp;
1981 	char *bp = where, *savebp;
1982 	char *ewhere;
1983 	int error;
1984 
1985 #define VPTRSZ	sizeof(struct vnode *)
1986 #define VNODESZ	sizeof(struct vnode)
1987 	if (where == NULL) {
1988 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
1989 		return (0);
1990 	}
1991 	ewhere = where + *sizep;
1992 
1993 	simple_lock(&mountlist_slock);
1994 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
1995 	     mp = nmp) {
1996 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
1997 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
1998 			continue;
1999 		}
2000 		savebp = bp;
2001 again:
2002 		simple_lock(&mntvnode_slock);
2003 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2004 		     vp != NULL;
2005 		     vp = nvp) {
2006 			/*
2007 			 * Check that the vp is still associated with
2008 			 * this filesystem.  RACE: could have been
2009 			 * recycled onto the same filesystem.
2010 			 */
2011 			if (vp->v_mount != mp) {
2012 				simple_unlock(&mntvnode_slock);
2013 				if (kinfo_vdebug)
2014 					printf("kinfo: vp changed\n");
2015 				bp = savebp;
2016 				goto again;
2017 			}
2018 			nvp = LIST_NEXT(vp, v_mntvnodes);
2019 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2020 				simple_unlock(&mntvnode_slock);
2021 				*sizep = bp - where;
2022 				return (ENOMEM);
2023 			}
2024 			simple_unlock(&mntvnode_slock);
2025 			if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
2026 			   (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
2027 				return (error);
2028 			bp += VPTRSZ + VNODESZ;
2029 			simple_lock(&mntvnode_slock);
2030 		}
2031 		simple_unlock(&mntvnode_slock);
2032 		simple_lock(&mountlist_slock);
2033 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
2034 		vfs_unbusy(mp);
2035 	}
2036 	simple_unlock(&mountlist_slock);
2037 
2038 	*sizep = bp - where;
2039 	return (0);
2040 }
2041 
2042 /*
2043  * Check to see if a filesystem is mounted on a block device.
2044  */
2045 int
2046 vfs_mountedon(vp)
2047 	struct vnode *vp;
2048 {
2049 	struct vnode *vq;
2050 	int error = 0;
2051 
2052 	if (vp->v_specmountpoint != NULL)
2053 		return (EBUSY);
2054 	if (vp->v_flag & VALIASED) {
2055 		simple_lock(&spechash_slock);
2056 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2057 			if (vq->v_rdev != vp->v_rdev ||
2058 			    vq->v_type != vp->v_type)
2059 				continue;
2060 			if (vq->v_specmountpoint != NULL) {
2061 				error = EBUSY;
2062 				break;
2063 			}
2064 		}
2065 		simple_unlock(&spechash_slock);
2066 	}
2067 	return (error);
2068 }
2069 
2070 /*
2071  * Build hash lists of net addresses and hang them off the mount point.
2072  * Called by ufs_mount() to set up the lists of export addresses.
2073  */
2074 static int
2075 vfs_hang_addrlist(mp, nep, argp)
2076 	struct mount *mp;
2077 	struct netexport *nep;
2078 	struct export_args *argp;
2079 {
2080 	struct netcred *np, *enp;
2081 	struct radix_node_head *rnh;
2082 	int i;
2083 	struct radix_node *rn;
2084 	struct sockaddr *saddr, *smask = 0;
2085 	struct domain *dom;
2086 	int error;
2087 
2088 	if (argp->ex_addrlen == 0) {
2089 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2090 			return (EPERM);
2091 		np = &nep->ne_defexported;
2092 		np->netc_exflags = argp->ex_flags;
2093 		crcvt(&np->netc_anon, &argp->ex_anon);
2094 		np->netc_anon.cr_ref = 1;
2095 		mp->mnt_flag |= MNT_DEFEXPORTED;
2096 		return (0);
2097 	}
2098 
2099 	if (argp->ex_addrlen > MLEN)
2100 		return (EINVAL);
2101 
2102 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2103 	np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK);
2104 	memset((caddr_t)np, 0, i);
2105 	saddr = (struct sockaddr *)(np + 1);
2106 	error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen);
2107 	if (error)
2108 		goto out;
2109 	if (saddr->sa_len > argp->ex_addrlen)
2110 		saddr->sa_len = argp->ex_addrlen;
2111 	if (argp->ex_masklen) {
2112 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2113 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2114 		if (error)
2115 			goto out;
2116 		if (smask->sa_len > argp->ex_masklen)
2117 			smask->sa_len = argp->ex_masklen;
2118 	}
2119 	i = saddr->sa_family;
2120 	if ((rnh = nep->ne_rtable[i]) == 0) {
2121 		/*
2122 		 * Seems silly to initialize every AF when most are not
2123 		 * used, do so on demand here
2124 		 */
2125 		for (dom = domains; dom; dom = dom->dom_next)
2126 			if (dom->dom_family == i && dom->dom_rtattach) {
2127 				dom->dom_rtattach((void **)&nep->ne_rtable[i],
2128 					dom->dom_rtoffset);
2129 				break;
2130 			}
2131 		if ((rnh = nep->ne_rtable[i]) == 0) {
2132 			error = ENOBUFS;
2133 			goto out;
2134 		}
2135 	}
2136 	rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh,
2137 		np->netc_rnodes);
2138 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
2139 		if (rn == 0) {
2140 			enp = (struct netcred *)(*rnh->rnh_lookup)(saddr,
2141 				smask, rnh);
2142 			if (enp == 0) {
2143 				error = EPERM;
2144 				goto out;
2145 			}
2146 		} else
2147 			enp = (struct netcred *)rn;
2148 
2149 		if (enp->netc_exflags != argp->ex_flags ||
2150 		    enp->netc_anon.cr_uid != argp->ex_anon.cr_uid ||
2151 		    enp->netc_anon.cr_gid != argp->ex_anon.cr_gid ||
2152 		    enp->netc_anon.cr_ngroups !=
2153 		    			(uint32_t) argp->ex_anon.cr_ngroups ||
2154 		    memcmp(&enp->netc_anon.cr_groups, &argp->ex_anon.cr_groups,
2155 			enp->netc_anon.cr_ngroups))
2156 				error = EPERM;
2157 		else
2158 			error = 0;
2159 		goto out;
2160 	}
2161 	np->netc_exflags = argp->ex_flags;
2162 	crcvt(&np->netc_anon, &argp->ex_anon);
2163 	np->netc_anon.cr_ref = 1;
2164 	return (0);
2165 out:
2166 	free(np, M_NETADDR);
2167 	return (error);
2168 }
2169 
2170 /* ARGSUSED */
2171 static int
2172 vfs_free_netcred(rn, w)
2173 	struct radix_node *rn;
2174 	void *w;
2175 {
2176 	struct radix_node_head *rnh = (struct radix_node_head *)w;
2177 
2178 	(*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh);
2179 	free((caddr_t)rn, M_NETADDR);
2180 	return (0);
2181 }
2182 
2183 /*
2184  * Free the net address hash lists that are hanging off the mount points.
2185  */
2186 static void
2187 vfs_free_addrlist(nep)
2188 	struct netexport *nep;
2189 {
2190 	int i;
2191 	struct radix_node_head *rnh;
2192 
2193 	for (i = 0; i <= AF_MAX; i++)
2194 		if ((rnh = nep->ne_rtable[i]) != NULL) {
2195 			(*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh);
2196 			free((caddr_t)rnh, M_RTABLE);
2197 			nep->ne_rtable[i] = 0;
2198 		}
2199 }
2200 
2201 int
2202 vfs_export(mp, nep, argp)
2203 	struct mount *mp;
2204 	struct netexport *nep;
2205 	struct export_args *argp;
2206 {
2207 	int error;
2208 
2209 	if (argp->ex_flags & MNT_DELEXPORT) {
2210 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2211 			vfs_setpublicfs(NULL, NULL, NULL);
2212 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2213 		}
2214 		vfs_free_addrlist(nep);
2215 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2216 	}
2217 	if (argp->ex_flags & MNT_EXPORTED) {
2218 		if (argp->ex_flags & MNT_EXPUBLIC) {
2219 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2220 				return (error);
2221 			mp->mnt_flag |= MNT_EXPUBLIC;
2222 		}
2223 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
2224 			return (error);
2225 		mp->mnt_flag |= MNT_EXPORTED;
2226 	}
2227 	return (0);
2228 }
2229 
2230 /*
2231  * Set the publicly exported filesystem (WebNFS). Currently, only
2232  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2233  */
2234 int
2235 vfs_setpublicfs(mp, nep, argp)
2236 	struct mount *mp;
2237 	struct netexport *nep;
2238 	struct export_args *argp;
2239 {
2240 	int error;
2241 	struct vnode *rvp;
2242 	char *cp;
2243 
2244 	/*
2245 	 * mp == NULL -> invalidate the current info, the FS is
2246 	 * no longer exported. May be called from either vfs_export
2247 	 * or unmount, so check if it hasn't already been done.
2248 	 */
2249 	if (mp == NULL) {
2250 		if (nfs_pub.np_valid) {
2251 			nfs_pub.np_valid = 0;
2252 			if (nfs_pub.np_index != NULL) {
2253 				FREE(nfs_pub.np_index, M_TEMP);
2254 				nfs_pub.np_index = NULL;
2255 			}
2256 		}
2257 		return (0);
2258 	}
2259 
2260 	/*
2261 	 * Only one allowed at a time.
2262 	 */
2263 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2264 		return (EBUSY);
2265 
2266 	/*
2267 	 * Get real filehandle for root of exported FS.
2268 	 */
2269 	memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle));
2270 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2271 
2272 	if ((error = VFS_ROOT(mp, &rvp)))
2273 		return (error);
2274 
2275 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2276 		return (error);
2277 
2278 	vput(rvp);
2279 
2280 	/*
2281 	 * If an indexfile was specified, pull it in.
2282 	 */
2283 	if (argp->ex_indexfile != NULL) {
2284 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2285 		    M_WAITOK);
2286 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2287 		    MAXNAMLEN, (size_t *)0);
2288 		if (!error) {
2289 			/*
2290 			 * Check for illegal filenames.
2291 			 */
2292 			for (cp = nfs_pub.np_index; *cp; cp++) {
2293 				if (*cp == '/') {
2294 					error = EINVAL;
2295 					break;
2296 				}
2297 			}
2298 		}
2299 		if (error) {
2300 			FREE(nfs_pub.np_index, M_TEMP);
2301 			return (error);
2302 		}
2303 	}
2304 
2305 	nfs_pub.np_mount = mp;
2306 	nfs_pub.np_valid = 1;
2307 	return (0);
2308 }
2309 
2310 struct netcred *
2311 vfs_export_lookup(mp, nep, nam)
2312 	struct mount *mp;
2313 	struct netexport *nep;
2314 	struct mbuf *nam;
2315 {
2316 	struct netcred *np;
2317 	struct radix_node_head *rnh;
2318 	struct sockaddr *saddr;
2319 
2320 	np = NULL;
2321 	if (mp->mnt_flag & MNT_EXPORTED) {
2322 		/*
2323 		 * Lookup in the export list first.
2324 		 */
2325 		if (nam != NULL) {
2326 			saddr = mtod(nam, struct sockaddr *);
2327 			rnh = nep->ne_rtable[saddr->sa_family];
2328 			if (rnh != NULL) {
2329 				np = (struct netcred *)
2330 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2331 							      rnh);
2332 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2333 					np = NULL;
2334 			}
2335 		}
2336 		/*
2337 		 * If no address match, use the default if it exists.
2338 		 */
2339 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2340 			np = &nep->ne_defexported;
2341 	}
2342 	return (np);
2343 }
2344 
2345 /*
2346  * Do the usual access checking.
2347  * file_mode, uid and gid are from the vnode in question,
2348  * while acc_mode and cred are from the VOP_ACCESS parameter list
2349  */
2350 int
2351 vaccess(type, file_mode, uid, gid, acc_mode, cred)
2352 	enum vtype type;
2353 	mode_t file_mode;
2354 	uid_t uid;
2355 	gid_t gid;
2356 	mode_t acc_mode;
2357 	struct ucred *cred;
2358 {
2359 	mode_t mask;
2360 
2361 	/*
2362 	 * Super-user always gets read/write access, but execute access depends
2363 	 * on at least one execute bit being set.
2364 	 */
2365 	if (cred->cr_uid == 0) {
2366 		if ((acc_mode & VEXEC) && type != VDIR &&
2367 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
2368 			return (EACCES);
2369 		return (0);
2370 	}
2371 
2372 	mask = 0;
2373 
2374 	/* Otherwise, check the owner. */
2375 	if (cred->cr_uid == uid) {
2376 		if (acc_mode & VEXEC)
2377 			mask |= S_IXUSR;
2378 		if (acc_mode & VREAD)
2379 			mask |= S_IRUSR;
2380 		if (acc_mode & VWRITE)
2381 			mask |= S_IWUSR;
2382 		return ((file_mode & mask) == mask ? 0 : EACCES);
2383 	}
2384 
2385 	/* Otherwise, check the groups. */
2386 	if (cred->cr_gid == gid || groupmember(gid, cred)) {
2387 		if (acc_mode & VEXEC)
2388 			mask |= S_IXGRP;
2389 		if (acc_mode & VREAD)
2390 			mask |= S_IRGRP;
2391 		if (acc_mode & VWRITE)
2392 			mask |= S_IWGRP;
2393 		return ((file_mode & mask) == mask ? 0 : EACCES);
2394 	}
2395 
2396 	/* Otherwise, check everyone else. */
2397 	if (acc_mode & VEXEC)
2398 		mask |= S_IXOTH;
2399 	if (acc_mode & VREAD)
2400 		mask |= S_IROTH;
2401 	if (acc_mode & VWRITE)
2402 		mask |= S_IWOTH;
2403 	return ((file_mode & mask) == mask ? 0 : EACCES);
2404 }
2405 
2406 /*
2407  * Unmount all file systems.
2408  * We traverse the list in reverse order under the assumption that doing so
2409  * will avoid needing to worry about dependencies.
2410  */
2411 void
2412 vfs_unmountall(p)
2413 	struct proc *p;
2414 {
2415 	struct mount *mp, *nmp;
2416 	int allerror, error;
2417 
2418 	for (allerror = 0,
2419 	     mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2420 		nmp = mp->mnt_list.cqe_prev;
2421 #ifdef DEBUG
2422 		printf("unmounting %s (%s)...\n",
2423 		    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
2424 #endif
2425 		/*
2426 		 * XXX Freeze syncer.  Must do this before locking the
2427 		 * mount point.  See dounmount() for details.
2428 		 */
2429 		lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL);
2430 		if (vfs_busy(mp, 0, 0)) {
2431 			lockmgr(&syncer_lock, LK_RELEASE, NULL);
2432 			continue;
2433 		}
2434 		if ((error = dounmount(mp, MNT_FORCE, p)) != 0) {
2435 			printf("unmount of %s failed with error %d\n",
2436 			    mp->mnt_stat.f_mntonname, error);
2437 			allerror = 1;
2438 		}
2439 	}
2440 	if (allerror)
2441 		printf("WARNING: some file systems would not unmount\n");
2442 }
2443 
2444 /*
2445  * Sync and unmount file systems before shutting down.
2446  */
2447 void
2448 vfs_shutdown()
2449 {
2450 	struct buf *bp;
2451 	int iter, nbusy, nbusy_prev = 0, dcount, s;
2452 	struct proc *p = curproc;
2453 
2454 	/* XXX we're certainly not running in proc0's context! */
2455 	if (p == NULL)
2456 		p = &proc0;
2457 
2458 	printf("syncing disks... ");
2459 
2460 	/* remove user process from run queue */
2461 	suspendsched();
2462 	(void) spl0();
2463 
2464 	/* avoid coming back this way again if we panic. */
2465 	doing_shutdown = 1;
2466 
2467 	sys_sync(p, NULL, NULL);
2468 
2469 	/* Wait for sync to finish. */
2470 	dcount = 10000;
2471 	for (iter = 0; iter < 20;) {
2472 		nbusy = 0;
2473 		for (bp = &buf[nbuf]; --bp >= buf; ) {
2474 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2475 				nbusy++;
2476 			/*
2477 			 * With soft updates, some buffers that are
2478 			 * written will be remarked as dirty until other
2479 			 * buffers are written.
2480 			 */
2481 			if (bp->b_vp && bp->b_vp->v_mount
2482 			    && (bp->b_vp->v_mount->mnt_flag & MNT_SOFTDEP)
2483 			    && (bp->b_flags & B_DELWRI)) {
2484 				s = splbio();
2485 				bremfree(bp);
2486 				bp->b_flags |= B_BUSY;
2487 				splx(s);
2488 				nbusy++;
2489 				bawrite(bp);
2490 				if (dcount-- <= 0) {
2491 					printf("softdep ");
2492 					goto fail;
2493 				}
2494 			}
2495 		}
2496 		if (nbusy == 0)
2497 			break;
2498 		if (nbusy_prev == 0)
2499 			nbusy_prev = nbusy;
2500 		printf("%d ", nbusy);
2501 		tsleep(&nbusy, PRIBIO, "bflush",
2502 		    (iter == 0) ? 1 : hz / 25 * iter);
2503 		if (nbusy >= nbusy_prev) /* we didn't flush anything */
2504 			iter++;
2505 		else
2506 			nbusy_prev = nbusy;
2507 	}
2508 	if (nbusy) {
2509 fail:
2510 #if defined(DEBUG) || defined(DEBUG_HALT_BUSY)
2511 		printf("giving up\nPrinting vnodes for busy buffers\n");
2512 		for (bp = &buf[nbuf]; --bp >= buf; )
2513 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2514 				vprint(NULL, bp->b_vp);
2515 
2516 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2517 		Debugger();
2518 #endif
2519 
2520 #else  /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2521 		printf("giving up\n");
2522 #endif /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2523 		return;
2524 	} else
2525 		printf("done\n");
2526 
2527 	/*
2528 	 * If we've panic'd, don't make the situation potentially
2529 	 * worse by unmounting the file systems.
2530 	 */
2531 	if (panicstr != NULL)
2532 		return;
2533 
2534 	/* Release inodes held by texts before update. */
2535 #ifdef notdef
2536 	vnshutdown();
2537 #endif
2538 	/* Unmount file systems. */
2539 	vfs_unmountall(p);
2540 }
2541 
2542 /*
2543  * Mount the root file system.  If the operator didn't specify a
2544  * file system to use, try all possible file systems until one
2545  * succeeds.
2546  */
2547 int
2548 vfs_mountroot()
2549 {
2550 	struct vfsops *v;
2551 
2552 	if (root_device == NULL)
2553 		panic("vfs_mountroot: root device unknown");
2554 
2555 	switch (root_device->dv_class) {
2556 	case DV_IFNET:
2557 		if (rootdev != NODEV)
2558 			panic("vfs_mountroot: rootdev set for DV_IFNET "
2559 			    "(0x%08x -> %d,%d)", rootdev,
2560 			    major(rootdev), minor(rootdev));
2561 		break;
2562 
2563 	case DV_DISK:
2564 		if (rootdev == NODEV)
2565 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2566 		break;
2567 
2568 	default:
2569 		printf("%s: inappropriate for root file system\n",
2570 		    root_device->dv_xname);
2571 		return (ENODEV);
2572 	}
2573 
2574 	/*
2575 	 * If user specified a file system, use it.
2576 	 */
2577 	if (mountroot != NULL)
2578 		return ((*mountroot)());
2579 
2580 	/*
2581 	 * Try each file system currently configured into the kernel.
2582 	 */
2583 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2584 		if (v->vfs_mountroot == NULL)
2585 			continue;
2586 #ifdef DEBUG
2587 		printf("mountroot: trying %s...\n", v->vfs_name);
2588 #endif
2589 		if ((*v->vfs_mountroot)() == 0) {
2590 			printf("root file system type: %s\n", v->vfs_name);
2591 			break;
2592 		}
2593 	}
2594 
2595 	if (v == NULL) {
2596 		printf("no file system for %s", root_device->dv_xname);
2597 		if (root_device->dv_class == DV_DISK)
2598 			printf(" (dev 0x%x)", rootdev);
2599 		printf("\n");
2600 		return (EFTYPE);
2601 	}
2602 	return (0);
2603 }
2604 
2605 /*
2606  * Given a file system name, look up the vfsops for that
2607  * file system, or return NULL if file system isn't present
2608  * in the kernel.
2609  */
2610 struct vfsops *
2611 vfs_getopsbyname(name)
2612 	const char *name;
2613 {
2614 	struct vfsops *v;
2615 
2616 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2617 		if (strcmp(v->vfs_name, name) == 0)
2618 			break;
2619 	}
2620 
2621 	return (v);
2622 }
2623 
2624 /*
2625  * Establish a file system and initialize it.
2626  */
2627 int
2628 vfs_attach(vfs)
2629 	struct vfsops *vfs;
2630 {
2631 	struct vfsops *v;
2632 	int error = 0;
2633 
2634 
2635 	/*
2636 	 * Make sure this file system doesn't already exist.
2637 	 */
2638 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2639 		if (strcmp(vfs->vfs_name, v->vfs_name) == 0) {
2640 			error = EEXIST;
2641 			goto out;
2642 		}
2643 	}
2644 
2645 	/*
2646 	 * Initialize the vnode operations for this file system.
2647 	 */
2648 	vfs_opv_init(vfs->vfs_opv_descs);
2649 
2650 	/*
2651 	 * Now initialize the file system itself.
2652 	 */
2653 	(*vfs->vfs_init)();
2654 
2655 	/*
2656 	 * ...and link it into the kernel's list.
2657 	 */
2658 	LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list);
2659 
2660 	/*
2661 	 * Sanity: make sure the reference count is 0.
2662 	 */
2663 	vfs->vfs_refcount = 0;
2664 
2665  out:
2666 	return (error);
2667 }
2668 
2669 /*
2670  * Remove a file system from the kernel.
2671  */
2672 int
2673 vfs_detach(vfs)
2674 	struct vfsops *vfs;
2675 {
2676 	struct vfsops *v;
2677 
2678 	/*
2679 	 * Make sure no one is using the filesystem.
2680 	 */
2681 	if (vfs->vfs_refcount != 0)
2682 		return (EBUSY);
2683 
2684 	/*
2685 	 * ...and remove it from the kernel's list.
2686 	 */
2687 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2688 		if (v == vfs) {
2689 			LIST_REMOVE(v, vfs_list);
2690 			break;
2691 		}
2692 	}
2693 
2694 	if (v == NULL)
2695 		return (ESRCH);
2696 
2697 	/*
2698 	 * Now run the file system-specific cleanups.
2699 	 */
2700 	(*vfs->vfs_done)();
2701 
2702 	/*
2703 	 * Free the vnode operations vector.
2704 	 */
2705 	vfs_opv_free(vfs->vfs_opv_descs);
2706 	return (0);
2707 }
2708 
2709 void
2710 vfs_reinit(void)
2711 {
2712 	struct vfsops *vfs;
2713 
2714 	LIST_FOREACH(vfs, &vfs_list, vfs_list) {
2715 		if (vfs->vfs_reinit) {
2716 			(*vfs->vfs_reinit)();
2717 		}
2718 	}
2719 }
2720 
2721 #ifdef DDB
2722 const char buf_flagbits[] =
2723 	"\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI"
2724 	"\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE"
2725 	"\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED"
2726 	"\32XXX\33VFLUSH";
2727 
2728 void
2729 vfs_buf_print(bp, full, pr)
2730 	struct buf *bp;
2731 	int full;
2732 	void (*pr) __P((const char *, ...));
2733 {
2734 	char buf[1024];
2735 
2736 	(*pr)("  vp %p lblkno 0x%x blkno 0x%x dev 0x%x\n",
2737 		  bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev);
2738 
2739 	bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf));
2740 	(*pr)("  error %d flags 0x%s\n", bp->b_error, buf);
2741 
2742 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
2743 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
2744 	(*pr)("  data %p saveaddr %p dep %p\n",
2745 		  bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep));
2746 	(*pr)("  iodone %p\n", bp->b_iodone);
2747 }
2748 
2749 
2750 const char vnode_flagbits[] =
2751 	"\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\5EXECMAP"
2752 	"\11XLOCK\12XWANT\13BWAIT\14ALIASED"
2753 	"\15DIROP\16LAYER\17ONWORKLIST\20DIRTY";
2754 
2755 const char *vnode_types[] = {
2756 	"VNON",
2757 	"VREG",
2758 	"VDIR",
2759 	"VBLK",
2760 	"VCHR",
2761 	"VLNK",
2762 	"VSOCK",
2763 	"VFIFO",
2764 	"VBAD",
2765 };
2766 
2767 const char *vnode_tags[] = {
2768 	"VT_NON",
2769 	"VT_UFS",
2770 	"VT_NFS",
2771 	"VT_MFS",
2772 	"VT_MSDOSFS",
2773 	"VT_LFS",
2774 	"VT_LOFS",
2775 	"VT_FDESC",
2776 	"VT_PORTAL",
2777 	"VT_NULL",
2778 	"VT_UMAP",
2779 	"VT_KERNFS",
2780 	"VT_PROCFS",
2781 	"VT_AFS",
2782 	"VT_ISOFS",
2783 	"VT_UNION",
2784 	"VT_ADOSFS",
2785 	"VT_EXT2FS",
2786 	"VT_CODA",
2787 	"VT_FILECORE",
2788 	"VT_NTFS",
2789 	"VT_VFS",
2790 	"VT_OVERLAY"
2791 };
2792 
2793 void
2794 vfs_vnode_print(vp, full, pr)
2795 	struct vnode *vp;
2796 	int full;
2797 	void (*pr) __P((const char *, ...));
2798 {
2799 	char buf[256];
2800 	const char *vtype, *vtag;
2801 	int tmp;
2802 
2803 	uvm_object_printit(&vp->v_uobj, full, pr);
2804 	bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf));
2805 	(*pr)("\nVNODE flags %s\n", buf);
2806 	(*pr)("mp %p numoutput %d size 0x%llx\n",
2807 	      vp->v_mount, vp->v_numoutput, vp->v_size);
2808 
2809 	(*pr)("data %p usecount %d writecount %ld holdcnt %ld numoutput %d\n",
2810 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2811 	      vp->v_holdcnt, vp->v_numoutput);
2812 
2813 	vtype = ((tmp = vp->v_type) >= 0 &&
2814 		 vp->v_type < sizeof(vnode_types) / sizeof(vnode_types[0])) ?
2815 		vnode_types[vp->v_type] : "UNKNOWN";
2816 	vtag = ((tmp = vp->v_tag) >= 0 &&
2817 		vp->v_tag < sizeof(vnode_tags) / sizeof(vnode_tags[0])) ?
2818 		vnode_tags[vp->v_tag] : "UNKNOWN";
2819 
2820 	(*pr)("type %s(%d) tag %s(%d) id 0x%lx mount %p typedata %p\n",
2821 	      vtype, vp->v_type, vtag, vp->v_tag,
2822 	      vp->v_id, vp->v_mount, vp->v_mountedhere);
2823 
2824 	if (full) {
2825 		struct buf *bp;
2826 
2827 		(*pr)("clean bufs:\n");
2828 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2829 			(*pr)(" bp %p\n", bp);
2830 			vfs_buf_print(bp, full, pr);
2831 		}
2832 
2833 		(*pr)("dirty bufs:\n");
2834 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2835 			(*pr)(" bp %p\n", bp);
2836 			vfs_buf_print(bp, full, pr);
2837 		}
2838 	}
2839 }
2840 #endif
2841