xref: /freebsd/sys/kern/vfs_export.c (revision 11afcc8f9f96d657b8e6f7547c02c1957331fc96)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
39  * $Id: vfs_subr.c,v 1.158 1998/07/11 07:45:43 bde Exp $
40  */
41 
42 /*
43  * External virtual filesystem routines
44  */
45 #include "opt_ddb.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/kernel.h>
50 #include <sys/proc.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/socket.h>
54 #include <sys/vnode.h>
55 #include <sys/stat.h>
56 #include <sys/buf.h>
57 #include <sys/domain.h>
58 #include <sys/dirent.h>
59 #include <sys/vmmeter.h>
60 
61 #include <machine/limits.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_extern.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_map.h>
68 #include <vm/vm_pager.h>
69 #include <vm/vnode_pager.h>
70 #include <vm/vm_zone.h>
71 #include <sys/sysctl.h>
72 
73 #include <miscfs/specfs/specdev.h>
74 
75 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
76 
77 static void	insmntque __P((struct vnode *vp, struct mount *mp));
78 #ifdef DDB
79 static void	printlockedvnodes __P((void));
80 #endif
81 static void	vclean __P((struct vnode *vp, int flags, struct proc *p));
82 static void	vfree __P((struct vnode *));
83 static void	vgonel __P((struct vnode *vp, struct proc *p));
84 static unsigned long	numvnodes;
85 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
86 
87 enum vtype iftovt_tab[16] = {
88 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
89 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
90 };
91 int vttoif_tab[9] = {
92 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
93 	S_IFSOCK, S_IFIFO, S_IFMT,
94 };
95 
96 /*
97  * Insq/Remq for the vnode usage lists.
98  */
99 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
100 #define	bufremvn(bp) {							\
101 	LIST_REMOVE(bp, b_vnbufs);					\
102 	(bp)->b_vnbufs.le_next = NOLIST;				\
103 }
104 
105 static TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
106 struct tobefreelist vnode_tobefree_list;	/* vnode free list */
107 
108 static u_long wantfreevnodes = 25;
109 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
110 static u_long freevnodes = 0;
111 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
112 
113 int vfs_ioopt = 0;
114 #ifdef ENABLE_VFS_IOOPT
115 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
116 #endif
117 
118 struct mntlist mountlist;	/* mounted filesystem list */
119 struct simplelock mountlist_slock;
120 static struct simplelock mntid_slock;
121 struct simplelock mntvnode_slock;
122 static struct simplelock vnode_free_list_slock;
123 static struct simplelock spechash_slock;
124 struct nfs_public nfs_pub;	/* publicly exported FS */
125 static vm_zone_t vnode_zone;
126 
127 /*
128  * The workitem queue.
129  */
130 #define SYNCER_MAXDELAY		32
131 int syncer_maxdelay =		SYNCER_MAXDELAY;	/* maximum delay time */
132 time_t syncdelay =		30;
133 int rushjob;				/* number of slots to run ASAP */
134 
135 static int syncer_delayno = 0;
136 static long syncer_mask;
137 LIST_HEAD(synclist, vnode);
138 static struct synclist *syncer_workitem_pending;
139 
140 int desiredvnodes;
141 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, "");
142 
143 static void	vfs_free_addrlist __P((struct netexport *nep));
144 static int	vfs_free_netcred __P((struct radix_node *rn, void *w));
145 static int	vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
146 				       struct export_args *argp));
147 
148 /*
149  * Initialize the vnode management data structures.
150  */
151 void
152 vntblinit()
153 {
154 
155 	desiredvnodes = maxproc + cnt.v_page_count / 4;
156 	simple_lock_init(&mntvnode_slock);
157 	simple_lock_init(&mntid_slock);
158 	simple_lock_init(&spechash_slock);
159 	TAILQ_INIT(&vnode_free_list);
160 	TAILQ_INIT(&vnode_tobefree_list);
161 	simple_lock_init(&vnode_free_list_slock);
162 	CIRCLEQ_INIT(&mountlist);
163 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
164 	/*
165 	 * Initialize the filesystem syncer.
166 	 */
167 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
168 		&syncer_mask);
169 	syncer_maxdelay = syncer_mask + 1;
170 }
171 
172 /*
173  * Mark a mount point as busy. Used to synchronize access and to delay
174  * unmounting. Interlock is not released on failure.
175  */
176 int
177 vfs_busy(mp, flags, interlkp, p)
178 	struct mount *mp;
179 	int flags;
180 	struct simplelock *interlkp;
181 	struct proc *p;
182 {
183 	int lkflags;
184 
185 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
186 		if (flags & LK_NOWAIT)
187 			return (ENOENT);
188 		mp->mnt_kern_flag |= MNTK_MWAIT;
189 		if (interlkp) {
190 			simple_unlock(interlkp);
191 		}
192 		/*
193 		 * Since all busy locks are shared except the exclusive
194 		 * lock granted when unmounting, the only place that a
195 		 * wakeup needs to be done is at the release of the
196 		 * exclusive lock at the end of dounmount.
197 		 */
198 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
199 		if (interlkp) {
200 			simple_lock(interlkp);
201 		}
202 		return (ENOENT);
203 	}
204 	lkflags = LK_SHARED | LK_NOPAUSE;
205 	if (interlkp)
206 		lkflags |= LK_INTERLOCK;
207 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
208 		panic("vfs_busy: unexpected lock failure");
209 	return (0);
210 }
211 
212 /*
213  * Free a busy filesystem.
214  */
215 void
216 vfs_unbusy(mp, p)
217 	struct mount *mp;
218 	struct proc *p;
219 {
220 
221 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
222 }
223 
224 /*
225  * Lookup a filesystem type, and if found allocate and initialize
226  * a mount structure for it.
227  *
228  * Devname is usually updated by mount(8) after booting.
229  */
230 int
231 vfs_rootmountalloc(fstypename, devname, mpp)
232 	char *fstypename;
233 	char *devname;
234 	struct mount **mpp;
235 {
236 	struct proc *p = curproc;	/* XXX */
237 	struct vfsconf *vfsp;
238 	struct mount *mp;
239 
240 	if (fstypename == NULL)
241 		return (ENODEV);
242 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
243 		if (!strcmp(vfsp->vfc_name, fstypename))
244 			break;
245 	if (vfsp == NULL)
246 		return (ENODEV);
247 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
248 	bzero((char *)mp, (u_long)sizeof(struct mount));
249 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
250 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
251 	LIST_INIT(&mp->mnt_vnodelist);
252 	mp->mnt_vfc = vfsp;
253 	mp->mnt_op = vfsp->vfc_vfsops;
254 	mp->mnt_flag = MNT_RDONLY;
255 	mp->mnt_vnodecovered = NULLVP;
256 	vfsp->vfc_refcount++;
257 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
258 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
259 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
260 	mp->mnt_stat.f_mntonname[0] = '/';
261 	mp->mnt_stat.f_mntonname[1] = 0;
262 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
263 	*mpp = mp;
264 	return (0);
265 }
266 
267 /*
268  * Find an appropriate filesystem to use for the root. If a filesystem
269  * has not been preselected, walk through the list of known filesystems
270  * trying those that have mountroot routines, and try them until one
271  * works or we have tried them all.
272  */
273 #ifdef notdef	/* XXX JH */
274 int
275 lite2_vfs_mountroot()
276 {
277 	struct vfsconf *vfsp;
278 	extern int (*lite2_mountroot) __P((void));
279 	int error;
280 
281 	if (lite2_mountroot != NULL)
282 		return ((*lite2_mountroot)());
283 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
284 		if (vfsp->vfc_mountroot == NULL)
285 			continue;
286 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
287 			return (0);
288 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
289 	}
290 	return (ENODEV);
291 }
292 #endif
293 
294 /*
295  * Lookup a mount point by filesystem identifier.
296  */
297 struct mount *
298 vfs_getvfs(fsid)
299 	fsid_t *fsid;
300 {
301 	register struct mount *mp;
302 
303 	simple_lock(&mountlist_slock);
304 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
305 	    mp = mp->mnt_list.cqe_next) {
306 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
307 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
308 			simple_unlock(&mountlist_slock);
309 			return (mp);
310 	    }
311 	}
312 	simple_unlock(&mountlist_slock);
313 	return ((struct mount *) 0);
314 }
315 
316 /*
317  * Get a new unique fsid
318  */
319 void
320 vfs_getnewfsid(mp)
321 	struct mount *mp;
322 {
323 	static u_short xxxfs_mntid;
324 
325 	fsid_t tfsid;
326 	int mtype;
327 
328 	simple_lock(&mntid_slock);
329 	mtype = mp->mnt_vfc->vfc_typenum;
330 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
331 	mp->mnt_stat.f_fsid.val[1] = mtype;
332 	if (xxxfs_mntid == 0)
333 		++xxxfs_mntid;
334 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
335 	tfsid.val[1] = mtype;
336 	if (mountlist.cqh_first != (void *)&mountlist) {
337 		while (vfs_getvfs(&tfsid)) {
338 			tfsid.val[0]++;
339 			xxxfs_mntid++;
340 		}
341 	}
342 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
343 	simple_unlock(&mntid_slock);
344 }
345 
346 /*
347  * Set vnode attributes to VNOVAL
348  */
349 void
350 vattr_null(vap)
351 	register struct vattr *vap;
352 {
353 
354 	vap->va_type = VNON;
355 	vap->va_size = VNOVAL;
356 	vap->va_bytes = VNOVAL;
357 	vap->va_mode = VNOVAL;
358 	vap->va_nlink = VNOVAL;
359 	vap->va_uid = VNOVAL;
360 	vap->va_gid = VNOVAL;
361 	vap->va_fsid = VNOVAL;
362 	vap->va_fileid = VNOVAL;
363 	vap->va_blocksize = VNOVAL;
364 	vap->va_rdev = VNOVAL;
365 	vap->va_atime.tv_sec = VNOVAL;
366 	vap->va_atime.tv_nsec = VNOVAL;
367 	vap->va_mtime.tv_sec = VNOVAL;
368 	vap->va_mtime.tv_nsec = VNOVAL;
369 	vap->va_ctime.tv_sec = VNOVAL;
370 	vap->va_ctime.tv_nsec = VNOVAL;
371 	vap->va_flags = VNOVAL;
372 	vap->va_gen = VNOVAL;
373 	vap->va_vaflags = 0;
374 }
375 
376 /*
377  * Routines having to do with the management of the vnode table.
378  */
379 extern vop_t **dead_vnodeop_p;
380 
381 /*
382  * Return the next vnode from the free list.
383  */
384 int
385 getnewvnode(tag, mp, vops, vpp)
386 	enum vtagtype tag;
387 	struct mount *mp;
388 	vop_t **vops;
389 	struct vnode **vpp;
390 {
391 	int s;
392 	struct proc *p = curproc;	/* XXX */
393 	struct vnode *vp, *tvp, *nvp;
394 	vm_object_t object;
395 	TAILQ_HEAD(freelst, vnode) vnode_tmp_list;
396 
397 	/*
398 	 * We take the least recently used vnode from the freelist
399 	 * if we can get it and it has no cached pages, and no
400 	 * namecache entries are relative to it.
401 	 * Otherwise we allocate a new vnode
402 	 */
403 
404 	s = splbio();
405 	simple_lock(&vnode_free_list_slock);
406 	TAILQ_INIT(&vnode_tmp_list);
407 
408 	for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) {
409 		nvp = TAILQ_NEXT(vp, v_freelist);
410 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
411 		if (vp->v_flag & VAGE) {
412 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
413 		} else {
414 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
415 		}
416 		vp->v_flag &= ~(VTBFREE|VAGE);
417 		vp->v_flag |= VFREE;
418 		if (vp->v_usecount)
419 			panic("tobe free vnode isn't");
420 		freevnodes++;
421 	}
422 
423 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
424 		vp = NULL;
425 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
426 		/*
427 		 * XXX: this is only here to be backwards compatible
428 		 */
429 		vp = NULL;
430 	} else {
431 		for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) {
432 
433 			nvp = TAILQ_NEXT(vp, v_freelist);
434 
435 			if (!simple_lock_try(&vp->v_interlock))
436 				continue;
437 			if (vp->v_usecount)
438 				panic("free vnode isn't");
439 
440 			object = vp->v_object;
441 			if (object && (object->resident_page_count || object->ref_count)) {
442 				printf("object inconsistant state: RPC: %d, RC: %d\n",
443 					object->resident_page_count, object->ref_count);
444 				/* Don't recycle if it's caching some pages */
445 				TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
446 				TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist);
447 				continue;
448 			} else if (LIST_FIRST(&vp->v_cache_src)) {
449 				/* Don't recycle if active in the namecache */
450 				simple_unlock(&vp->v_interlock);
451 				continue;
452 			} else {
453 				break;
454 			}
455 		}
456 	}
457 
458 	for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) {
459 		nvp = TAILQ_NEXT(tvp, v_freelist);
460 		TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist);
461 		TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist);
462 		simple_unlock(&tvp->v_interlock);
463 	}
464 
465 	if (vp) {
466 		vp->v_flag |= VDOOMED;
467 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
468 		freevnodes--;
469 		simple_unlock(&vnode_free_list_slock);
470 		cache_purge(vp);
471 		vp->v_lease = NULL;
472 		if (vp->v_type != VBAD) {
473 			vgonel(vp, p);
474 		} else {
475 			simple_unlock(&vp->v_interlock);
476 		}
477 
478 #ifdef DIAGNOSTIC
479 		{
480 			int s;
481 
482 			if (vp->v_data)
483 				panic("cleaned vnode isn't");
484 			s = splbio();
485 			if (vp->v_numoutput)
486 				panic("Clean vnode has pending I/O's");
487 			splx(s);
488 		}
489 #endif
490 		vp->v_flag = 0;
491 		vp->v_lastr = 0;
492 		vp->v_lastw = 0;
493 		vp->v_lasta = 0;
494 		vp->v_cstart = 0;
495 		vp->v_clen = 0;
496 		vp->v_socket = 0;
497 		vp->v_writecount = 0;	/* XXX */
498 		vp->v_maxio = 0;
499 	} else {
500 		simple_unlock(&vnode_free_list_slock);
501 		vp = (struct vnode *) zalloc(vnode_zone);
502 		bzero((char *) vp, sizeof *vp);
503 		simple_lock_init(&vp->v_interlock);
504 		vp->v_dd = vp;
505 		cache_purge(vp);
506 		LIST_INIT(&vp->v_cache_src);
507 		TAILQ_INIT(&vp->v_cache_dst);
508 		numvnodes++;
509 	}
510 
511 	vp->v_type = VNON;
512 	vp->v_tag = tag;
513 	vp->v_op = vops;
514 	insmntque(vp, mp);
515 	*vpp = vp;
516 	vp->v_usecount = 1;
517 	vp->v_data = 0;
518 	splx(s);
519 
520 	vfs_object_create(vp, p, p->p_ucred, TRUE);
521 	return (0);
522 }
523 
524 /*
525  * Move a vnode from one mount queue to another.
526  */
527 static void
528 insmntque(vp, mp)
529 	register struct vnode *vp;
530 	register struct mount *mp;
531 {
532 
533 	simple_lock(&mntvnode_slock);
534 	/*
535 	 * Delete from old mount point vnode list, if on one.
536 	 */
537 	if (vp->v_mount != NULL)
538 		LIST_REMOVE(vp, v_mntvnodes);
539 	/*
540 	 * Insert into list of vnodes for the new mount point, if available.
541 	 */
542 	if ((vp->v_mount = mp) == NULL) {
543 		simple_unlock(&mntvnode_slock);
544 		return;
545 	}
546 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
547 	simple_unlock(&mntvnode_slock);
548 }
549 
550 /*
551  * Update outstanding I/O count and do wakeup if requested.
552  */
553 void
554 vwakeup(bp)
555 	register struct buf *bp;
556 {
557 	register struct vnode *vp;
558 
559 	bp->b_flags &= ~B_WRITEINPROG;
560 	if ((vp = bp->b_vp)) {
561 		vp->v_numoutput--;
562 		if (vp->v_numoutput < 0)
563 			panic("vwakeup: neg numoutput");
564 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
565 			vp->v_flag &= ~VBWAIT;
566 			wakeup((caddr_t) &vp->v_numoutput);
567 		}
568 	}
569 }
570 
571 /*
572  * Flush out and invalidate all buffers associated with a vnode.
573  * Called with the underlying object locked.
574  */
575 int
576 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
577 	register struct vnode *vp;
578 	int flags;
579 	struct ucred *cred;
580 	struct proc *p;
581 	int slpflag, slptimeo;
582 {
583 	register struct buf *bp;
584 	struct buf *nbp, *blist;
585 	int s, error;
586 	vm_object_t object;
587 
588 	if (flags & V_SAVE) {
589 		s = splbio();
590 		while (vp->v_numoutput) {
591 			vp->v_flag |= VBWAIT;
592 			tsleep((caddr_t)&vp->v_numoutput,
593 				slpflag | (PRIBIO + 1),
594 				"vinvlbuf", slptimeo);
595 		}
596 		if (vp->v_dirtyblkhd.lh_first != NULL) {
597 			splx(s);
598 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
599 				return (error);
600 			s = splbio();
601 			if (vp->v_numoutput > 0 ||
602 			    vp->v_dirtyblkhd.lh_first != NULL)
603 				panic("vinvalbuf: dirty bufs");
604 		}
605 		splx(s);
606   	}
607 	s = splbio();
608 	for (;;) {
609 		if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA))
610 			while (blist && blist->b_lblkno < 0)
611 				blist = blist->b_vnbufs.le_next;
612 		if (!blist && (blist = vp->v_dirtyblkhd.lh_first) &&
613 		    (flags & V_SAVEMETA))
614 			while (blist && blist->b_lblkno < 0)
615 				blist = blist->b_vnbufs.le_next;
616 		if (!blist)
617 			break;
618 
619 		for (bp = blist; bp; bp = nbp) {
620 			nbp = bp->b_vnbufs.le_next;
621 			if ((flags & V_SAVEMETA) && bp->b_lblkno < 0)
622 				continue;
623 			if (bp->b_flags & B_BUSY) {
624 				bp->b_flags |= B_WANTED;
625 				error = tsleep((caddr_t) bp,
626 				    slpflag | (PRIBIO + 4), "vinvalbuf",
627 				    slptimeo);
628 				if (error) {
629 					splx(s);
630 					return (error);
631 				}
632 				break;
633 			}
634 			/*
635 			 * XXX Since there are no node locks for NFS, I
636 			 * believe there is a slight chance that a delayed
637 			 * write will occur while sleeping just above, so
638 			 * check for it.  Note that vfs_bio_awrite expects
639 			 * buffers to reside on a queue, while VOP_BWRITE and
640 			 * brelse do not.
641 			 */
642 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
643 				(flags & V_SAVE)) {
644 
645 				if (bp->b_vp == vp) {
646 					if (bp->b_flags & B_CLUSTEROK) {
647 						vfs_bio_awrite(bp);
648 					} else {
649 						bremfree(bp);
650 						bp->b_flags |= (B_BUSY | B_ASYNC);
651 						VOP_BWRITE(bp);
652 					}
653 				} else {
654 					bremfree(bp);
655 					bp->b_flags |= B_BUSY;
656 					(void) VOP_BWRITE(bp);
657 				}
658 				break;
659 			}
660 			bremfree(bp);
661 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF | B_BUSY);
662 			bp->b_flags &= ~B_ASYNC;
663 			brelse(bp);
664 		}
665 	}
666 
667 	while (vp->v_numoutput > 0) {
668 		vp->v_flag |= VBWAIT;
669 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
670 	}
671 
672 	splx(s);
673 
674 	/*
675 	 * Destroy the copy in the VM cache, too.
676 	 */
677 	simple_lock(&vp->v_interlock);
678 	object = vp->v_object;
679 	if (object != NULL) {
680 		if (flags & V_SAVEMETA)
681 			vm_object_page_remove(object, 0, object->size,
682 				(flags & V_SAVE) ? TRUE : FALSE);
683 		else
684 			vm_object_page_remove(object, 0, 0,
685 				(flags & V_SAVE) ? TRUE : FALSE);
686 	}
687 	simple_unlock(&vp->v_interlock);
688 
689 	if (!(flags & V_SAVEMETA) &&
690 	    (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first))
691 		panic("vinvalbuf: flush failed");
692 	return (0);
693 }
694 
695 /*
696  * Truncate a file's buffer and pages to a specified length.  This
697  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
698  * sync activity.
699  */
700 int
701 vtruncbuf(vp, cred, p, length, blksize)
702 	register struct vnode *vp;
703 	struct ucred *cred;
704 	struct proc *p;
705 	off_t length;
706 	int blksize;
707 {
708 	register struct buf *bp;
709 	struct buf *nbp, *blist;
710 	int s, error, anyfreed;
711 	vm_object_t object;
712 	int trunclbn;
713 
714 	/*
715 	 * Round up to the *next* lbn.
716 	 */
717 	trunclbn = (length + blksize - 1) / blksize;
718 
719 	s = splbio();
720 restart:
721 	anyfreed = 1;
722 	for (;anyfreed;) {
723 		anyfreed = 0;
724 		for ( bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
725 
726 			nbp = LIST_NEXT(bp, b_vnbufs);
727 
728 			if (bp->b_lblkno >= trunclbn) {
729 				if (bp->b_flags & B_BUSY) {
730 					bp->b_flags |= B_WANTED;
731 					tsleep(bp, PRIBIO + 4, "vtrb1", 0);
732 					goto restart;
733 				} else {
734 					bremfree(bp);
735 					bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
736 					bp->b_flags &= ~B_ASYNC;
737 					brelse(bp);
738 					anyfreed = 1;
739 				}
740 				if (nbp &&
741 					((LIST_NEXT(nbp, b_vnbufs) == NOLIST) ||
742 					 (nbp->b_vp != vp) ||
743 					 (nbp->b_flags & B_DELWRI))) {
744 					goto restart;
745 				}
746 			}
747 		}
748 
749 		for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
750 
751 			nbp = LIST_NEXT(bp, b_vnbufs);
752 
753 			if (bp->b_lblkno >= trunclbn) {
754 				if (bp->b_flags & B_BUSY) {
755 					bp->b_flags |= B_WANTED;
756 					tsleep(bp, PRIBIO + 4, "vtrb2", 0);
757 					goto restart;
758 				} else {
759 					bremfree(bp);
760 					bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
761 					bp->b_flags &= ~B_ASYNC;
762 					brelse(bp);
763 					anyfreed = 1;
764 				}
765 				if (nbp &&
766 					((LIST_NEXT(nbp, b_vnbufs) == NOLIST) ||
767 					 (nbp->b_vp != vp) ||
768 					 (nbp->b_flags & B_DELWRI) == 0)) {
769 					goto restart;
770 				}
771 			}
772 		}
773 	}
774 
775 	if (length > 0) {
776 restartsync:
777 		for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
778 
779 			nbp = LIST_NEXT(bp, b_vnbufs);
780 
781 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
782 				if (bp->b_flags & B_BUSY) {
783 					bp->b_flags |= B_WANTED;
784 					tsleep(bp, PRIBIO, "vtrb3", 0);
785 				} else {
786 					bremfree(bp);
787 					bp->b_flags |= B_BUSY;
788 					if (bp->b_vp == vp) {
789 						bp->b_flags |= B_ASYNC;
790 					} else {
791 						bp->b_flags &= ~B_ASYNC;
792 					}
793 					VOP_BWRITE(bp);
794 				}
795 				goto restartsync;
796 			}
797 
798 		}
799 	}
800 
801 	while (vp->v_numoutput > 0) {
802 		vp->v_flag |= VBWAIT;
803 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
804 	}
805 
806 	splx(s);
807 
808 	vnode_pager_setsize(vp, length);
809 
810 	return (0);
811 }
812 
813 /*
814  * Associate a buffer with a vnode.
815  */
816 void
817 bgetvp(vp, bp)
818 	register struct vnode *vp;
819 	register struct buf *bp;
820 {
821 	int s;
822 
823 #if defined(DIAGNOSTIC)
824 	if (bp->b_vp)
825 		panic("bgetvp: not free");
826 #endif
827 	vhold(vp);
828 	bp->b_vp = vp;
829 	if (vp->v_type == VBLK || vp->v_type == VCHR)
830 		bp->b_dev = vp->v_rdev;
831 	else
832 		bp->b_dev = NODEV;
833 	/*
834 	 * Insert onto list for new vnode.
835 	 */
836 	s = splbio();
837 	bufinsvn(bp, &vp->v_cleanblkhd);
838 	splx(s);
839 }
840 
841 /*
842  * Disassociate a buffer from a vnode.
843  */
844 void
845 brelvp(bp)
846 	register struct buf *bp;
847 {
848 	struct vnode *vp;
849 	int s;
850 
851 #if defined(DIAGNOSTIC)
852 	if (bp->b_vp == (struct vnode *) 0)
853 		panic("brelvp: NULL");
854 #endif
855 
856 	/*
857 	 * Delete from old vnode list, if on one.
858 	 */
859 	vp = bp->b_vp;
860 	s = splbio();
861 	if (bp->b_vnbufs.le_next != NOLIST)
862 		bufremvn(bp);
863 	if ((vp->v_flag & VONWORKLST) && (LIST_FIRST(&vp->v_dirtyblkhd) == NULL)) {
864 		vp->v_flag &= ~VONWORKLST;
865 		LIST_REMOVE(vp, v_synclist);
866 	}
867 	splx(s);
868 	bp->b_vp = (struct vnode *) 0;
869 	vdrop(vp);
870 }
871 
872 /*
873  * The workitem queue.
874  *
875  * It is useful to delay writes of file data and filesystem metadata
876  * for tens of seconds so that quickly created and deleted files need
877  * not waste disk bandwidth being created and removed. To realize this,
878  * we append vnodes to a "workitem" queue. When running with a soft
879  * updates implementation, most pending metadata dependencies should
880  * not wait for more than a few seconds. Thus, mounted on block devices
881  * are delayed only about a half the time that file data is delayed.
882  * Similarly, directory updates are more critical, so are only delayed
883  * about a third the time that file data is delayed. Thus, there are
884  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
885  * one each second (driven off the filesystem syner process). The
886  * syncer_delayno variable indicates the next queue that is to be processed.
887  * Items that need to be processed soon are placed in this queue:
888  *
889  *	syncer_workitem_pending[syncer_delayno]
890  *
891  * A delay of fifteen seconds is done by placing the request fifteen
892  * entries later in the queue:
893  *
894  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
895  *
896  */
897 
898 /*
899  * Add an item to the syncer work queue.
900  */
901 void
902 vn_syncer_add_to_worklist(vp, delay)
903 	struct vnode *vp;
904 	int delay;
905 {
906 	int s, slot;
907 
908 	s = splbio();
909 
910 	if (vp->v_flag & VONWORKLST) {
911 		LIST_REMOVE(vp, v_synclist);
912 	}
913 
914 	if (delay > syncer_maxdelay - 2)
915 		delay = syncer_maxdelay - 2;
916 	slot = (syncer_delayno + delay) & syncer_mask;
917 
918 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
919 	vp->v_flag |= VONWORKLST;
920 	splx(s);
921 }
922 
923 static void sched_sync __P((void));
924 static struct	proc *updateproc;
925 static struct kproc_desc up_kp = {
926 	"syncer",
927 	sched_sync,
928 	&updateproc
929 };
930 SYSINIT_KT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
931 
932 /*
933  * System filesystem synchronizer daemon.
934  */
935 void
936 sched_sync(void)
937 {
938 	struct synclist *slp;
939 	struct vnode *vp;
940 	long starttime;
941 	int s;
942 	struct proc *p = updateproc;
943 
944 	for (;;) {
945 		starttime = time_second;
946 
947 		/*
948 		 * Push files whose dirty time has expired.
949 		 */
950 		s = splbio();
951 		slp = &syncer_workitem_pending[syncer_delayno];
952 		syncer_delayno += 1;
953 		if (syncer_delayno == syncer_maxdelay)
954 			syncer_delayno = 0;
955 		splx(s);
956 
957 		while ((vp = LIST_FIRST(slp)) != NULL) {
958 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
959 			(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
960 			VOP_UNLOCK(vp, 0, p);
961 			if (LIST_FIRST(slp) == vp) {
962 				if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL &&
963 				    vp->v_type != VBLK)
964 					panic("sched_sync: fsync failed");
965 				/*
966 				 * Move ourselves to the back of the sync list.
967 				 */
968 				LIST_REMOVE(vp, v_synclist);
969 				vn_syncer_add_to_worklist(vp, syncdelay);
970 			}
971 		}
972 
973 		/*
974 		 * Do soft update processing.
975 		 */
976 		if (bioops.io_sync)
977 			(*bioops.io_sync)(NULL);
978 
979 		/*
980 		 * The variable rushjob allows the kernel to speed up the
981 		 * processing of the filesystem syncer process. A rushjob
982 		 * value of N tells the filesystem syncer to process the next
983 		 * N seconds worth of work on its queue ASAP. Currently rushjob
984 		 * is used by the soft update code to speed up the filesystem
985 		 * syncer process when the incore state is getting so far
986 		 * ahead of the disk that the kernel memory pool is being
987 		 * threatened with exhaustion.
988 		 */
989 		if (rushjob > 0) {
990 			rushjob -= 1;
991 			continue;
992 		}
993 		/*
994 		 * If it has taken us less than a second to process the
995 		 * current work, then wait. Otherwise start right over
996 		 * again. We can still lose time if any single round
997 		 * takes more than two seconds, but it does not really
998 		 * matter as we are just trying to generally pace the
999 		 * filesystem activity.
1000 		 */
1001 		if (time_second == starttime)
1002 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1003 	}
1004 }
1005 
1006 /*
1007  * Associate a p-buffer with a vnode.
1008  */
1009 void
1010 pbgetvp(vp, bp)
1011 	register struct vnode *vp;
1012 	register struct buf *bp;
1013 {
1014 #if defined(DIAGNOSTIC)
1015 	if (bp->b_vp)
1016 		panic("pbgetvp: not free");
1017 #endif
1018 	bp->b_vp = vp;
1019 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1020 		bp->b_dev = vp->v_rdev;
1021 	else
1022 		bp->b_dev = NODEV;
1023 }
1024 
1025 /*
1026  * Disassociate a p-buffer from a vnode.
1027  */
1028 void
1029 pbrelvp(bp)
1030 	register struct buf *bp;
1031 {
1032 
1033 #if defined(DIAGNOSTIC)
1034 	if (bp->b_vp == (struct vnode *) 0)
1035 		panic("pbrelvp: NULL");
1036 #endif
1037 
1038 	bp->b_vp = (struct vnode *) 0;
1039 }
1040 
1041 /*
1042  * Reassign a buffer from one vnode to another.
1043  * Used to assign file specific control information
1044  * (indirect blocks) to the vnode to which they belong.
1045  */
1046 void
1047 reassignbuf(bp, newvp)
1048 	register struct buf *bp;
1049 	register struct vnode *newvp;
1050 {
1051 	struct buflists *listheadp;
1052 	int delay;
1053 	int s;
1054 
1055 	if (newvp == NULL) {
1056 		printf("reassignbuf: NULL");
1057 		return;
1058 	}
1059 
1060 	s = splbio();
1061 	/*
1062 	 * Delete from old vnode list, if on one.
1063 	 */
1064 	if (bp->b_vnbufs.le_next != NOLIST) {
1065 		bufremvn(bp);
1066 		vdrop(bp->b_vp);
1067 	}
1068 	/*
1069 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1070 	 * of clean buffers.
1071 	 */
1072 	if (bp->b_flags & B_DELWRI) {
1073 		struct buf *tbp;
1074 
1075 		listheadp = &newvp->v_dirtyblkhd;
1076 		if ((newvp->v_flag & VONWORKLST) == 0) {
1077 			switch (newvp->v_type) {
1078 			case VDIR:
1079 				delay = syncdelay / 3;
1080 				break;
1081 			case VBLK:
1082 				if (newvp->v_specmountpoint != NULL) {
1083 					delay = syncdelay / 2;
1084 					break;
1085 				}
1086 				/* fall through */
1087 			default:
1088 				delay = syncdelay;
1089 			}
1090 			vn_syncer_add_to_worklist(newvp, delay);
1091 		}
1092 		tbp = listheadp->lh_first;
1093 		if (!tbp || (tbp->b_lblkno > bp->b_lblkno)) {
1094 			bufinsvn(bp, listheadp);
1095 		} else {
1096 			while (tbp->b_vnbufs.le_next &&
1097 			    (tbp->b_vnbufs.le_next->b_lblkno < bp->b_lblkno)) {
1098 				tbp = tbp->b_vnbufs.le_next;
1099 			}
1100 			LIST_INSERT_AFTER(tbp, bp, b_vnbufs);
1101 		}
1102 	} else {
1103 		bufinsvn(bp, &newvp->v_cleanblkhd);
1104 		if ((newvp->v_flag & VONWORKLST) &&
1105 			LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
1106 			newvp->v_flag &= ~VONWORKLST;
1107 			LIST_REMOVE(newvp, v_synclist);
1108 		}
1109 	}
1110 	bp->b_vp = newvp;
1111 	vhold(bp->b_vp);
1112 	splx(s);
1113 }
1114 
1115 #ifndef SLICE
1116 /*
1117  * Create a vnode for a block device.
1118  * Used for mounting the root file system.
1119  */
1120 int
1121 bdevvp(dev, vpp)
1122 	dev_t dev;
1123 	struct vnode **vpp;
1124 {
1125 	register struct vnode *vp;
1126 	struct vnode *nvp;
1127 	int error;
1128 
1129 	if (dev == NODEV)
1130 		return (0);
1131 	error = getnewvnode(VT_NON, (struct mount *) 0, spec_vnodeop_p, &nvp);
1132 	if (error) {
1133 		*vpp = 0;
1134 		return (error);
1135 	}
1136 	vp = nvp;
1137 	vp->v_type = VBLK;
1138 	if ((nvp = checkalias(vp, dev, (struct mount *) 0))) {
1139 		vput(vp);
1140 		vp = nvp;
1141 	}
1142 	*vpp = vp;
1143 	return (0);
1144 }
1145 #endif	/* !SLICE */
1146 
1147 /*
1148  * Check to see if the new vnode represents a special device
1149  * for which we already have a vnode (either because of
1150  * bdevvp() or because of a different vnode representing
1151  * the same block device). If such an alias exists, deallocate
1152  * the existing contents and return the aliased vnode. The
1153  * caller is responsible for filling it with its new contents.
1154  */
1155 struct vnode *
1156 checkalias(nvp, nvp_rdev, mp)
1157 	register struct vnode *nvp;
1158 	dev_t nvp_rdev;
1159 	struct mount *mp;
1160 {
1161 	struct proc *p = curproc;	/* XXX */
1162 	struct vnode *vp;
1163 	struct vnode **vpp;
1164 
1165 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1166 		return (NULLVP);
1167 
1168 	vpp = &speclisth[SPECHASH(nvp_rdev)];
1169 loop:
1170 	simple_lock(&spechash_slock);
1171 	for (vp = *vpp; vp; vp = vp->v_specnext) {
1172 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1173 			continue;
1174 		/*
1175 		 * Alias, but not in use, so flush it out.
1176 		 * Only alias active device nodes.
1177 		 * Not sure why we don't re-use this like we do below.
1178 		 */
1179 		simple_lock(&vp->v_interlock);
1180 		if (vp->v_usecount == 0) {
1181 			simple_unlock(&spechash_slock);
1182 			vgonel(vp, p);
1183 			goto loop;
1184 		}
1185 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) {
1186 			/*
1187 			 * It dissappeared, and we may have slept.
1188 			 * Restart from the beginning
1189 			 */
1190 			simple_unlock(&spechash_slock);
1191 			goto loop;
1192 		}
1193 		break;
1194 	}
1195 	/*
1196 	 * It would be a lot clearer what is going on here if
1197 	 * this had been expressed as:
1198 	 * if ( vp && (vp->v_tag == VT_NULL))
1199 	 * and the clauses had been swapped.
1200 	 */
1201 	if (vp == NULL || vp->v_tag != VT_NON) {
1202 		/*
1203 		 * Put the new vnode into the hash chain.
1204 		 * and if there was an alias, connect them.
1205 		 */
1206 		MALLOC(nvp->v_specinfo, struct specinfo *,
1207 		    sizeof(struct specinfo), M_VNODE, M_WAITOK);
1208 		nvp->v_rdev = nvp_rdev;
1209 		nvp->v_hashchain = vpp;
1210 		nvp->v_specnext = *vpp;
1211 		nvp->v_specmountpoint = NULL;
1212 		simple_unlock(&spechash_slock);
1213 		*vpp = nvp;
1214 		if (vp != NULLVP) {
1215 			nvp->v_flag |= VALIASED;
1216 			vp->v_flag |= VALIASED;
1217 			vput(vp);
1218 		}
1219 		return (NULLVP);
1220 	}
1221 	/*
1222 	 * if ( vp && (vp->v_tag == VT_NULL))
1223 	 * We have a vnode alias, but it is a trashed.
1224 	 * Make it look like it's newley allocated. (by getnewvnode())
1225 	 * The caller should use this instead.
1226 	 */
1227 	simple_unlock(&spechash_slock);
1228 	VOP_UNLOCK(vp, 0, p);
1229 	simple_lock(&vp->v_interlock);
1230 	vclean(vp, 0, p);
1231 	vp->v_op = nvp->v_op;
1232 	vp->v_tag = nvp->v_tag;
1233 	nvp->v_type = VNON;
1234 	insmntque(vp, mp);
1235 	return (vp);
1236 }
1237 
1238 /*
1239  * Grab a particular vnode from the free list, increment its
1240  * reference count and lock it. The vnode lock bit is set the
1241  * vnode is being eliminated in vgone. The process is awakened
1242  * when the transition is completed, and an error returned to
1243  * indicate that the vnode is no longer usable (possibly having
1244  * been changed to a new file system type).
1245  */
1246 int
1247 vget(vp, flags, p)
1248 	register struct vnode *vp;
1249 	int flags;
1250 	struct proc *p;
1251 {
1252 	int error;
1253 
1254 	/*
1255 	 * If the vnode is in the process of being cleaned out for
1256 	 * another use, we wait for the cleaning to finish and then
1257 	 * return failure. Cleaning is determined by checking that
1258 	 * the VXLOCK flag is set.
1259 	 */
1260 	if ((flags & LK_INTERLOCK) == 0) {
1261 		simple_lock(&vp->v_interlock);
1262 	}
1263 	if (vp->v_flag & VXLOCK) {
1264 		vp->v_flag |= VXWANT;
1265 		simple_unlock(&vp->v_interlock);
1266 		tsleep((caddr_t)vp, PINOD, "vget", 0);
1267 		return (ENOENT);
1268 	}
1269 
1270 	vp->v_usecount++;
1271 
1272 	if (VSHOULDBUSY(vp))
1273 		vbusy(vp);
1274 	if (flags & LK_TYPE_MASK) {
1275 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1276 			/*
1277 			 * must expand vrele here because we do not want
1278 			 * to call VOP_INACTIVE if the reference count
1279 			 * drops back to zero since it was never really
1280 			 * active. We must remove it from the free list
1281 			 * before sleeping so that multiple processes do
1282 			 * not try to recycle it.
1283 			 */
1284 			simple_lock(&vp->v_interlock);
1285 			vp->v_usecount--;
1286 			if (VSHOULDFREE(vp))
1287 				vfree(vp);
1288 			simple_unlock(&vp->v_interlock);
1289 		}
1290 		return (error);
1291 	}
1292 	simple_unlock(&vp->v_interlock);
1293 	return (0);
1294 }
1295 
1296 void
1297 vref(struct vnode *vp)
1298 {
1299 	simple_lock(&vp->v_interlock);
1300 	vp->v_usecount++;
1301 	simple_unlock(&vp->v_interlock);
1302 }
1303 
1304 /*
1305  * Vnode put/release.
1306  * If count drops to zero, call inactive routine and return to freelist.
1307  */
1308 void
1309 vrele(vp)
1310 	struct vnode *vp;
1311 {
1312 	struct proc *p = curproc;	/* XXX */
1313 
1314 #ifdef DIAGNOSTIC
1315 	if (vp == NULL)
1316 		panic("vrele: null vp");
1317 #endif
1318 	simple_lock(&vp->v_interlock);
1319 
1320 	if (vp->v_usecount > 1) {
1321 
1322 		vp->v_usecount--;
1323 		simple_unlock(&vp->v_interlock);
1324 
1325 		return;
1326 	}
1327 
1328 	if (vp->v_usecount == 1) {
1329 
1330 		vp->v_usecount--;
1331 
1332 		if (VSHOULDFREE(vp))
1333 			vfree(vp);
1334 	/*
1335 	 * If we are doing a vput, the node is already locked, and we must
1336 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1337 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1338 	 */
1339 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1340 			VOP_INACTIVE(vp, p);
1341 		}
1342 
1343 	} else {
1344 #ifdef DIAGNOSTIC
1345 		vprint("vrele: negative ref count", vp);
1346 		simple_unlock(&vp->v_interlock);
1347 #endif
1348 		panic("vrele: negative ref cnt");
1349 	}
1350 }
1351 
1352 void
1353 vput(vp)
1354 	struct vnode *vp;
1355 {
1356 	struct proc *p = curproc;	/* XXX */
1357 
1358 #ifdef DIAGNOSTIC
1359 	if (vp == NULL)
1360 		panic("vput: null vp");
1361 #endif
1362 
1363 	simple_lock(&vp->v_interlock);
1364 
1365 	if (vp->v_usecount > 1) {
1366 
1367 		vp->v_usecount--;
1368 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1369 		return;
1370 
1371 	}
1372 
1373 	if (vp->v_usecount == 1) {
1374 
1375 		vp->v_usecount--;
1376 		if (VSHOULDFREE(vp))
1377 			vfree(vp);
1378 	/*
1379 	 * If we are doing a vput, the node is already locked, and we must
1380 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1381 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1382 	 */
1383 		simple_unlock(&vp->v_interlock);
1384 		VOP_INACTIVE(vp, p);
1385 
1386 	} else {
1387 #ifdef DIAGNOSTIC
1388 		vprint("vput: negative ref count", vp);
1389 #endif
1390 		panic("vput: negative ref cnt");
1391 	}
1392 }
1393 
1394 /*
1395  * Somebody doesn't want the vnode recycled.
1396  */
1397 void
1398 vhold(vp)
1399 	register struct vnode *vp;
1400 {
1401 	int s;
1402 
1403   	s = splbio();
1404 	vp->v_holdcnt++;
1405 	if (VSHOULDBUSY(vp))
1406 		vbusy(vp);
1407 	splx(s);
1408 }
1409 
1410 /*
1411  * One less who cares about this vnode.
1412  */
1413 void
1414 vdrop(vp)
1415 	register struct vnode *vp;
1416 {
1417 	int s;
1418 
1419 	s = splbio();
1420 	if (vp->v_holdcnt <= 0)
1421 		panic("vdrop: holdcnt");
1422 	vp->v_holdcnt--;
1423 	if (VSHOULDFREE(vp))
1424 		vfree(vp);
1425 	splx(s);
1426 }
1427 
1428 /*
1429  * Remove any vnodes in the vnode table belonging to mount point mp.
1430  *
1431  * If MNT_NOFORCE is specified, there should not be any active ones,
1432  * return error if any are found (nb: this is a user error, not a
1433  * system error). If MNT_FORCE is specified, detach any active vnodes
1434  * that are found.
1435  */
1436 #ifdef DIAGNOSTIC
1437 static int busyprt = 0;		/* print out busy vnodes */
1438 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1439 #endif
1440 
1441 int
1442 vflush(mp, skipvp, flags)
1443 	struct mount *mp;
1444 	struct vnode *skipvp;
1445 	int flags;
1446 {
1447 	struct proc *p = curproc;	/* XXX */
1448 	struct vnode *vp, *nvp;
1449 	int busy = 0;
1450 
1451 	simple_lock(&mntvnode_slock);
1452 loop:
1453 	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
1454 		/*
1455 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1456 		 * Start over if it has (it won't be on the list anymore).
1457 		 */
1458 		if (vp->v_mount != mp)
1459 			goto loop;
1460 		nvp = vp->v_mntvnodes.le_next;
1461 		/*
1462 		 * Skip over a selected vnode.
1463 		 */
1464 		if (vp == skipvp)
1465 			continue;
1466 
1467 		simple_lock(&vp->v_interlock);
1468 		/*
1469 		 * Skip over a vnodes marked VSYSTEM.
1470 		 */
1471 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1472 			simple_unlock(&vp->v_interlock);
1473 			continue;
1474 		}
1475 		/*
1476 		 * If WRITECLOSE is set, only flush out regular file vnodes
1477 		 * open for writing.
1478 		 */
1479 		if ((flags & WRITECLOSE) &&
1480 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1481 			simple_unlock(&vp->v_interlock);
1482 			continue;
1483 		}
1484 
1485 		/*
1486 		 * With v_usecount == 0, all we need to do is clear out the
1487 		 * vnode data structures and we are done.
1488 		 */
1489 		if (vp->v_usecount == 0) {
1490 			simple_unlock(&mntvnode_slock);
1491 			vgonel(vp, p);
1492 			simple_lock(&mntvnode_slock);
1493 			continue;
1494 		}
1495 
1496 		/*
1497 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1498 		 * or character devices, revert to an anonymous device. For
1499 		 * all other files, just kill them.
1500 		 */
1501 		if (flags & FORCECLOSE) {
1502 			simple_unlock(&mntvnode_slock);
1503 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1504 				vgonel(vp, p);
1505 			} else {
1506 				vclean(vp, 0, p);
1507 				vp->v_op = spec_vnodeop_p;
1508 				insmntque(vp, (struct mount *) 0);
1509 			}
1510 			simple_lock(&mntvnode_slock);
1511 			continue;
1512 		}
1513 #ifdef DIAGNOSTIC
1514 		if (busyprt)
1515 			vprint("vflush: busy vnode", vp);
1516 #endif
1517 		simple_unlock(&vp->v_interlock);
1518 		busy++;
1519 	}
1520 	simple_unlock(&mntvnode_slock);
1521 	if (busy)
1522 		return (EBUSY);
1523 	return (0);
1524 }
1525 
1526 /*
1527  * Disassociate the underlying file system from a vnode.
1528  */
1529 static void
1530 vclean(vp, flags, p)
1531 	struct vnode *vp;
1532 	int flags;
1533 	struct proc *p;
1534 {
1535 	int active;
1536 	vm_object_t obj;
1537 
1538 	/*
1539 	 * Check to see if the vnode is in use. If so we have to reference it
1540 	 * before we clean it out so that its count cannot fall to zero and
1541 	 * generate a race against ourselves to recycle it.
1542 	 */
1543 	if ((active = vp->v_usecount))
1544 		vp->v_usecount++;
1545 
1546 	/*
1547 	 * Prevent the vnode from being recycled or brought into use while we
1548 	 * clean it out.
1549 	 */
1550 	if (vp->v_flag & VXLOCK)
1551 		panic("vclean: deadlock");
1552 	vp->v_flag |= VXLOCK;
1553 	/*
1554 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1555 	 * have the object locked while it cleans it out. The VOP_LOCK
1556 	 * ensures that the VOP_INACTIVE routine is done with its work.
1557 	 * For active vnodes, it ensures that no other activity can
1558 	 * occur while the underlying object is being cleaned out.
1559 	 */
1560 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1561 
1562 	/*
1563 	 * Clean out any buffers associated with the vnode.
1564 	 */
1565 	vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1566 	if (obj = vp->v_object) {
1567 		if (obj->ref_count == 0) {
1568 			/*
1569 			 * This is a normal way of shutting down the object/vnode
1570 			 * association.
1571 			 */
1572 			vm_object_terminate(obj);
1573 		} else {
1574 			/*
1575 			 * Woe to the process that tries to page now :-).
1576 			 */
1577 			vm_pager_deallocate(obj);
1578 		}
1579 	}
1580 
1581 	/*
1582 	 * If purging an active vnode, it must be closed and
1583 	 * deactivated before being reclaimed. Note that the
1584 	 * VOP_INACTIVE will unlock the vnode.
1585 	 */
1586 	if (active) {
1587 		if (flags & DOCLOSE)
1588 			VOP_CLOSE(vp, IO_NDELAY, NOCRED, p);
1589 		VOP_INACTIVE(vp, p);
1590 	} else {
1591 		/*
1592 		 * Any other processes trying to obtain this lock must first
1593 		 * wait for VXLOCK to clear, then call the new lock operation.
1594 		 */
1595 		VOP_UNLOCK(vp, 0, p);
1596 	}
1597 	/*
1598 	 * Reclaim the vnode.
1599 	 */
1600 	if (VOP_RECLAIM(vp, p))
1601 		panic("vclean: cannot reclaim");
1602 
1603 	if (active)
1604 		vrele(vp);
1605 
1606 	cache_purge(vp);
1607 	if (vp->v_vnlock) {
1608 #if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */
1609 #ifdef DIAGNOSTIC
1610 		if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0)
1611 			vprint("vclean: lock not drained", vp);
1612 #endif
1613 #endif
1614 		FREE(vp->v_vnlock, M_VNODE);
1615 		vp->v_vnlock = NULL;
1616 	}
1617 
1618 	if (VSHOULDFREE(vp))
1619 		vfree(vp);
1620 
1621 	/*
1622 	 * Done with purge, notify sleepers of the grim news.
1623 	 */
1624 	vp->v_op = dead_vnodeop_p;
1625 	vn_pollgone(vp);
1626 	vp->v_tag = VT_NON;
1627 	vp->v_flag &= ~VXLOCK;
1628 	if (vp->v_flag & VXWANT) {
1629 		vp->v_flag &= ~VXWANT;
1630 		wakeup((caddr_t) vp);
1631 	}
1632 }
1633 
1634 /*
1635  * Eliminate all activity associated with the requested vnode
1636  * and with all vnodes aliased to the requested vnode.
1637  */
1638 int
1639 vop_revoke(ap)
1640 	struct vop_revoke_args /* {
1641 		struct vnode *a_vp;
1642 		int a_flags;
1643 	} */ *ap;
1644 {
1645 	struct vnode *vp, *vq;
1646 	struct proc *p = curproc;	/* XXX */
1647 
1648 #ifdef DIAGNOSTIC
1649 	if ((ap->a_flags & REVOKEALL) == 0)
1650 		panic("vop_revoke");
1651 #endif
1652 
1653 	vp = ap->a_vp;
1654 	simple_lock(&vp->v_interlock);
1655 
1656 	if (vp->v_flag & VALIASED) {
1657 		/*
1658 		 * If a vgone (or vclean) is already in progress,
1659 		 * wait until it is done and return.
1660 		 */
1661 		if (vp->v_flag & VXLOCK) {
1662 			vp->v_flag |= VXWANT;
1663 			simple_unlock(&vp->v_interlock);
1664 			tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1665 			return (0);
1666 		}
1667 		/*
1668 		 * Ensure that vp will not be vgone'd while we
1669 		 * are eliminating its aliases.
1670 		 */
1671 		vp->v_flag |= VXLOCK;
1672 		simple_unlock(&vp->v_interlock);
1673 		while (vp->v_flag & VALIASED) {
1674 			simple_lock(&spechash_slock);
1675 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1676 				if (vq->v_rdev != vp->v_rdev ||
1677 				    vq->v_type != vp->v_type || vp == vq)
1678 					continue;
1679 				simple_unlock(&spechash_slock);
1680 				vgone(vq);
1681 				break;
1682 			}
1683 			if (vq == NULLVP) {
1684 				simple_unlock(&spechash_slock);
1685 			}
1686 		}
1687 		/*
1688 		 * Remove the lock so that vgone below will
1689 		 * really eliminate the vnode after which time
1690 		 * vgone will awaken any sleepers.
1691 		 */
1692 		simple_lock(&vp->v_interlock);
1693 		vp->v_flag &= ~VXLOCK;
1694 		if (vp->v_flag & VXWANT) {
1695 			vp->v_flag &= ~VXWANT;
1696 			wakeup(vp);
1697 		}
1698 	}
1699 	vgonel(vp, p);
1700 	return (0);
1701 }
1702 
1703 /*
1704  * Recycle an unused vnode to the front of the free list.
1705  * Release the passed interlock if the vnode will be recycled.
1706  */
1707 int
1708 vrecycle(vp, inter_lkp, p)
1709 	struct vnode *vp;
1710 	struct simplelock *inter_lkp;
1711 	struct proc *p;
1712 {
1713 
1714 	simple_lock(&vp->v_interlock);
1715 	if (vp->v_usecount == 0) {
1716 		if (inter_lkp) {
1717 			simple_unlock(inter_lkp);
1718 		}
1719 		vgonel(vp, p);
1720 		return (1);
1721 	}
1722 	simple_unlock(&vp->v_interlock);
1723 	return (0);
1724 }
1725 
1726 /*
1727  * Eliminate all activity associated with a vnode
1728  * in preparation for reuse.
1729  */
1730 void
1731 vgone(vp)
1732 	register struct vnode *vp;
1733 {
1734 	struct proc *p = curproc;	/* XXX */
1735 
1736 	simple_lock(&vp->v_interlock);
1737 	vgonel(vp, p);
1738 }
1739 
1740 /*
1741  * vgone, with the vp interlock held.
1742  */
1743 static void
1744 vgonel(vp, p)
1745 	struct vnode *vp;
1746 	struct proc *p;
1747 {
1748 	int s;
1749 	struct vnode *vq;
1750 	struct vnode *vx;
1751 
1752 	/*
1753 	 * If a vgone (or vclean) is already in progress,
1754 	 * wait until it is done and return.
1755 	 */
1756 	if (vp->v_flag & VXLOCK) {
1757 		vp->v_flag |= VXWANT;
1758 		simple_unlock(&vp->v_interlock);
1759 		tsleep((caddr_t)vp, PINOD, "vgone", 0);
1760 		return;
1761 	}
1762 
1763 	/*
1764 	 * Clean out the filesystem specific data.
1765 	 */
1766 	vclean(vp, DOCLOSE, p);
1767 	simple_lock(&vp->v_interlock);
1768 
1769 	/*
1770 	 * Delete from old mount point vnode list, if on one.
1771 	 */
1772 	if (vp->v_mount != NULL)
1773 		insmntque(vp, (struct mount *)0);
1774 	/*
1775 	 * If special device, remove it from special device alias list
1776 	 * if it is on one.
1777 	 */
1778 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1779 		simple_lock(&spechash_slock);
1780 		if (*vp->v_hashchain == vp) {
1781 			*vp->v_hashchain = vp->v_specnext;
1782 		} else {
1783 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1784 				if (vq->v_specnext != vp)
1785 					continue;
1786 				vq->v_specnext = vp->v_specnext;
1787 				break;
1788 			}
1789 			if (vq == NULL)
1790 				panic("missing bdev");
1791 		}
1792 		if (vp->v_flag & VALIASED) {
1793 			vx = NULL;
1794 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1795 				if (vq->v_rdev != vp->v_rdev ||
1796 				    vq->v_type != vp->v_type)
1797 					continue;
1798 				if (vx)
1799 					break;
1800 				vx = vq;
1801 			}
1802 			if (vx == NULL)
1803 				panic("missing alias");
1804 			if (vq == NULL)
1805 				vx->v_flag &= ~VALIASED;
1806 			vp->v_flag &= ~VALIASED;
1807 		}
1808 		simple_unlock(&spechash_slock);
1809 		FREE(vp->v_specinfo, M_VNODE);
1810 		vp->v_specinfo = NULL;
1811 	}
1812 
1813 	/*
1814 	 * If it is on the freelist and not already at the head,
1815 	 * move it to the head of the list. The test of the back
1816 	 * pointer and the reference count of zero is because
1817 	 * it will be removed from the free list by getnewvnode,
1818 	 * but will not have its reference count incremented until
1819 	 * after calling vgone. If the reference count were
1820 	 * incremented first, vgone would (incorrectly) try to
1821 	 * close the previous instance of the underlying object.
1822 	 */
1823 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1824 		s = splbio();
1825 		simple_lock(&vnode_free_list_slock);
1826 		if (vp->v_flag & VFREE) {
1827 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1828 		} else if (vp->v_flag & VTBFREE) {
1829 			TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
1830 			vp->v_flag &= ~VTBFREE;
1831 			freevnodes++;
1832 		} else
1833 			freevnodes++;
1834 		vp->v_flag |= VFREE;
1835 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1836 		simple_unlock(&vnode_free_list_slock);
1837 		splx(s);
1838 	}
1839 
1840 	vp->v_type = VBAD;
1841 	simple_unlock(&vp->v_interlock);
1842 }
1843 
1844 /*
1845  * Lookup a vnode by device number.
1846  */
1847 int
1848 vfinddev(dev, type, vpp)
1849 	dev_t dev;
1850 	enum vtype type;
1851 	struct vnode **vpp;
1852 {
1853 	register struct vnode *vp;
1854 	int rc = 0;
1855 
1856 	simple_lock(&spechash_slock);
1857 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1858 		if (dev != vp->v_rdev || type != vp->v_type)
1859 			continue;
1860 		*vpp = vp;
1861 		rc = 1;
1862 		break;
1863 	}
1864 	simple_unlock(&spechash_slock);
1865 	return (rc);
1866 }
1867 
1868 /*
1869  * Calculate the total number of references to a special device.
1870  */
1871 int
1872 vcount(vp)
1873 	register struct vnode *vp;
1874 {
1875 	struct vnode *vq, *vnext;
1876 	int count;
1877 
1878 loop:
1879 	if ((vp->v_flag & VALIASED) == 0)
1880 		return (vp->v_usecount);
1881 	simple_lock(&spechash_slock);
1882 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1883 		vnext = vq->v_specnext;
1884 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1885 			continue;
1886 		/*
1887 		 * Alias, but not in use, so flush it out.
1888 		 */
1889 		if (vq->v_usecount == 0 && vq != vp) {
1890 			simple_unlock(&spechash_slock);
1891 			vgone(vq);
1892 			goto loop;
1893 		}
1894 		count += vq->v_usecount;
1895 	}
1896 	simple_unlock(&spechash_slock);
1897 	return (count);
1898 }
1899 /*
1900  * Print out a description of a vnode.
1901  */
1902 static char *typename[] =
1903 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1904 
1905 void
1906 vprint(label, vp)
1907 	char *label;
1908 	register struct vnode *vp;
1909 {
1910 	char buf[64];
1911 
1912 	if (label != NULL)
1913 		printf("%s: %p: ", label, (void *)vp);
1914 	else
1915 		printf("%p: ", (void *)vp);
1916 	printf("type %s, usecount %d, writecount %d, refcount %d,",
1917 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1918 	    vp->v_holdcnt);
1919 	buf[0] = '\0';
1920 	if (vp->v_flag & VROOT)
1921 		strcat(buf, "|VROOT");
1922 	if (vp->v_flag & VTEXT)
1923 		strcat(buf, "|VTEXT");
1924 	if (vp->v_flag & VSYSTEM)
1925 		strcat(buf, "|VSYSTEM");
1926 	if (vp->v_flag & VXLOCK)
1927 		strcat(buf, "|VXLOCK");
1928 	if (vp->v_flag & VXWANT)
1929 		strcat(buf, "|VXWANT");
1930 	if (vp->v_flag & VBWAIT)
1931 		strcat(buf, "|VBWAIT");
1932 	if (vp->v_flag & VALIASED)
1933 		strcat(buf, "|VALIASED");
1934 	if (vp->v_flag & VDOOMED)
1935 		strcat(buf, "|VDOOMED");
1936 	if (vp->v_flag & VFREE)
1937 		strcat(buf, "|VFREE");
1938 	if (vp->v_flag & VOBJBUF)
1939 		strcat(buf, "|VOBJBUF");
1940 	if (buf[0] != '\0')
1941 		printf(" flags (%s)", &buf[1]);
1942 	if (vp->v_data == NULL) {
1943 		printf("\n");
1944 	} else {
1945 		printf("\n\t");
1946 		VOP_PRINT(vp);
1947 	}
1948 }
1949 
1950 #ifdef DDB
1951 /*
1952  * List all of the locked vnodes in the system.
1953  * Called when debugging the kernel.
1954  */
1955 static void
1956 printlockedvnodes()
1957 {
1958 	struct proc *p = curproc;	/* XXX */
1959 	struct mount *mp, *nmp;
1960 	struct vnode *vp;
1961 
1962 	printf("Locked vnodes\n");
1963 	simple_lock(&mountlist_slock);
1964 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
1965 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
1966 			nmp = mp->mnt_list.cqe_next;
1967 			continue;
1968 		}
1969 		for (vp = mp->mnt_vnodelist.lh_first;
1970 		     vp != NULL;
1971 		     vp = vp->v_mntvnodes.le_next) {
1972 			if (VOP_ISLOCKED(vp))
1973 				vprint((char *)0, vp);
1974 		}
1975 		simple_lock(&mountlist_slock);
1976 		nmp = mp->mnt_list.cqe_next;
1977 		vfs_unbusy(mp, p);
1978 	}
1979 	simple_unlock(&mountlist_slock);
1980 }
1981 #endif
1982 
1983 /*
1984  * Top level filesystem related information gathering.
1985  */
1986 static int	sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS);
1987 
1988 static int
1989 vfs_sysctl SYSCTL_HANDLER_ARGS
1990 {
1991 	int *name = (int *)arg1 - 1;	/* XXX */
1992 	u_int namelen = arg2 + 1;	/* XXX */
1993 	struct vfsconf *vfsp;
1994 
1995 #ifndef NO_COMPAT_PRELITE2
1996 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1997 	if (namelen == 1)
1998 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1999 #endif
2000 
2001 #ifdef notyet
2002 	/* all sysctl names at this level are at least name and field */
2003 	if (namelen < 2)
2004 		return (ENOTDIR);		/* overloaded */
2005 	if (name[0] != VFS_GENERIC) {
2006 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2007 			if (vfsp->vfc_typenum == name[0])
2008 				break;
2009 		if (vfsp == NULL)
2010 			return (EOPNOTSUPP);
2011 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2012 		    oldp, oldlenp, newp, newlen, p));
2013 	}
2014 #endif
2015 	switch (name[1]) {
2016 	case VFS_MAXTYPENUM:
2017 		if (namelen != 2)
2018 			return (ENOTDIR);
2019 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2020 	case VFS_CONF:
2021 		if (namelen != 3)
2022 			return (ENOTDIR);	/* overloaded */
2023 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2024 			if (vfsp->vfc_typenum == name[2])
2025 				break;
2026 		if (vfsp == NULL)
2027 			return (EOPNOTSUPP);
2028 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2029 	}
2030 	return (EOPNOTSUPP);
2031 }
2032 
2033 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2034 	"Generic filesystem");
2035 
2036 #ifndef NO_COMPAT_PRELITE2
2037 
2038 static int
2039 sysctl_ovfs_conf SYSCTL_HANDLER_ARGS
2040 {
2041 	int error;
2042 	struct vfsconf *vfsp;
2043 	struct ovfsconf ovfs;
2044 
2045 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2046 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2047 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2048 		ovfs.vfc_index = vfsp->vfc_typenum;
2049 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2050 		ovfs.vfc_flags = vfsp->vfc_flags;
2051 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2052 		if (error)
2053 			return error;
2054 	}
2055 	return 0;
2056 }
2057 
2058 #endif /* !NO_COMPAT_PRELITE2 */
2059 
2060 static volatile int kinfo_vdebug = 1;
2061 
2062 #if 0
2063 #define KINFO_VNODESLOP	10
2064 /*
2065  * Dump vnode list (via sysctl).
2066  * Copyout address of vnode followed by vnode.
2067  */
2068 /* ARGSUSED */
2069 static int
2070 sysctl_vnode SYSCTL_HANDLER_ARGS
2071 {
2072 	struct proc *p = curproc;	/* XXX */
2073 	struct mount *mp, *nmp;
2074 	struct vnode *nvp, *vp;
2075 	int error;
2076 
2077 #define VPTRSZ	sizeof (struct vnode *)
2078 #define VNODESZ	sizeof (struct vnode)
2079 
2080 	req->lock = 0;
2081 	if (!req->oldptr) /* Make an estimate */
2082 		return (SYSCTL_OUT(req, 0,
2083 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2084 
2085 	simple_lock(&mountlist_slock);
2086 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
2087 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2088 			nmp = mp->mnt_list.cqe_next;
2089 			continue;
2090 		}
2091 again:
2092 		simple_lock(&mntvnode_slock);
2093 		for (vp = mp->mnt_vnodelist.lh_first;
2094 		     vp != NULL;
2095 		     vp = nvp) {
2096 			/*
2097 			 * Check that the vp is still associated with
2098 			 * this filesystem.  RACE: could have been
2099 			 * recycled onto the same filesystem.
2100 			 */
2101 			if (vp->v_mount != mp) {
2102 				simple_unlock(&mntvnode_slock);
2103 				if (kinfo_vdebug)
2104 					printf("kinfo: vp changed\n");
2105 				goto again;
2106 			}
2107 			nvp = vp->v_mntvnodes.le_next;
2108 			simple_unlock(&mntvnode_slock);
2109 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2110 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2111 				return (error);
2112 			simple_lock(&mntvnode_slock);
2113 		}
2114 		simple_unlock(&mntvnode_slock);
2115 		simple_lock(&mountlist_slock);
2116 		nmp = mp->mnt_list.cqe_next;
2117 		vfs_unbusy(mp, p);
2118 	}
2119 	simple_unlock(&mountlist_slock);
2120 
2121 	return (0);
2122 }
2123 #endif
2124 
2125 /*
2126  * XXX
2127  * Exporting the vnode list on large systems causes them to crash.
2128  * Exporting the vnode list on medium systems causes sysctl to coredump.
2129  */
2130 #if 0
2131 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2132 	0, 0, sysctl_vnode, "S,vnode", "");
2133 #endif
2134 
2135 /*
2136  * Check to see if a filesystem is mounted on a block device.
2137  */
2138 int
2139 vfs_mountedon(vp)
2140 	struct vnode *vp;
2141 {
2142 	struct vnode *vq;
2143 	int error = 0;
2144 
2145 	if (vp->v_specmountpoint != NULL)
2146 		return (EBUSY);
2147 	if (vp->v_flag & VALIASED) {
2148 		simple_lock(&spechash_slock);
2149 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2150 			if (vq->v_rdev != vp->v_rdev ||
2151 			    vq->v_type != vp->v_type)
2152 				continue;
2153 			if (vq->v_specmountpoint != NULL) {
2154 				error = EBUSY;
2155 				break;
2156 			}
2157 		}
2158 		simple_unlock(&spechash_slock);
2159 	}
2160 	return (error);
2161 }
2162 
2163 /*
2164  * Unmount all filesystems. The list is traversed in reverse order
2165  * of mounting to avoid dependencies.
2166  */
2167 void
2168 vfs_unmountall()
2169 {
2170 	struct mount *mp, *nmp;
2171 	struct proc *p;
2172 	int error;
2173 
2174 	if (curproc != NULL)
2175 		p = curproc;
2176 	else
2177 		p = initproc;	/* XXX XXX should this be proc0? */
2178 	/*
2179 	 * Since this only runs when rebooting, it is not interlocked.
2180 	 */
2181 	for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2182 		nmp = mp->mnt_list.cqe_prev;
2183 		error = dounmount(mp, MNT_FORCE, p);
2184 		if (error) {
2185 			printf("unmount of %s failed (",
2186 			    mp->mnt_stat.f_mntonname);
2187 			if (error == EBUSY)
2188 				printf("BUSY)\n");
2189 			else
2190 				printf("%d)\n", error);
2191 		}
2192 	}
2193 }
2194 
2195 /*
2196  * Build hash lists of net addresses and hang them off the mount point.
2197  * Called by ufs_mount() to set up the lists of export addresses.
2198  */
2199 static int
2200 vfs_hang_addrlist(mp, nep, argp)
2201 	struct mount *mp;
2202 	struct netexport *nep;
2203 	struct export_args *argp;
2204 {
2205 	register struct netcred *np;
2206 	register struct radix_node_head *rnh;
2207 	register int i;
2208 	struct radix_node *rn;
2209 	struct sockaddr *saddr, *smask = 0;
2210 	struct domain *dom;
2211 	int error;
2212 
2213 	if (argp->ex_addrlen == 0) {
2214 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2215 			return (EPERM);
2216 		np = &nep->ne_defexported;
2217 		np->netc_exflags = argp->ex_flags;
2218 		np->netc_anon = argp->ex_anon;
2219 		np->netc_anon.cr_ref = 1;
2220 		mp->mnt_flag |= MNT_DEFEXPORTED;
2221 		return (0);
2222 	}
2223 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2224 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2225 	bzero((caddr_t) np, i);
2226 	saddr = (struct sockaddr *) (np + 1);
2227 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2228 		goto out;
2229 	if (saddr->sa_len > argp->ex_addrlen)
2230 		saddr->sa_len = argp->ex_addrlen;
2231 	if (argp->ex_masklen) {
2232 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2233 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2234 		if (error)
2235 			goto out;
2236 		if (smask->sa_len > argp->ex_masklen)
2237 			smask->sa_len = argp->ex_masklen;
2238 	}
2239 	i = saddr->sa_family;
2240 	if ((rnh = nep->ne_rtable[i]) == 0) {
2241 		/*
2242 		 * Seems silly to initialize every AF when most are not used,
2243 		 * do so on demand here
2244 		 */
2245 		for (dom = domains; dom; dom = dom->dom_next)
2246 			if (dom->dom_family == i && dom->dom_rtattach) {
2247 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2248 				    dom->dom_rtoffset);
2249 				break;
2250 			}
2251 		if ((rnh = nep->ne_rtable[i]) == 0) {
2252 			error = ENOBUFS;
2253 			goto out;
2254 		}
2255 	}
2256 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2257 	    np->netc_rnodes);
2258 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2259 		error = EPERM;
2260 		goto out;
2261 	}
2262 	np->netc_exflags = argp->ex_flags;
2263 	np->netc_anon = argp->ex_anon;
2264 	np->netc_anon.cr_ref = 1;
2265 	return (0);
2266 out:
2267 	free(np, M_NETADDR);
2268 	return (error);
2269 }
2270 
2271 /* ARGSUSED */
2272 static int
2273 vfs_free_netcred(rn, w)
2274 	struct radix_node *rn;
2275 	void *w;
2276 {
2277 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
2278 
2279 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2280 	free((caddr_t) rn, M_NETADDR);
2281 	return (0);
2282 }
2283 
2284 /*
2285  * Free the net address hash lists that are hanging off the mount points.
2286  */
2287 static void
2288 vfs_free_addrlist(nep)
2289 	struct netexport *nep;
2290 {
2291 	register int i;
2292 	register struct radix_node_head *rnh;
2293 
2294 	for (i = 0; i <= AF_MAX; i++)
2295 		if ((rnh = nep->ne_rtable[i])) {
2296 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2297 			    (caddr_t) rnh);
2298 			free((caddr_t) rnh, M_RTABLE);
2299 			nep->ne_rtable[i] = 0;
2300 		}
2301 }
2302 
2303 int
2304 vfs_export(mp, nep, argp)
2305 	struct mount *mp;
2306 	struct netexport *nep;
2307 	struct export_args *argp;
2308 {
2309 	int error;
2310 
2311 	if (argp->ex_flags & MNT_DELEXPORT) {
2312 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2313 			vfs_setpublicfs(NULL, NULL, NULL);
2314 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2315 		}
2316 		vfs_free_addrlist(nep);
2317 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2318 	}
2319 	if (argp->ex_flags & MNT_EXPORTED) {
2320 		if (argp->ex_flags & MNT_EXPUBLIC) {
2321 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2322 				return (error);
2323 			mp->mnt_flag |= MNT_EXPUBLIC;
2324 		}
2325 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2326 			return (error);
2327 		mp->mnt_flag |= MNT_EXPORTED;
2328 	}
2329 	return (0);
2330 }
2331 
2332 
2333 /*
2334  * Set the publicly exported filesystem (WebNFS). Currently, only
2335  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2336  */
2337 int
2338 vfs_setpublicfs(mp, nep, argp)
2339 	struct mount *mp;
2340 	struct netexport *nep;
2341 	struct export_args *argp;
2342 {
2343 	int error;
2344 	struct vnode *rvp;
2345 	char *cp;
2346 
2347 	/*
2348 	 * mp == NULL -> invalidate the current info, the FS is
2349 	 * no longer exported. May be called from either vfs_export
2350 	 * or unmount, so check if it hasn't already been done.
2351 	 */
2352 	if (mp == NULL) {
2353 		if (nfs_pub.np_valid) {
2354 			nfs_pub.np_valid = 0;
2355 			if (nfs_pub.np_index != NULL) {
2356 				FREE(nfs_pub.np_index, M_TEMP);
2357 				nfs_pub.np_index = NULL;
2358 			}
2359 		}
2360 		return (0);
2361 	}
2362 
2363 	/*
2364 	 * Only one allowed at a time.
2365 	 */
2366 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2367 		return (EBUSY);
2368 
2369 	/*
2370 	 * Get real filehandle for root of exported FS.
2371 	 */
2372 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2373 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2374 
2375 	if ((error = VFS_ROOT(mp, &rvp)))
2376 		return (error);
2377 
2378 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2379 		return (error);
2380 
2381 	vput(rvp);
2382 
2383 	/*
2384 	 * If an indexfile was specified, pull it in.
2385 	 */
2386 	if (argp->ex_indexfile != NULL) {
2387 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2388 		    M_WAITOK);
2389 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2390 		    MAXNAMLEN, (size_t *)0);
2391 		if (!error) {
2392 			/*
2393 			 * Check for illegal filenames.
2394 			 */
2395 			for (cp = nfs_pub.np_index; *cp; cp++) {
2396 				if (*cp == '/') {
2397 					error = EINVAL;
2398 					break;
2399 				}
2400 			}
2401 		}
2402 		if (error) {
2403 			FREE(nfs_pub.np_index, M_TEMP);
2404 			return (error);
2405 		}
2406 	}
2407 
2408 	nfs_pub.np_mount = mp;
2409 	nfs_pub.np_valid = 1;
2410 	return (0);
2411 }
2412 
2413 struct netcred *
2414 vfs_export_lookup(mp, nep, nam)
2415 	register struct mount *mp;
2416 	struct netexport *nep;
2417 	struct sockaddr *nam;
2418 {
2419 	register struct netcred *np;
2420 	register struct radix_node_head *rnh;
2421 	struct sockaddr *saddr;
2422 
2423 	np = NULL;
2424 	if (mp->mnt_flag & MNT_EXPORTED) {
2425 		/*
2426 		 * Lookup in the export list first.
2427 		 */
2428 		if (nam != NULL) {
2429 			saddr = nam;
2430 			rnh = nep->ne_rtable[saddr->sa_family];
2431 			if (rnh != NULL) {
2432 				np = (struct netcred *)
2433 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2434 							      rnh);
2435 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2436 					np = NULL;
2437 			}
2438 		}
2439 		/*
2440 		 * If no address match, use the default if it exists.
2441 		 */
2442 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2443 			np = &nep->ne_defexported;
2444 	}
2445 	return (np);
2446 }
2447 
2448 /*
2449  * perform msync on all vnodes under a mount point
2450  * the mount point must be locked.
2451  */
2452 void
2453 vfs_msync(struct mount *mp, int flags) {
2454 	struct vnode *vp, *nvp;
2455 	struct vm_object *obj;
2456 	int anyio, tries;
2457 
2458 	tries = 5;
2459 loop:
2460 	anyio = 0;
2461 	for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
2462 
2463 		nvp = vp->v_mntvnodes.le_next;
2464 
2465 		if (vp->v_mount != mp) {
2466 			goto loop;
2467 		}
2468 
2469 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2470 			continue;
2471 
2472 		if (flags != MNT_WAIT) {
2473 			obj = vp->v_object;
2474 			if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2475 				continue;
2476 			if (VOP_ISLOCKED(vp))
2477 				continue;
2478 		}
2479 
2480 		simple_lock(&vp->v_interlock);
2481 		if (vp->v_object &&
2482 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
2483 			if (!vget(vp,
2484 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2485 				if (vp->v_object) {
2486 					vm_object_page_clean(vp->v_object, 0, 0, TRUE);
2487 					anyio = 1;
2488 				}
2489 				vput(vp);
2490 			}
2491 		} else {
2492 			simple_unlock(&vp->v_interlock);
2493 		}
2494 	}
2495 	if (anyio && (--tries > 0))
2496 		goto loop;
2497 }
2498 
2499 /*
2500  * Create the VM object needed for VMIO and mmap support.  This
2501  * is done for all VREG files in the system.  Some filesystems might
2502  * afford the additional metadata buffering capability of the
2503  * VMIO code by making the device node be VMIO mode also.
2504  *
2505  * If !waslocked, must be called with interlock.
2506  */
2507 int
2508 vfs_object_create(vp, p, cred, waslocked)
2509 	struct vnode *vp;
2510 	struct proc *p;
2511 	struct ucred *cred;
2512 	int waslocked;
2513 {
2514 	struct vattr vat;
2515 	vm_object_t object;
2516 	int error = 0;
2517 
2518 	if ((vp->v_type != VREG) && (vp->v_type != VBLK)) {
2519 		if (!waslocked)
2520 			simple_unlock(&vp->v_interlock);
2521 		return 0;
2522 	}
2523 
2524 	if (!waslocked)
2525 		vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY, p);
2526 
2527 retry:
2528 	if ((object = vp->v_object) == NULL) {
2529 		if (vp->v_type == VREG) {
2530 			if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
2531 				goto retn;
2532 			object = vnode_pager_alloc(vp,
2533 				OFF_TO_IDX(round_page(vat.va_size)), 0, 0);
2534 		} else if (major(vp->v_rdev) < nblkdev) {
2535 			/*
2536 			 * This simply allocates the biggest object possible
2537 			 * for a VBLK vnode.  This should be fixed, but doesn't
2538 			 * cause any problems (yet).
2539 			 */
2540 			object = vnode_pager_alloc(vp, INT_MAX, 0, 0);
2541 		}
2542 		object->ref_count--;
2543 		vp->v_usecount--;
2544 	} else {
2545 		if (object->flags & OBJ_DEAD) {
2546 			VOP_UNLOCK(vp, 0, p);
2547 			tsleep(object, PVM, "vodead", 0);
2548 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
2549 			goto retry;
2550 		}
2551 	}
2552 
2553 	if (vp->v_object) {
2554 		vp->v_flag |= VOBJBUF;
2555 	}
2556 
2557 retn:
2558 	if (!waslocked) {
2559 		simple_lock(&vp->v_interlock);
2560 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
2561 	}
2562 
2563 	return error;
2564 }
2565 
2566 static void
2567 vfree(vp)
2568 	struct vnode *vp;
2569 {
2570 	int s;
2571 
2572 	s = splbio();
2573 	simple_lock(&vnode_free_list_slock);
2574 	if (vp->v_flag & VTBFREE) {
2575 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2576 		vp->v_flag &= ~VTBFREE;
2577 	}
2578 	if (vp->v_flag & VAGE) {
2579 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2580 	} else {
2581 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2582 	}
2583 	freevnodes++;
2584 	simple_unlock(&vnode_free_list_slock);
2585 	vp->v_flag &= ~VAGE;
2586 	vp->v_flag |= VFREE;
2587 	splx(s);
2588 }
2589 
2590 void
2591 vbusy(vp)
2592 	struct vnode *vp;
2593 {
2594 	int s;
2595 
2596 	s = splbio();
2597 	simple_lock(&vnode_free_list_slock);
2598 	if (vp->v_flag & VTBFREE) {
2599 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2600 		vp->v_flag &= ~VTBFREE;
2601 	} else {
2602 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2603 		freevnodes--;
2604 	}
2605 	simple_unlock(&vnode_free_list_slock);
2606 	vp->v_flag &= ~(VFREE|VAGE);
2607 	splx(s);
2608 }
2609 
2610 /*
2611  * Record a process's interest in events which might happen to
2612  * a vnode.  Because poll uses the historic select-style interface
2613  * internally, this routine serves as both the ``check for any
2614  * pending events'' and the ``record my interest in future events''
2615  * functions.  (These are done together, while the lock is held,
2616  * to avoid race conditions.)
2617  */
2618 int
2619 vn_pollrecord(vp, p, events)
2620 	struct vnode *vp;
2621 	struct proc *p;
2622 	short events;
2623 {
2624 	simple_lock(&vp->v_pollinfo.vpi_lock);
2625 	if (vp->v_pollinfo.vpi_revents & events) {
2626 		/*
2627 		 * This leaves events we are not interested
2628 		 * in available for the other process which
2629 		 * which presumably had requested them
2630 		 * (otherwise they would never have been
2631 		 * recorded).
2632 		 */
2633 		events &= vp->v_pollinfo.vpi_revents;
2634 		vp->v_pollinfo.vpi_revents &= ~events;
2635 
2636 		simple_unlock(&vp->v_pollinfo.vpi_lock);
2637 		return events;
2638 	}
2639 	vp->v_pollinfo.vpi_events |= events;
2640 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2641 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2642 	return 0;
2643 }
2644 
2645 /*
2646  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2647  * it is possible for us to miss an event due to race conditions, but
2648  * that condition is expected to be rare, so for the moment it is the
2649  * preferred interface.
2650  */
2651 void
2652 vn_pollevent(vp, events)
2653 	struct vnode *vp;
2654 	short events;
2655 {
2656 	simple_lock(&vp->v_pollinfo.vpi_lock);
2657 	if (vp->v_pollinfo.vpi_events & events) {
2658 		/*
2659 		 * We clear vpi_events so that we don't
2660 		 * call selwakeup() twice if two events are
2661 		 * posted before the polling process(es) is
2662 		 * awakened.  This also ensures that we take at
2663 		 * most one selwakeup() if the polling process
2664 		 * is no longer interested.  However, it does
2665 		 * mean that only one event can be noticed at
2666 		 * a time.  (Perhaps we should only clear those
2667 		 * event bits which we note?) XXX
2668 		 */
2669 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2670 		vp->v_pollinfo.vpi_revents |= events;
2671 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2672 	}
2673 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2674 }
2675 
2676 /*
2677  * Wake up anyone polling on vp because it is being revoked.
2678  * This depends on dead_poll() returning POLLHUP for correct
2679  * behavior.
2680  */
2681 void
2682 vn_pollgone(vp)
2683 	struct vnode *vp;
2684 {
2685 	simple_lock(&vp->v_pollinfo.vpi_lock);
2686 	if (vp->v_pollinfo.vpi_events) {
2687 		vp->v_pollinfo.vpi_events = 0;
2688 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2689 	}
2690 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2691 }
2692 
2693 
2694 
2695 /*
2696  * Routine to create and manage a filesystem syncer vnode.
2697  */
2698 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2699 int	sync_fsync __P((struct  vop_fsync_args *));
2700 int	sync_inactive __P((struct  vop_inactive_args *));
2701 int	sync_reclaim  __P((struct  vop_reclaim_args *));
2702 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2703 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2704 int	sync_print __P((struct vop_print_args *));
2705 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2706 
2707 vop_t **sync_vnodeop_p;
2708 struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2709 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2710 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2711 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2712 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2713 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2714 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2715 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2716 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2717 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2718 	{ NULL, NULL }
2719 };
2720 struct vnodeopv_desc sync_vnodeop_opv_desc =
2721 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2722 
2723 VNODEOP_SET(sync_vnodeop_opv_desc);
2724 
2725 /*
2726  * Create a new filesystem syncer vnode for the specified mount point.
2727  */
2728 int
2729 vfs_allocate_syncvnode(mp)
2730 	struct mount *mp;
2731 {
2732 	struct vnode *vp;
2733 	static long start, incr, next;
2734 	int error;
2735 
2736 	/* Allocate a new vnode */
2737 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2738 		mp->mnt_syncer = NULL;
2739 		return (error);
2740 	}
2741 	vp->v_type = VNON;
2742 	/*
2743 	 * Place the vnode onto the syncer worklist. We attempt to
2744 	 * scatter them about on the list so that they will go off
2745 	 * at evenly distributed times even if all the filesystems
2746 	 * are mounted at once.
2747 	 */
2748 	next += incr;
2749 	if (next == 0 || next > syncer_maxdelay) {
2750 		start /= 2;
2751 		incr /= 2;
2752 		if (start == 0) {
2753 			start = syncer_maxdelay / 2;
2754 			incr = syncer_maxdelay;
2755 		}
2756 		next = start;
2757 	}
2758 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2759 	mp->mnt_syncer = vp;
2760 	return (0);
2761 }
2762 
2763 /*
2764  * Do a lazy sync of the filesystem.
2765  */
2766 int
2767 sync_fsync(ap)
2768 	struct vop_fsync_args /* {
2769 		struct vnode *a_vp;
2770 		struct ucred *a_cred;
2771 		int a_waitfor;
2772 		struct proc *a_p;
2773 	} */ *ap;
2774 {
2775 	struct vnode *syncvp = ap->a_vp;
2776 	struct mount *mp = syncvp->v_mount;
2777 	struct proc *p = ap->a_p;
2778 	int asyncflag;
2779 
2780 	/*
2781 	 * We only need to do something if this is a lazy evaluation.
2782 	 */
2783 	if (ap->a_waitfor != MNT_LAZY)
2784 		return (0);
2785 
2786 	/*
2787 	 * Move ourselves to the back of the sync list.
2788 	 */
2789 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2790 
2791 	/*
2792 	 * Walk the list of vnodes pushing all that are dirty and
2793 	 * not already on the sync list.
2794 	 */
2795 	simple_lock(&mountlist_slock);
2796 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) {
2797 		simple_unlock(&mountlist_slock);
2798 		return (0);
2799 	}
2800 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2801 	mp->mnt_flag &= ~MNT_ASYNC;
2802 	vfs_msync(mp, MNT_NOWAIT);
2803 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2804 	if (asyncflag)
2805 		mp->mnt_flag |= MNT_ASYNC;
2806 	vfs_unbusy(mp, p);
2807 	return (0);
2808 }
2809 
2810 /*
2811  * The syncer vnode is no referenced.
2812  */
2813 int
2814 sync_inactive(ap)
2815 	struct vop_inactive_args /* {
2816 		struct vnode *a_vp;
2817 		struct proc *a_p;
2818 	} */ *ap;
2819 {
2820 
2821 	vgone(ap->a_vp);
2822 	return (0);
2823 }
2824 
2825 /*
2826  * The syncer vnode is no longer needed and is being decommissioned.
2827  */
2828 int
2829 sync_reclaim(ap)
2830 	struct vop_reclaim_args /* {
2831 		struct vnode *a_vp;
2832 	} */ *ap;
2833 {
2834 	struct vnode *vp = ap->a_vp;
2835 
2836 	vp->v_mount->mnt_syncer = NULL;
2837 	if (vp->v_flag & VONWORKLST) {
2838 		LIST_REMOVE(vp, v_synclist);
2839 		vp->v_flag &= ~VONWORKLST;
2840 	}
2841 
2842 	return (0);
2843 }
2844 
2845 /*
2846  * Print out a syncer vnode.
2847  */
2848 int
2849 sync_print(ap)
2850 	struct vop_print_args /* {
2851 		struct vnode *a_vp;
2852 	} */ *ap;
2853 {
2854 	struct vnode *vp = ap->a_vp;
2855 
2856 	printf("syncer vnode");
2857 	if (vp->v_vnlock != NULL)
2858 		lockmgr_printinfo(vp->v_vnlock);
2859 	printf("\n");
2860 	return (0);
2861 }
2862