xref: /freebsd/sys/kern/vfs_export.c (revision 2ad872c5794e4c26fdf6ed219ad3f09ca0d5304a)
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 1.181 1999/01/08 17:31:17 eivind 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/conf.h>
50 #include <sys/fcntl.h>
51 #include <sys/kernel.h>
52 #include <sys/proc.h>
53 #include <sys/malloc.h>
54 #include <sys/mount.h>
55 #include <sys/socket.h>
56 #include <sys/vnode.h>
57 #include <sys/stat.h>
58 #include <sys/buf.h>
59 #include <sys/domain.h>
60 #include <sys/dirent.h>
61 #include <sys/vmmeter.h>
62 
63 #include <machine/limits.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_extern.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_pager.h>
71 #include <vm/vnode_pager.h>
72 #include <vm/vm_zone.h>
73 #include <sys/sysctl.h>
74 
75 #include <miscfs/specfs/specdev.h>
76 
77 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
78 
79 static void	insmntque __P((struct vnode *vp, struct mount *mp));
80 static void	vclean __P((struct vnode *vp, int flags, struct proc *p));
81 static void	vfree __P((struct vnode *));
82 static void	vgonel __P((struct vnode *vp, struct proc *p));
83 static unsigned long	numvnodes;
84 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
85 
86 enum vtype iftovt_tab[16] = {
87 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
88 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
89 };
90 int vttoif_tab[9] = {
91 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
92 	S_IFSOCK, S_IFIFO, S_IFMT,
93 };
94 
95 static TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
96 struct tobefreelist vnode_tobefree_list;	/* vnode free list */
97 
98 static u_long wantfreevnodes = 25;
99 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
100 static u_long freevnodes = 0;
101 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
102 
103 int vfs_ioopt = 0;
104 #ifdef ENABLE_VFS_IOOPT
105 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
106 #endif
107 
108 struct mntlist mountlist;	/* mounted filesystem list */
109 struct simplelock mountlist_slock;
110 struct simplelock mntvnode_slock;
111 int	nfs_mount_type = -1;
112 #ifndef NULL_SIMPLELOCKS
113 static struct simplelock mntid_slock;
114 static struct simplelock vnode_free_list_slock;
115 static struct simplelock spechash_slock;
116 #endif
117 struct nfs_public nfs_pub;	/* publicly exported FS */
118 static vm_zone_t vnode_zone;
119 
120 /*
121  * The workitem queue.
122  */
123 #define SYNCER_MAXDELAY		32
124 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
125 time_t syncdelay =		30;
126 int rushjob;				/* number of slots to run ASAP */
127 
128 static int syncer_delayno = 0;
129 static long syncer_mask;
130 LIST_HEAD(synclist, vnode);
131 static struct synclist *syncer_workitem_pending;
132 
133 int desiredvnodes;
134 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, "");
135 
136 static void	vfs_free_addrlist __P((struct netexport *nep));
137 static int	vfs_free_netcred __P((struct radix_node *rn, void *w));
138 static int	vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
139 				       struct export_args *argp));
140 
141 /*
142  * Initialize the vnode management data structures.
143  */
144 void
145 vntblinit()
146 {
147 
148 	desiredvnodes = maxproc + cnt.v_page_count / 4;
149 	simple_lock_init(&mntvnode_slock);
150 	simple_lock_init(&mntid_slock);
151 	simple_lock_init(&spechash_slock);
152 	TAILQ_INIT(&vnode_free_list);
153 	TAILQ_INIT(&vnode_tobefree_list);
154 	simple_lock_init(&vnode_free_list_slock);
155 	CIRCLEQ_INIT(&mountlist);
156 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
157 	/*
158 	 * Initialize the filesystem syncer.
159 	 */
160 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
161 		&syncer_mask);
162 	syncer_maxdelay = syncer_mask + 1;
163 }
164 
165 /*
166  * Mark a mount point as busy. Used to synchronize access and to delay
167  * unmounting. Interlock is not released on failure.
168  */
169 int
170 vfs_busy(mp, flags, interlkp, p)
171 	struct mount *mp;
172 	int flags;
173 	struct simplelock *interlkp;
174 	struct proc *p;
175 {
176 	int lkflags;
177 
178 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
179 		if (flags & LK_NOWAIT)
180 			return (ENOENT);
181 		mp->mnt_kern_flag |= MNTK_MWAIT;
182 		if (interlkp) {
183 			simple_unlock(interlkp);
184 		}
185 		/*
186 		 * Since all busy locks are shared except the exclusive
187 		 * lock granted when unmounting, the only place that a
188 		 * wakeup needs to be done is at the release of the
189 		 * exclusive lock at the end of dounmount.
190 		 */
191 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
192 		if (interlkp) {
193 			simple_lock(interlkp);
194 		}
195 		return (ENOENT);
196 	}
197 	lkflags = LK_SHARED | LK_NOPAUSE;
198 	if (interlkp)
199 		lkflags |= LK_INTERLOCK;
200 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
201 		panic("vfs_busy: unexpected lock failure");
202 	return (0);
203 }
204 
205 /*
206  * Free a busy filesystem.
207  */
208 void
209 vfs_unbusy(mp, p)
210 	struct mount *mp;
211 	struct proc *p;
212 {
213 
214 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
215 }
216 
217 /*
218  * Lookup a filesystem type, and if found allocate and initialize
219  * a mount structure for it.
220  *
221  * Devname is usually updated by mount(8) after booting.
222  */
223 int
224 vfs_rootmountalloc(fstypename, devname, mpp)
225 	char *fstypename;
226 	char *devname;
227 	struct mount **mpp;
228 {
229 	struct proc *p = curproc;	/* XXX */
230 	struct vfsconf *vfsp;
231 	struct mount *mp;
232 
233 	if (fstypename == NULL)
234 		return (ENODEV);
235 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
236 		if (!strcmp(vfsp->vfc_name, fstypename))
237 			break;
238 	if (vfsp == NULL)
239 		return (ENODEV);
240 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
241 	bzero((char *)mp, (u_long)sizeof(struct mount));
242 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
243 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
244 	LIST_INIT(&mp->mnt_vnodelist);
245 	mp->mnt_vfc = vfsp;
246 	mp->mnt_op = vfsp->vfc_vfsops;
247 	mp->mnt_flag = MNT_RDONLY;
248 	mp->mnt_vnodecovered = NULLVP;
249 	vfsp->vfc_refcount++;
250 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
251 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
252 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
253 	mp->mnt_stat.f_mntonname[0] = '/';
254 	mp->mnt_stat.f_mntonname[1] = 0;
255 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
256 	*mpp = mp;
257 	return (0);
258 }
259 
260 /*
261  * Find an appropriate filesystem to use for the root. If a filesystem
262  * has not been preselected, walk through the list of known filesystems
263  * trying those that have mountroot routines, and try them until one
264  * works or we have tried them all.
265  */
266 #ifdef notdef	/* XXX JH */
267 int
268 lite2_vfs_mountroot()
269 {
270 	struct vfsconf *vfsp;
271 	extern int (*lite2_mountroot) __P((void));
272 	int error;
273 
274 	if (lite2_mountroot != NULL)
275 		return ((*lite2_mountroot)());
276 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
277 		if (vfsp->vfc_mountroot == NULL)
278 			continue;
279 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
280 			return (0);
281 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
282 	}
283 	return (ENODEV);
284 }
285 #endif
286 
287 /*
288  * Lookup a mount point by filesystem identifier.
289  */
290 struct mount *
291 vfs_getvfs(fsid)
292 	fsid_t *fsid;
293 {
294 	register struct mount *mp;
295 
296 	simple_lock(&mountlist_slock);
297 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
298 	    mp = mp->mnt_list.cqe_next) {
299 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
300 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
301 			simple_unlock(&mountlist_slock);
302 			return (mp);
303 	    }
304 	}
305 	simple_unlock(&mountlist_slock);
306 	return ((struct mount *) 0);
307 }
308 
309 /*
310  * Get a new unique fsid
311  */
312 void
313 vfs_getnewfsid(mp)
314 	struct mount *mp;
315 {
316 	static u_short xxxfs_mntid;
317 
318 	fsid_t tfsid;
319 	int mtype;
320 
321 	simple_lock(&mntid_slock);
322 	mtype = mp->mnt_vfc->vfc_typenum;
323 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
324 	mp->mnt_stat.f_fsid.val[1] = mtype;
325 	if (xxxfs_mntid == 0)
326 		++xxxfs_mntid;
327 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
328 	tfsid.val[1] = mtype;
329 	if (mountlist.cqh_first != (void *)&mountlist) {
330 		while (vfs_getvfs(&tfsid)) {
331 			tfsid.val[0]++;
332 			xxxfs_mntid++;
333 		}
334 	}
335 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
336 	simple_unlock(&mntid_slock);
337 }
338 
339 /*
340  * Set vnode attributes to VNOVAL
341  */
342 void
343 vattr_null(vap)
344 	register struct vattr *vap;
345 {
346 
347 	vap->va_type = VNON;
348 	vap->va_size = VNOVAL;
349 	vap->va_bytes = VNOVAL;
350 	vap->va_mode = VNOVAL;
351 	vap->va_nlink = VNOVAL;
352 	vap->va_uid = VNOVAL;
353 	vap->va_gid = VNOVAL;
354 	vap->va_fsid = VNOVAL;
355 	vap->va_fileid = VNOVAL;
356 	vap->va_blocksize = VNOVAL;
357 	vap->va_rdev = VNOVAL;
358 	vap->va_atime.tv_sec = VNOVAL;
359 	vap->va_atime.tv_nsec = VNOVAL;
360 	vap->va_mtime.tv_sec = VNOVAL;
361 	vap->va_mtime.tv_nsec = VNOVAL;
362 	vap->va_ctime.tv_sec = VNOVAL;
363 	vap->va_ctime.tv_nsec = VNOVAL;
364 	vap->va_flags = VNOVAL;
365 	vap->va_gen = VNOVAL;
366 	vap->va_vaflags = 0;
367 }
368 
369 /*
370  * Routines having to do with the management of the vnode table.
371  */
372 extern vop_t **dead_vnodeop_p;
373 
374 /*
375  * Return the next vnode from the free list.
376  */
377 int
378 getnewvnode(tag, mp, vops, vpp)
379 	enum vtagtype tag;
380 	struct mount *mp;
381 	vop_t **vops;
382 	struct vnode **vpp;
383 {
384 	int s;
385 	struct proc *p = curproc;	/* XXX */
386 	struct vnode *vp, *tvp, *nvp;
387 	vm_object_t object;
388 	TAILQ_HEAD(freelst, vnode) vnode_tmp_list;
389 
390 	/*
391 	 * We take the least recently used vnode from the freelist
392 	 * if we can get it and it has no cached pages, and no
393 	 * namecache entries are relative to it.
394 	 * Otherwise we allocate a new vnode
395 	 */
396 
397 	s = splbio();
398 	simple_lock(&vnode_free_list_slock);
399 	TAILQ_INIT(&vnode_tmp_list);
400 
401 	for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) {
402 		nvp = TAILQ_NEXT(vp, v_freelist);
403 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
404 		if (vp->v_flag & VAGE) {
405 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
406 		} else {
407 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
408 		}
409 		vp->v_flag &= ~(VTBFREE|VAGE);
410 		vp->v_flag |= VFREE;
411 		if (vp->v_usecount)
412 			panic("tobe free vnode isn't");
413 		freevnodes++;
414 	}
415 
416 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
417 		vp = NULL;
418 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
419 		/*
420 		 * XXX: this is only here to be backwards compatible
421 		 */
422 		vp = NULL;
423 	} else {
424 		for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) {
425 			nvp = TAILQ_NEXT(vp, v_freelist);
426 			if (!simple_lock_try(&vp->v_interlock))
427 				continue;
428 			if (vp->v_usecount)
429 				panic("free vnode isn't");
430 
431 			object = vp->v_object;
432 			if (object && (object->resident_page_count || object->ref_count)) {
433 				printf("object inconsistant state: RPC: %d, RC: %d\n",
434 					object->resident_page_count, object->ref_count);
435 				/* Don't recycle if it's caching some pages */
436 				TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
437 				TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist);
438 				continue;
439 			} else if (LIST_FIRST(&vp->v_cache_src)) {
440 				/* Don't recycle if active in the namecache */
441 				simple_unlock(&vp->v_interlock);
442 				continue;
443 			} else {
444 				break;
445 			}
446 		}
447 	}
448 
449 	for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) {
450 		nvp = TAILQ_NEXT(tvp, v_freelist);
451 		TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist);
452 		TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist);
453 		simple_unlock(&tvp->v_interlock);
454 	}
455 
456 	if (vp) {
457 		vp->v_flag |= VDOOMED;
458 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
459 		freevnodes--;
460 		simple_unlock(&vnode_free_list_slock);
461 		cache_purge(vp);
462 		vp->v_lease = NULL;
463 		if (vp->v_type != VBAD) {
464 			vgonel(vp, p);
465 		} else {
466 			simple_unlock(&vp->v_interlock);
467 		}
468 
469 #ifdef INVARIANTS
470 		{
471 			int s;
472 
473 			if (vp->v_data)
474 				panic("cleaned vnode isn't");
475 			s = splbio();
476 			if (vp->v_numoutput)
477 				panic("Clean vnode has pending I/O's");
478 			splx(s);
479 		}
480 #endif
481 		vp->v_flag = 0;
482 		vp->v_lastr = 0;
483 		vp->v_lastw = 0;
484 		vp->v_lasta = 0;
485 		vp->v_cstart = 0;
486 		vp->v_clen = 0;
487 		vp->v_socket = 0;
488 		vp->v_writecount = 0;	/* XXX */
489 		vp->v_maxio = 0;
490 	} else {
491 		simple_unlock(&vnode_free_list_slock);
492 		vp = (struct vnode *) zalloc(vnode_zone);
493 		bzero((char *) vp, sizeof *vp);
494 		simple_lock_init(&vp->v_interlock);
495 		vp->v_dd = vp;
496 		cache_purge(vp);
497 		LIST_INIT(&vp->v_cache_src);
498 		TAILQ_INIT(&vp->v_cache_dst);
499 		numvnodes++;
500 	}
501 
502 	TAILQ_INIT(&vp->v_cleanblkhd);
503 	TAILQ_INIT(&vp->v_dirtyblkhd);
504 	vp->v_type = VNON;
505 	vp->v_tag = tag;
506 	vp->v_op = vops;
507 	insmntque(vp, mp);
508 	*vpp = vp;
509 	vp->v_usecount = 1;
510 	vp->v_data = 0;
511 	splx(s);
512 
513 	vfs_object_create(vp, p, p->p_ucred);
514 	return (0);
515 }
516 
517 /*
518  * Move a vnode from one mount queue to another.
519  */
520 static void
521 insmntque(vp, mp)
522 	register struct vnode *vp;
523 	register struct mount *mp;
524 {
525 
526 	simple_lock(&mntvnode_slock);
527 	/*
528 	 * Delete from old mount point vnode list, if on one.
529 	 */
530 	if (vp->v_mount != NULL)
531 		LIST_REMOVE(vp, v_mntvnodes);
532 	/*
533 	 * Insert into list of vnodes for the new mount point, if available.
534 	 */
535 	if ((vp->v_mount = mp) == NULL) {
536 		simple_unlock(&mntvnode_slock);
537 		return;
538 	}
539 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
540 	simple_unlock(&mntvnode_slock);
541 }
542 
543 /*
544  * Update outstanding I/O count and do wakeup if requested.
545  */
546 void
547 vwakeup(bp)
548 	register struct buf *bp;
549 {
550 	register struct vnode *vp;
551 
552 	bp->b_flags &= ~B_WRITEINPROG;
553 	if ((vp = bp->b_vp)) {
554 		vp->v_numoutput--;
555 		if (vp->v_numoutput < 0)
556 			panic("vwakeup: neg numoutput");
557 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
558 			vp->v_flag &= ~VBWAIT;
559 			wakeup((caddr_t) &vp->v_numoutput);
560 		}
561 	}
562 }
563 
564 /*
565  * Flush out and invalidate all buffers associated with a vnode.
566  * Called with the underlying object locked.
567  */
568 int
569 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
570 	register struct vnode *vp;
571 	int flags;
572 	struct ucred *cred;
573 	struct proc *p;
574 	int slpflag, slptimeo;
575 {
576 	register struct buf *bp;
577 	struct buf *nbp, *blist;
578 	int s, error;
579 	vm_object_t object;
580 
581 	if (flags & V_SAVE) {
582 		s = splbio();
583 		while (vp->v_numoutput) {
584 			vp->v_flag |= VBWAIT;
585 			error = tsleep((caddr_t)&vp->v_numoutput,
586 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
587 			if (error) {
588 				splx(s);
589 				return (error);
590 			}
591 		}
592 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
593 			splx(s);
594 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
595 				return (error);
596 			s = splbio();
597 			if (vp->v_numoutput > 0 ||
598 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
599 				panic("vinvalbuf: dirty bufs");
600 		}
601 		splx(s);
602   	}
603 	s = splbio();
604 	for (;;) {
605 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
606 		if (!blist)
607 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
608 		if (!blist)
609 			break;
610 
611 		for (bp = blist; bp; bp = nbp) {
612 			nbp = TAILQ_NEXT(bp, b_vnbufs);
613 			if (bp->b_flags & B_BUSY) {
614 				bp->b_flags |= B_WANTED;
615 				error = tsleep((caddr_t) bp,
616 				    slpflag | (PRIBIO + 4), "vinvalbuf",
617 				    slptimeo);
618 				if (error) {
619 					splx(s);
620 					return (error);
621 				}
622 				break;
623 			}
624 			/*
625 			 * XXX Since there are no node locks for NFS, I
626 			 * believe there is a slight chance that a delayed
627 			 * write will occur while sleeping just above, so
628 			 * check for it.  Note that vfs_bio_awrite expects
629 			 * buffers to reside on a queue, while VOP_BWRITE and
630 			 * brelse do not.
631 			 */
632 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
633 				(flags & V_SAVE)) {
634 
635 				if (bp->b_vp == vp) {
636 					if (bp->b_flags & B_CLUSTEROK) {
637 						vfs_bio_awrite(bp);
638 					} else {
639 						bremfree(bp);
640 						bp->b_flags |= (B_BUSY | B_ASYNC);
641 						VOP_BWRITE(bp);
642 					}
643 				} else {
644 					bremfree(bp);
645 					bp->b_flags |= B_BUSY;
646 					(void) VOP_BWRITE(bp);
647 				}
648 				break;
649 			}
650 			bremfree(bp);
651 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF | B_BUSY);
652 			bp->b_flags &= ~B_ASYNC;
653 			brelse(bp);
654 		}
655 	}
656 
657 	while (vp->v_numoutput > 0) {
658 		vp->v_flag |= VBWAIT;
659 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
660 	}
661 
662 	splx(s);
663 
664 	/*
665 	 * Destroy the copy in the VM cache, too.
666 	 */
667 	simple_lock(&vp->v_interlock);
668 	object = vp->v_object;
669 	if (object != NULL) {
670 		vm_object_page_remove(object, 0, 0,
671 			(flags & V_SAVE) ? TRUE : FALSE);
672 	}
673 	simple_unlock(&vp->v_interlock);
674 
675 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
676 		panic("vinvalbuf: flush failed");
677 	return (0);
678 }
679 
680 /*
681  * Truncate a file's buffer and pages to a specified length.  This
682  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
683  * sync activity.
684  */
685 int
686 vtruncbuf(vp, cred, p, length, blksize)
687 	register struct vnode *vp;
688 	struct ucred *cred;
689 	struct proc *p;
690 	off_t length;
691 	int blksize;
692 {
693 	register struct buf *bp;
694 	struct buf *nbp;
695 	int s, anyfreed;
696 	int trunclbn;
697 
698 	/*
699 	 * Round up to the *next* lbn.
700 	 */
701 	trunclbn = (length + blksize - 1) / blksize;
702 
703 	s = splbio();
704 restart:
705 	anyfreed = 1;
706 	for (;anyfreed;) {
707 		anyfreed = 0;
708 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
709 			nbp = TAILQ_NEXT(bp, b_vnbufs);
710 			if (bp->b_lblkno >= trunclbn) {
711 				if (bp->b_flags & B_BUSY) {
712 					bp->b_flags |= B_WANTED;
713 					tsleep(bp, PRIBIO + 4, "vtrb1", 0);
714 					goto restart;
715 				} else {
716 					bremfree(bp);
717 					bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
718 					bp->b_flags &= ~B_ASYNC;
719 					brelse(bp);
720 					anyfreed = 1;
721 				}
722 				if (nbp && (((nbp->b_xflags & B_VNCLEAN) == 0)||
723 					 (nbp->b_vp != vp) ||
724 					 (nbp->b_flags & B_DELWRI))) {
725 					goto restart;
726 				}
727 			}
728 		}
729 
730 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
731 			nbp = TAILQ_NEXT(bp, b_vnbufs);
732 			if (bp->b_lblkno >= trunclbn) {
733 				if (bp->b_flags & B_BUSY) {
734 					bp->b_flags |= B_WANTED;
735 					tsleep(bp, PRIBIO + 4, "vtrb2", 0);
736 					goto restart;
737 				} else {
738 					bremfree(bp);
739 					bp->b_flags |= (B_BUSY | B_INVAL | B_RELBUF);
740 					bp->b_flags &= ~B_ASYNC;
741 					brelse(bp);
742 					anyfreed = 1;
743 				}
744 				if (nbp && (((nbp->b_xflags & B_VNDIRTY) == 0)||
745 					 (nbp->b_vp != vp) ||
746 					 (nbp->b_flags & B_DELWRI) == 0)) {
747 					goto restart;
748 				}
749 			}
750 		}
751 	}
752 
753 	if (length > 0) {
754 restartsync:
755 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
756 			nbp = TAILQ_NEXT(bp, b_vnbufs);
757 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
758 				if (bp->b_flags & B_BUSY) {
759 					bp->b_flags |= B_WANTED;
760 					tsleep(bp, PRIBIO, "vtrb3", 0);
761 				} else {
762 					bremfree(bp);
763 					bp->b_flags |= B_BUSY;
764 					if (bp->b_vp == vp) {
765 						bp->b_flags |= B_ASYNC;
766 					} else {
767 						bp->b_flags &= ~B_ASYNC;
768 					}
769 					VOP_BWRITE(bp);
770 				}
771 				goto restartsync;
772 			}
773 
774 		}
775 	}
776 
777 	while (vp->v_numoutput > 0) {
778 		vp->v_flag |= VBWAIT;
779 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
780 	}
781 
782 	splx(s);
783 
784 	vnode_pager_setsize(vp, length);
785 
786 	return (0);
787 }
788 
789 /*
790  * Associate a buffer with a vnode.
791  */
792 void
793 bgetvp(vp, bp)
794 	register struct vnode *vp;
795 	register struct buf *bp;
796 {
797 	int s;
798 
799 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
800 
801 	vhold(vp);
802 	bp->b_vp = vp;
803 	if (vp->v_type == VBLK || vp->v_type == VCHR)
804 		bp->b_dev = vp->v_rdev;
805 	else
806 		bp->b_dev = NODEV;
807 	/*
808 	 * Insert onto list for new vnode.
809 	 */
810 	s = splbio();
811 	bp->b_xflags |= B_VNCLEAN;
812 	bp->b_xflags &= ~B_VNDIRTY;
813 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
814 	splx(s);
815 }
816 
817 /*
818  * Disassociate a buffer from a vnode.
819  */
820 void
821 brelvp(bp)
822 	register struct buf *bp;
823 {
824 	struct vnode *vp;
825 	struct buflists *listheadp;
826 	int s;
827 
828 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
829 
830 	/*
831 	 * Delete from old vnode list, if on one.
832 	 */
833 	vp = bp->b_vp;
834 	s = splbio();
835 	if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
836 		if (bp->b_xflags & B_VNDIRTY)
837 			listheadp = &vp->v_dirtyblkhd;
838 		else
839 			listheadp = &vp->v_cleanblkhd;
840 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
841 		bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
842 	}
843 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
844 		vp->v_flag &= ~VONWORKLST;
845 		LIST_REMOVE(vp, v_synclist);
846 	}
847 	splx(s);
848 	bp->b_vp = (struct vnode *) 0;
849 	vdrop(vp);
850 }
851 
852 /*
853  * The workitem queue.
854  *
855  * It is useful to delay writes of file data and filesystem metadata
856  * for tens of seconds so that quickly created and deleted files need
857  * not waste disk bandwidth being created and removed. To realize this,
858  * we append vnodes to a "workitem" queue. When running with a soft
859  * updates implementation, most pending metadata dependencies should
860  * not wait for more than a few seconds. Thus, mounted on block devices
861  * are delayed only about a half the time that file data is delayed.
862  * Similarly, directory updates are more critical, so are only delayed
863  * about a third the time that file data is delayed. Thus, there are
864  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
865  * one each second (driven off the filesystem syner process). The
866  * syncer_delayno variable indicates the next queue that is to be processed.
867  * Items that need to be processed soon are placed in this queue:
868  *
869  *	syncer_workitem_pending[syncer_delayno]
870  *
871  * A delay of fifteen seconds is done by placing the request fifteen
872  * entries later in the queue:
873  *
874  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
875  *
876  */
877 
878 /*
879  * Add an item to the syncer work queue.
880  */
881 void
882 vn_syncer_add_to_worklist(vp, delay)
883 	struct vnode *vp;
884 	int delay;
885 {
886 	int s, slot;
887 
888 	s = splbio();
889 
890 	if (vp->v_flag & VONWORKLST) {
891 		LIST_REMOVE(vp, v_synclist);
892 	}
893 
894 	if (delay > syncer_maxdelay - 2)
895 		delay = syncer_maxdelay - 2;
896 	slot = (syncer_delayno + delay) & syncer_mask;
897 
898 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
899 	vp->v_flag |= VONWORKLST;
900 	splx(s);
901 }
902 
903 static void sched_sync __P((void));
904 static struct	proc *updateproc;
905 static struct kproc_desc up_kp = {
906 	"syncer",
907 	sched_sync,
908 	&updateproc
909 };
910 SYSINIT_KT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
911 
912 /*
913  * System filesystem synchronizer daemon.
914  */
915 void
916 sched_sync(void)
917 {
918 	struct synclist *slp;
919 	struct vnode *vp;
920 	long starttime;
921 	int s;
922 	struct proc *p = updateproc;
923 
924 	for (;;) {
925 		starttime = time_second;
926 
927 		/*
928 		 * Push files whose dirty time has expired.
929 		 */
930 		s = splbio();
931 		slp = &syncer_workitem_pending[syncer_delayno];
932 		syncer_delayno += 1;
933 		if (syncer_delayno == syncer_maxdelay)
934 			syncer_delayno = 0;
935 		splx(s);
936 
937 		while ((vp = LIST_FIRST(slp)) != NULL) {
938 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
939 			(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
940 			VOP_UNLOCK(vp, 0, p);
941 			if (LIST_FIRST(slp) == vp) {
942 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
943 				    vp->v_type != VBLK)
944 					panic("sched_sync: fsync failed");
945 				/*
946 				 * Move ourselves to the back of the sync list.
947 				 */
948 				LIST_REMOVE(vp, v_synclist);
949 				vn_syncer_add_to_worklist(vp, syncdelay);
950 			}
951 		}
952 
953 		/*
954 		 * Do soft update processing.
955 		 */
956 		if (bioops.io_sync)
957 			(*bioops.io_sync)(NULL);
958 
959 		/*
960 		 * The variable rushjob allows the kernel to speed up the
961 		 * processing of the filesystem syncer process. A rushjob
962 		 * value of N tells the filesystem syncer to process the next
963 		 * N seconds worth of work on its queue ASAP. Currently rushjob
964 		 * is used by the soft update code to speed up the filesystem
965 		 * syncer process when the incore state is getting so far
966 		 * ahead of the disk that the kernel memory pool is being
967 		 * threatened with exhaustion.
968 		 */
969 		if (rushjob > 0) {
970 			rushjob -= 1;
971 			continue;
972 		}
973 		/*
974 		 * If it has taken us less than a second to process the
975 		 * current work, then wait. Otherwise start right over
976 		 * again. We can still lose time if any single round
977 		 * takes more than two seconds, but it does not really
978 		 * matter as we are just trying to generally pace the
979 		 * filesystem activity.
980 		 */
981 		if (time_second == starttime)
982 			tsleep(&lbolt, PPAUSE, "syncer", 0);
983 	}
984 }
985 
986 /*
987  * Associate a p-buffer with a vnode.
988  */
989 void
990 pbgetvp(vp, bp)
991 	register struct vnode *vp;
992 	register struct buf *bp;
993 {
994 
995 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
996 
997 	bp->b_vp = vp;
998 	if (vp->v_type == VBLK || vp->v_type == VCHR)
999 		bp->b_dev = vp->v_rdev;
1000 	else
1001 		bp->b_dev = NODEV;
1002 }
1003 
1004 /*
1005  * Disassociate a p-buffer from a vnode.
1006  */
1007 void
1008 pbrelvp(bp)
1009 	register struct buf *bp;
1010 {
1011 
1012 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1013 
1014 	bp->b_vp = (struct vnode *) 0;
1015 }
1016 
1017 /*
1018  * Reassign a buffer from one vnode to another.
1019  * Used to assign file specific control information
1020  * (indirect blocks) to the vnode to which they belong.
1021  */
1022 void
1023 reassignbuf(bp, newvp)
1024 	register struct buf *bp;
1025 	register struct vnode *newvp;
1026 {
1027 	struct buflists *listheadp;
1028 	struct vnode *oldvp;
1029 	int delay;
1030 	int s;
1031 
1032 	if (newvp == NULL) {
1033 		printf("reassignbuf: NULL");
1034 		return;
1035 	}
1036 
1037 	s = splbio();
1038 	/*
1039 	 * Delete from old vnode list, if on one.
1040 	 */
1041 	if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
1042 		oldvp = bp->b_vp;
1043 		if (bp->b_xflags & B_VNDIRTY)
1044 			listheadp = &oldvp->v_dirtyblkhd;
1045 		else
1046 			listheadp = &oldvp->v_cleanblkhd;
1047 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1048 		bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
1049 		vdrop(oldvp);
1050 	}
1051 	/*
1052 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1053 	 * of clean buffers.
1054 	 */
1055 	if (bp->b_flags & B_DELWRI) {
1056 		struct buf *tbp;
1057 
1058 		listheadp = &newvp->v_dirtyblkhd;
1059 		if ((newvp->v_flag & VONWORKLST) == 0) {
1060 			switch (newvp->v_type) {
1061 			case VDIR:
1062 				delay = syncdelay / 3;
1063 				break;
1064 			case VBLK:
1065 				if (newvp->v_specmountpoint != NULL) {
1066 					delay = syncdelay / 2;
1067 					break;
1068 				}
1069 				/* fall through */
1070 			default:
1071 				delay = syncdelay;
1072 			}
1073 			vn_syncer_add_to_worklist(newvp, delay);
1074 		}
1075 		bp->b_xflags |= B_VNDIRTY;
1076 		tbp = TAILQ_FIRST(listheadp);
1077 		if (tbp == NULL ||
1078 		    (bp->b_lblkno >= 0 && tbp->b_lblkno > bp->b_lblkno)) {
1079 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1080 		} else {
1081 			if (bp->b_lblkno >= 0) {
1082 				struct buf *ttbp;
1083 				while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1084 				    (ttbp->b_lblkno < bp->b_lblkno)) {
1085 					tbp = ttbp;
1086 				}
1087 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1088 			} else {
1089 				TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1090 			}
1091 		}
1092 	} else {
1093 		bp->b_xflags |= B_VNCLEAN;
1094 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1095 		if ((newvp->v_flag & VONWORKLST) &&
1096 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1097 			newvp->v_flag &= ~VONWORKLST;
1098 			LIST_REMOVE(newvp, v_synclist);
1099 		}
1100 	}
1101 	bp->b_vp = newvp;
1102 	vhold(bp->b_vp);
1103 	splx(s);
1104 }
1105 
1106 /*
1107  * Create a vnode for a block device.
1108  * Used for mounting the root file system.
1109  */
1110 int
1111 bdevvp(dev, vpp)
1112 	dev_t dev;
1113 	struct vnode **vpp;
1114 {
1115 	register struct vnode *vp;
1116 	struct vnode *nvp;
1117 	int error;
1118 
1119 	/* XXX 255 is for mfs. */
1120 	if (dev == NODEV || (major(dev) != 255 && (major(dev) >= nblkdev ||
1121 	    bdevsw[major(dev)] == NULL))) {
1122 		*vpp = NULLVP;
1123 		return (ENXIO);
1124 	}
1125 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1126 	if (error) {
1127 		*vpp = NULLVP;
1128 		return (error);
1129 	}
1130 	vp = nvp;
1131 	vp->v_type = VBLK;
1132 	if ((nvp = checkalias(vp, dev, (struct mount *)0)) != NULL) {
1133 		vput(vp);
1134 		vp = nvp;
1135 	}
1136 	*vpp = vp;
1137 	return (0);
1138 }
1139 
1140 /*
1141  * Check to see if the new vnode represents a special device
1142  * for which we already have a vnode (either because of
1143  * bdevvp() or because of a different vnode representing
1144  * the same block device). If such an alias exists, deallocate
1145  * the existing contents and return the aliased vnode. The
1146  * caller is responsible for filling it with its new contents.
1147  */
1148 struct vnode *
1149 checkalias(nvp, nvp_rdev, mp)
1150 	register struct vnode *nvp;
1151 	dev_t nvp_rdev;
1152 	struct mount *mp;
1153 {
1154 	struct proc *p = curproc;	/* XXX */
1155 	struct vnode *vp;
1156 	struct vnode **vpp;
1157 
1158 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1159 		return (NULLVP);
1160 
1161 	vpp = &speclisth[SPECHASH(nvp_rdev)];
1162 loop:
1163 	simple_lock(&spechash_slock);
1164 	for (vp = *vpp; vp; vp = vp->v_specnext) {
1165 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1166 			continue;
1167 		/*
1168 		 * Alias, but not in use, so flush it out.
1169 		 * Only alias active device nodes.
1170 		 * Not sure why we don't re-use this like we do below.
1171 		 */
1172 		simple_lock(&vp->v_interlock);
1173 		if (vp->v_usecount == 0) {
1174 			simple_unlock(&spechash_slock);
1175 			vgonel(vp, p);
1176 			goto loop;
1177 		}
1178 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) {
1179 			/*
1180 			 * It dissappeared, and we may have slept.
1181 			 * Restart from the beginning
1182 			 */
1183 			simple_unlock(&spechash_slock);
1184 			goto loop;
1185 		}
1186 		break;
1187 	}
1188 	/*
1189 	 * It would be a lot clearer what is going on here if
1190 	 * this had been expressed as:
1191 	 * if ( vp && (vp->v_tag == VT_NULL))
1192 	 * and the clauses had been swapped.
1193 	 */
1194 	if (vp == NULL || vp->v_tag != VT_NON) {
1195 		/*
1196 		 * Put the new vnode into the hash chain.
1197 		 * and if there was an alias, connect them.
1198 		 */
1199 		MALLOC(nvp->v_specinfo, struct specinfo *,
1200 		    sizeof(struct specinfo), M_VNODE, M_WAITOK);
1201 		nvp->v_rdev = nvp_rdev;
1202 		nvp->v_hashchain = vpp;
1203 		nvp->v_specnext = *vpp;
1204 		nvp->v_specmountpoint = NULL;
1205 		simple_unlock(&spechash_slock);
1206 		*vpp = nvp;
1207 		if (vp != NULLVP) {
1208 			nvp->v_flag |= VALIASED;
1209 			vp->v_flag |= VALIASED;
1210 			vput(vp);
1211 		}
1212 		return (NULLVP);
1213 	}
1214 	/*
1215 	 * if ( vp && (vp->v_tag == VT_NULL))
1216 	 * We have a vnode alias, but it is a trashed.
1217 	 * Make it look like it's newley allocated. (by getnewvnode())
1218 	 * The caller should use this instead.
1219 	 */
1220 	simple_unlock(&spechash_slock);
1221 	VOP_UNLOCK(vp, 0, p);
1222 	simple_lock(&vp->v_interlock);
1223 	vclean(vp, 0, p);
1224 	vp->v_op = nvp->v_op;
1225 	vp->v_tag = nvp->v_tag;
1226 	nvp->v_type = VNON;
1227 	insmntque(vp, mp);
1228 	return (vp);
1229 }
1230 
1231 /*
1232  * Grab a particular vnode from the free list, increment its
1233  * reference count and lock it. The vnode lock bit is set the
1234  * vnode is being eliminated in vgone. The process is awakened
1235  * when the transition is completed, and an error returned to
1236  * indicate that the vnode is no longer usable (possibly having
1237  * been changed to a new file system type).
1238  */
1239 int
1240 vget(vp, flags, p)
1241 	register struct vnode *vp;
1242 	int flags;
1243 	struct proc *p;
1244 {
1245 	int error;
1246 
1247 	/*
1248 	 * If the vnode is in the process of being cleaned out for
1249 	 * another use, we wait for the cleaning to finish and then
1250 	 * return failure. Cleaning is determined by checking that
1251 	 * the VXLOCK flag is set.
1252 	 */
1253 	if ((flags & LK_INTERLOCK) == 0) {
1254 		simple_lock(&vp->v_interlock);
1255 	}
1256 	if (vp->v_flag & VXLOCK) {
1257 		vp->v_flag |= VXWANT;
1258 		simple_unlock(&vp->v_interlock);
1259 		tsleep((caddr_t)vp, PINOD, "vget", 0);
1260 		return (ENOENT);
1261 	}
1262 
1263 	vp->v_usecount++;
1264 
1265 	if (VSHOULDBUSY(vp))
1266 		vbusy(vp);
1267 	if (flags & LK_TYPE_MASK) {
1268 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1269 			/*
1270 			 * must expand vrele here because we do not want
1271 			 * to call VOP_INACTIVE if the reference count
1272 			 * drops back to zero since it was never really
1273 			 * active. We must remove it from the free list
1274 			 * before sleeping so that multiple processes do
1275 			 * not try to recycle it.
1276 			 */
1277 			simple_lock(&vp->v_interlock);
1278 			vp->v_usecount--;
1279 			if (VSHOULDFREE(vp))
1280 				vfree(vp);
1281 			simple_unlock(&vp->v_interlock);
1282 		}
1283 		return (error);
1284 	}
1285 	simple_unlock(&vp->v_interlock);
1286 	return (0);
1287 }
1288 
1289 void
1290 vref(struct vnode *vp)
1291 {
1292 	simple_lock(&vp->v_interlock);
1293 	vp->v_usecount++;
1294 	simple_unlock(&vp->v_interlock);
1295 }
1296 
1297 /*
1298  * Vnode put/release.
1299  * If count drops to zero, call inactive routine and return to freelist.
1300  */
1301 void
1302 vrele(vp)
1303 	struct vnode *vp;
1304 {
1305 	struct proc *p = curproc;	/* XXX */
1306 
1307 	KASSERT(vp != NULL, ("vrele: null vp"));
1308 
1309 	simple_lock(&vp->v_interlock);
1310 
1311 	if (vp->v_usecount > 1) {
1312 
1313 		vp->v_usecount--;
1314 		simple_unlock(&vp->v_interlock);
1315 
1316 		return;
1317 	}
1318 
1319 	if (vp->v_usecount == 1) {
1320 
1321 		vp->v_usecount--;
1322 		if (VSHOULDFREE(vp))
1323 			vfree(vp);
1324 	/*
1325 	 * If we are doing a vput, the node is already locked, and we must
1326 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1327 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1328 	 */
1329 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1330 			VOP_INACTIVE(vp, p);
1331 		}
1332 
1333 	} else {
1334 #ifdef DIAGNOSTIC
1335 		vprint("vrele: negative ref count", vp);
1336 		simple_unlock(&vp->v_interlock);
1337 #endif
1338 		panic("vrele: negative ref cnt");
1339 	}
1340 }
1341 
1342 void
1343 vput(vp)
1344 	struct vnode *vp;
1345 {
1346 	struct proc *p = curproc;	/* XXX */
1347 
1348 	KASSERT(vp != NULL, ("vput: null vp"));
1349 
1350 	simple_lock(&vp->v_interlock);
1351 
1352 	if (vp->v_usecount > 1) {
1353 
1354 		vp->v_usecount--;
1355 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1356 		return;
1357 
1358 	}
1359 
1360 	if (vp->v_usecount == 1) {
1361 
1362 		vp->v_usecount--;
1363 		if (VSHOULDFREE(vp))
1364 			vfree(vp);
1365 	/*
1366 	 * If we are doing a vput, the node is already locked, and we must
1367 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1368 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1369 	 */
1370 		simple_unlock(&vp->v_interlock);
1371 		VOP_INACTIVE(vp, p);
1372 
1373 	} else {
1374 #ifdef DIAGNOSTIC
1375 		vprint("vput: negative ref count", vp);
1376 #endif
1377 		panic("vput: negative ref cnt");
1378 	}
1379 }
1380 
1381 /*
1382  * Somebody doesn't want the vnode recycled.
1383  */
1384 void
1385 vhold(vp)
1386 	register struct vnode *vp;
1387 {
1388 	int s;
1389 
1390   	s = splbio();
1391 	vp->v_holdcnt++;
1392 	if (VSHOULDBUSY(vp))
1393 		vbusy(vp);
1394 	splx(s);
1395 }
1396 
1397 /*
1398  * One less who cares about this vnode.
1399  */
1400 void
1401 vdrop(vp)
1402 	register struct vnode *vp;
1403 {
1404 	int s;
1405 
1406 	s = splbio();
1407 	if (vp->v_holdcnt <= 0)
1408 		panic("vdrop: holdcnt");
1409 	vp->v_holdcnt--;
1410 	if (VSHOULDFREE(vp))
1411 		vfree(vp);
1412 	splx(s);
1413 }
1414 
1415 /*
1416  * Remove any vnodes in the vnode table belonging to mount point mp.
1417  *
1418  * If MNT_NOFORCE is specified, there should not be any active ones,
1419  * return error if any are found (nb: this is a user error, not a
1420  * system error). If MNT_FORCE is specified, detach any active vnodes
1421  * that are found.
1422  */
1423 #ifdef DIAGNOSTIC
1424 static int busyprt = 0;		/* print out busy vnodes */
1425 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1426 #endif
1427 
1428 int
1429 vflush(mp, skipvp, flags)
1430 	struct mount *mp;
1431 	struct vnode *skipvp;
1432 	int flags;
1433 {
1434 	struct proc *p = curproc;	/* XXX */
1435 	struct vnode *vp, *nvp;
1436 	int busy = 0;
1437 
1438 	simple_lock(&mntvnode_slock);
1439 loop:
1440 	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
1441 		/*
1442 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1443 		 * Start over if it has (it won't be on the list anymore).
1444 		 */
1445 		if (vp->v_mount != mp)
1446 			goto loop;
1447 		nvp = vp->v_mntvnodes.le_next;
1448 		/*
1449 		 * Skip over a selected vnode.
1450 		 */
1451 		if (vp == skipvp)
1452 			continue;
1453 
1454 		simple_lock(&vp->v_interlock);
1455 		/*
1456 		 * Skip over a vnodes marked VSYSTEM.
1457 		 */
1458 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1459 			simple_unlock(&vp->v_interlock);
1460 			continue;
1461 		}
1462 		/*
1463 		 * If WRITECLOSE is set, only flush out regular file vnodes
1464 		 * open for writing.
1465 		 */
1466 		if ((flags & WRITECLOSE) &&
1467 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1468 			simple_unlock(&vp->v_interlock);
1469 			continue;
1470 		}
1471 
1472 		/*
1473 		 * With v_usecount == 0, all we need to do is clear out the
1474 		 * vnode data structures and we are done.
1475 		 */
1476 		if (vp->v_usecount == 0) {
1477 			simple_unlock(&mntvnode_slock);
1478 			vgonel(vp, p);
1479 			simple_lock(&mntvnode_slock);
1480 			continue;
1481 		}
1482 
1483 		/*
1484 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1485 		 * or character devices, revert to an anonymous device. For
1486 		 * all other files, just kill them.
1487 		 */
1488 		if (flags & FORCECLOSE) {
1489 			simple_unlock(&mntvnode_slock);
1490 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1491 				vgonel(vp, p);
1492 			} else {
1493 				vclean(vp, 0, p);
1494 				vp->v_op = spec_vnodeop_p;
1495 				insmntque(vp, (struct mount *) 0);
1496 			}
1497 			simple_lock(&mntvnode_slock);
1498 			continue;
1499 		}
1500 #ifdef DIAGNOSTIC
1501 		if (busyprt)
1502 			vprint("vflush: busy vnode", vp);
1503 #endif
1504 		simple_unlock(&vp->v_interlock);
1505 		busy++;
1506 	}
1507 	simple_unlock(&mntvnode_slock);
1508 	if (busy)
1509 		return (EBUSY);
1510 	return (0);
1511 }
1512 
1513 /*
1514  * Disassociate the underlying file system from a vnode.
1515  */
1516 static void
1517 vclean(vp, flags, p)
1518 	struct vnode *vp;
1519 	int flags;
1520 	struct proc *p;
1521 {
1522 	int active;
1523 	vm_object_t obj;
1524 
1525 	/*
1526 	 * Check to see if the vnode is in use. If so we have to reference it
1527 	 * before we clean it out so that its count cannot fall to zero and
1528 	 * generate a race against ourselves to recycle it.
1529 	 */
1530 	if ((active = vp->v_usecount))
1531 		vp->v_usecount++;
1532 
1533 	/*
1534 	 * Prevent the vnode from being recycled or brought into use while we
1535 	 * clean it out.
1536 	 */
1537 	if (vp->v_flag & VXLOCK)
1538 		panic("vclean: deadlock");
1539 	vp->v_flag |= VXLOCK;
1540 	/*
1541 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1542 	 * have the object locked while it cleans it out. The VOP_LOCK
1543 	 * ensures that the VOP_INACTIVE routine is done with its work.
1544 	 * For active vnodes, it ensures that no other activity can
1545 	 * occur while the underlying object is being cleaned out.
1546 	 */
1547 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1548 
1549 	/*
1550 	 * Clean out any buffers associated with the vnode.
1551 	 */
1552 	vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1553 	if (obj = vp->v_object) {
1554 		if (obj->ref_count == 0) {
1555 			/*
1556 			 * This is a normal way of shutting down the object/vnode
1557 			 * association.
1558 			 */
1559 			vm_object_terminate(obj);
1560 		} else {
1561 			/*
1562 			 * Woe to the process that tries to page now :-).
1563 			 */
1564 			vm_pager_deallocate(obj);
1565 		}
1566 	}
1567 
1568 	/*
1569 	 * If purging an active vnode, it must be closed and
1570 	 * deactivated before being reclaimed. Note that the
1571 	 * VOP_INACTIVE will unlock the vnode.
1572 	 */
1573 	if (active) {
1574 		if (flags & DOCLOSE)
1575 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1576 		VOP_INACTIVE(vp, p);
1577 	} else {
1578 		/*
1579 		 * Any other processes trying to obtain this lock must first
1580 		 * wait for VXLOCK to clear, then call the new lock operation.
1581 		 */
1582 		VOP_UNLOCK(vp, 0, p);
1583 	}
1584 	/*
1585 	 * Reclaim the vnode.
1586 	 */
1587 	if (VOP_RECLAIM(vp, p))
1588 		panic("vclean: cannot reclaim");
1589 
1590 	if (active)
1591 		vrele(vp);
1592 
1593 	cache_purge(vp);
1594 	if (vp->v_vnlock) {
1595 #if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */
1596 #ifdef DIAGNOSTIC
1597 		if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0)
1598 			vprint("vclean: lock not drained", vp);
1599 #endif
1600 #endif
1601 		FREE(vp->v_vnlock, M_VNODE);
1602 		vp->v_vnlock = NULL;
1603 	}
1604 
1605 	if (VSHOULDFREE(vp))
1606 		vfree(vp);
1607 
1608 	/*
1609 	 * Done with purge, notify sleepers of the grim news.
1610 	 */
1611 	vp->v_op = dead_vnodeop_p;
1612 	vn_pollgone(vp);
1613 	vp->v_tag = VT_NON;
1614 	vp->v_flag &= ~VXLOCK;
1615 	if (vp->v_flag & VXWANT) {
1616 		vp->v_flag &= ~VXWANT;
1617 		wakeup((caddr_t) vp);
1618 	}
1619 }
1620 
1621 /*
1622  * Eliminate all activity associated with the requested vnode
1623  * and with all vnodes aliased to the requested vnode.
1624  */
1625 int
1626 vop_revoke(ap)
1627 	struct vop_revoke_args /* {
1628 		struct vnode *a_vp;
1629 		int a_flags;
1630 	} */ *ap;
1631 {
1632 	struct vnode *vp, *vq;
1633 	struct proc *p = curproc;	/* XXX */
1634 
1635 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1636 
1637 	vp = ap->a_vp;
1638 	simple_lock(&vp->v_interlock);
1639 
1640 	if (vp->v_flag & VALIASED) {
1641 		/*
1642 		 * If a vgone (or vclean) is already in progress,
1643 		 * wait until it is done and return.
1644 		 */
1645 		if (vp->v_flag & VXLOCK) {
1646 			vp->v_flag |= VXWANT;
1647 			simple_unlock(&vp->v_interlock);
1648 			tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1649 			return (0);
1650 		}
1651 		/*
1652 		 * Ensure that vp will not be vgone'd while we
1653 		 * are eliminating its aliases.
1654 		 */
1655 		vp->v_flag |= VXLOCK;
1656 		simple_unlock(&vp->v_interlock);
1657 		while (vp->v_flag & VALIASED) {
1658 			simple_lock(&spechash_slock);
1659 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1660 				if (vq->v_rdev != vp->v_rdev ||
1661 				    vq->v_type != vp->v_type || vp == vq)
1662 					continue;
1663 				simple_unlock(&spechash_slock);
1664 				vgone(vq);
1665 				break;
1666 			}
1667 			if (vq == NULLVP) {
1668 				simple_unlock(&spechash_slock);
1669 			}
1670 		}
1671 		/*
1672 		 * Remove the lock so that vgone below will
1673 		 * really eliminate the vnode after which time
1674 		 * vgone will awaken any sleepers.
1675 		 */
1676 		simple_lock(&vp->v_interlock);
1677 		vp->v_flag &= ~VXLOCK;
1678 		if (vp->v_flag & VXWANT) {
1679 			vp->v_flag &= ~VXWANT;
1680 			wakeup(vp);
1681 		}
1682 	}
1683 	vgonel(vp, p);
1684 	return (0);
1685 }
1686 
1687 /*
1688  * Recycle an unused vnode to the front of the free list.
1689  * Release the passed interlock if the vnode will be recycled.
1690  */
1691 int
1692 vrecycle(vp, inter_lkp, p)
1693 	struct vnode *vp;
1694 	struct simplelock *inter_lkp;
1695 	struct proc *p;
1696 {
1697 
1698 	simple_lock(&vp->v_interlock);
1699 	if (vp->v_usecount == 0) {
1700 		if (inter_lkp) {
1701 			simple_unlock(inter_lkp);
1702 		}
1703 		vgonel(vp, p);
1704 		return (1);
1705 	}
1706 	simple_unlock(&vp->v_interlock);
1707 	return (0);
1708 }
1709 
1710 /*
1711  * Eliminate all activity associated with a vnode
1712  * in preparation for reuse.
1713  */
1714 void
1715 vgone(vp)
1716 	register struct vnode *vp;
1717 {
1718 	struct proc *p = curproc;	/* XXX */
1719 
1720 	simple_lock(&vp->v_interlock);
1721 	vgonel(vp, p);
1722 }
1723 
1724 /*
1725  * vgone, with the vp interlock held.
1726  */
1727 static void
1728 vgonel(vp, p)
1729 	struct vnode *vp;
1730 	struct proc *p;
1731 {
1732 	int s;
1733 	struct vnode *vq;
1734 	struct vnode *vx;
1735 
1736 	/*
1737 	 * If a vgone (or vclean) is already in progress,
1738 	 * wait until it is done and return.
1739 	 */
1740 	if (vp->v_flag & VXLOCK) {
1741 		vp->v_flag |= VXWANT;
1742 		simple_unlock(&vp->v_interlock);
1743 		tsleep((caddr_t)vp, PINOD, "vgone", 0);
1744 		return;
1745 	}
1746 
1747 	/*
1748 	 * Clean out the filesystem specific data.
1749 	 */
1750 	vclean(vp, DOCLOSE, p);
1751 	simple_lock(&vp->v_interlock);
1752 
1753 	/*
1754 	 * Delete from old mount point vnode list, if on one.
1755 	 */
1756 	if (vp->v_mount != NULL)
1757 		insmntque(vp, (struct mount *)0);
1758 	/*
1759 	 * If special device, remove it from special device alias list
1760 	 * if it is on one.
1761 	 */
1762 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1763 		simple_lock(&spechash_slock);
1764 		if (*vp->v_hashchain == vp) {
1765 			*vp->v_hashchain = vp->v_specnext;
1766 		} else {
1767 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1768 				if (vq->v_specnext != vp)
1769 					continue;
1770 				vq->v_specnext = vp->v_specnext;
1771 				break;
1772 			}
1773 			if (vq == NULL)
1774 				panic("missing bdev");
1775 		}
1776 		if (vp->v_flag & VALIASED) {
1777 			vx = NULL;
1778 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1779 				if (vq->v_rdev != vp->v_rdev ||
1780 				    vq->v_type != vp->v_type)
1781 					continue;
1782 				if (vx)
1783 					break;
1784 				vx = vq;
1785 			}
1786 			if (vx == NULL)
1787 				panic("missing alias");
1788 			if (vq == NULL)
1789 				vx->v_flag &= ~VALIASED;
1790 			vp->v_flag &= ~VALIASED;
1791 		}
1792 		simple_unlock(&spechash_slock);
1793 		FREE(vp->v_specinfo, M_VNODE);
1794 		vp->v_specinfo = NULL;
1795 	}
1796 
1797 	/*
1798 	 * If it is on the freelist and not already at the head,
1799 	 * move it to the head of the list. The test of the back
1800 	 * pointer and the reference count of zero is because
1801 	 * it will be removed from the free list by getnewvnode,
1802 	 * but will not have its reference count incremented until
1803 	 * after calling vgone. If the reference count were
1804 	 * incremented first, vgone would (incorrectly) try to
1805 	 * close the previous instance of the underlying object.
1806 	 */
1807 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1808 		s = splbio();
1809 		simple_lock(&vnode_free_list_slock);
1810 		if (vp->v_flag & VFREE) {
1811 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1812 		} else if (vp->v_flag & VTBFREE) {
1813 			TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
1814 			vp->v_flag &= ~VTBFREE;
1815 			freevnodes++;
1816 		} else
1817 			freevnodes++;
1818 		vp->v_flag |= VFREE;
1819 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1820 		simple_unlock(&vnode_free_list_slock);
1821 		splx(s);
1822 	}
1823 
1824 	vp->v_type = VBAD;
1825 	simple_unlock(&vp->v_interlock);
1826 }
1827 
1828 /*
1829  * Lookup a vnode by device number.
1830  */
1831 int
1832 vfinddev(dev, type, vpp)
1833 	dev_t dev;
1834 	enum vtype type;
1835 	struct vnode **vpp;
1836 {
1837 	register struct vnode *vp;
1838 	int rc = 0;
1839 
1840 	simple_lock(&spechash_slock);
1841 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1842 		if (dev != vp->v_rdev || type != vp->v_type)
1843 			continue;
1844 		*vpp = vp;
1845 		rc = 1;
1846 		break;
1847 	}
1848 	simple_unlock(&spechash_slock);
1849 	return (rc);
1850 }
1851 
1852 /*
1853  * Calculate the total number of references to a special device.
1854  */
1855 int
1856 vcount(vp)
1857 	register struct vnode *vp;
1858 {
1859 	struct vnode *vq, *vnext;
1860 	int count;
1861 
1862 loop:
1863 	if ((vp->v_flag & VALIASED) == 0)
1864 		return (vp->v_usecount);
1865 	simple_lock(&spechash_slock);
1866 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1867 		vnext = vq->v_specnext;
1868 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1869 			continue;
1870 		/*
1871 		 * Alias, but not in use, so flush it out.
1872 		 */
1873 		if (vq->v_usecount == 0 && vq != vp) {
1874 			simple_unlock(&spechash_slock);
1875 			vgone(vq);
1876 			goto loop;
1877 		}
1878 		count += vq->v_usecount;
1879 	}
1880 	simple_unlock(&spechash_slock);
1881 	return (count);
1882 }
1883 /*
1884  * Print out a description of a vnode.
1885  */
1886 static char *typename[] =
1887 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1888 
1889 void
1890 vprint(label, vp)
1891 	char *label;
1892 	register struct vnode *vp;
1893 {
1894 	char buf[96];
1895 
1896 	if (label != NULL)
1897 		printf("%s: %p: ", label, (void *)vp);
1898 	else
1899 		printf("%p: ", (void *)vp);
1900 	printf("type %s, usecount %d, writecount %d, refcount %d,",
1901 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1902 	    vp->v_holdcnt);
1903 	buf[0] = '\0';
1904 	if (vp->v_flag & VROOT)
1905 		strcat(buf, "|VROOT");
1906 	if (vp->v_flag & VTEXT)
1907 		strcat(buf, "|VTEXT");
1908 	if (vp->v_flag & VSYSTEM)
1909 		strcat(buf, "|VSYSTEM");
1910 	if (vp->v_flag & VXLOCK)
1911 		strcat(buf, "|VXLOCK");
1912 	if (vp->v_flag & VXWANT)
1913 		strcat(buf, "|VXWANT");
1914 	if (vp->v_flag & VBWAIT)
1915 		strcat(buf, "|VBWAIT");
1916 	if (vp->v_flag & VALIASED)
1917 		strcat(buf, "|VALIASED");
1918 	if (vp->v_flag & VDOOMED)
1919 		strcat(buf, "|VDOOMED");
1920 	if (vp->v_flag & VFREE)
1921 		strcat(buf, "|VFREE");
1922 	if (vp->v_flag & VOBJBUF)
1923 		strcat(buf, "|VOBJBUF");
1924 	if (buf[0] != '\0')
1925 		printf(" flags (%s)", &buf[1]);
1926 	if (vp->v_data == NULL) {
1927 		printf("\n");
1928 	} else {
1929 		printf("\n\t");
1930 		VOP_PRINT(vp);
1931 	}
1932 }
1933 
1934 #ifdef DDB
1935 #include <ddb/ddb.h>
1936 /*
1937  * List all of the locked vnodes in the system.
1938  * Called when debugging the kernel.
1939  */
1940 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1941 {
1942 	struct proc *p = curproc;	/* XXX */
1943 	struct mount *mp, *nmp;
1944 	struct vnode *vp;
1945 
1946 	printf("Locked vnodes\n");
1947 	simple_lock(&mountlist_slock);
1948 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
1949 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
1950 			nmp = mp->mnt_list.cqe_next;
1951 			continue;
1952 		}
1953 		for (vp = mp->mnt_vnodelist.lh_first;
1954 		     vp != NULL;
1955 		     vp = vp->v_mntvnodes.le_next) {
1956 			if (VOP_ISLOCKED(vp))
1957 				vprint((char *)0, vp);
1958 		}
1959 		simple_lock(&mountlist_slock);
1960 		nmp = mp->mnt_list.cqe_next;
1961 		vfs_unbusy(mp, p);
1962 	}
1963 	simple_unlock(&mountlist_slock);
1964 }
1965 #endif
1966 
1967 /*
1968  * Top level filesystem related information gathering.
1969  */
1970 static int	sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS);
1971 
1972 static int
1973 vfs_sysctl SYSCTL_HANDLER_ARGS
1974 {
1975 	int *name = (int *)arg1 - 1;	/* XXX */
1976 	u_int namelen = arg2 + 1;	/* XXX */
1977 	struct vfsconf *vfsp;
1978 
1979 #if 1 || defined(COMPAT_PRELITE2)
1980 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1981 	if (namelen == 1)
1982 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1983 #endif
1984 
1985 #ifdef notyet
1986 	/* all sysctl names at this level are at least name and field */
1987 	if (namelen < 2)
1988 		return (ENOTDIR);		/* overloaded */
1989 	if (name[0] != VFS_GENERIC) {
1990 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1991 			if (vfsp->vfc_typenum == name[0])
1992 				break;
1993 		if (vfsp == NULL)
1994 			return (EOPNOTSUPP);
1995 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1996 		    oldp, oldlenp, newp, newlen, p));
1997 	}
1998 #endif
1999 	switch (name[1]) {
2000 	case VFS_MAXTYPENUM:
2001 		if (namelen != 2)
2002 			return (ENOTDIR);
2003 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2004 	case VFS_CONF:
2005 		if (namelen != 3)
2006 			return (ENOTDIR);	/* overloaded */
2007 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2008 			if (vfsp->vfc_typenum == name[2])
2009 				break;
2010 		if (vfsp == NULL)
2011 			return (EOPNOTSUPP);
2012 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2013 	}
2014 	return (EOPNOTSUPP);
2015 }
2016 
2017 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2018 	"Generic filesystem");
2019 
2020 #if 1 || defined(COMPAT_PRELITE2)
2021 
2022 static int
2023 sysctl_ovfs_conf SYSCTL_HANDLER_ARGS
2024 {
2025 	int error;
2026 	struct vfsconf *vfsp;
2027 	struct ovfsconf ovfs;
2028 
2029 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2030 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2031 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2032 		ovfs.vfc_index = vfsp->vfc_typenum;
2033 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2034 		ovfs.vfc_flags = vfsp->vfc_flags;
2035 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2036 		if (error)
2037 			return error;
2038 	}
2039 	return 0;
2040 }
2041 
2042 #endif /* 1 || COMPAT_PRELITE2 */
2043 
2044 #if 0
2045 #define KINFO_VNODESLOP	10
2046 /*
2047  * Dump vnode list (via sysctl).
2048  * Copyout address of vnode followed by vnode.
2049  */
2050 /* ARGSUSED */
2051 static int
2052 sysctl_vnode SYSCTL_HANDLER_ARGS
2053 {
2054 	struct proc *p = curproc;	/* XXX */
2055 	struct mount *mp, *nmp;
2056 	struct vnode *nvp, *vp;
2057 	int error;
2058 
2059 #define VPTRSZ	sizeof (struct vnode *)
2060 #define VNODESZ	sizeof (struct vnode)
2061 
2062 	req->lock = 0;
2063 	if (!req->oldptr) /* Make an estimate */
2064 		return (SYSCTL_OUT(req, 0,
2065 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2066 
2067 	simple_lock(&mountlist_slock);
2068 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
2069 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2070 			nmp = mp->mnt_list.cqe_next;
2071 			continue;
2072 		}
2073 again:
2074 		simple_lock(&mntvnode_slock);
2075 		for (vp = mp->mnt_vnodelist.lh_first;
2076 		     vp != NULL;
2077 		     vp = nvp) {
2078 			/*
2079 			 * Check that the vp is still associated with
2080 			 * this filesystem.  RACE: could have been
2081 			 * recycled onto the same filesystem.
2082 			 */
2083 			if (vp->v_mount != mp) {
2084 				simple_unlock(&mntvnode_slock);
2085 				goto again;
2086 			}
2087 			nvp = vp->v_mntvnodes.le_next;
2088 			simple_unlock(&mntvnode_slock);
2089 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2090 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2091 				return (error);
2092 			simple_lock(&mntvnode_slock);
2093 		}
2094 		simple_unlock(&mntvnode_slock);
2095 		simple_lock(&mountlist_slock);
2096 		nmp = mp->mnt_list.cqe_next;
2097 		vfs_unbusy(mp, p);
2098 	}
2099 	simple_unlock(&mountlist_slock);
2100 
2101 	return (0);
2102 }
2103 #endif
2104 
2105 /*
2106  * XXX
2107  * Exporting the vnode list on large systems causes them to crash.
2108  * Exporting the vnode list on medium systems causes sysctl to coredump.
2109  */
2110 #if 0
2111 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2112 	0, 0, sysctl_vnode, "S,vnode", "");
2113 #endif
2114 
2115 /*
2116  * Check to see if a filesystem is mounted on a block device.
2117  */
2118 int
2119 vfs_mountedon(vp)
2120 	struct vnode *vp;
2121 {
2122 	struct vnode *vq;
2123 	int error = 0;
2124 
2125 	if (vp->v_specmountpoint != NULL)
2126 		return (EBUSY);
2127 	if (vp->v_flag & VALIASED) {
2128 		simple_lock(&spechash_slock);
2129 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2130 			if (vq->v_rdev != vp->v_rdev ||
2131 			    vq->v_type != vp->v_type)
2132 				continue;
2133 			if (vq->v_specmountpoint != NULL) {
2134 				error = EBUSY;
2135 				break;
2136 			}
2137 		}
2138 		simple_unlock(&spechash_slock);
2139 	}
2140 	return (error);
2141 }
2142 
2143 /*
2144  * Unmount all filesystems. The list is traversed in reverse order
2145  * of mounting to avoid dependencies.
2146  */
2147 void
2148 vfs_unmountall()
2149 {
2150 	struct mount *mp, *nmp;
2151 	struct proc *p;
2152 	int error;
2153 
2154 	if (curproc != NULL)
2155 		p = curproc;
2156 	else
2157 		p = initproc;	/* XXX XXX should this be proc0? */
2158 	/*
2159 	 * Since this only runs when rebooting, it is not interlocked.
2160 	 */
2161 	for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2162 		nmp = mp->mnt_list.cqe_prev;
2163 		error = dounmount(mp, MNT_FORCE, p);
2164 		if (error) {
2165 			printf("unmount of %s failed (",
2166 			    mp->mnt_stat.f_mntonname);
2167 			if (error == EBUSY)
2168 				printf("BUSY)\n");
2169 			else
2170 				printf("%d)\n", error);
2171 		}
2172 	}
2173 }
2174 
2175 /*
2176  * Build hash lists of net addresses and hang them off the mount point.
2177  * Called by ufs_mount() to set up the lists of export addresses.
2178  */
2179 static int
2180 vfs_hang_addrlist(mp, nep, argp)
2181 	struct mount *mp;
2182 	struct netexport *nep;
2183 	struct export_args *argp;
2184 {
2185 	register struct netcred *np;
2186 	register struct radix_node_head *rnh;
2187 	register int i;
2188 	struct radix_node *rn;
2189 	struct sockaddr *saddr, *smask = 0;
2190 	struct domain *dom;
2191 	int error;
2192 
2193 	if (argp->ex_addrlen == 0) {
2194 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2195 			return (EPERM);
2196 		np = &nep->ne_defexported;
2197 		np->netc_exflags = argp->ex_flags;
2198 		np->netc_anon = argp->ex_anon;
2199 		np->netc_anon.cr_ref = 1;
2200 		mp->mnt_flag |= MNT_DEFEXPORTED;
2201 		return (0);
2202 	}
2203 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2204 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2205 	bzero((caddr_t) np, i);
2206 	saddr = (struct sockaddr *) (np + 1);
2207 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2208 		goto out;
2209 	if (saddr->sa_len > argp->ex_addrlen)
2210 		saddr->sa_len = argp->ex_addrlen;
2211 	if (argp->ex_masklen) {
2212 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2213 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2214 		if (error)
2215 			goto out;
2216 		if (smask->sa_len > argp->ex_masklen)
2217 			smask->sa_len = argp->ex_masklen;
2218 	}
2219 	i = saddr->sa_family;
2220 	if ((rnh = nep->ne_rtable[i]) == 0) {
2221 		/*
2222 		 * Seems silly to initialize every AF when most are not used,
2223 		 * do so on demand here
2224 		 */
2225 		for (dom = domains; dom; dom = dom->dom_next)
2226 			if (dom->dom_family == i && dom->dom_rtattach) {
2227 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2228 				    dom->dom_rtoffset);
2229 				break;
2230 			}
2231 		if ((rnh = nep->ne_rtable[i]) == 0) {
2232 			error = ENOBUFS;
2233 			goto out;
2234 		}
2235 	}
2236 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2237 	    np->netc_rnodes);
2238 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2239 		error = EPERM;
2240 		goto out;
2241 	}
2242 	np->netc_exflags = argp->ex_flags;
2243 	np->netc_anon = argp->ex_anon;
2244 	np->netc_anon.cr_ref = 1;
2245 	return (0);
2246 out:
2247 	free(np, M_NETADDR);
2248 	return (error);
2249 }
2250 
2251 /* ARGSUSED */
2252 static int
2253 vfs_free_netcred(rn, w)
2254 	struct radix_node *rn;
2255 	void *w;
2256 {
2257 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
2258 
2259 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2260 	free((caddr_t) rn, M_NETADDR);
2261 	return (0);
2262 }
2263 
2264 /*
2265  * Free the net address hash lists that are hanging off the mount points.
2266  */
2267 static void
2268 vfs_free_addrlist(nep)
2269 	struct netexport *nep;
2270 {
2271 	register int i;
2272 	register struct radix_node_head *rnh;
2273 
2274 	for (i = 0; i <= AF_MAX; i++)
2275 		if ((rnh = nep->ne_rtable[i])) {
2276 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2277 			    (caddr_t) rnh);
2278 			free((caddr_t) rnh, M_RTABLE);
2279 			nep->ne_rtable[i] = 0;
2280 		}
2281 }
2282 
2283 int
2284 vfs_export(mp, nep, argp)
2285 	struct mount *mp;
2286 	struct netexport *nep;
2287 	struct export_args *argp;
2288 {
2289 	int error;
2290 
2291 	if (argp->ex_flags & MNT_DELEXPORT) {
2292 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2293 			vfs_setpublicfs(NULL, NULL, NULL);
2294 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2295 		}
2296 		vfs_free_addrlist(nep);
2297 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2298 	}
2299 	if (argp->ex_flags & MNT_EXPORTED) {
2300 		if (argp->ex_flags & MNT_EXPUBLIC) {
2301 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2302 				return (error);
2303 			mp->mnt_flag |= MNT_EXPUBLIC;
2304 		}
2305 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2306 			return (error);
2307 		mp->mnt_flag |= MNT_EXPORTED;
2308 	}
2309 	return (0);
2310 }
2311 
2312 
2313 /*
2314  * Set the publicly exported filesystem (WebNFS). Currently, only
2315  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2316  */
2317 int
2318 vfs_setpublicfs(mp, nep, argp)
2319 	struct mount *mp;
2320 	struct netexport *nep;
2321 	struct export_args *argp;
2322 {
2323 	int error;
2324 	struct vnode *rvp;
2325 	char *cp;
2326 
2327 	/*
2328 	 * mp == NULL -> invalidate the current info, the FS is
2329 	 * no longer exported. May be called from either vfs_export
2330 	 * or unmount, so check if it hasn't already been done.
2331 	 */
2332 	if (mp == NULL) {
2333 		if (nfs_pub.np_valid) {
2334 			nfs_pub.np_valid = 0;
2335 			if (nfs_pub.np_index != NULL) {
2336 				FREE(nfs_pub.np_index, M_TEMP);
2337 				nfs_pub.np_index = NULL;
2338 			}
2339 		}
2340 		return (0);
2341 	}
2342 
2343 	/*
2344 	 * Only one allowed at a time.
2345 	 */
2346 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2347 		return (EBUSY);
2348 
2349 	/*
2350 	 * Get real filehandle for root of exported FS.
2351 	 */
2352 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2353 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2354 
2355 	if ((error = VFS_ROOT(mp, &rvp)))
2356 		return (error);
2357 
2358 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2359 		return (error);
2360 
2361 	vput(rvp);
2362 
2363 	/*
2364 	 * If an indexfile was specified, pull it in.
2365 	 */
2366 	if (argp->ex_indexfile != NULL) {
2367 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2368 		    M_WAITOK);
2369 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2370 		    MAXNAMLEN, (size_t *)0);
2371 		if (!error) {
2372 			/*
2373 			 * Check for illegal filenames.
2374 			 */
2375 			for (cp = nfs_pub.np_index; *cp; cp++) {
2376 				if (*cp == '/') {
2377 					error = EINVAL;
2378 					break;
2379 				}
2380 			}
2381 		}
2382 		if (error) {
2383 			FREE(nfs_pub.np_index, M_TEMP);
2384 			return (error);
2385 		}
2386 	}
2387 
2388 	nfs_pub.np_mount = mp;
2389 	nfs_pub.np_valid = 1;
2390 	return (0);
2391 }
2392 
2393 struct netcred *
2394 vfs_export_lookup(mp, nep, nam)
2395 	register struct mount *mp;
2396 	struct netexport *nep;
2397 	struct sockaddr *nam;
2398 {
2399 	register struct netcred *np;
2400 	register struct radix_node_head *rnh;
2401 	struct sockaddr *saddr;
2402 
2403 	np = NULL;
2404 	if (mp->mnt_flag & MNT_EXPORTED) {
2405 		/*
2406 		 * Lookup in the export list first.
2407 		 */
2408 		if (nam != NULL) {
2409 			saddr = nam;
2410 			rnh = nep->ne_rtable[saddr->sa_family];
2411 			if (rnh != NULL) {
2412 				np = (struct netcred *)
2413 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2414 							      rnh);
2415 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2416 					np = NULL;
2417 			}
2418 		}
2419 		/*
2420 		 * If no address match, use the default if it exists.
2421 		 */
2422 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2423 			np = &nep->ne_defexported;
2424 	}
2425 	return (np);
2426 }
2427 
2428 /*
2429  * perform msync on all vnodes under a mount point
2430  * the mount point must be locked.
2431  */
2432 void
2433 vfs_msync(struct mount *mp, int flags) {
2434 	struct vnode *vp, *nvp;
2435 	struct vm_object *obj;
2436 	int anyio, tries;
2437 
2438 	tries = 5;
2439 loop:
2440 	anyio = 0;
2441 	for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
2442 
2443 		nvp = vp->v_mntvnodes.le_next;
2444 
2445 		if (vp->v_mount != mp) {
2446 			goto loop;
2447 		}
2448 
2449 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2450 			continue;
2451 
2452 		if (flags != MNT_WAIT) {
2453 			obj = vp->v_object;
2454 			if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2455 				continue;
2456 			if (VOP_ISLOCKED(vp))
2457 				continue;
2458 		}
2459 
2460 		simple_lock(&vp->v_interlock);
2461 		if (vp->v_object &&
2462 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
2463 			if (!vget(vp,
2464 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2465 				if (vp->v_object) {
2466 					vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : 0);
2467 					anyio = 1;
2468 				}
2469 				vput(vp);
2470 			}
2471 		} else {
2472 			simple_unlock(&vp->v_interlock);
2473 		}
2474 	}
2475 	if (anyio && (--tries > 0))
2476 		goto loop;
2477 }
2478 
2479 /*
2480  * Create the VM object needed for VMIO and mmap support.  This
2481  * is done for all VREG files in the system.  Some filesystems might
2482  * afford the additional metadata buffering capability of the
2483  * VMIO code by making the device node be VMIO mode also.
2484  *
2485  * vp must be locked when vfs_object_create is called.
2486  */
2487 int
2488 vfs_object_create(vp, p, cred)
2489 	struct vnode *vp;
2490 	struct proc *p;
2491 	struct ucred *cred;
2492 {
2493 	struct vattr vat;
2494 	vm_object_t object;
2495 	int error = 0;
2496 
2497 	if ((vp->v_type != VREG) && (vp->v_type != VBLK))
2498 		return 0;
2499 
2500 retry:
2501 	if ((object = vp->v_object) == NULL) {
2502 		if (vp->v_type == VREG) {
2503 			if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
2504 				goto retn;
2505 			object = vnode_pager_alloc(vp, vat.va_size, 0, 0);
2506 		} else if (major(vp->v_rdev) < nblkdev &&
2507 		    bdevsw[major(vp->v_rdev)] != NULL) {
2508 			/*
2509 			 * This simply allocates the biggest object possible
2510 			 * for a VBLK vnode.  This should be fixed, but doesn't
2511 			 * cause any problems (yet).
2512 			 */
2513 			object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0);
2514 		}
2515 		object->ref_count--;
2516 		vp->v_usecount--;
2517 	} else {
2518 		if (object->flags & OBJ_DEAD) {
2519 			VOP_UNLOCK(vp, 0, p);
2520 			tsleep(object, PVM, "vodead", 0);
2521 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
2522 			goto retry;
2523 		}
2524 	}
2525 
2526 	if (vp->v_object)
2527 		vp->v_flag |= VOBJBUF;
2528 
2529 retn:
2530 	return error;
2531 }
2532 
2533 static void
2534 vfree(vp)
2535 	struct vnode *vp;
2536 {
2537 	int s;
2538 
2539 	s = splbio();
2540 	simple_lock(&vnode_free_list_slock);
2541 	if (vp->v_flag & VTBFREE) {
2542 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2543 		vp->v_flag &= ~VTBFREE;
2544 	}
2545 	if (vp->v_flag & VAGE) {
2546 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2547 	} else {
2548 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2549 	}
2550 	freevnodes++;
2551 	simple_unlock(&vnode_free_list_slock);
2552 	vp->v_flag &= ~VAGE;
2553 	vp->v_flag |= VFREE;
2554 	splx(s);
2555 }
2556 
2557 void
2558 vbusy(vp)
2559 	struct vnode *vp;
2560 {
2561 	int s;
2562 
2563 	s = splbio();
2564 	simple_lock(&vnode_free_list_slock);
2565 	if (vp->v_flag & VTBFREE) {
2566 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2567 		vp->v_flag &= ~VTBFREE;
2568 	} else {
2569 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2570 		freevnodes--;
2571 	}
2572 	simple_unlock(&vnode_free_list_slock);
2573 	vp->v_flag &= ~(VFREE|VAGE);
2574 	splx(s);
2575 }
2576 
2577 /*
2578  * Record a process's interest in events which might happen to
2579  * a vnode.  Because poll uses the historic select-style interface
2580  * internally, this routine serves as both the ``check for any
2581  * pending events'' and the ``record my interest in future events''
2582  * functions.  (These are done together, while the lock is held,
2583  * to avoid race conditions.)
2584  */
2585 int
2586 vn_pollrecord(vp, p, events)
2587 	struct vnode *vp;
2588 	struct proc *p;
2589 	short events;
2590 {
2591 	simple_lock(&vp->v_pollinfo.vpi_lock);
2592 	if (vp->v_pollinfo.vpi_revents & events) {
2593 		/*
2594 		 * This leaves events we are not interested
2595 		 * in available for the other process which
2596 		 * which presumably had requested them
2597 		 * (otherwise they would never have been
2598 		 * recorded).
2599 		 */
2600 		events &= vp->v_pollinfo.vpi_revents;
2601 		vp->v_pollinfo.vpi_revents &= ~events;
2602 
2603 		simple_unlock(&vp->v_pollinfo.vpi_lock);
2604 		return events;
2605 	}
2606 	vp->v_pollinfo.vpi_events |= events;
2607 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2608 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2609 	return 0;
2610 }
2611 
2612 /*
2613  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2614  * it is possible for us to miss an event due to race conditions, but
2615  * that condition is expected to be rare, so for the moment it is the
2616  * preferred interface.
2617  */
2618 void
2619 vn_pollevent(vp, events)
2620 	struct vnode *vp;
2621 	short events;
2622 {
2623 	simple_lock(&vp->v_pollinfo.vpi_lock);
2624 	if (vp->v_pollinfo.vpi_events & events) {
2625 		/*
2626 		 * We clear vpi_events so that we don't
2627 		 * call selwakeup() twice if two events are
2628 		 * posted before the polling process(es) is
2629 		 * awakened.  This also ensures that we take at
2630 		 * most one selwakeup() if the polling process
2631 		 * is no longer interested.  However, it does
2632 		 * mean that only one event can be noticed at
2633 		 * a time.  (Perhaps we should only clear those
2634 		 * event bits which we note?) XXX
2635 		 */
2636 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2637 		vp->v_pollinfo.vpi_revents |= events;
2638 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2639 	}
2640 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2641 }
2642 
2643 /*
2644  * Wake up anyone polling on vp because it is being revoked.
2645  * This depends on dead_poll() returning POLLHUP for correct
2646  * behavior.
2647  */
2648 void
2649 vn_pollgone(vp)
2650 	struct vnode *vp;
2651 {
2652 	simple_lock(&vp->v_pollinfo.vpi_lock);
2653 	if (vp->v_pollinfo.vpi_events) {
2654 		vp->v_pollinfo.vpi_events = 0;
2655 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2656 	}
2657 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2658 }
2659 
2660 
2661 
2662 /*
2663  * Routine to create and manage a filesystem syncer vnode.
2664  */
2665 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2666 static int	sync_fsync __P((struct  vop_fsync_args *));
2667 static int	sync_inactive __P((struct  vop_inactive_args *));
2668 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2669 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2670 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2671 static int	sync_print __P((struct vop_print_args *));
2672 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2673 
2674 static vop_t **sync_vnodeop_p;
2675 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2676 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2677 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2678 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2679 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2680 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2681 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2682 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2683 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2684 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2685 	{ NULL, NULL }
2686 };
2687 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2688 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2689 
2690 VNODEOP_SET(sync_vnodeop_opv_desc);
2691 
2692 /*
2693  * Create a new filesystem syncer vnode for the specified mount point.
2694  */
2695 int
2696 vfs_allocate_syncvnode(mp)
2697 	struct mount *mp;
2698 {
2699 	struct vnode *vp;
2700 	static long start, incr, next;
2701 	int error;
2702 
2703 	/* Allocate a new vnode */
2704 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2705 		mp->mnt_syncer = NULL;
2706 		return (error);
2707 	}
2708 	vp->v_type = VNON;
2709 	/*
2710 	 * Place the vnode onto the syncer worklist. We attempt to
2711 	 * scatter them about on the list so that they will go off
2712 	 * at evenly distributed times even if all the filesystems
2713 	 * are mounted at once.
2714 	 */
2715 	next += incr;
2716 	if (next == 0 || next > syncer_maxdelay) {
2717 		start /= 2;
2718 		incr /= 2;
2719 		if (start == 0) {
2720 			start = syncer_maxdelay / 2;
2721 			incr = syncer_maxdelay;
2722 		}
2723 		next = start;
2724 	}
2725 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2726 	mp->mnt_syncer = vp;
2727 	return (0);
2728 }
2729 
2730 /*
2731  * Do a lazy sync of the filesystem.
2732  */
2733 static int
2734 sync_fsync(ap)
2735 	struct vop_fsync_args /* {
2736 		struct vnode *a_vp;
2737 		struct ucred *a_cred;
2738 		int a_waitfor;
2739 		struct proc *a_p;
2740 	} */ *ap;
2741 {
2742 	struct vnode *syncvp = ap->a_vp;
2743 	struct mount *mp = syncvp->v_mount;
2744 	struct proc *p = ap->a_p;
2745 	int asyncflag;
2746 
2747 	/*
2748 	 * We only need to do something if this is a lazy evaluation.
2749 	 */
2750 	if (ap->a_waitfor != MNT_LAZY)
2751 		return (0);
2752 
2753 	/*
2754 	 * Move ourselves to the back of the sync list.
2755 	 */
2756 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2757 
2758 	/*
2759 	 * Walk the list of vnodes pushing all that are dirty and
2760 	 * not already on the sync list.
2761 	 */
2762 	simple_lock(&mountlist_slock);
2763 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) {
2764 		simple_unlock(&mountlist_slock);
2765 		return (0);
2766 	}
2767 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2768 	mp->mnt_flag &= ~MNT_ASYNC;
2769 	vfs_msync(mp, MNT_NOWAIT);
2770 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2771 	if (asyncflag)
2772 		mp->mnt_flag |= MNT_ASYNC;
2773 	vfs_unbusy(mp, p);
2774 	return (0);
2775 }
2776 
2777 /*
2778  * The syncer vnode is no referenced.
2779  */
2780 static int
2781 sync_inactive(ap)
2782 	struct vop_inactive_args /* {
2783 		struct vnode *a_vp;
2784 		struct proc *a_p;
2785 	} */ *ap;
2786 {
2787 
2788 	vgone(ap->a_vp);
2789 	return (0);
2790 }
2791 
2792 /*
2793  * The syncer vnode is no longer needed and is being decommissioned.
2794  */
2795 static int
2796 sync_reclaim(ap)
2797 	struct vop_reclaim_args /* {
2798 		struct vnode *a_vp;
2799 	} */ *ap;
2800 {
2801 	struct vnode *vp = ap->a_vp;
2802 
2803 	vp->v_mount->mnt_syncer = NULL;
2804 	if (vp->v_flag & VONWORKLST) {
2805 		LIST_REMOVE(vp, v_synclist);
2806 		vp->v_flag &= ~VONWORKLST;
2807 	}
2808 
2809 	return (0);
2810 }
2811 
2812 /*
2813  * Print out a syncer vnode.
2814  */
2815 static int
2816 sync_print(ap)
2817 	struct vop_print_args /* {
2818 		struct vnode *a_vp;
2819 	} */ *ap;
2820 {
2821 	struct vnode *vp = ap->a_vp;
2822 
2823 	printf("syncer vnode");
2824 	if (vp->v_vnlock != NULL)
2825 		lockmgr_printinfo(vp->v_vnlock);
2826 	printf("\n");
2827 	return (0);
2828 }
2829