xref: /freebsd/sys/kern/vfs_export.c (revision 5129159789cc9d7bc514e4546b88e3427695002d)
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  * $FreeBSD$
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/fcntl.h>
50 #include <sys/kernel.h>
51 #include <sys/proc.h>
52 #include <sys/kthread.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 #include <sys/conf.h>
63 
64 #include <machine/limits.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_pager.h>
73 #include <vm/vnode_pager.h>
74 #include <vm/vm_zone.h>
75 #include <sys/sysctl.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 static int reassignbufcalls;
104 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
105 static int reassignbufloops;
106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
107 static int reassignbufsortgood;
108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
109 static int reassignbufsortbad;
110 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
111 static int reassignbufmethod = 1;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
113 
114 #ifdef ENABLE_VFS_IOOPT
115 int vfs_ioopt = 0;
116 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
117 #endif
118 
119 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
120 struct simplelock mountlist_slock;
121 struct simplelock mntvnode_slock;
122 int	nfs_mount_type = -1;
123 #ifndef NULL_SIMPLELOCKS
124 static struct simplelock mntid_slock;
125 static struct simplelock vnode_free_list_slock;
126 static struct simplelock spechash_slock;
127 #endif
128 struct nfs_public nfs_pub;	/* publicly exported FS */
129 static vm_zone_t vnode_zone;
130 
131 /*
132  * The workitem queue.
133  */
134 #define SYNCER_MAXDELAY		32
135 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
136 time_t syncdelay = 30;		/* max time to delay syncing data */
137 time_t filedelay = 30;		/* time to delay syncing files */
138 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
139 time_t dirdelay = 29;		/* time to delay syncing directories */
140 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
141 time_t metadelay = 28;		/* time to delay syncing metadata */
142 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
143 static int rushjob;			/* number of slots to run ASAP */
144 static int stat_rush_requests;	/* number of times I/O speeded up */
145 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
146 
147 static int syncer_delayno = 0;
148 static long syncer_mask;
149 LIST_HEAD(synclist, vnode);
150 static struct synclist *syncer_workitem_pending;
151 
152 int desiredvnodes;
153 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
154     &desiredvnodes, 0, "Maximum number of vnodes");
155 
156 static void	vfs_free_addrlist __P((struct netexport *nep));
157 static int	vfs_free_netcred __P((struct radix_node *rn, void *w));
158 static int	vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
159 				       struct export_args *argp));
160 
161 /*
162  * Initialize the vnode management data structures.
163  */
164 void
165 vntblinit()
166 {
167 
168 	desiredvnodes = maxproc + cnt.v_page_count / 4;
169 	simple_lock_init(&mntvnode_slock);
170 	simple_lock_init(&mntid_slock);
171 	simple_lock_init(&spechash_slock);
172 	TAILQ_INIT(&vnode_free_list);
173 	TAILQ_INIT(&vnode_tobefree_list);
174 	simple_lock_init(&vnode_free_list_slock);
175 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
176 	/*
177 	 * Initialize the filesystem syncer.
178 	 */
179 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
180 		&syncer_mask);
181 	syncer_maxdelay = syncer_mask + 1;
182 }
183 
184 /*
185  * Mark a mount point as busy. Used to synchronize access and to delay
186  * unmounting. Interlock is not released on failure.
187  */
188 int
189 vfs_busy(mp, flags, interlkp, p)
190 	struct mount *mp;
191 	int flags;
192 	struct simplelock *interlkp;
193 	struct proc *p;
194 {
195 	int lkflags;
196 
197 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
198 		if (flags & LK_NOWAIT)
199 			return (ENOENT);
200 		mp->mnt_kern_flag |= MNTK_MWAIT;
201 		if (interlkp) {
202 			simple_unlock(interlkp);
203 		}
204 		/*
205 		 * Since all busy locks are shared except the exclusive
206 		 * lock granted when unmounting, the only place that a
207 		 * wakeup needs to be done is at the release of the
208 		 * exclusive lock at the end of dounmount.
209 		 */
210 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
211 		if (interlkp) {
212 			simple_lock(interlkp);
213 		}
214 		return (ENOENT);
215 	}
216 	lkflags = LK_SHARED | LK_NOPAUSE;
217 	if (interlkp)
218 		lkflags |= LK_INTERLOCK;
219 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
220 		panic("vfs_busy: unexpected lock failure");
221 	return (0);
222 }
223 
224 /*
225  * Free a busy filesystem.
226  */
227 void
228 vfs_unbusy(mp, p)
229 	struct mount *mp;
230 	struct proc *p;
231 {
232 
233 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
234 }
235 
236 /*
237  * Lookup a filesystem type, and if found allocate and initialize
238  * a mount structure for it.
239  *
240  * Devname is usually updated by mount(8) after booting.
241  */
242 int
243 vfs_rootmountalloc(fstypename, devname, mpp)
244 	char *fstypename;
245 	char *devname;
246 	struct mount **mpp;
247 {
248 	struct proc *p = curproc;	/* XXX */
249 	struct vfsconf *vfsp;
250 	struct mount *mp;
251 
252 	if (fstypename == NULL)
253 		return (ENODEV);
254 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
255 		if (!strcmp(vfsp->vfc_name, fstypename))
256 			break;
257 	if (vfsp == NULL)
258 		return (ENODEV);
259 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
260 	bzero((char *)mp, (u_long)sizeof(struct mount));
261 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
262 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
263 	LIST_INIT(&mp->mnt_vnodelist);
264 	mp->mnt_vfc = vfsp;
265 	mp->mnt_op = vfsp->vfc_vfsops;
266 	mp->mnt_flag = MNT_RDONLY;
267 	mp->mnt_vnodecovered = NULLVP;
268 	vfsp->vfc_refcount++;
269 	mp->mnt_iosize_max = DFLTPHYS;
270 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
271 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
272 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
273 	mp->mnt_stat.f_mntonname[0] = '/';
274 	mp->mnt_stat.f_mntonname[1] = 0;
275 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
276 	*mpp = mp;
277 	return (0);
278 }
279 
280 /*
281  * Find an appropriate filesystem to use for the root. If a filesystem
282  * has not been preselected, walk through the list of known filesystems
283  * trying those that have mountroot routines, and try them until one
284  * works or we have tried them all.
285  */
286 #ifdef notdef	/* XXX JH */
287 int
288 lite2_vfs_mountroot()
289 {
290 	struct vfsconf *vfsp;
291 	extern int (*lite2_mountroot) __P((void));
292 	int error;
293 
294 	if (lite2_mountroot != NULL)
295 		return ((*lite2_mountroot)());
296 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
297 		if (vfsp->vfc_mountroot == NULL)
298 			continue;
299 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
300 			return (0);
301 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
302 	}
303 	return (ENODEV);
304 }
305 #endif
306 
307 /*
308  * Lookup a mount point by filesystem identifier.
309  */
310 struct mount *
311 vfs_getvfs(fsid)
312 	fsid_t *fsid;
313 {
314 	register struct mount *mp;
315 
316 	simple_lock(&mountlist_slock);
317 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
318 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
319 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
320 			simple_unlock(&mountlist_slock);
321 			return (mp);
322 	    }
323 	}
324 	simple_unlock(&mountlist_slock);
325 	return ((struct mount *) 0);
326 }
327 
328 /*
329  * Get a new unique fsid
330  *
331  * Keep in mind that several mounts may be running in parallel,
332  * so always increment mntid_base even if lower numbers are available.
333  */
334 
335 static u_short mntid_base;
336 
337 void
338 vfs_getnewfsid(mp)
339 	struct mount *mp;
340 {
341 	fsid_t tfsid;
342 	int mtype;
343 
344 	simple_lock(&mntid_slock);
345 
346 	mtype = mp->mnt_vfc->vfc_typenum;
347 	for (;;) {
348 		tfsid.val[0] = makeudev(255, mtype + (mntid_base << 16));
349 		tfsid.val[1] = mtype;
350 		++mntid_base;
351 		if (vfs_getvfs(&tfsid) == NULL)
352 			break;
353 	}
354 
355 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
356 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
357 
358 	simple_unlock(&mntid_slock);
359 }
360 
361 /*
362  * Knob to control the precision of file timestamps:
363  *
364  *   0 = seconds only; nanoseconds zeroed.
365  *   1 = seconds and nanoseconds, accurate within 1/HZ.
366  *   2 = seconds and nanoseconds, truncated to microseconds.
367  * >=3 = seconds and nanoseconds, maximum precision.
368  */
369 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
370 
371 static int timestamp_precision = TSP_SEC;
372 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
373     &timestamp_precision, 0, "");
374 
375 /*
376  * Get a current timestamp.
377  */
378 void
379 vfs_timestamp(tsp)
380 	struct timespec *tsp;
381 {
382 	struct timeval tv;
383 
384 	switch (timestamp_precision) {
385 	case TSP_SEC:
386 		tsp->tv_sec = time_second;
387 		tsp->tv_nsec = 0;
388 		break;
389 	case TSP_HZ:
390 		getnanotime(tsp);
391 		break;
392 	case TSP_USEC:
393 		microtime(&tv);
394 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
395 		break;
396 	case TSP_NSEC:
397 	default:
398 		nanotime(tsp);
399 		break;
400 	}
401 }
402 
403 /*
404  * Set vnode attributes to VNOVAL
405  */
406 void
407 vattr_null(vap)
408 	register struct vattr *vap;
409 {
410 
411 	vap->va_type = VNON;
412 	vap->va_size = VNOVAL;
413 	vap->va_bytes = VNOVAL;
414 	vap->va_mode = VNOVAL;
415 	vap->va_nlink = VNOVAL;
416 	vap->va_uid = VNOVAL;
417 	vap->va_gid = VNOVAL;
418 	vap->va_fsid = VNOVAL;
419 	vap->va_fileid = VNOVAL;
420 	vap->va_blocksize = VNOVAL;
421 	vap->va_rdev = VNOVAL;
422 	vap->va_atime.tv_sec = VNOVAL;
423 	vap->va_atime.tv_nsec = VNOVAL;
424 	vap->va_mtime.tv_sec = VNOVAL;
425 	vap->va_mtime.tv_nsec = VNOVAL;
426 	vap->va_ctime.tv_sec = VNOVAL;
427 	vap->va_ctime.tv_nsec = VNOVAL;
428 	vap->va_flags = VNOVAL;
429 	vap->va_gen = VNOVAL;
430 	vap->va_vaflags = 0;
431 }
432 
433 /*
434  * Routines having to do with the management of the vnode table.
435  */
436 extern vop_t **dead_vnodeop_p;
437 
438 /*
439  * Return the next vnode from the free list.
440  */
441 int
442 getnewvnode(tag, mp, vops, vpp)
443 	enum vtagtype tag;
444 	struct mount *mp;
445 	vop_t **vops;
446 	struct vnode **vpp;
447 {
448 	int s;
449 	struct proc *p = curproc;	/* XXX */
450 	struct vnode *vp, *tvp, *nvp;
451 	vm_object_t object;
452 	TAILQ_HEAD(freelst, vnode) vnode_tmp_list;
453 
454 	/*
455 	 * We take the least recently used vnode from the freelist
456 	 * if we can get it and it has no cached pages, and no
457 	 * namecache entries are relative to it.
458 	 * Otherwise we allocate a new vnode
459 	 */
460 
461 	s = splbio();
462 	simple_lock(&vnode_free_list_slock);
463 	TAILQ_INIT(&vnode_tmp_list);
464 
465 	for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) {
466 		nvp = TAILQ_NEXT(vp, v_freelist);
467 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
468 		if (vp->v_flag & VAGE) {
469 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
470 		} else {
471 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
472 		}
473 		vp->v_flag &= ~(VTBFREE|VAGE);
474 		vp->v_flag |= VFREE;
475 		if (vp->v_usecount)
476 			panic("tobe free vnode isn't");
477 		freevnodes++;
478 	}
479 
480 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
481 		vp = NULL;
482 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
483 		/*
484 		 * XXX: this is only here to be backwards compatible
485 		 */
486 		vp = NULL;
487 	} else {
488 		for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) {
489 			nvp = TAILQ_NEXT(vp, v_freelist);
490 			if (!simple_lock_try(&vp->v_interlock))
491 				continue;
492 			if (vp->v_usecount)
493 				panic("free vnode isn't");
494 
495 			object = vp->v_object;
496 			if (object && (object->resident_page_count || object->ref_count)) {
497 				printf("object inconsistant state: RPC: %d, RC: %d\n",
498 					object->resident_page_count, object->ref_count);
499 				/* Don't recycle if it's caching some pages */
500 				TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
501 				TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist);
502 				continue;
503 			} else if (LIST_FIRST(&vp->v_cache_src)) {
504 				/* Don't recycle if active in the namecache */
505 				simple_unlock(&vp->v_interlock);
506 				continue;
507 			} else {
508 				break;
509 			}
510 		}
511 	}
512 
513 	for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) {
514 		nvp = TAILQ_NEXT(tvp, v_freelist);
515 		TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist);
516 		TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist);
517 		simple_unlock(&tvp->v_interlock);
518 	}
519 
520 	if (vp) {
521 		vp->v_flag |= VDOOMED;
522 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
523 		freevnodes--;
524 		simple_unlock(&vnode_free_list_slock);
525 		cache_purge(vp);
526 		vp->v_lease = NULL;
527 		if (vp->v_type != VBAD) {
528 			vgonel(vp, p);
529 		} else {
530 			simple_unlock(&vp->v_interlock);
531 		}
532 
533 #ifdef INVARIANTS
534 		{
535 			int s;
536 
537 			if (vp->v_data)
538 				panic("cleaned vnode isn't");
539 			s = splbio();
540 			if (vp->v_numoutput)
541 				panic("Clean vnode has pending I/O's");
542 			splx(s);
543 		}
544 #endif
545 		vp->v_flag = 0;
546 		vp->v_lastw = 0;
547 		vp->v_lasta = 0;
548 		vp->v_cstart = 0;
549 		vp->v_clen = 0;
550 		vp->v_socket = 0;
551 		vp->v_writecount = 0;	/* XXX */
552 	} else {
553 		simple_unlock(&vnode_free_list_slock);
554 		vp = (struct vnode *) zalloc(vnode_zone);
555 		bzero((char *) vp, sizeof *vp);
556 		simple_lock_init(&vp->v_interlock);
557 		vp->v_dd = vp;
558 		cache_purge(vp);
559 		LIST_INIT(&vp->v_cache_src);
560 		TAILQ_INIT(&vp->v_cache_dst);
561 		numvnodes++;
562 	}
563 
564 	TAILQ_INIT(&vp->v_cleanblkhd);
565 	TAILQ_INIT(&vp->v_dirtyblkhd);
566 	vp->v_type = VNON;
567 	vp->v_tag = tag;
568 	vp->v_op = vops;
569 	insmntque(vp, mp);
570 	*vpp = vp;
571 	vp->v_usecount = 1;
572 	vp->v_data = 0;
573 	splx(s);
574 
575 	vfs_object_create(vp, p, p->p_ucred);
576 	return (0);
577 }
578 
579 /*
580  * Move a vnode from one mount queue to another.
581  */
582 static void
583 insmntque(vp, mp)
584 	register struct vnode *vp;
585 	register struct mount *mp;
586 {
587 
588 	simple_lock(&mntvnode_slock);
589 	/*
590 	 * Delete from old mount point vnode list, if on one.
591 	 */
592 	if (vp->v_mount != NULL)
593 		LIST_REMOVE(vp, v_mntvnodes);
594 	/*
595 	 * Insert into list of vnodes for the new mount point, if available.
596 	 */
597 	if ((vp->v_mount = mp) == NULL) {
598 		simple_unlock(&mntvnode_slock);
599 		return;
600 	}
601 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
602 	simple_unlock(&mntvnode_slock);
603 }
604 
605 /*
606  * Update outstanding I/O count and do wakeup if requested.
607  */
608 void
609 vwakeup(bp)
610 	register struct buf *bp;
611 {
612 	register struct vnode *vp;
613 
614 	bp->b_flags &= ~B_WRITEINPROG;
615 	if ((vp = bp->b_vp)) {
616 		vp->v_numoutput--;
617 		if (vp->v_numoutput < 0)
618 			panic("vwakeup: neg numoutput");
619 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
620 			vp->v_flag &= ~VBWAIT;
621 			wakeup((caddr_t) &vp->v_numoutput);
622 		}
623 	}
624 }
625 
626 /*
627  * Flush out and invalidate all buffers associated with a vnode.
628  * Called with the underlying object locked.
629  */
630 int
631 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
632 	register struct vnode *vp;
633 	int flags;
634 	struct ucred *cred;
635 	struct proc *p;
636 	int slpflag, slptimeo;
637 {
638 	register struct buf *bp;
639 	struct buf *nbp, *blist;
640 	int s, error;
641 	vm_object_t object;
642 
643 	if (flags & V_SAVE) {
644 		s = splbio();
645 		while (vp->v_numoutput) {
646 			vp->v_flag |= VBWAIT;
647 			error = tsleep((caddr_t)&vp->v_numoutput,
648 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
649 			if (error) {
650 				splx(s);
651 				return (error);
652 			}
653 		}
654 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
655 			splx(s);
656 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
657 				return (error);
658 			s = splbio();
659 			if (vp->v_numoutput > 0 ||
660 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
661 				panic("vinvalbuf: dirty bufs");
662 		}
663 		splx(s);
664   	}
665 	s = splbio();
666 	for (;;) {
667 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
668 		if (!blist)
669 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
670 		if (!blist)
671 			break;
672 
673 		for (bp = blist; bp; bp = nbp) {
674 			nbp = TAILQ_NEXT(bp, b_vnbufs);
675 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
676 				error = BUF_TIMELOCK(bp,
677 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
678 				    "vinvalbuf", slpflag, slptimeo);
679 				if (error == ENOLCK)
680 					break;
681 				splx(s);
682 				return (error);
683 			}
684 			/*
685 			 * XXX Since there are no node locks for NFS, I
686 			 * believe there is a slight chance that a delayed
687 			 * write will occur while sleeping just above, so
688 			 * check for it.  Note that vfs_bio_awrite expects
689 			 * buffers to reside on a queue, while VOP_BWRITE and
690 			 * brelse do not.
691 			 */
692 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
693 				(flags & V_SAVE)) {
694 
695 				if (bp->b_vp == vp) {
696 					if (bp->b_flags & B_CLUSTEROK) {
697 						BUF_UNLOCK(bp);
698 						vfs_bio_awrite(bp);
699 					} else {
700 						bremfree(bp);
701 						bp->b_flags |= B_ASYNC;
702 						VOP_BWRITE(bp->b_vp, bp);
703 					}
704 				} else {
705 					bremfree(bp);
706 					(void) VOP_BWRITE(bp->b_vp, bp);
707 				}
708 				break;
709 			}
710 			bremfree(bp);
711 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
712 			bp->b_flags &= ~B_ASYNC;
713 			brelse(bp);
714 		}
715 	}
716 
717 	while (vp->v_numoutput > 0) {
718 		vp->v_flag |= VBWAIT;
719 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
720 	}
721 
722 	splx(s);
723 
724 	/*
725 	 * Destroy the copy in the VM cache, too.
726 	 */
727 	simple_lock(&vp->v_interlock);
728 	object = vp->v_object;
729 	if (object != NULL) {
730 		vm_object_page_remove(object, 0, 0,
731 			(flags & V_SAVE) ? TRUE : FALSE);
732 	}
733 	simple_unlock(&vp->v_interlock);
734 
735 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
736 		panic("vinvalbuf: flush failed");
737 	return (0);
738 }
739 
740 /*
741  * Truncate a file's buffer and pages to a specified length.  This
742  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
743  * sync activity.
744  */
745 int
746 vtruncbuf(vp, cred, p, length, blksize)
747 	register struct vnode *vp;
748 	struct ucred *cred;
749 	struct proc *p;
750 	off_t length;
751 	int blksize;
752 {
753 	register struct buf *bp;
754 	struct buf *nbp;
755 	int s, anyfreed;
756 	int trunclbn;
757 
758 	/*
759 	 * Round up to the *next* lbn.
760 	 */
761 	trunclbn = (length + blksize - 1) / blksize;
762 
763 	s = splbio();
764 restart:
765 	anyfreed = 1;
766 	for (;anyfreed;) {
767 		anyfreed = 0;
768 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
769 			nbp = TAILQ_NEXT(bp, b_vnbufs);
770 			if (bp->b_lblkno >= trunclbn) {
771 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
772 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
773 					goto restart;
774 				} else {
775 					bremfree(bp);
776 					bp->b_flags |= (B_INVAL | B_RELBUF);
777 					bp->b_flags &= ~B_ASYNC;
778 					brelse(bp);
779 					anyfreed = 1;
780 				}
781 				if (nbp && (((nbp->b_xflags & B_VNCLEAN) == 0)||
782 					 (nbp->b_vp != vp) ||
783 					 (nbp->b_flags & B_DELWRI))) {
784 					goto restart;
785 				}
786 			}
787 		}
788 
789 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
790 			nbp = TAILQ_NEXT(bp, b_vnbufs);
791 			if (bp->b_lblkno >= trunclbn) {
792 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
793 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
794 					goto restart;
795 				} else {
796 					bremfree(bp);
797 					bp->b_flags |= (B_INVAL | B_RELBUF);
798 					bp->b_flags &= ~B_ASYNC;
799 					brelse(bp);
800 					anyfreed = 1;
801 				}
802 				if (nbp && (((nbp->b_xflags & B_VNDIRTY) == 0)||
803 					 (nbp->b_vp != vp) ||
804 					 (nbp->b_flags & B_DELWRI) == 0)) {
805 					goto restart;
806 				}
807 			}
808 		}
809 	}
810 
811 	if (length > 0) {
812 restartsync:
813 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
814 			nbp = TAILQ_NEXT(bp, b_vnbufs);
815 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
816 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
817 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
818 					goto restart;
819 				} else {
820 					bremfree(bp);
821 					if (bp->b_vp == vp) {
822 						bp->b_flags |= B_ASYNC;
823 					} else {
824 						bp->b_flags &= ~B_ASYNC;
825 					}
826 					VOP_BWRITE(bp->b_vp, bp);
827 				}
828 				goto restartsync;
829 			}
830 
831 		}
832 	}
833 
834 	while (vp->v_numoutput > 0) {
835 		vp->v_flag |= VBWAIT;
836 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
837 	}
838 
839 	splx(s);
840 
841 	vnode_pager_setsize(vp, length);
842 
843 	return (0);
844 }
845 
846 /*
847  * Associate a buffer with a vnode.
848  */
849 void
850 bgetvp(vp, bp)
851 	register struct vnode *vp;
852 	register struct buf *bp;
853 {
854 	int s;
855 
856 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
857 
858 	vhold(vp);
859 	bp->b_vp = vp;
860 	bp->b_dev = vn_todev(vp);
861 	/*
862 	 * Insert onto list for new vnode.
863 	 */
864 	s = splbio();
865 	bp->b_xflags |= B_VNCLEAN;
866 	bp->b_xflags &= ~B_VNDIRTY;
867 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
868 	splx(s);
869 }
870 
871 /*
872  * Disassociate a buffer from a vnode.
873  */
874 void
875 brelvp(bp)
876 	register struct buf *bp;
877 {
878 	struct vnode *vp;
879 	struct buflists *listheadp;
880 	int s;
881 
882 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
883 
884 	/*
885 	 * Delete from old vnode list, if on one.
886 	 */
887 	vp = bp->b_vp;
888 	s = splbio();
889 	if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
890 		if (bp->b_xflags & B_VNDIRTY)
891 			listheadp = &vp->v_dirtyblkhd;
892 		else
893 			listheadp = &vp->v_cleanblkhd;
894 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
895 		bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
896 	}
897 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
898 		vp->v_flag &= ~VONWORKLST;
899 		LIST_REMOVE(vp, v_synclist);
900 	}
901 	splx(s);
902 	bp->b_vp = (struct vnode *) 0;
903 	vdrop(vp);
904 }
905 
906 /*
907  * The workitem queue.
908  *
909  * It is useful to delay writes of file data and filesystem metadata
910  * for tens of seconds so that quickly created and deleted files need
911  * not waste disk bandwidth being created and removed. To realize this,
912  * we append vnodes to a "workitem" queue. When running with a soft
913  * updates implementation, most pending metadata dependencies should
914  * not wait for more than a few seconds. Thus, mounted on block devices
915  * are delayed only about a half the time that file data is delayed.
916  * Similarly, directory updates are more critical, so are only delayed
917  * about a third the time that file data is delayed. Thus, there are
918  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
919  * one each second (driven off the filesystem syncer process). The
920  * syncer_delayno variable indicates the next queue that is to be processed.
921  * Items that need to be processed soon are placed in this queue:
922  *
923  *	syncer_workitem_pending[syncer_delayno]
924  *
925  * A delay of fifteen seconds is done by placing the request fifteen
926  * entries later in the queue:
927  *
928  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
929  *
930  */
931 
932 /*
933  * Add an item to the syncer work queue.
934  */
935 static void
936 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
937 {
938 	int s, slot;
939 
940 	s = splbio();
941 
942 	if (vp->v_flag & VONWORKLST) {
943 		LIST_REMOVE(vp, v_synclist);
944 	}
945 
946 	if (delay > syncer_maxdelay - 2)
947 		delay = syncer_maxdelay - 2;
948 	slot = (syncer_delayno + delay) & syncer_mask;
949 
950 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
951 	vp->v_flag |= VONWORKLST;
952 	splx(s);
953 }
954 
955 struct  proc *updateproc;
956 static void sched_sync __P((void));
957 static struct kproc_desc up_kp = {
958 	"syncer",
959 	sched_sync,
960 	&updateproc
961 };
962 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
963 
964 /*
965  * System filesystem synchronizer daemon.
966  */
967 void
968 sched_sync(void)
969 {
970 	struct synclist *slp;
971 	struct vnode *vp;
972 	long starttime;
973 	int s;
974 	struct proc *p = updateproc;
975 
976 	p->p_flag |= P_BUFEXHAUST;
977 
978 	for (;;) {
979 		starttime = time_second;
980 
981 		/*
982 		 * Push files whose dirty time has expired.  Be careful
983 		 * of interrupt race on slp queue.
984 		 */
985 		s = splbio();
986 		slp = &syncer_workitem_pending[syncer_delayno];
987 		syncer_delayno += 1;
988 		if (syncer_delayno == syncer_maxdelay)
989 			syncer_delayno = 0;
990 		splx(s);
991 
992 		while ((vp = LIST_FIRST(slp)) != NULL) {
993 			if (VOP_ISLOCKED(vp, NULL) == 0) {
994 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
995 				(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
996 				VOP_UNLOCK(vp, 0, p);
997 			}
998 			s = splbio();
999 			if (LIST_FIRST(slp) == vp) {
1000 				/*
1001 				 * Note: v_tag VT_VFS vps can remain on the
1002 				 * worklist too with no dirty blocks, but
1003 				 * since sync_fsync() moves it to a different
1004 				 * slot we are safe.
1005 				 */
1006 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1007 				    !vn_isdisk(vp))
1008 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1009 				/*
1010 				 * Put us back on the worklist.  The worklist
1011 				 * routine will remove us from our current
1012 				 * position and then add us back in at a later
1013 				 * position.
1014 				 */
1015 				vn_syncer_add_to_worklist(vp, syncdelay);
1016 			}
1017 			splx(s);
1018 		}
1019 
1020 		/*
1021 		 * Do soft update processing.
1022 		 */
1023 		if (bioops.io_sync)
1024 			(*bioops.io_sync)(NULL);
1025 
1026 		/*
1027 		 * The variable rushjob allows the kernel to speed up the
1028 		 * processing of the filesystem syncer process. A rushjob
1029 		 * value of N tells the filesystem syncer to process the next
1030 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1031 		 * is used by the soft update code to speed up the filesystem
1032 		 * syncer process when the incore state is getting so far
1033 		 * ahead of the disk that the kernel memory pool is being
1034 		 * threatened with exhaustion.
1035 		 */
1036 		if (rushjob > 0) {
1037 			rushjob -= 1;
1038 			continue;
1039 		}
1040 		/*
1041 		 * If it has taken us less than a second to process the
1042 		 * current work, then wait. Otherwise start right over
1043 		 * again. We can still lose time if any single round
1044 		 * takes more than two seconds, but it does not really
1045 		 * matter as we are just trying to generally pace the
1046 		 * filesystem activity.
1047 		 */
1048 		if (time_second == starttime)
1049 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1050 	}
1051 }
1052 
1053 /*
1054  * Request the syncer daemon to speed up its work.
1055  * We never push it to speed up more than half of its
1056  * normal turn time, otherwise it could take over the cpu.
1057  */
1058 int
1059 speedup_syncer()
1060 {
1061 	int s;
1062 
1063 	s = splhigh();
1064 	if (updateproc->p_wchan == &lbolt)
1065 		setrunnable(updateproc);
1066 	splx(s);
1067 	if (rushjob < syncdelay / 2) {
1068 		rushjob += 1;
1069 		stat_rush_requests += 1;
1070 		return (1);
1071 	}
1072 	return(0);
1073 }
1074 
1075 /*
1076  * Associate a p-buffer with a vnode.
1077  *
1078  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1079  * with the buffer.  i.e. the bp has not been linked into the vnode or
1080  * ref-counted.
1081  */
1082 void
1083 pbgetvp(vp, bp)
1084 	register struct vnode *vp;
1085 	register struct buf *bp;
1086 {
1087 
1088 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1089 
1090 	bp->b_vp = vp;
1091 	bp->b_flags |= B_PAGING;
1092 	bp->b_dev = vn_todev(vp);
1093 }
1094 
1095 /*
1096  * Disassociate a p-buffer from a vnode.
1097  */
1098 void
1099 pbrelvp(bp)
1100 	register struct buf *bp;
1101 {
1102 
1103 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1104 
1105 #if !defined(MAX_PERF)
1106 	/* XXX REMOVE ME */
1107 	if (bp->b_vnbufs.tqe_next != NULL) {
1108 		panic(
1109 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1110 		    bp,
1111 		    (int)bp->b_flags
1112 		);
1113 	}
1114 #endif
1115 	bp->b_vp = (struct vnode *) 0;
1116 	bp->b_flags &= ~B_PAGING;
1117 }
1118 
1119 void
1120 pbreassignbuf(bp, newvp)
1121 	struct buf *bp;
1122 	struct vnode *newvp;
1123 {
1124 #if !defined(MAX_PERF)
1125 	if ((bp->b_flags & B_PAGING) == 0) {
1126 		panic(
1127 		    "pbreassignbuf() on non phys bp %p",
1128 		    bp
1129 		);
1130 	}
1131 #endif
1132 	bp->b_vp = newvp;
1133 }
1134 
1135 /*
1136  * Reassign a buffer from one vnode to another.
1137  * Used to assign file specific control information
1138  * (indirect blocks) to the vnode to which they belong.
1139  */
1140 void
1141 reassignbuf(bp, newvp)
1142 	register struct buf *bp;
1143 	register struct vnode *newvp;
1144 {
1145 	struct buflists *listheadp;
1146 	int delay;
1147 	int s;
1148 
1149 	if (newvp == NULL) {
1150 		printf("reassignbuf: NULL");
1151 		return;
1152 	}
1153 	++reassignbufcalls;
1154 
1155 #if !defined(MAX_PERF)
1156 	/*
1157 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1158 	 * is not fully linked in.
1159 	 */
1160 	if (bp->b_flags & B_PAGING)
1161 		panic("cannot reassign paging buffer");
1162 #endif
1163 
1164 	s = splbio();
1165 	/*
1166 	 * Delete from old vnode list, if on one.
1167 	 */
1168 	if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
1169 		if (bp->b_xflags & B_VNDIRTY)
1170 			listheadp = &bp->b_vp->v_dirtyblkhd;
1171 		else
1172 			listheadp = &bp->b_vp->v_cleanblkhd;
1173 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1174 		bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
1175 		if (bp->b_vp != newvp) {
1176 			vdrop(bp->b_vp);
1177 			bp->b_vp = NULL;	/* for clarification */
1178 		}
1179 	}
1180 	/*
1181 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1182 	 * of clean buffers.
1183 	 */
1184 	if (bp->b_flags & B_DELWRI) {
1185 		struct buf *tbp;
1186 
1187 		listheadp = &newvp->v_dirtyblkhd;
1188 		if ((newvp->v_flag & VONWORKLST) == 0) {
1189 			switch (newvp->v_type) {
1190 			case VDIR:
1191 				delay = dirdelay;
1192 				break;
1193 			case VCHR:
1194 			case VBLK:
1195 				if (newvp->v_specmountpoint != NULL) {
1196 					delay = metadelay;
1197 					break;
1198 				}
1199 				/* fall through */
1200 			default:
1201 				delay = filedelay;
1202 			}
1203 			vn_syncer_add_to_worklist(newvp, delay);
1204 		}
1205 		bp->b_xflags |= B_VNDIRTY;
1206 		tbp = TAILQ_FIRST(listheadp);
1207 		if (tbp == NULL ||
1208 		    bp->b_lblkno == 0 ||
1209 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1210 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1211 			++reassignbufsortgood;
1212 		} else if (bp->b_lblkno < 0) {
1213 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1214 			++reassignbufsortgood;
1215 		} else if (reassignbufmethod == 1) {
1216 			/*
1217 			 * New sorting algorithm, only handle sequential case,
1218 			 * otherwise guess.
1219 			 */
1220 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1221 			    (tbp->b_xflags & B_VNDIRTY)) {
1222 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1223 				++reassignbufsortgood;
1224 			} else {
1225 				TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1226 				++reassignbufsortbad;
1227 			}
1228 		} else {
1229 			/*
1230 			 * Old sorting algorithm, scan queue and insert
1231 			 */
1232 			struct buf *ttbp;
1233 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1234 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1235 				++reassignbufloops;
1236 				tbp = ttbp;
1237 			}
1238 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1239 		}
1240 	} else {
1241 		bp->b_xflags |= B_VNCLEAN;
1242 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1243 		if ((newvp->v_flag & VONWORKLST) &&
1244 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1245 			newvp->v_flag &= ~VONWORKLST;
1246 			LIST_REMOVE(newvp, v_synclist);
1247 		}
1248 	}
1249 	if (bp->b_vp != newvp) {
1250 		bp->b_vp = newvp;
1251 		vhold(bp->b_vp);
1252 	}
1253 	splx(s);
1254 }
1255 
1256 /*
1257  * Create a vnode for a block device.
1258  * Used for mounting the root file system.
1259  */
1260 int
1261 bdevvp(dev, vpp)
1262 	dev_t dev;
1263 	struct vnode **vpp;
1264 {
1265 	register struct vnode *vp;
1266 	struct vnode *nvp;
1267 	int error;
1268 
1269 	if (dev == NODEV) {
1270 		*vpp = NULLVP;
1271 		return (ENXIO);
1272 	}
1273 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1274 	if (error) {
1275 		*vpp = NULLVP;
1276 		return (error);
1277 	}
1278 	vp = nvp;
1279 	vp->v_type = VBLK;
1280 	addalias(vp, dev);
1281 	*vpp = vp;
1282 	return (0);
1283 }
1284 
1285 /*
1286  * Add vnode to the alias list hung off the dev_t.
1287  *
1288  * The reason for this gunk is that multiple vnodes can reference
1289  * the same physical device, so checking vp->v_usecount to see
1290  * how many users there are is inadequate; the v_usecount for
1291  * the vnodes need to be accumulated.  vcount() does that.
1292  */
1293 void
1294 addaliasu(nvp, nvp_rdev)
1295 	struct vnode *nvp;
1296 	udev_t nvp_rdev;
1297 {
1298 
1299 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1300 		panic("addaliasu on non-special vnode");
1301 	addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0));
1302 }
1303 
1304 void
1305 addalias(nvp, dev)
1306 	struct vnode *nvp;
1307 	dev_t dev;
1308 {
1309 
1310 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1311 		panic("addalias on non-special vnode");
1312 
1313 	nvp->v_rdev = dev;
1314 	simple_lock(&spechash_slock);
1315 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1316 	simple_unlock(&spechash_slock);
1317 }
1318 
1319 /*
1320  * Grab a particular vnode from the free list, increment its
1321  * reference count and lock it. The vnode lock bit is set if the
1322  * vnode is being eliminated in vgone. The process is awakened
1323  * when the transition is completed, and an error returned to
1324  * indicate that the vnode is no longer usable (possibly having
1325  * been changed to a new file system type).
1326  */
1327 int
1328 vget(vp, flags, p)
1329 	register struct vnode *vp;
1330 	int flags;
1331 	struct proc *p;
1332 {
1333 	int error;
1334 
1335 	/*
1336 	 * If the vnode is in the process of being cleaned out for
1337 	 * another use, we wait for the cleaning to finish and then
1338 	 * return failure. Cleaning is determined by checking that
1339 	 * the VXLOCK flag is set.
1340 	 */
1341 	if ((flags & LK_INTERLOCK) == 0) {
1342 		simple_lock(&vp->v_interlock);
1343 	}
1344 	if (vp->v_flag & VXLOCK) {
1345 		vp->v_flag |= VXWANT;
1346 		simple_unlock(&vp->v_interlock);
1347 		tsleep((caddr_t)vp, PINOD, "vget", 0);
1348 		return (ENOENT);
1349 	}
1350 
1351 	vp->v_usecount++;
1352 
1353 	if (VSHOULDBUSY(vp))
1354 		vbusy(vp);
1355 	if (flags & LK_TYPE_MASK) {
1356 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1357 			/*
1358 			 * must expand vrele here because we do not want
1359 			 * to call VOP_INACTIVE if the reference count
1360 			 * drops back to zero since it was never really
1361 			 * active. We must remove it from the free list
1362 			 * before sleeping so that multiple processes do
1363 			 * not try to recycle it.
1364 			 */
1365 			simple_lock(&vp->v_interlock);
1366 			vp->v_usecount--;
1367 			if (VSHOULDFREE(vp))
1368 				vfree(vp);
1369 			simple_unlock(&vp->v_interlock);
1370 		}
1371 		return (error);
1372 	}
1373 	simple_unlock(&vp->v_interlock);
1374 	return (0);
1375 }
1376 
1377 void
1378 vref(struct vnode *vp)
1379 {
1380 	simple_lock(&vp->v_interlock);
1381 	vp->v_usecount++;
1382 	simple_unlock(&vp->v_interlock);
1383 }
1384 
1385 /*
1386  * Vnode put/release.
1387  * If count drops to zero, call inactive routine and return to freelist.
1388  */
1389 void
1390 vrele(vp)
1391 	struct vnode *vp;
1392 {
1393 	struct proc *p = curproc;	/* XXX */
1394 
1395 	KASSERT(vp != NULL, ("vrele: null vp"));
1396 
1397 	simple_lock(&vp->v_interlock);
1398 
1399 	if (vp->v_usecount > 1) {
1400 
1401 		vp->v_usecount--;
1402 		simple_unlock(&vp->v_interlock);
1403 
1404 		return;
1405 	}
1406 
1407 	if (vp->v_usecount == 1) {
1408 
1409 		vp->v_usecount--;
1410 		if (VSHOULDFREE(vp))
1411 			vfree(vp);
1412 	/*
1413 	 * If we are doing a vput, the node is already locked, and we must
1414 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1415 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1416 	 */
1417 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1418 			VOP_INACTIVE(vp, p);
1419 		}
1420 
1421 	} else {
1422 #ifdef DIAGNOSTIC
1423 		vprint("vrele: negative ref count", vp);
1424 		simple_unlock(&vp->v_interlock);
1425 #endif
1426 		panic("vrele: negative ref cnt");
1427 	}
1428 }
1429 
1430 void
1431 vput(vp)
1432 	struct vnode *vp;
1433 {
1434 	struct proc *p = curproc;	/* XXX */
1435 
1436 	KASSERT(vp != NULL, ("vput: null vp"));
1437 
1438 	simple_lock(&vp->v_interlock);
1439 
1440 	if (vp->v_usecount > 1) {
1441 
1442 		vp->v_usecount--;
1443 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1444 		return;
1445 
1446 	}
1447 
1448 	if (vp->v_usecount == 1) {
1449 
1450 		vp->v_usecount--;
1451 		if (VSHOULDFREE(vp))
1452 			vfree(vp);
1453 	/*
1454 	 * If we are doing a vput, the node is already locked, and we must
1455 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1456 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1457 	 */
1458 		simple_unlock(&vp->v_interlock);
1459 		VOP_INACTIVE(vp, p);
1460 
1461 	} else {
1462 #ifdef DIAGNOSTIC
1463 		vprint("vput: negative ref count", vp);
1464 #endif
1465 		panic("vput: negative ref cnt");
1466 	}
1467 }
1468 
1469 /*
1470  * Somebody doesn't want the vnode recycled.
1471  */
1472 void
1473 vhold(vp)
1474 	register struct vnode *vp;
1475 {
1476 	int s;
1477 
1478   	s = splbio();
1479 	vp->v_holdcnt++;
1480 	if (VSHOULDBUSY(vp))
1481 		vbusy(vp);
1482 	splx(s);
1483 }
1484 
1485 /*
1486  * One less who cares about this vnode.
1487  */
1488 void
1489 vdrop(vp)
1490 	register struct vnode *vp;
1491 {
1492 	int s;
1493 
1494 	s = splbio();
1495 	if (vp->v_holdcnt <= 0)
1496 		panic("vdrop: holdcnt");
1497 	vp->v_holdcnt--;
1498 	if (VSHOULDFREE(vp))
1499 		vfree(vp);
1500 	splx(s);
1501 }
1502 
1503 /*
1504  * Remove any vnodes in the vnode table belonging to mount point mp.
1505  *
1506  * If MNT_NOFORCE is specified, there should not be any active ones,
1507  * return error if any are found (nb: this is a user error, not a
1508  * system error). If MNT_FORCE is specified, detach any active vnodes
1509  * that are found.
1510  */
1511 #ifdef DIAGNOSTIC
1512 static int busyprt = 0;		/* print out busy vnodes */
1513 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1514 #endif
1515 
1516 int
1517 vflush(mp, skipvp, flags)
1518 	struct mount *mp;
1519 	struct vnode *skipvp;
1520 	int flags;
1521 {
1522 	struct proc *p = curproc;	/* XXX */
1523 	struct vnode *vp, *nvp;
1524 	int busy = 0;
1525 
1526 	simple_lock(&mntvnode_slock);
1527 loop:
1528 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1529 		/*
1530 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1531 		 * Start over if it has (it won't be on the list anymore).
1532 		 */
1533 		if (vp->v_mount != mp)
1534 			goto loop;
1535 		nvp = LIST_NEXT(vp, v_mntvnodes);
1536 		/*
1537 		 * Skip over a selected vnode.
1538 		 */
1539 		if (vp == skipvp)
1540 			continue;
1541 
1542 		simple_lock(&vp->v_interlock);
1543 		/*
1544 		 * Skip over a vnodes marked VSYSTEM.
1545 		 */
1546 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1547 			simple_unlock(&vp->v_interlock);
1548 			continue;
1549 		}
1550 		/*
1551 		 * If WRITECLOSE is set, only flush out regular file vnodes
1552 		 * open for writing.
1553 		 */
1554 		if ((flags & WRITECLOSE) &&
1555 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1556 			simple_unlock(&vp->v_interlock);
1557 			continue;
1558 		}
1559 
1560 		/*
1561 		 * With v_usecount == 0, all we need to do is clear out the
1562 		 * vnode data structures and we are done.
1563 		 */
1564 		if (vp->v_usecount == 0) {
1565 			simple_unlock(&mntvnode_slock);
1566 			vgonel(vp, p);
1567 			simple_lock(&mntvnode_slock);
1568 			continue;
1569 		}
1570 
1571 		/*
1572 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1573 		 * or character devices, revert to an anonymous device. For
1574 		 * all other files, just kill them.
1575 		 */
1576 		if (flags & FORCECLOSE) {
1577 			simple_unlock(&mntvnode_slock);
1578 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1579 				vgonel(vp, p);
1580 			} else {
1581 				vclean(vp, 0, p);
1582 				vp->v_op = spec_vnodeop_p;
1583 				insmntque(vp, (struct mount *) 0);
1584 			}
1585 			simple_lock(&mntvnode_slock);
1586 			continue;
1587 		}
1588 #ifdef DIAGNOSTIC
1589 		if (busyprt)
1590 			vprint("vflush: busy vnode", vp);
1591 #endif
1592 		simple_unlock(&vp->v_interlock);
1593 		busy++;
1594 	}
1595 	simple_unlock(&mntvnode_slock);
1596 	if (busy)
1597 		return (EBUSY);
1598 	return (0);
1599 }
1600 
1601 /*
1602  * Disassociate the underlying file system from a vnode.
1603  */
1604 static void
1605 vclean(vp, flags, p)
1606 	struct vnode *vp;
1607 	int flags;
1608 	struct proc *p;
1609 {
1610 	int active;
1611 	vm_object_t obj;
1612 
1613 	/*
1614 	 * Check to see if the vnode is in use. If so we have to reference it
1615 	 * before we clean it out so that its count cannot fall to zero and
1616 	 * generate a race against ourselves to recycle it.
1617 	 */
1618 	if ((active = vp->v_usecount))
1619 		vp->v_usecount++;
1620 
1621 	/*
1622 	 * Prevent the vnode from being recycled or brought into use while we
1623 	 * clean it out.
1624 	 */
1625 	if (vp->v_flag & VXLOCK)
1626 		panic("vclean: deadlock");
1627 	vp->v_flag |= VXLOCK;
1628 	/*
1629 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1630 	 * have the object locked while it cleans it out. The VOP_LOCK
1631 	 * ensures that the VOP_INACTIVE routine is done with its work.
1632 	 * For active vnodes, it ensures that no other activity can
1633 	 * occur while the underlying object is being cleaned out.
1634 	 */
1635 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1636 
1637 	/*
1638 	 * Clean out any buffers associated with the vnode.
1639 	 */
1640 	vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1641 	if ((obj = vp->v_object) != NULL) {
1642 		if (obj->ref_count == 0) {
1643 			/*
1644 			 * vclean() may be called twice.  The first time removes the
1645 			 * primary reference to the object, the second time goes
1646 			 * one further and is a special-case to terminate the object.
1647 			 */
1648 			vm_object_terminate(obj);
1649 		} else {
1650 			/*
1651 			 * Woe to the process that tries to page now :-).
1652 			 */
1653 			vm_pager_deallocate(obj);
1654 		}
1655 	}
1656 
1657 	/*
1658 	 * If purging an active vnode, it must be closed and
1659 	 * deactivated before being reclaimed. Note that the
1660 	 * VOP_INACTIVE will unlock the vnode.
1661 	 */
1662 	if (active) {
1663 		if (flags & DOCLOSE)
1664 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1665 		VOP_INACTIVE(vp, p);
1666 	} else {
1667 		/*
1668 		 * Any other processes trying to obtain this lock must first
1669 		 * wait for VXLOCK to clear, then call the new lock operation.
1670 		 */
1671 		VOP_UNLOCK(vp, 0, p);
1672 	}
1673 	/*
1674 	 * Reclaim the vnode.
1675 	 */
1676 	if (VOP_RECLAIM(vp, p))
1677 		panic("vclean: cannot reclaim");
1678 
1679 	if (active)
1680 		vrele(vp);
1681 
1682 	cache_purge(vp);
1683 	if (vp->v_vnlock) {
1684 		FREE(vp->v_vnlock, M_VNODE);
1685 		vp->v_vnlock = NULL;
1686 	}
1687 
1688 	if (VSHOULDFREE(vp))
1689 		vfree(vp);
1690 
1691 	/*
1692 	 * Done with purge, notify sleepers of the grim news.
1693 	 */
1694 	vp->v_op = dead_vnodeop_p;
1695 	vn_pollgone(vp);
1696 	vp->v_tag = VT_NON;
1697 	vp->v_flag &= ~VXLOCK;
1698 	if (vp->v_flag & VXWANT) {
1699 		vp->v_flag &= ~VXWANT;
1700 		wakeup((caddr_t) vp);
1701 	}
1702 }
1703 
1704 /*
1705  * Eliminate all activity associated with the requested vnode
1706  * and with all vnodes aliased to the requested vnode.
1707  */
1708 int
1709 vop_revoke(ap)
1710 	struct vop_revoke_args /* {
1711 		struct vnode *a_vp;
1712 		int a_flags;
1713 	} */ *ap;
1714 {
1715 	struct vnode *vp, *vq;
1716 	dev_t dev;
1717 
1718 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1719 
1720 	vp = ap->a_vp;
1721 	/*
1722 	 * If a vgone (or vclean) is already in progress,
1723 	 * wait until it is done and return.
1724 	 */
1725 	if (vp->v_flag & VXLOCK) {
1726 		vp->v_flag |= VXWANT;
1727 		simple_unlock(&vp->v_interlock);
1728 		tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1729 		return (0);
1730 	}
1731 	dev = vp->v_rdev;
1732 	for (;;) {
1733 		simple_lock(&spechash_slock);
1734 		vq = SLIST_FIRST(&dev->si_hlist);
1735 		simple_unlock(&spechash_slock);
1736 		if (!vq)
1737 			break;
1738 		vgone(vq);
1739 	}
1740 	return (0);
1741 }
1742 
1743 /*
1744  * Recycle an unused vnode to the front of the free list.
1745  * Release the passed interlock if the vnode will be recycled.
1746  */
1747 int
1748 vrecycle(vp, inter_lkp, p)
1749 	struct vnode *vp;
1750 	struct simplelock *inter_lkp;
1751 	struct proc *p;
1752 {
1753 
1754 	simple_lock(&vp->v_interlock);
1755 	if (vp->v_usecount == 0) {
1756 		if (inter_lkp) {
1757 			simple_unlock(inter_lkp);
1758 		}
1759 		vgonel(vp, p);
1760 		return (1);
1761 	}
1762 	simple_unlock(&vp->v_interlock);
1763 	return (0);
1764 }
1765 
1766 /*
1767  * Eliminate all activity associated with a vnode
1768  * in preparation for reuse.
1769  */
1770 void
1771 vgone(vp)
1772 	register struct vnode *vp;
1773 {
1774 	struct proc *p = curproc;	/* XXX */
1775 
1776 	simple_lock(&vp->v_interlock);
1777 	vgonel(vp, p);
1778 }
1779 
1780 /*
1781  * vgone, with the vp interlock held.
1782  */
1783 static void
1784 vgonel(vp, p)
1785 	struct vnode *vp;
1786 	struct proc *p;
1787 {
1788 	int s;
1789 
1790 	/*
1791 	 * If a vgone (or vclean) is already in progress,
1792 	 * wait until it is done and return.
1793 	 */
1794 	if (vp->v_flag & VXLOCK) {
1795 		vp->v_flag |= VXWANT;
1796 		simple_unlock(&vp->v_interlock);
1797 		tsleep((caddr_t)vp, PINOD, "vgone", 0);
1798 		return;
1799 	}
1800 
1801 	/*
1802 	 * Clean out the filesystem specific data.
1803 	 */
1804 	vclean(vp, DOCLOSE, p);
1805 	simple_lock(&vp->v_interlock);
1806 
1807 	/*
1808 	 * Delete from old mount point vnode list, if on one.
1809 	 */
1810 	if (vp->v_mount != NULL)
1811 		insmntque(vp, (struct mount *)0);
1812 	/*
1813 	 * If special device, remove it from special device alias list
1814 	 * if it is on one.
1815 	 */
1816 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1817 		simple_lock(&spechash_slock);
1818 		SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
1819 		freedev(vp->v_rdev);
1820 		simple_unlock(&spechash_slock);
1821 		vp->v_rdev = NULL;
1822 	}
1823 
1824 	/*
1825 	 * If it is on the freelist and not already at the head,
1826 	 * move it to the head of the list. The test of the back
1827 	 * pointer and the reference count of zero is because
1828 	 * it will be removed from the free list by getnewvnode,
1829 	 * but will not have its reference count incremented until
1830 	 * after calling vgone. If the reference count were
1831 	 * incremented first, vgone would (incorrectly) try to
1832 	 * close the previous instance of the underlying object.
1833 	 */
1834 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1835 		s = splbio();
1836 		simple_lock(&vnode_free_list_slock);
1837 		if (vp->v_flag & VFREE) {
1838 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1839 		} else if (vp->v_flag & VTBFREE) {
1840 			TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
1841 			vp->v_flag &= ~VTBFREE;
1842 			freevnodes++;
1843 		} else
1844 			freevnodes++;
1845 		vp->v_flag |= VFREE;
1846 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1847 		simple_unlock(&vnode_free_list_slock);
1848 		splx(s);
1849 	}
1850 
1851 	vp->v_type = VBAD;
1852 	simple_unlock(&vp->v_interlock);
1853 }
1854 
1855 /*
1856  * Lookup a vnode by device number.
1857  */
1858 int
1859 vfinddev(dev, type, vpp)
1860 	dev_t dev;
1861 	enum vtype type;
1862 	struct vnode **vpp;
1863 {
1864 	struct vnode *vp;
1865 
1866 	simple_lock(&spechash_slock);
1867 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1868 		if (type == vp->v_type) {
1869 			*vpp = vp;
1870 			simple_unlock(&spechash_slock);
1871 			return (1);
1872 		}
1873 	}
1874 	simple_unlock(&spechash_slock);
1875 	return (0);
1876 }
1877 
1878 /*
1879  * Calculate the total number of references to a special device.
1880  */
1881 int
1882 vcount(vp)
1883 	struct vnode *vp;
1884 {
1885 	struct vnode *vq;
1886 	int count;
1887 
1888 	count = 0;
1889 	simple_lock(&spechash_slock);
1890 	SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
1891 		count += vq->v_usecount;
1892 	simple_unlock(&spechash_slock);
1893 	return (count);
1894 }
1895 
1896 /*
1897  * Print out a description of a vnode.
1898  */
1899 static char *typename[] =
1900 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1901 
1902 void
1903 vprint(label, vp)
1904 	char *label;
1905 	struct vnode *vp;
1906 {
1907 	char buf[96];
1908 
1909 	if (label != NULL)
1910 		printf("%s: %p: ", label, (void *)vp);
1911 	else
1912 		printf("%p: ", (void *)vp);
1913 	printf("type %s, usecount %d, writecount %d, refcount %d,",
1914 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1915 	    vp->v_holdcnt);
1916 	buf[0] = '\0';
1917 	if (vp->v_flag & VROOT)
1918 		strcat(buf, "|VROOT");
1919 	if (vp->v_flag & VTEXT)
1920 		strcat(buf, "|VTEXT");
1921 	if (vp->v_flag & VSYSTEM)
1922 		strcat(buf, "|VSYSTEM");
1923 	if (vp->v_flag & VXLOCK)
1924 		strcat(buf, "|VXLOCK");
1925 	if (vp->v_flag & VXWANT)
1926 		strcat(buf, "|VXWANT");
1927 	if (vp->v_flag & VBWAIT)
1928 		strcat(buf, "|VBWAIT");
1929 	if (vp->v_flag & VDOOMED)
1930 		strcat(buf, "|VDOOMED");
1931 	if (vp->v_flag & VFREE)
1932 		strcat(buf, "|VFREE");
1933 	if (vp->v_flag & VOBJBUF)
1934 		strcat(buf, "|VOBJBUF");
1935 	if (buf[0] != '\0')
1936 		printf(" flags (%s)", &buf[1]);
1937 	if (vp->v_data == NULL) {
1938 		printf("\n");
1939 	} else {
1940 		printf("\n\t");
1941 		VOP_PRINT(vp);
1942 	}
1943 }
1944 
1945 #ifdef DDB
1946 #include <ddb/ddb.h>
1947 /*
1948  * List all of the locked vnodes in the system.
1949  * Called when debugging the kernel.
1950  */
1951 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1952 {
1953 	struct proc *p = curproc;	/* XXX */
1954 	struct mount *mp, *nmp;
1955 	struct vnode *vp;
1956 
1957 	printf("Locked vnodes\n");
1958 	simple_lock(&mountlist_slock);
1959 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
1960 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
1961 			nmp = TAILQ_NEXT(mp, mnt_list);
1962 			continue;
1963 		}
1964 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1965 			if (VOP_ISLOCKED(vp, NULL))
1966 				vprint((char *)0, vp);
1967 		}
1968 		simple_lock(&mountlist_slock);
1969 		nmp = TAILQ_NEXT(mp, mnt_list);
1970 		vfs_unbusy(mp, p);
1971 	}
1972 	simple_unlock(&mountlist_slock);
1973 }
1974 #endif
1975 
1976 /*
1977  * Top level filesystem related information gathering.
1978  */
1979 static int	sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS);
1980 
1981 static int
1982 vfs_sysctl SYSCTL_HANDLER_ARGS
1983 {
1984 	int *name = (int *)arg1 - 1;	/* XXX */
1985 	u_int namelen = arg2 + 1;	/* XXX */
1986 	struct vfsconf *vfsp;
1987 
1988 #if 1 || defined(COMPAT_PRELITE2)
1989 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1990 	if (namelen == 1)
1991 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1992 #endif
1993 
1994 #ifdef notyet
1995 	/* all sysctl names at this level are at least name and field */
1996 	if (namelen < 2)
1997 		return (ENOTDIR);		/* overloaded */
1998 	if (name[0] != VFS_GENERIC) {
1999 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2000 			if (vfsp->vfc_typenum == name[0])
2001 				break;
2002 		if (vfsp == NULL)
2003 			return (EOPNOTSUPP);
2004 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2005 		    oldp, oldlenp, newp, newlen, p));
2006 	}
2007 #endif
2008 	switch (name[1]) {
2009 	case VFS_MAXTYPENUM:
2010 		if (namelen != 2)
2011 			return (ENOTDIR);
2012 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2013 	case VFS_CONF:
2014 		if (namelen != 3)
2015 			return (ENOTDIR);	/* overloaded */
2016 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2017 			if (vfsp->vfc_typenum == name[2])
2018 				break;
2019 		if (vfsp == NULL)
2020 			return (EOPNOTSUPP);
2021 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2022 	}
2023 	return (EOPNOTSUPP);
2024 }
2025 
2026 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2027 	"Generic filesystem");
2028 
2029 #if 1 || defined(COMPAT_PRELITE2)
2030 
2031 static int
2032 sysctl_ovfs_conf SYSCTL_HANDLER_ARGS
2033 {
2034 	int error;
2035 	struct vfsconf *vfsp;
2036 	struct ovfsconf ovfs;
2037 
2038 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2039 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2040 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2041 		ovfs.vfc_index = vfsp->vfc_typenum;
2042 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2043 		ovfs.vfc_flags = vfsp->vfc_flags;
2044 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2045 		if (error)
2046 			return error;
2047 	}
2048 	return 0;
2049 }
2050 
2051 #endif /* 1 || COMPAT_PRELITE2 */
2052 
2053 #if 0
2054 #define KINFO_VNODESLOP	10
2055 /*
2056  * Dump vnode list (via sysctl).
2057  * Copyout address of vnode followed by vnode.
2058  */
2059 /* ARGSUSED */
2060 static int
2061 sysctl_vnode SYSCTL_HANDLER_ARGS
2062 {
2063 	struct proc *p = curproc;	/* XXX */
2064 	struct mount *mp, *nmp;
2065 	struct vnode *nvp, *vp;
2066 	int error;
2067 
2068 #define VPTRSZ	sizeof (struct vnode *)
2069 #define VNODESZ	sizeof (struct vnode)
2070 
2071 	req->lock = 0;
2072 	if (!req->oldptr) /* Make an estimate */
2073 		return (SYSCTL_OUT(req, 0,
2074 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2075 
2076 	simple_lock(&mountlist_slock);
2077 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2078 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2079 			nmp = TAILQ_NEXT(mp, mnt_list);
2080 			continue;
2081 		}
2082 again:
2083 		simple_lock(&mntvnode_slock);
2084 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2085 		     vp != NULL;
2086 		     vp = nvp) {
2087 			/*
2088 			 * Check that the vp is still associated with
2089 			 * this filesystem.  RACE: could have been
2090 			 * recycled onto the same filesystem.
2091 			 */
2092 			if (vp->v_mount != mp) {
2093 				simple_unlock(&mntvnode_slock);
2094 				goto again;
2095 			}
2096 			nvp = LIST_NEXT(vp, v_mntvnodes);
2097 			simple_unlock(&mntvnode_slock);
2098 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2099 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2100 				return (error);
2101 			simple_lock(&mntvnode_slock);
2102 		}
2103 		simple_unlock(&mntvnode_slock);
2104 		simple_lock(&mountlist_slock);
2105 		nmp = TAILQ_NEXT(mp, mnt_list);
2106 		vfs_unbusy(mp, p);
2107 	}
2108 	simple_unlock(&mountlist_slock);
2109 
2110 	return (0);
2111 }
2112 #endif
2113 
2114 /*
2115  * XXX
2116  * Exporting the vnode list on large systems causes them to crash.
2117  * Exporting the vnode list on medium systems causes sysctl to coredump.
2118  */
2119 #if 0
2120 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2121 	0, 0, sysctl_vnode, "S,vnode", "");
2122 #endif
2123 
2124 /*
2125  * Check to see if a filesystem is mounted on a block device.
2126  */
2127 int
2128 vfs_mountedon(vp)
2129 	struct vnode *vp;
2130 {
2131 
2132 	if (vp->v_specmountpoint != NULL)
2133 		return (EBUSY);
2134 	return (0);
2135 }
2136 
2137 /*
2138  * Unmount all filesystems. The list is traversed in reverse order
2139  * of mounting to avoid dependencies.
2140  */
2141 void
2142 vfs_unmountall()
2143 {
2144 	struct mount *mp;
2145 	struct proc *p;
2146 	int error;
2147 
2148 	if (curproc != NULL)
2149 		p = curproc;
2150 	else
2151 		p = initproc;	/* XXX XXX should this be proc0? */
2152 	/*
2153 	 * Since this only runs when rebooting, it is not interlocked.
2154 	 */
2155 	while(!TAILQ_EMPTY(&mountlist)) {
2156 		mp = TAILQ_LAST(&mountlist, mntlist);
2157 		error = dounmount(mp, MNT_FORCE, p);
2158 		if (error) {
2159 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2160 			printf("unmount of %s failed (",
2161 			    mp->mnt_stat.f_mntonname);
2162 			if (error == EBUSY)
2163 				printf("BUSY)\n");
2164 			else
2165 				printf("%d)\n", error);
2166 		} else {
2167 			/* The unmount has removed mp from the mountlist */
2168 		}
2169 	}
2170 }
2171 
2172 /*
2173  * Build hash lists of net addresses and hang them off the mount point.
2174  * Called by ufs_mount() to set up the lists of export addresses.
2175  */
2176 static int
2177 vfs_hang_addrlist(mp, nep, argp)
2178 	struct mount *mp;
2179 	struct netexport *nep;
2180 	struct export_args *argp;
2181 {
2182 	register struct netcred *np;
2183 	register struct radix_node_head *rnh;
2184 	register int i;
2185 	struct radix_node *rn;
2186 	struct sockaddr *saddr, *smask = 0;
2187 	struct domain *dom;
2188 	int error;
2189 
2190 	if (argp->ex_addrlen == 0) {
2191 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2192 			return (EPERM);
2193 		np = &nep->ne_defexported;
2194 		np->netc_exflags = argp->ex_flags;
2195 		np->netc_anon = argp->ex_anon;
2196 		np->netc_anon.cr_ref = 1;
2197 		mp->mnt_flag |= MNT_DEFEXPORTED;
2198 		return (0);
2199 	}
2200 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2201 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2202 	bzero((caddr_t) np, i);
2203 	saddr = (struct sockaddr *) (np + 1);
2204 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2205 		goto out;
2206 	if (saddr->sa_len > argp->ex_addrlen)
2207 		saddr->sa_len = argp->ex_addrlen;
2208 	if (argp->ex_masklen) {
2209 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2210 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2211 		if (error)
2212 			goto out;
2213 		if (smask->sa_len > argp->ex_masklen)
2214 			smask->sa_len = argp->ex_masklen;
2215 	}
2216 	i = saddr->sa_family;
2217 	if ((rnh = nep->ne_rtable[i]) == 0) {
2218 		/*
2219 		 * Seems silly to initialize every AF when most are not used,
2220 		 * do so on demand here
2221 		 */
2222 		for (dom = domains; dom; dom = dom->dom_next)
2223 			if (dom->dom_family == i && dom->dom_rtattach) {
2224 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2225 				    dom->dom_rtoffset);
2226 				break;
2227 			}
2228 		if ((rnh = nep->ne_rtable[i]) == 0) {
2229 			error = ENOBUFS;
2230 			goto out;
2231 		}
2232 	}
2233 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2234 	    np->netc_rnodes);
2235 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2236 		error = EPERM;
2237 		goto out;
2238 	}
2239 	np->netc_exflags = argp->ex_flags;
2240 	np->netc_anon = argp->ex_anon;
2241 	np->netc_anon.cr_ref = 1;
2242 	return (0);
2243 out:
2244 	free(np, M_NETADDR);
2245 	return (error);
2246 }
2247 
2248 /* ARGSUSED */
2249 static int
2250 vfs_free_netcred(rn, w)
2251 	struct radix_node *rn;
2252 	void *w;
2253 {
2254 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
2255 
2256 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2257 	free((caddr_t) rn, M_NETADDR);
2258 	return (0);
2259 }
2260 
2261 /*
2262  * Free the net address hash lists that are hanging off the mount points.
2263  */
2264 static void
2265 vfs_free_addrlist(nep)
2266 	struct netexport *nep;
2267 {
2268 	register int i;
2269 	register struct radix_node_head *rnh;
2270 
2271 	for (i = 0; i <= AF_MAX; i++)
2272 		if ((rnh = nep->ne_rtable[i])) {
2273 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2274 			    (caddr_t) rnh);
2275 			free((caddr_t) rnh, M_RTABLE);
2276 			nep->ne_rtable[i] = 0;
2277 		}
2278 }
2279 
2280 int
2281 vfs_export(mp, nep, argp)
2282 	struct mount *mp;
2283 	struct netexport *nep;
2284 	struct export_args *argp;
2285 {
2286 	int error;
2287 
2288 	if (argp->ex_flags & MNT_DELEXPORT) {
2289 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2290 			vfs_setpublicfs(NULL, NULL, NULL);
2291 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2292 		}
2293 		vfs_free_addrlist(nep);
2294 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2295 	}
2296 	if (argp->ex_flags & MNT_EXPORTED) {
2297 		if (argp->ex_flags & MNT_EXPUBLIC) {
2298 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2299 				return (error);
2300 			mp->mnt_flag |= MNT_EXPUBLIC;
2301 		}
2302 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2303 			return (error);
2304 		mp->mnt_flag |= MNT_EXPORTED;
2305 	}
2306 	return (0);
2307 }
2308 
2309 
2310 /*
2311  * Set the publicly exported filesystem (WebNFS). Currently, only
2312  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2313  */
2314 int
2315 vfs_setpublicfs(mp, nep, argp)
2316 	struct mount *mp;
2317 	struct netexport *nep;
2318 	struct export_args *argp;
2319 {
2320 	int error;
2321 	struct vnode *rvp;
2322 	char *cp;
2323 
2324 	/*
2325 	 * mp == NULL -> invalidate the current info, the FS is
2326 	 * no longer exported. May be called from either vfs_export
2327 	 * or unmount, so check if it hasn't already been done.
2328 	 */
2329 	if (mp == NULL) {
2330 		if (nfs_pub.np_valid) {
2331 			nfs_pub.np_valid = 0;
2332 			if (nfs_pub.np_index != NULL) {
2333 				FREE(nfs_pub.np_index, M_TEMP);
2334 				nfs_pub.np_index = NULL;
2335 			}
2336 		}
2337 		return (0);
2338 	}
2339 
2340 	/*
2341 	 * Only one allowed at a time.
2342 	 */
2343 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2344 		return (EBUSY);
2345 
2346 	/*
2347 	 * Get real filehandle for root of exported FS.
2348 	 */
2349 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2350 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2351 
2352 	if ((error = VFS_ROOT(mp, &rvp)))
2353 		return (error);
2354 
2355 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2356 		return (error);
2357 
2358 	vput(rvp);
2359 
2360 	/*
2361 	 * If an indexfile was specified, pull it in.
2362 	 */
2363 	if (argp->ex_indexfile != NULL) {
2364 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2365 		    M_WAITOK);
2366 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2367 		    MAXNAMLEN, (size_t *)0);
2368 		if (!error) {
2369 			/*
2370 			 * Check for illegal filenames.
2371 			 */
2372 			for (cp = nfs_pub.np_index; *cp; cp++) {
2373 				if (*cp == '/') {
2374 					error = EINVAL;
2375 					break;
2376 				}
2377 			}
2378 		}
2379 		if (error) {
2380 			FREE(nfs_pub.np_index, M_TEMP);
2381 			return (error);
2382 		}
2383 	}
2384 
2385 	nfs_pub.np_mount = mp;
2386 	nfs_pub.np_valid = 1;
2387 	return (0);
2388 }
2389 
2390 struct netcred *
2391 vfs_export_lookup(mp, nep, nam)
2392 	register struct mount *mp;
2393 	struct netexport *nep;
2394 	struct sockaddr *nam;
2395 {
2396 	register struct netcred *np;
2397 	register struct radix_node_head *rnh;
2398 	struct sockaddr *saddr;
2399 
2400 	np = NULL;
2401 	if (mp->mnt_flag & MNT_EXPORTED) {
2402 		/*
2403 		 * Lookup in the export list first.
2404 		 */
2405 		if (nam != NULL) {
2406 			saddr = nam;
2407 			rnh = nep->ne_rtable[saddr->sa_family];
2408 			if (rnh != NULL) {
2409 				np = (struct netcred *)
2410 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2411 							      rnh);
2412 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2413 					np = NULL;
2414 			}
2415 		}
2416 		/*
2417 		 * If no address match, use the default if it exists.
2418 		 */
2419 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2420 			np = &nep->ne_defexported;
2421 	}
2422 	return (np);
2423 }
2424 
2425 /*
2426  * perform msync on all vnodes under a mount point
2427  * the mount point must be locked.
2428  */
2429 void
2430 vfs_msync(struct mount *mp, int flags) {
2431 	struct vnode *vp, *nvp;
2432 	struct vm_object *obj;
2433 	int anyio, tries;
2434 
2435 	tries = 5;
2436 loop:
2437 	anyio = 0;
2438 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
2439 
2440 		nvp = LIST_NEXT(vp, v_mntvnodes);
2441 
2442 		if (vp->v_mount != mp) {
2443 			goto loop;
2444 		}
2445 
2446 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2447 			continue;
2448 
2449 		if (flags != MNT_WAIT) {
2450 			obj = vp->v_object;
2451 			if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2452 				continue;
2453 			if (VOP_ISLOCKED(vp, NULL))
2454 				continue;
2455 		}
2456 
2457 		simple_lock(&vp->v_interlock);
2458 		if (vp->v_object &&
2459 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
2460 			if (!vget(vp,
2461 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2462 				if (vp->v_object) {
2463 					vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2464 					anyio = 1;
2465 				}
2466 				vput(vp);
2467 			}
2468 		} else {
2469 			simple_unlock(&vp->v_interlock);
2470 		}
2471 	}
2472 	if (anyio && (--tries > 0))
2473 		goto loop;
2474 }
2475 
2476 /*
2477  * Create the VM object needed for VMIO and mmap support.  This
2478  * is done for all VREG files in the system.  Some filesystems might
2479  * afford the additional metadata buffering capability of the
2480  * VMIO code by making the device node be VMIO mode also.
2481  *
2482  * vp must be locked when vfs_object_create is called.
2483  */
2484 int
2485 vfs_object_create(vp, p, cred)
2486 	struct vnode *vp;
2487 	struct proc *p;
2488 	struct ucred *cred;
2489 {
2490 	struct vattr vat;
2491 	vm_object_t object;
2492 	int error = 0;
2493 
2494 	if (!vn_isdisk(vp) && vn_canvmio(vp) == FALSE)
2495 		return 0;
2496 
2497 retry:
2498 	if ((object = vp->v_object) == NULL) {
2499 		if (vp->v_type == VREG || vp->v_type == VDIR) {
2500 			if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
2501 				goto retn;
2502 			object = vnode_pager_alloc(vp, vat.va_size, 0, 0);
2503 		} else if (devsw(vp->v_rdev) != NULL) {
2504 			/*
2505 			 * This simply allocates the biggest object possible
2506 			 * for a disk vnode.  This should be fixed, but doesn't
2507 			 * cause any problems (yet).
2508 			 */
2509 			object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0);
2510 		} else {
2511 			goto retn;
2512 		}
2513 		/*
2514 		 * Dereference the reference we just created.  This assumes
2515 		 * that the object is associated with the vp.
2516 		 */
2517 		object->ref_count--;
2518 		vp->v_usecount--;
2519 	} else {
2520 		if (object->flags & OBJ_DEAD) {
2521 			VOP_UNLOCK(vp, 0, p);
2522 			tsleep(object, PVM, "vodead", 0);
2523 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
2524 			goto retry;
2525 		}
2526 	}
2527 
2528 	KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object"));
2529 	vp->v_flag |= VOBJBUF;
2530 
2531 retn:
2532 	return error;
2533 }
2534 
2535 static void
2536 vfree(vp)
2537 	struct vnode *vp;
2538 {
2539 	int s;
2540 
2541 	s = splbio();
2542 	simple_lock(&vnode_free_list_slock);
2543 	if (vp->v_flag & VTBFREE) {
2544 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2545 		vp->v_flag &= ~VTBFREE;
2546 	}
2547 	if (vp->v_flag & VAGE) {
2548 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2549 	} else {
2550 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2551 	}
2552 	freevnodes++;
2553 	simple_unlock(&vnode_free_list_slock);
2554 	vp->v_flag &= ~VAGE;
2555 	vp->v_flag |= VFREE;
2556 	splx(s);
2557 }
2558 
2559 void
2560 vbusy(vp)
2561 	struct vnode *vp;
2562 {
2563 	int s;
2564 
2565 	s = splbio();
2566 	simple_lock(&vnode_free_list_slock);
2567 	if (vp->v_flag & VTBFREE) {
2568 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2569 		vp->v_flag &= ~VTBFREE;
2570 	} else {
2571 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2572 		freevnodes--;
2573 	}
2574 	simple_unlock(&vnode_free_list_slock);
2575 	vp->v_flag &= ~(VFREE|VAGE);
2576 	splx(s);
2577 }
2578 
2579 /*
2580  * Record a process's interest in events which might happen to
2581  * a vnode.  Because poll uses the historic select-style interface
2582  * internally, this routine serves as both the ``check for any
2583  * pending events'' and the ``record my interest in future events''
2584  * functions.  (These are done together, while the lock is held,
2585  * to avoid race conditions.)
2586  */
2587 int
2588 vn_pollrecord(vp, p, events)
2589 	struct vnode *vp;
2590 	struct proc *p;
2591 	short events;
2592 {
2593 	simple_lock(&vp->v_pollinfo.vpi_lock);
2594 	if (vp->v_pollinfo.vpi_revents & events) {
2595 		/*
2596 		 * This leaves events we are not interested
2597 		 * in available for the other process which
2598 		 * which presumably had requested them
2599 		 * (otherwise they would never have been
2600 		 * recorded).
2601 		 */
2602 		events &= vp->v_pollinfo.vpi_revents;
2603 		vp->v_pollinfo.vpi_revents &= ~events;
2604 
2605 		simple_unlock(&vp->v_pollinfo.vpi_lock);
2606 		return events;
2607 	}
2608 	vp->v_pollinfo.vpi_events |= events;
2609 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2610 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2611 	return 0;
2612 }
2613 
2614 /*
2615  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2616  * it is possible for us to miss an event due to race conditions, but
2617  * that condition is expected to be rare, so for the moment it is the
2618  * preferred interface.
2619  */
2620 void
2621 vn_pollevent(vp, events)
2622 	struct vnode *vp;
2623 	short events;
2624 {
2625 	simple_lock(&vp->v_pollinfo.vpi_lock);
2626 	if (vp->v_pollinfo.vpi_events & events) {
2627 		/*
2628 		 * We clear vpi_events so that we don't
2629 		 * call selwakeup() twice if two events are
2630 		 * posted before the polling process(es) is
2631 		 * awakened.  This also ensures that we take at
2632 		 * most one selwakeup() if the polling process
2633 		 * is no longer interested.  However, it does
2634 		 * mean that only one event can be noticed at
2635 		 * a time.  (Perhaps we should only clear those
2636 		 * event bits which we note?) XXX
2637 		 */
2638 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2639 		vp->v_pollinfo.vpi_revents |= events;
2640 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2641 	}
2642 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2643 }
2644 
2645 /*
2646  * Wake up anyone polling on vp because it is being revoked.
2647  * This depends on dead_poll() returning POLLHUP for correct
2648  * behavior.
2649  */
2650 void
2651 vn_pollgone(vp)
2652 	struct vnode *vp;
2653 {
2654 	simple_lock(&vp->v_pollinfo.vpi_lock);
2655 	if (vp->v_pollinfo.vpi_events) {
2656 		vp->v_pollinfo.vpi_events = 0;
2657 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2658 	}
2659 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2660 }
2661 
2662 
2663 
2664 /*
2665  * Routine to create and manage a filesystem syncer vnode.
2666  */
2667 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2668 static int	sync_fsync __P((struct  vop_fsync_args *));
2669 static int	sync_inactive __P((struct  vop_inactive_args *));
2670 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2671 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2672 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2673 static int	sync_print __P((struct vop_print_args *));
2674 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2675 
2676 static vop_t **sync_vnodeop_p;
2677 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2678 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2679 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2680 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2681 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2682 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2683 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2684 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2685 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2686 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2687 	{ NULL, NULL }
2688 };
2689 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2690 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2691 
2692 VNODEOP_SET(sync_vnodeop_opv_desc);
2693 
2694 /*
2695  * Create a new filesystem syncer vnode for the specified mount point.
2696  */
2697 int
2698 vfs_allocate_syncvnode(mp)
2699 	struct mount *mp;
2700 {
2701 	struct vnode *vp;
2702 	static long start, incr, next;
2703 	int error;
2704 
2705 	/* Allocate a new vnode */
2706 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2707 		mp->mnt_syncer = NULL;
2708 		return (error);
2709 	}
2710 	vp->v_type = VNON;
2711 	/*
2712 	 * Place the vnode onto the syncer worklist. We attempt to
2713 	 * scatter them about on the list so that they will go off
2714 	 * at evenly distributed times even if all the filesystems
2715 	 * are mounted at once.
2716 	 */
2717 	next += incr;
2718 	if (next == 0 || next > syncer_maxdelay) {
2719 		start /= 2;
2720 		incr /= 2;
2721 		if (start == 0) {
2722 			start = syncer_maxdelay / 2;
2723 			incr = syncer_maxdelay;
2724 		}
2725 		next = start;
2726 	}
2727 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2728 	mp->mnt_syncer = vp;
2729 	return (0);
2730 }
2731 
2732 /*
2733  * Do a lazy sync of the filesystem.
2734  */
2735 static int
2736 sync_fsync(ap)
2737 	struct vop_fsync_args /* {
2738 		struct vnode *a_vp;
2739 		struct ucred *a_cred;
2740 		int a_waitfor;
2741 		struct proc *a_p;
2742 	} */ *ap;
2743 {
2744 	struct vnode *syncvp = ap->a_vp;
2745 	struct mount *mp = syncvp->v_mount;
2746 	struct proc *p = ap->a_p;
2747 	int asyncflag;
2748 
2749 	/*
2750 	 * We only need to do something if this is a lazy evaluation.
2751 	 */
2752 	if (ap->a_waitfor != MNT_LAZY)
2753 		return (0);
2754 
2755 	/*
2756 	 * Move ourselves to the back of the sync list.
2757 	 */
2758 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2759 
2760 	/*
2761 	 * Walk the list of vnodes pushing all that are dirty and
2762 	 * not already on the sync list.
2763 	 */
2764 	simple_lock(&mountlist_slock);
2765 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) {
2766 		simple_unlock(&mountlist_slock);
2767 		return (0);
2768 	}
2769 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2770 	mp->mnt_flag &= ~MNT_ASYNC;
2771 	vfs_msync(mp, MNT_NOWAIT);
2772 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2773 	if (asyncflag)
2774 		mp->mnt_flag |= MNT_ASYNC;
2775 	vfs_unbusy(mp, p);
2776 	return (0);
2777 }
2778 
2779 /*
2780  * The syncer vnode is no referenced.
2781  */
2782 static int
2783 sync_inactive(ap)
2784 	struct vop_inactive_args /* {
2785 		struct vnode *a_vp;
2786 		struct proc *a_p;
2787 	} */ *ap;
2788 {
2789 
2790 	vgone(ap->a_vp);
2791 	return (0);
2792 }
2793 
2794 /*
2795  * The syncer vnode is no longer needed and is being decommissioned.
2796  *
2797  * Modifications to the worklist must be protected at splbio().
2798  */
2799 static int
2800 sync_reclaim(ap)
2801 	struct vop_reclaim_args /* {
2802 		struct vnode *a_vp;
2803 	} */ *ap;
2804 {
2805 	struct vnode *vp = ap->a_vp;
2806 	int s;
2807 
2808 	s = splbio();
2809 	vp->v_mount->mnt_syncer = NULL;
2810 	if (vp->v_flag & VONWORKLST) {
2811 		LIST_REMOVE(vp, v_synclist);
2812 		vp->v_flag &= ~VONWORKLST;
2813 	}
2814 	splx(s);
2815 
2816 	return (0);
2817 }
2818 
2819 /*
2820  * Print out a syncer vnode.
2821  */
2822 static int
2823 sync_print(ap)
2824 	struct vop_print_args /* {
2825 		struct vnode *a_vp;
2826 	} */ *ap;
2827 {
2828 	struct vnode *vp = ap->a_vp;
2829 
2830 	printf("syncer vnode");
2831 	if (vp->v_vnlock != NULL)
2832 		lockmgr_printinfo(vp->v_vnlock);
2833 	printf("\n");
2834 	return (0);
2835 }
2836 
2837 /*
2838  * extract the dev_t from a VBLK or VCHR
2839  */
2840 dev_t
2841 vn_todev(vp)
2842 	struct vnode *vp;
2843 {
2844 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2845 		return (NODEV);
2846 	return (vp->v_rdev);
2847 }
2848 
2849 /*
2850  * Check if vnode represents a disk device
2851  */
2852 int
2853 vn_isdisk(vp)
2854 	struct vnode *vp;
2855 {
2856 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2857 		return (0);
2858 	if (!devsw(vp->v_rdev))
2859 		return (0);
2860 	if (!(devsw(vp->v_rdev)->d_flags & D_DISK))
2861 		return (0);
2862 	return (1);
2863 }
2864 
2865