xref: /freebsd/sys/kern/vfs_subr.c (revision 23f282aa31e9b6fceacd449020e936e98d6f2298)
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/buf.h>
50 #include <sys/conf.h>
51 #include <sys/dirent.h>
52 #include <sys/domain.h>
53 #include <sys/eventhandler.h>
54 #include <sys/fcntl.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/namei.h>
60 #include <sys/proc.h>
61 #include <sys/reboot.h>
62 #include <sys/socket.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/vmmeter.h>
66 #include <sys/vnode.h>
67 
68 #include <machine/limits.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_extern.h>
73 #include <vm/pmap.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_pager.h>
77 #include <vm/vnode_pager.h>
78 #include <vm/vm_zone.h>
79 
80 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
81 
82 static void	insmntque __P((struct vnode *vp, struct mount *mp));
83 static void	vclean __P((struct vnode *vp, int flags, struct proc *p));
84 static void	vfree __P((struct vnode *));
85 static unsigned long	numvnodes;
86 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
87 
88 enum vtype iftovt_tab[16] = {
89 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
90 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
91 };
92 int vttoif_tab[9] = {
93 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
94 	S_IFSOCK, S_IFIFO, S_IFMT,
95 };
96 
97 static TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
98 struct tobefreelist vnode_tobefree_list;	/* vnode free list */
99 
100 static u_long wantfreevnodes = 25;
101 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
102 static u_long freevnodes = 0;
103 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
104 
105 static int reassignbufcalls;
106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
107 static int reassignbufloops;
108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
109 static int reassignbufsortgood;
110 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
111 static int reassignbufsortbad;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
113 static int reassignbufmethod = 1;
114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
115 
116 #ifdef ENABLE_VFS_IOOPT
117 int vfs_ioopt = 0;
118 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
119 #endif
120 
121 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
122 struct simplelock mountlist_slock;
123 struct simplelock mntvnode_slock;
124 int	nfs_mount_type = -1;
125 #ifndef NULL_SIMPLELOCKS
126 static struct simplelock mntid_slock;
127 static struct simplelock vnode_free_list_slock;
128 static struct simplelock spechash_slock;
129 #endif
130 struct nfs_public nfs_pub;	/* publicly exported FS */
131 static vm_zone_t vnode_zone;
132 
133 /*
134  * The workitem queue.
135  */
136 #define SYNCER_MAXDELAY		32
137 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
138 time_t syncdelay = 30;		/* max time to delay syncing data */
139 time_t filedelay = 30;		/* time to delay syncing files */
140 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
141 time_t dirdelay = 29;		/* time to delay syncing directories */
142 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
143 time_t metadelay = 28;		/* time to delay syncing metadata */
144 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
145 static int rushjob;			/* number of slots to run ASAP */
146 static int stat_rush_requests;	/* number of times I/O speeded up */
147 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
148 
149 static int syncer_delayno = 0;
150 static long syncer_mask;
151 LIST_HEAD(synclist, vnode);
152 static struct synclist *syncer_workitem_pending;
153 
154 int desiredvnodes;
155 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
156     &desiredvnodes, 0, "Maximum number of vnodes");
157 
158 static void	vfs_free_addrlist __P((struct netexport *nep));
159 static int	vfs_free_netcred __P((struct radix_node *rn, void *w));
160 static int	vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
161 				       struct export_args *argp));
162 
163 /*
164  * Initialize the vnode management data structures.
165  */
166 void
167 vntblinit()
168 {
169 
170 	desiredvnodes = maxproc + cnt.v_page_count / 4;
171 	simple_lock_init(&mntvnode_slock);
172 	simple_lock_init(&mntid_slock);
173 	simple_lock_init(&spechash_slock);
174 	TAILQ_INIT(&vnode_free_list);
175 	TAILQ_INIT(&vnode_tobefree_list);
176 	simple_lock_init(&vnode_free_list_slock);
177 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
178 	/*
179 	 * Initialize the filesystem syncer.
180 	 */
181 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
182 		&syncer_mask);
183 	syncer_maxdelay = syncer_mask + 1;
184 }
185 
186 /*
187  * Mark a mount point as busy. Used to synchronize access and to delay
188  * unmounting. Interlock is not released on failure.
189  */
190 int
191 vfs_busy(mp, flags, interlkp, p)
192 	struct mount *mp;
193 	int flags;
194 	struct simplelock *interlkp;
195 	struct proc *p;
196 {
197 	int lkflags;
198 
199 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
200 		if (flags & LK_NOWAIT)
201 			return (ENOENT);
202 		mp->mnt_kern_flag |= MNTK_MWAIT;
203 		if (interlkp) {
204 			simple_unlock(interlkp);
205 		}
206 		/*
207 		 * Since all busy locks are shared except the exclusive
208 		 * lock granted when unmounting, the only place that a
209 		 * wakeup needs to be done is at the release of the
210 		 * exclusive lock at the end of dounmount.
211 		 */
212 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
213 		if (interlkp) {
214 			simple_lock(interlkp);
215 		}
216 		return (ENOENT);
217 	}
218 	lkflags = LK_SHARED | LK_NOPAUSE;
219 	if (interlkp)
220 		lkflags |= LK_INTERLOCK;
221 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
222 		panic("vfs_busy: unexpected lock failure");
223 	return (0);
224 }
225 
226 /*
227  * Free a busy filesystem.
228  */
229 void
230 vfs_unbusy(mp, p)
231 	struct mount *mp;
232 	struct proc *p;
233 {
234 
235 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
236 }
237 
238 /*
239  * Lookup a filesystem type, and if found allocate and initialize
240  * a mount structure for it.
241  *
242  * Devname is usually updated by mount(8) after booting.
243  */
244 int
245 vfs_rootmountalloc(fstypename, devname, mpp)
246 	char *fstypename;
247 	char *devname;
248 	struct mount **mpp;
249 {
250 	struct proc *p = curproc;	/* XXX */
251 	struct vfsconf *vfsp;
252 	struct mount *mp;
253 
254 	if (fstypename == NULL)
255 		return (ENODEV);
256 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
257 		if (!strcmp(vfsp->vfc_name, fstypename))
258 			break;
259 	if (vfsp == NULL)
260 		return (ENODEV);
261 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
262 	bzero((char *)mp, (u_long)sizeof(struct mount));
263 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
264 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
265 	LIST_INIT(&mp->mnt_vnodelist);
266 	mp->mnt_vfc = vfsp;
267 	mp->mnt_op = vfsp->vfc_vfsops;
268 	mp->mnt_flag = MNT_RDONLY;
269 	mp->mnt_vnodecovered = NULLVP;
270 	vfsp->vfc_refcount++;
271 	mp->mnt_iosize_max = DFLTPHYS;
272 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
273 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
274 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
275 	mp->mnt_stat.f_mntonname[0] = '/';
276 	mp->mnt_stat.f_mntonname[1] = 0;
277 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
278 	*mpp = mp;
279 	return (0);
280 }
281 
282 /*
283  * Find an appropriate filesystem to use for the root. If a filesystem
284  * has not been preselected, walk through the list of known filesystems
285  * trying those that have mountroot routines, and try them until one
286  * works or we have tried them all.
287  */
288 #ifdef notdef	/* XXX JH */
289 int
290 lite2_vfs_mountroot()
291 {
292 	struct vfsconf *vfsp;
293 	extern int (*lite2_mountroot) __P((void));
294 	int error;
295 
296 	if (lite2_mountroot != NULL)
297 		return ((*lite2_mountroot)());
298 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
299 		if (vfsp->vfc_mountroot == NULL)
300 			continue;
301 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
302 			return (0);
303 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
304 	}
305 	return (ENODEV);
306 }
307 #endif
308 
309 /*
310  * Lookup a mount point by filesystem identifier.
311  */
312 struct mount *
313 vfs_getvfs(fsid)
314 	fsid_t *fsid;
315 {
316 	register struct mount *mp;
317 
318 	simple_lock(&mountlist_slock);
319 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
320 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
321 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
322 			simple_unlock(&mountlist_slock);
323 			return (mp);
324 	    }
325 	}
326 	simple_unlock(&mountlist_slock);
327 	return ((struct mount *) 0);
328 }
329 
330 /*
331  * Get a new unique fsid.  Try to make its val[0] unique, since this value
332  * will be used to create fake device numbers for stat().  Also try (but
333  * not so hard) make its val[0] unique mod 2^16, since some emulators only
334  * support 16-bit device numbers.  We end up with unique val[0]'s for the
335  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
336  *
337  * Keep in mind that several mounts may be running in parallel.  Starting
338  * the search one past where the previous search terminated is both a
339  * micro-optimization and a defense against returning the same fsid to
340  * different mounts.
341  */
342 void
343 vfs_getnewfsid(mp)
344 	struct mount *mp;
345 {
346 	static u_int16_t mntid_base;
347 	fsid_t tfsid;
348 	int mtype;
349 
350 	simple_lock(&mntid_slock);
351 	mtype = mp->mnt_vfc->vfc_typenum;
352 	tfsid.val[1] = mtype;
353 	mtype = (mtype & 0xFF) << 16;
354 	for (;;) {
355 		tfsid.val[0] = makeudev(255, mtype | mntid_base++);
356 		if (vfs_getvfs(&tfsid) == NULL)
357 			break;
358 	}
359 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
360 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
361 	simple_unlock(&mntid_slock);
362 }
363 
364 /*
365  * Knob to control the precision of file timestamps:
366  *
367  *   0 = seconds only; nanoseconds zeroed.
368  *   1 = seconds and nanoseconds, accurate within 1/HZ.
369  *   2 = seconds and nanoseconds, truncated to microseconds.
370  * >=3 = seconds and nanoseconds, maximum precision.
371  */
372 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
373 
374 static int timestamp_precision = TSP_SEC;
375 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
376     &timestamp_precision, 0, "");
377 
378 /*
379  * Get a current timestamp.
380  */
381 void
382 vfs_timestamp(tsp)
383 	struct timespec *tsp;
384 {
385 	struct timeval tv;
386 
387 	switch (timestamp_precision) {
388 	case TSP_SEC:
389 		tsp->tv_sec = time_second;
390 		tsp->tv_nsec = 0;
391 		break;
392 	case TSP_HZ:
393 		getnanotime(tsp);
394 		break;
395 	case TSP_USEC:
396 		microtime(&tv);
397 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
398 		break;
399 	case TSP_NSEC:
400 	default:
401 		nanotime(tsp);
402 		break;
403 	}
404 }
405 
406 /*
407  * Set vnode attributes to VNOVAL
408  */
409 void
410 vattr_null(vap)
411 	register struct vattr *vap;
412 {
413 
414 	vap->va_type = VNON;
415 	vap->va_size = VNOVAL;
416 	vap->va_bytes = VNOVAL;
417 	vap->va_mode = VNOVAL;
418 	vap->va_nlink = VNOVAL;
419 	vap->va_uid = VNOVAL;
420 	vap->va_gid = VNOVAL;
421 	vap->va_fsid = VNOVAL;
422 	vap->va_fileid = VNOVAL;
423 	vap->va_blocksize = VNOVAL;
424 	vap->va_rdev = VNOVAL;
425 	vap->va_atime.tv_sec = VNOVAL;
426 	vap->va_atime.tv_nsec = VNOVAL;
427 	vap->va_mtime.tv_sec = VNOVAL;
428 	vap->va_mtime.tv_nsec = VNOVAL;
429 	vap->va_ctime.tv_sec = VNOVAL;
430 	vap->va_ctime.tv_nsec = VNOVAL;
431 	vap->va_flags = VNOVAL;
432 	vap->va_gen = VNOVAL;
433 	vap->va_vaflags = 0;
434 }
435 
436 /*
437  * Routines having to do with the management of the vnode table.
438  */
439 extern vop_t **dead_vnodeop_p;
440 
441 /*
442  * Return the next vnode from the free list.
443  */
444 int
445 getnewvnode(tag, mp, vops, vpp)
446 	enum vtagtype tag;
447 	struct mount *mp;
448 	vop_t **vops;
449 	struct vnode **vpp;
450 {
451 	int s;
452 	struct proc *p = curproc;	/* XXX */
453 	struct vnode *vp, *tvp, *nvp;
454 	vm_object_t object;
455 	TAILQ_HEAD(freelst, vnode) vnode_tmp_list;
456 
457 	/*
458 	 * We take the least recently used vnode from the freelist
459 	 * if we can get it and it has no cached pages, and no
460 	 * namecache entries are relative to it.
461 	 * Otherwise we allocate a new vnode
462 	 */
463 
464 	s = splbio();
465 	simple_lock(&vnode_free_list_slock);
466 	TAILQ_INIT(&vnode_tmp_list);
467 
468 	for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) {
469 		nvp = TAILQ_NEXT(vp, v_freelist);
470 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
471 		if (vp->v_flag & VAGE) {
472 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
473 		} else {
474 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
475 		}
476 		vp->v_flag &= ~(VTBFREE|VAGE);
477 		vp->v_flag |= VFREE;
478 		if (vp->v_usecount)
479 			panic("tobe free vnode isn't");
480 		freevnodes++;
481 	}
482 
483 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
484 		vp = NULL;
485 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
486 		/*
487 		 * XXX: this is only here to be backwards compatible
488 		 */
489 		vp = NULL;
490 	} else {
491 		for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) {
492 			nvp = TAILQ_NEXT(vp, v_freelist);
493 			if (!simple_lock_try(&vp->v_interlock))
494 				continue;
495 			if (vp->v_usecount)
496 				panic("free vnode isn't");
497 
498 			object = vp->v_object;
499 			if (object && (object->resident_page_count || object->ref_count)) {
500 				printf("object inconsistant state: RPC: %d, RC: %d\n",
501 					object->resident_page_count, object->ref_count);
502 				/* Don't recycle if it's caching some pages */
503 				TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
504 				TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist);
505 				continue;
506 			} else if (LIST_FIRST(&vp->v_cache_src)) {
507 				/* Don't recycle if active in the namecache */
508 				simple_unlock(&vp->v_interlock);
509 				continue;
510 			} else {
511 				break;
512 			}
513 		}
514 	}
515 
516 	for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) {
517 		nvp = TAILQ_NEXT(tvp, v_freelist);
518 		TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist);
519 		TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist);
520 		simple_unlock(&tvp->v_interlock);
521 	}
522 
523 	if (vp) {
524 		vp->v_flag |= VDOOMED;
525 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
526 		freevnodes--;
527 		simple_unlock(&vnode_free_list_slock);
528 		cache_purge(vp);
529 		vp->v_lease = NULL;
530 		if (vp->v_type != VBAD) {
531 			vgonel(vp, p);
532 		} else {
533 			simple_unlock(&vp->v_interlock);
534 		}
535 
536 #ifdef INVARIANTS
537 		{
538 			int s;
539 
540 			if (vp->v_data)
541 				panic("cleaned vnode isn't");
542 			s = splbio();
543 			if (vp->v_numoutput)
544 				panic("Clean vnode has pending I/O's");
545 			splx(s);
546 		}
547 #endif
548 		vp->v_flag = 0;
549 		vp->v_lastw = 0;
550 		vp->v_lasta = 0;
551 		vp->v_cstart = 0;
552 		vp->v_clen = 0;
553 		vp->v_socket = 0;
554 		vp->v_writecount = 0;	/* XXX */
555 	} else {
556 		simple_unlock(&vnode_free_list_slock);
557 		vp = (struct vnode *) zalloc(vnode_zone);
558 		bzero((char *) vp, sizeof *vp);
559 		simple_lock_init(&vp->v_interlock);
560 		vp->v_dd = vp;
561 		cache_purge(vp);
562 		LIST_INIT(&vp->v_cache_src);
563 		TAILQ_INIT(&vp->v_cache_dst);
564 		numvnodes++;
565 	}
566 
567 	TAILQ_INIT(&vp->v_cleanblkhd);
568 	TAILQ_INIT(&vp->v_dirtyblkhd);
569 	vp->v_type = VNON;
570 	vp->v_tag = tag;
571 	vp->v_op = vops;
572 	insmntque(vp, mp);
573 	*vpp = vp;
574 	vp->v_usecount = 1;
575 	vp->v_data = 0;
576 	splx(s);
577 
578 	vfs_object_create(vp, p, p->p_ucred);
579 	return (0);
580 }
581 
582 /*
583  * Move a vnode from one mount queue to another.
584  */
585 static void
586 insmntque(vp, mp)
587 	register struct vnode *vp;
588 	register struct mount *mp;
589 {
590 
591 	simple_lock(&mntvnode_slock);
592 	/*
593 	 * Delete from old mount point vnode list, if on one.
594 	 */
595 	if (vp->v_mount != NULL)
596 		LIST_REMOVE(vp, v_mntvnodes);
597 	/*
598 	 * Insert into list of vnodes for the new mount point, if available.
599 	 */
600 	if ((vp->v_mount = mp) == NULL) {
601 		simple_unlock(&mntvnode_slock);
602 		return;
603 	}
604 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
605 	simple_unlock(&mntvnode_slock);
606 }
607 
608 /*
609  * Update outstanding I/O count and do wakeup if requested.
610  */
611 void
612 vwakeup(bp)
613 	register struct buf *bp;
614 {
615 	register struct vnode *vp;
616 
617 	bp->b_flags &= ~B_WRITEINPROG;
618 	if ((vp = bp->b_vp)) {
619 		vp->v_numoutput--;
620 		if (vp->v_numoutput < 0)
621 			panic("vwakeup: neg numoutput");
622 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
623 			vp->v_flag &= ~VBWAIT;
624 			wakeup((caddr_t) &vp->v_numoutput);
625 		}
626 	}
627 }
628 
629 /*
630  * Flush out and invalidate all buffers associated with a vnode.
631  * Called with the underlying object locked.
632  */
633 int
634 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
635 	register struct vnode *vp;
636 	int flags;
637 	struct ucred *cred;
638 	struct proc *p;
639 	int slpflag, slptimeo;
640 {
641 	register struct buf *bp;
642 	struct buf *nbp, *blist;
643 	int s, error;
644 	vm_object_t object;
645 
646 	if (flags & V_SAVE) {
647 		s = splbio();
648 		while (vp->v_numoutput) {
649 			vp->v_flag |= VBWAIT;
650 			error = tsleep((caddr_t)&vp->v_numoutput,
651 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
652 			if (error) {
653 				splx(s);
654 				return (error);
655 			}
656 		}
657 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
658 			splx(s);
659 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
660 				return (error);
661 			s = splbio();
662 			if (vp->v_numoutput > 0 ||
663 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
664 				panic("vinvalbuf: dirty bufs");
665 		}
666 		splx(s);
667   	}
668 	s = splbio();
669 	for (;;) {
670 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
671 		if (!blist)
672 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
673 		if (!blist)
674 			break;
675 
676 		for (bp = blist; bp; bp = nbp) {
677 			nbp = TAILQ_NEXT(bp, b_vnbufs);
678 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
679 				error = BUF_TIMELOCK(bp,
680 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
681 				    "vinvalbuf", slpflag, slptimeo);
682 				if (error == ENOLCK)
683 					break;
684 				splx(s);
685 				return (error);
686 			}
687 			/*
688 			 * XXX Since there are no node locks for NFS, I
689 			 * believe there is a slight chance that a delayed
690 			 * write will occur while sleeping just above, so
691 			 * check for it.  Note that vfs_bio_awrite expects
692 			 * buffers to reside on a queue, while VOP_BWRITE and
693 			 * brelse do not.
694 			 */
695 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
696 				(flags & V_SAVE)) {
697 
698 				if (bp->b_vp == vp) {
699 					if (bp->b_flags & B_CLUSTEROK) {
700 						BUF_UNLOCK(bp);
701 						vfs_bio_awrite(bp);
702 					} else {
703 						bremfree(bp);
704 						bp->b_flags |= B_ASYNC;
705 						BUF_WRITE(bp);
706 					}
707 				} else {
708 					bremfree(bp);
709 					(void) BUF_WRITE(bp);
710 				}
711 				break;
712 			}
713 			bremfree(bp);
714 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
715 			bp->b_flags &= ~B_ASYNC;
716 			brelse(bp);
717 		}
718 	}
719 
720 	while (vp->v_numoutput > 0) {
721 		vp->v_flag |= VBWAIT;
722 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
723 	}
724 
725 	splx(s);
726 
727 	/*
728 	 * Destroy the copy in the VM cache, too.
729 	 */
730 	simple_lock(&vp->v_interlock);
731 	object = vp->v_object;
732 	if (object != NULL) {
733 		vm_object_page_remove(object, 0, 0,
734 			(flags & V_SAVE) ? TRUE : FALSE);
735 	}
736 	simple_unlock(&vp->v_interlock);
737 
738 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
739 		panic("vinvalbuf: flush failed");
740 	return (0);
741 }
742 
743 /*
744  * Truncate a file's buffer and pages to a specified length.  This
745  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
746  * sync activity.
747  */
748 int
749 vtruncbuf(vp, cred, p, length, blksize)
750 	register struct vnode *vp;
751 	struct ucred *cred;
752 	struct proc *p;
753 	off_t length;
754 	int blksize;
755 {
756 	register struct buf *bp;
757 	struct buf *nbp;
758 	int s, anyfreed;
759 	int trunclbn;
760 
761 	/*
762 	 * Round up to the *next* lbn.
763 	 */
764 	trunclbn = (length + blksize - 1) / blksize;
765 
766 	s = splbio();
767 restart:
768 	anyfreed = 1;
769 	for (;anyfreed;) {
770 		anyfreed = 0;
771 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
772 			nbp = TAILQ_NEXT(bp, b_vnbufs);
773 			if (bp->b_lblkno >= trunclbn) {
774 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
775 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
776 					goto restart;
777 				} else {
778 					bremfree(bp);
779 					bp->b_flags |= (B_INVAL | B_RELBUF);
780 					bp->b_flags &= ~B_ASYNC;
781 					brelse(bp);
782 					anyfreed = 1;
783 				}
784 				if (nbp &&
785 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
786 				    (nbp->b_vp != vp) ||
787 				    (nbp->b_flags & B_DELWRI))) {
788 					goto restart;
789 				}
790 			}
791 		}
792 
793 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
794 			nbp = TAILQ_NEXT(bp, b_vnbufs);
795 			if (bp->b_lblkno >= trunclbn) {
796 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
797 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
798 					goto restart;
799 				} else {
800 					bremfree(bp);
801 					bp->b_flags |= (B_INVAL | B_RELBUF);
802 					bp->b_flags &= ~B_ASYNC;
803 					brelse(bp);
804 					anyfreed = 1;
805 				}
806 				if (nbp &&
807 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
808 				    (nbp->b_vp != vp) ||
809 				    (nbp->b_flags & B_DELWRI) == 0)) {
810 					goto restart;
811 				}
812 			}
813 		}
814 	}
815 
816 	if (length > 0) {
817 restartsync:
818 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
819 			nbp = TAILQ_NEXT(bp, b_vnbufs);
820 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
821 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
822 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
823 					goto restart;
824 				} else {
825 					bremfree(bp);
826 					if (bp->b_vp == vp) {
827 						bp->b_flags |= B_ASYNC;
828 					} else {
829 						bp->b_flags &= ~B_ASYNC;
830 					}
831 					BUF_WRITE(bp);
832 				}
833 				goto restartsync;
834 			}
835 
836 		}
837 	}
838 
839 	while (vp->v_numoutput > 0) {
840 		vp->v_flag |= VBWAIT;
841 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
842 	}
843 
844 	splx(s);
845 
846 	vnode_pager_setsize(vp, length);
847 
848 	return (0);
849 }
850 
851 /*
852  * Associate a buffer with a vnode.
853  */
854 void
855 bgetvp(vp, bp)
856 	register struct vnode *vp;
857 	register struct buf *bp;
858 {
859 	int s;
860 
861 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
862 
863 	vhold(vp);
864 	bp->b_vp = vp;
865 	bp->b_dev = vn_todev(vp);
866 	/*
867 	 * Insert onto list for new vnode.
868 	 */
869 	s = splbio();
870 	bp->b_xflags |= BX_VNCLEAN;
871 	bp->b_xflags &= ~BX_VNDIRTY;
872 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
873 	splx(s);
874 }
875 
876 /*
877  * Disassociate a buffer from a vnode.
878  */
879 void
880 brelvp(bp)
881 	register struct buf *bp;
882 {
883 	struct vnode *vp;
884 	struct buflists *listheadp;
885 	int s;
886 
887 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
888 
889 	/*
890 	 * Delete from old vnode list, if on one.
891 	 */
892 	vp = bp->b_vp;
893 	s = splbio();
894 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
895 		if (bp->b_xflags & BX_VNDIRTY)
896 			listheadp = &vp->v_dirtyblkhd;
897 		else
898 			listheadp = &vp->v_cleanblkhd;
899 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
900 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
901 	}
902 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
903 		vp->v_flag &= ~VONWORKLST;
904 		LIST_REMOVE(vp, v_synclist);
905 	}
906 	splx(s);
907 	bp->b_vp = (struct vnode *) 0;
908 	vdrop(vp);
909 }
910 
911 /*
912  * The workitem queue.
913  *
914  * It is useful to delay writes of file data and filesystem metadata
915  * for tens of seconds so that quickly created and deleted files need
916  * not waste disk bandwidth being created and removed. To realize this,
917  * we append vnodes to a "workitem" queue. When running with a soft
918  * updates implementation, most pending metadata dependencies should
919  * not wait for more than a few seconds. Thus, mounted on block devices
920  * are delayed only about a half the time that file data is delayed.
921  * Similarly, directory updates are more critical, so are only delayed
922  * about a third the time that file data is delayed. Thus, there are
923  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
924  * one each second (driven off the filesystem syncer process). The
925  * syncer_delayno variable indicates the next queue that is to be processed.
926  * Items that need to be processed soon are placed in this queue:
927  *
928  *	syncer_workitem_pending[syncer_delayno]
929  *
930  * A delay of fifteen seconds is done by placing the request fifteen
931  * entries later in the queue:
932  *
933  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
934  *
935  */
936 
937 /*
938  * Add an item to the syncer work queue.
939  */
940 static void
941 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
942 {
943 	int s, slot;
944 
945 	s = splbio();
946 
947 	if (vp->v_flag & VONWORKLST) {
948 		LIST_REMOVE(vp, v_synclist);
949 	}
950 
951 	if (delay > syncer_maxdelay - 2)
952 		delay = syncer_maxdelay - 2;
953 	slot = (syncer_delayno + delay) & syncer_mask;
954 
955 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
956 	vp->v_flag |= VONWORKLST;
957 	splx(s);
958 }
959 
960 struct  proc *updateproc;
961 static void sched_sync __P((void));
962 static struct kproc_desc up_kp = {
963 	"syncer",
964 	sched_sync,
965 	&updateproc
966 };
967 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
968 
969 /*
970  * System filesystem synchronizer daemon.
971  */
972 void
973 sched_sync(void)
974 {
975 	struct synclist *slp;
976 	struct vnode *vp;
977 	long starttime;
978 	int s;
979 	struct proc *p = updateproc;
980 
981 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, p,
982 	    SHUTDOWN_PRI_LAST);
983 
984 	for (;;) {
985 		kproc_suspend_loop(p);
986 
987 		starttime = time_second;
988 
989 		/*
990 		 * Push files whose dirty time has expired.  Be careful
991 		 * of interrupt race on slp queue.
992 		 */
993 		s = splbio();
994 		slp = &syncer_workitem_pending[syncer_delayno];
995 		syncer_delayno += 1;
996 		if (syncer_delayno == syncer_maxdelay)
997 			syncer_delayno = 0;
998 		splx(s);
999 
1000 		while ((vp = LIST_FIRST(slp)) != NULL) {
1001 			if (VOP_ISLOCKED(vp, NULL) == 0) {
1002 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1003 				(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
1004 				VOP_UNLOCK(vp, 0, p);
1005 			}
1006 			s = splbio();
1007 			if (LIST_FIRST(slp) == vp) {
1008 				/*
1009 				 * Note: v_tag VT_VFS vps can remain on the
1010 				 * worklist too with no dirty blocks, but
1011 				 * since sync_fsync() moves it to a different
1012 				 * slot we are safe.
1013 				 */
1014 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1015 				    !vn_isdisk(vp, NULL))
1016 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1017 				/*
1018 				 * Put us back on the worklist.  The worklist
1019 				 * routine will remove us from our current
1020 				 * position and then add us back in at a later
1021 				 * position.
1022 				 */
1023 				vn_syncer_add_to_worklist(vp, syncdelay);
1024 			}
1025 			splx(s);
1026 		}
1027 
1028 		/*
1029 		 * Do soft update processing.
1030 		 */
1031 		if (bioops.io_sync)
1032 			(*bioops.io_sync)(NULL);
1033 
1034 		/*
1035 		 * The variable rushjob allows the kernel to speed up the
1036 		 * processing of the filesystem syncer process. A rushjob
1037 		 * value of N tells the filesystem syncer to process the next
1038 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1039 		 * is used by the soft update code to speed up the filesystem
1040 		 * syncer process when the incore state is getting so far
1041 		 * ahead of the disk that the kernel memory pool is being
1042 		 * threatened with exhaustion.
1043 		 */
1044 		if (rushjob > 0) {
1045 			rushjob -= 1;
1046 			continue;
1047 		}
1048 		/*
1049 		 * If it has taken us less than a second to process the
1050 		 * current work, then wait. Otherwise start right over
1051 		 * again. We can still lose time if any single round
1052 		 * takes more than two seconds, but it does not really
1053 		 * matter as we are just trying to generally pace the
1054 		 * filesystem activity.
1055 		 */
1056 		if (time_second == starttime)
1057 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1058 	}
1059 }
1060 
1061 /*
1062  * Request the syncer daemon to speed up its work.
1063  * We never push it to speed up more than half of its
1064  * normal turn time, otherwise it could take over the cpu.
1065  */
1066 int
1067 speedup_syncer()
1068 {
1069 	int s;
1070 
1071 	s = splhigh();
1072 	if (updateproc->p_wchan == &lbolt)
1073 		setrunnable(updateproc);
1074 	splx(s);
1075 	if (rushjob < syncdelay / 2) {
1076 		rushjob += 1;
1077 		stat_rush_requests += 1;
1078 		return (1);
1079 	}
1080 	return(0);
1081 }
1082 
1083 /*
1084  * Associate a p-buffer with a vnode.
1085  *
1086  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1087  * with the buffer.  i.e. the bp has not been linked into the vnode or
1088  * ref-counted.
1089  */
1090 void
1091 pbgetvp(vp, bp)
1092 	register struct vnode *vp;
1093 	register struct buf *bp;
1094 {
1095 
1096 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1097 
1098 	bp->b_vp = vp;
1099 	bp->b_flags |= B_PAGING;
1100 	bp->b_dev = vn_todev(vp);
1101 }
1102 
1103 /*
1104  * Disassociate a p-buffer from a vnode.
1105  */
1106 void
1107 pbrelvp(bp)
1108 	register struct buf *bp;
1109 {
1110 
1111 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1112 
1113 	/* XXX REMOVE ME */
1114 	if (bp->b_vnbufs.tqe_next != NULL) {
1115 		panic(
1116 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1117 		    bp,
1118 		    (int)bp->b_flags
1119 		);
1120 	}
1121 	bp->b_vp = (struct vnode *) 0;
1122 	bp->b_flags &= ~B_PAGING;
1123 }
1124 
1125 void
1126 pbreassignbuf(bp, newvp)
1127 	struct buf *bp;
1128 	struct vnode *newvp;
1129 {
1130 	if ((bp->b_flags & B_PAGING) == 0) {
1131 		panic(
1132 		    "pbreassignbuf() on non phys bp %p",
1133 		    bp
1134 		);
1135 	}
1136 	bp->b_vp = newvp;
1137 }
1138 
1139 /*
1140  * Reassign a buffer from one vnode to another.
1141  * Used to assign file specific control information
1142  * (indirect blocks) to the vnode to which they belong.
1143  */
1144 void
1145 reassignbuf(bp, newvp)
1146 	register struct buf *bp;
1147 	register struct vnode *newvp;
1148 {
1149 	struct buflists *listheadp;
1150 	int delay;
1151 	int s;
1152 
1153 	if (newvp == NULL) {
1154 		printf("reassignbuf: NULL");
1155 		return;
1156 	}
1157 	++reassignbufcalls;
1158 
1159 	/*
1160 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1161 	 * is not fully linked in.
1162 	 */
1163 	if (bp->b_flags & B_PAGING)
1164 		panic("cannot reassign paging buffer");
1165 
1166 	s = splbio();
1167 	/*
1168 	 * Delete from old vnode list, if on one.
1169 	 */
1170 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1171 		if (bp->b_xflags & BX_VNDIRTY)
1172 			listheadp = &bp->b_vp->v_dirtyblkhd;
1173 		else
1174 			listheadp = &bp->b_vp->v_cleanblkhd;
1175 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1176 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1177 		if (bp->b_vp != newvp) {
1178 			vdrop(bp->b_vp);
1179 			bp->b_vp = NULL;	/* for clarification */
1180 		}
1181 	}
1182 	/*
1183 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1184 	 * of clean buffers.
1185 	 */
1186 	if (bp->b_flags & B_DELWRI) {
1187 		struct buf *tbp;
1188 
1189 		listheadp = &newvp->v_dirtyblkhd;
1190 		if ((newvp->v_flag & VONWORKLST) == 0) {
1191 			switch (newvp->v_type) {
1192 			case VDIR:
1193 				delay = dirdelay;
1194 				break;
1195 			case VCHR:
1196 			case VBLK:
1197 				if (newvp->v_specmountpoint != NULL) {
1198 					delay = metadelay;
1199 					break;
1200 				}
1201 				/* fall through */
1202 			default:
1203 				delay = filedelay;
1204 			}
1205 			vn_syncer_add_to_worklist(newvp, delay);
1206 		}
1207 		bp->b_xflags |= BX_VNDIRTY;
1208 		tbp = TAILQ_FIRST(listheadp);
1209 		if (tbp == NULL ||
1210 		    bp->b_lblkno == 0 ||
1211 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1212 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1213 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1214 			++reassignbufsortgood;
1215 		} else if (bp->b_lblkno < 0) {
1216 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1217 			++reassignbufsortgood;
1218 		} else if (reassignbufmethod == 1) {
1219 			/*
1220 			 * New sorting algorithm, only handle sequential case,
1221 			 * otherwise append to end (but before metadata)
1222 			 */
1223 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1224 			    (tbp->b_xflags & BX_VNDIRTY)) {
1225 				/*
1226 				 * Found the best place to insert the buffer
1227 				 */
1228 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1229 				++reassignbufsortgood;
1230 			} else {
1231 				/*
1232 				 * Missed, append to end, but before meta-data.
1233 				 * We know that the head buffer in the list is
1234 				 * not meta-data due to prior conditionals.
1235 				 *
1236 				 * Indirect effects:  NFS second stage write
1237 				 * tends to wind up here, giving maximum
1238 				 * distance between the unstable write and the
1239 				 * commit rpc.
1240 				 */
1241 				tbp = TAILQ_LAST(listheadp, buflists);
1242 				while (tbp && tbp->b_lblkno < 0)
1243 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1244 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1245 				++reassignbufsortbad;
1246 			}
1247 		} else {
1248 			/*
1249 			 * Old sorting algorithm, scan queue and insert
1250 			 */
1251 			struct buf *ttbp;
1252 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1253 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1254 				++reassignbufloops;
1255 				tbp = ttbp;
1256 			}
1257 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1258 		}
1259 	} else {
1260 		bp->b_xflags |= BX_VNCLEAN;
1261 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1262 		if ((newvp->v_flag & VONWORKLST) &&
1263 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1264 			newvp->v_flag &= ~VONWORKLST;
1265 			LIST_REMOVE(newvp, v_synclist);
1266 		}
1267 	}
1268 	if (bp->b_vp != newvp) {
1269 		bp->b_vp = newvp;
1270 		vhold(bp->b_vp);
1271 	}
1272 	splx(s);
1273 }
1274 
1275 /*
1276  * Create a vnode for a block device.
1277  * Used for mounting the root file system.
1278  */
1279 int
1280 bdevvp(dev, vpp)
1281 	dev_t dev;
1282 	struct vnode **vpp;
1283 {
1284 	register struct vnode *vp;
1285 	struct vnode *nvp;
1286 	int error;
1287 
1288 	if (dev == NODEV) {
1289 		*vpp = NULLVP;
1290 		return (ENXIO);
1291 	}
1292 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1293 	if (error) {
1294 		*vpp = NULLVP;
1295 		return (error);
1296 	}
1297 	vp = nvp;
1298 	vp->v_type = VBLK;
1299 	addalias(vp, dev);
1300 	*vpp = vp;
1301 	return (0);
1302 }
1303 
1304 /*
1305  * Add vnode to the alias list hung off the dev_t.
1306  *
1307  * The reason for this gunk is that multiple vnodes can reference
1308  * the same physical device, so checking vp->v_usecount to see
1309  * how many users there are is inadequate; the v_usecount for
1310  * the vnodes need to be accumulated.  vcount() does that.
1311  */
1312 void
1313 addaliasu(nvp, nvp_rdev)
1314 	struct vnode *nvp;
1315 	udev_t nvp_rdev;
1316 {
1317 
1318 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1319 		panic("addaliasu on non-special vnode");
1320 	addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0));
1321 }
1322 
1323 void
1324 addalias(nvp, dev)
1325 	struct vnode *nvp;
1326 	dev_t dev;
1327 {
1328 
1329 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1330 		panic("addalias on non-special vnode");
1331 
1332 	nvp->v_rdev = dev;
1333 	simple_lock(&spechash_slock);
1334 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1335 	simple_unlock(&spechash_slock);
1336 }
1337 
1338 /*
1339  * Grab a particular vnode from the free list, increment its
1340  * reference count and lock it. The vnode lock bit is set if the
1341  * vnode is being eliminated in vgone. The process is awakened
1342  * when the transition is completed, and an error returned to
1343  * indicate that the vnode is no longer usable (possibly having
1344  * been changed to a new file system type).
1345  */
1346 int
1347 vget(vp, flags, p)
1348 	register struct vnode *vp;
1349 	int flags;
1350 	struct proc *p;
1351 {
1352 	int error;
1353 
1354 	/*
1355 	 * If the vnode is in the process of being cleaned out for
1356 	 * another use, we wait for the cleaning to finish and then
1357 	 * return failure. Cleaning is determined by checking that
1358 	 * the VXLOCK flag is set.
1359 	 */
1360 	if ((flags & LK_INTERLOCK) == 0) {
1361 		simple_lock(&vp->v_interlock);
1362 	}
1363 	if (vp->v_flag & VXLOCK) {
1364 		vp->v_flag |= VXWANT;
1365 		simple_unlock(&vp->v_interlock);
1366 		tsleep((caddr_t)vp, PINOD, "vget", 0);
1367 		return (ENOENT);
1368 	}
1369 
1370 	vp->v_usecount++;
1371 
1372 	if (VSHOULDBUSY(vp))
1373 		vbusy(vp);
1374 	if (flags & LK_TYPE_MASK) {
1375 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1376 			/*
1377 			 * must expand vrele here because we do not want
1378 			 * to call VOP_INACTIVE if the reference count
1379 			 * drops back to zero since it was never really
1380 			 * active. We must remove it from the free list
1381 			 * before sleeping so that multiple processes do
1382 			 * not try to recycle it.
1383 			 */
1384 			simple_lock(&vp->v_interlock);
1385 			vp->v_usecount--;
1386 			if (VSHOULDFREE(vp))
1387 				vfree(vp);
1388 			simple_unlock(&vp->v_interlock);
1389 		}
1390 		return (error);
1391 	}
1392 	simple_unlock(&vp->v_interlock);
1393 	return (0);
1394 }
1395 
1396 void
1397 vref(struct vnode *vp)
1398 {
1399 	simple_lock(&vp->v_interlock);
1400 	vp->v_usecount++;
1401 	simple_unlock(&vp->v_interlock);
1402 }
1403 
1404 /*
1405  * Vnode put/release.
1406  * If count drops to zero, call inactive routine and return to freelist.
1407  */
1408 void
1409 vrele(vp)
1410 	struct vnode *vp;
1411 {
1412 	struct proc *p = curproc;	/* XXX */
1413 
1414 	KASSERT(vp != NULL, ("vrele: null vp"));
1415 
1416 	simple_lock(&vp->v_interlock);
1417 
1418 	if (vp->v_usecount > 1) {
1419 
1420 		vp->v_usecount--;
1421 		simple_unlock(&vp->v_interlock);
1422 
1423 		return;
1424 	}
1425 
1426 	if (vp->v_usecount == 1) {
1427 
1428 		vp->v_usecount--;
1429 		if (VSHOULDFREE(vp))
1430 			vfree(vp);
1431 	/*
1432 	 * If we are doing a vput, the node is already locked, and we must
1433 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1434 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1435 	 */
1436 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1437 			VOP_INACTIVE(vp, p);
1438 		}
1439 
1440 	} else {
1441 #ifdef DIAGNOSTIC
1442 		vprint("vrele: negative ref count", vp);
1443 		simple_unlock(&vp->v_interlock);
1444 #endif
1445 		panic("vrele: negative ref cnt");
1446 	}
1447 }
1448 
1449 void
1450 vput(vp)
1451 	struct vnode *vp;
1452 {
1453 	struct proc *p = curproc;	/* XXX */
1454 
1455 	KASSERT(vp != NULL, ("vput: null vp"));
1456 
1457 	simple_lock(&vp->v_interlock);
1458 
1459 	if (vp->v_usecount > 1) {
1460 
1461 		vp->v_usecount--;
1462 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1463 		return;
1464 
1465 	}
1466 
1467 	if (vp->v_usecount == 1) {
1468 
1469 		vp->v_usecount--;
1470 		if (VSHOULDFREE(vp))
1471 			vfree(vp);
1472 	/*
1473 	 * If we are doing a vput, the node is already locked, and we must
1474 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1475 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1476 	 */
1477 		simple_unlock(&vp->v_interlock);
1478 		VOP_INACTIVE(vp, p);
1479 
1480 	} else {
1481 #ifdef DIAGNOSTIC
1482 		vprint("vput: negative ref count", vp);
1483 #endif
1484 		panic("vput: negative ref cnt");
1485 	}
1486 }
1487 
1488 /*
1489  * Somebody doesn't want the vnode recycled.
1490  */
1491 void
1492 vhold(vp)
1493 	register struct vnode *vp;
1494 {
1495 	int s;
1496 
1497   	s = splbio();
1498 	vp->v_holdcnt++;
1499 	if (VSHOULDBUSY(vp))
1500 		vbusy(vp);
1501 	splx(s);
1502 }
1503 
1504 /*
1505  * One less who cares about this vnode.
1506  */
1507 void
1508 vdrop(vp)
1509 	register struct vnode *vp;
1510 {
1511 	int s;
1512 
1513 	s = splbio();
1514 	if (vp->v_holdcnt <= 0)
1515 		panic("vdrop: holdcnt");
1516 	vp->v_holdcnt--;
1517 	if (VSHOULDFREE(vp))
1518 		vfree(vp);
1519 	splx(s);
1520 }
1521 
1522 /*
1523  * Remove any vnodes in the vnode table belonging to mount point mp.
1524  *
1525  * If MNT_NOFORCE is specified, there should not be any active ones,
1526  * return error if any are found (nb: this is a user error, not a
1527  * system error). If MNT_FORCE is specified, detach any active vnodes
1528  * that are found.
1529  */
1530 #ifdef DIAGNOSTIC
1531 static int busyprt = 0;		/* print out busy vnodes */
1532 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1533 #endif
1534 
1535 int
1536 vflush(mp, skipvp, flags)
1537 	struct mount *mp;
1538 	struct vnode *skipvp;
1539 	int flags;
1540 {
1541 	struct proc *p = curproc;	/* XXX */
1542 	struct vnode *vp, *nvp;
1543 	int busy = 0;
1544 
1545 	simple_lock(&mntvnode_slock);
1546 loop:
1547 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1548 		/*
1549 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1550 		 * Start over if it has (it won't be on the list anymore).
1551 		 */
1552 		if (vp->v_mount != mp)
1553 			goto loop;
1554 		nvp = LIST_NEXT(vp, v_mntvnodes);
1555 		/*
1556 		 * Skip over a selected vnode.
1557 		 */
1558 		if (vp == skipvp)
1559 			continue;
1560 
1561 		simple_lock(&vp->v_interlock);
1562 		/*
1563 		 * Skip over a vnodes marked VSYSTEM.
1564 		 */
1565 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1566 			simple_unlock(&vp->v_interlock);
1567 			continue;
1568 		}
1569 		/*
1570 		 * If WRITECLOSE is set, only flush out regular file vnodes
1571 		 * open for writing.
1572 		 */
1573 		if ((flags & WRITECLOSE) &&
1574 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1575 			simple_unlock(&vp->v_interlock);
1576 			continue;
1577 		}
1578 
1579 		/*
1580 		 * With v_usecount == 0, all we need to do is clear out the
1581 		 * vnode data structures and we are done.
1582 		 */
1583 		if (vp->v_usecount == 0) {
1584 			simple_unlock(&mntvnode_slock);
1585 			vgonel(vp, p);
1586 			simple_lock(&mntvnode_slock);
1587 			continue;
1588 		}
1589 
1590 		/*
1591 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1592 		 * or character devices, revert to an anonymous device. For
1593 		 * all other files, just kill them.
1594 		 */
1595 		if (flags & FORCECLOSE) {
1596 			simple_unlock(&mntvnode_slock);
1597 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1598 				vgonel(vp, p);
1599 			} else {
1600 				vclean(vp, 0, p);
1601 				vp->v_op = spec_vnodeop_p;
1602 				insmntque(vp, (struct mount *) 0);
1603 			}
1604 			simple_lock(&mntvnode_slock);
1605 			continue;
1606 		}
1607 #ifdef DIAGNOSTIC
1608 		if (busyprt)
1609 			vprint("vflush: busy vnode", vp);
1610 #endif
1611 		simple_unlock(&vp->v_interlock);
1612 		busy++;
1613 	}
1614 	simple_unlock(&mntvnode_slock);
1615 	if (busy)
1616 		return (EBUSY);
1617 	return (0);
1618 }
1619 
1620 /*
1621  * Disassociate the underlying file system from a vnode.
1622  */
1623 static void
1624 vclean(vp, flags, p)
1625 	struct vnode *vp;
1626 	int flags;
1627 	struct proc *p;
1628 {
1629 	int active;
1630 	vm_object_t obj;
1631 
1632 	/*
1633 	 * Check to see if the vnode is in use. If so we have to reference it
1634 	 * before we clean it out so that its count cannot fall to zero and
1635 	 * generate a race against ourselves to recycle it.
1636 	 */
1637 	if ((active = vp->v_usecount))
1638 		vp->v_usecount++;
1639 
1640 	/*
1641 	 * Prevent the vnode from being recycled or brought into use while we
1642 	 * clean it out.
1643 	 */
1644 	if (vp->v_flag & VXLOCK)
1645 		panic("vclean: deadlock");
1646 	vp->v_flag |= VXLOCK;
1647 	/*
1648 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1649 	 * have the object locked while it cleans it out. The VOP_LOCK
1650 	 * ensures that the VOP_INACTIVE routine is done with its work.
1651 	 * For active vnodes, it ensures that no other activity can
1652 	 * occur while the underlying object is being cleaned out.
1653 	 */
1654 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1655 
1656 	/*
1657 	 * Clean out any buffers associated with the vnode.
1658 	 */
1659 	vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1660 	if ((obj = vp->v_object) != NULL) {
1661 		if (obj->ref_count == 0) {
1662 			/*
1663 			 * vclean() may be called twice.  The first time removes the
1664 			 * primary reference to the object, the second time goes
1665 			 * one further and is a special-case to terminate the object.
1666 			 */
1667 			vm_object_terminate(obj);
1668 		} else {
1669 			/*
1670 			 * Woe to the process that tries to page now :-).
1671 			 */
1672 			vm_pager_deallocate(obj);
1673 		}
1674 	}
1675 
1676 	/*
1677 	 * If purging an active vnode, it must be closed and
1678 	 * deactivated before being reclaimed. Note that the
1679 	 * VOP_INACTIVE will unlock the vnode.
1680 	 */
1681 	if (active) {
1682 		if (flags & DOCLOSE)
1683 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1684 		VOP_INACTIVE(vp, p);
1685 	} else {
1686 		/*
1687 		 * Any other processes trying to obtain this lock must first
1688 		 * wait for VXLOCK to clear, then call the new lock operation.
1689 		 */
1690 		VOP_UNLOCK(vp, 0, p);
1691 	}
1692 	/*
1693 	 * Reclaim the vnode.
1694 	 */
1695 	if (VOP_RECLAIM(vp, p))
1696 		panic("vclean: cannot reclaim");
1697 
1698 	if (active) {
1699 		/*
1700 		 * Inline copy of vrele() since VOP_INACTIVE
1701 		 * has already been called.
1702 		 */
1703 		simple_lock(&vp->v_interlock);
1704 		if (--vp->v_usecount <= 0) {
1705 #ifdef DIAGNOSTIC
1706 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1707 				vprint("vclean: bad ref count", vp);
1708 				panic("vclean: ref cnt");
1709 			}
1710 #endif
1711 			vfree(vp);
1712 		}
1713 		simple_unlock(&vp->v_interlock);
1714 	}
1715 
1716 	cache_purge(vp);
1717 	if (vp->v_vnlock) {
1718 		FREE(vp->v_vnlock, M_VNODE);
1719 		vp->v_vnlock = NULL;
1720 	}
1721 
1722 	if (VSHOULDFREE(vp))
1723 		vfree(vp);
1724 
1725 	/*
1726 	 * Done with purge, notify sleepers of the grim news.
1727 	 */
1728 	vp->v_op = dead_vnodeop_p;
1729 	vn_pollgone(vp);
1730 	vp->v_tag = VT_NON;
1731 	vp->v_flag &= ~VXLOCK;
1732 	if (vp->v_flag & VXWANT) {
1733 		vp->v_flag &= ~VXWANT;
1734 		wakeup((caddr_t) vp);
1735 	}
1736 }
1737 
1738 /*
1739  * Eliminate all activity associated with the requested vnode
1740  * and with all vnodes aliased to the requested vnode.
1741  */
1742 int
1743 vop_revoke(ap)
1744 	struct vop_revoke_args /* {
1745 		struct vnode *a_vp;
1746 		int a_flags;
1747 	} */ *ap;
1748 {
1749 	struct vnode *vp, *vq;
1750 	dev_t dev;
1751 
1752 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1753 
1754 	vp = ap->a_vp;
1755 	/*
1756 	 * If a vgone (or vclean) is already in progress,
1757 	 * wait until it is done and return.
1758 	 */
1759 	if (vp->v_flag & VXLOCK) {
1760 		vp->v_flag |= VXWANT;
1761 		simple_unlock(&vp->v_interlock);
1762 		tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1763 		return (0);
1764 	}
1765 	dev = vp->v_rdev;
1766 	for (;;) {
1767 		simple_lock(&spechash_slock);
1768 		vq = SLIST_FIRST(&dev->si_hlist);
1769 		simple_unlock(&spechash_slock);
1770 		if (!vq)
1771 			break;
1772 		vgone(vq);
1773 	}
1774 	return (0);
1775 }
1776 
1777 /*
1778  * Recycle an unused vnode to the front of the free list.
1779  * Release the passed interlock if the vnode will be recycled.
1780  */
1781 int
1782 vrecycle(vp, inter_lkp, p)
1783 	struct vnode *vp;
1784 	struct simplelock *inter_lkp;
1785 	struct proc *p;
1786 {
1787 
1788 	simple_lock(&vp->v_interlock);
1789 	if (vp->v_usecount == 0) {
1790 		if (inter_lkp) {
1791 			simple_unlock(inter_lkp);
1792 		}
1793 		vgonel(vp, p);
1794 		return (1);
1795 	}
1796 	simple_unlock(&vp->v_interlock);
1797 	return (0);
1798 }
1799 
1800 /*
1801  * Eliminate all activity associated with a vnode
1802  * in preparation for reuse.
1803  */
1804 void
1805 vgone(vp)
1806 	register struct vnode *vp;
1807 {
1808 	struct proc *p = curproc;	/* XXX */
1809 
1810 	simple_lock(&vp->v_interlock);
1811 	vgonel(vp, p);
1812 }
1813 
1814 /*
1815  * vgone, with the vp interlock held.
1816  */
1817 void
1818 vgonel(vp, p)
1819 	struct vnode *vp;
1820 	struct proc *p;
1821 {
1822 	int s;
1823 
1824 	/*
1825 	 * If a vgone (or vclean) is already in progress,
1826 	 * wait until it is done and return.
1827 	 */
1828 	if (vp->v_flag & VXLOCK) {
1829 		vp->v_flag |= VXWANT;
1830 		simple_unlock(&vp->v_interlock);
1831 		tsleep((caddr_t)vp, PINOD, "vgone", 0);
1832 		return;
1833 	}
1834 
1835 	/*
1836 	 * Clean out the filesystem specific data.
1837 	 */
1838 	vclean(vp, DOCLOSE, p);
1839 	simple_lock(&vp->v_interlock);
1840 
1841 	/*
1842 	 * Delete from old mount point vnode list, if on one.
1843 	 */
1844 	if (vp->v_mount != NULL)
1845 		insmntque(vp, (struct mount *)0);
1846 	/*
1847 	 * If special device, remove it from special device alias list
1848 	 * if it is on one.
1849 	 */
1850 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1851 		simple_lock(&spechash_slock);
1852 		SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
1853 		freedev(vp->v_rdev);
1854 		simple_unlock(&spechash_slock);
1855 		vp->v_rdev = NULL;
1856 	}
1857 
1858 	/*
1859 	 * If it is on the freelist and not already at the head,
1860 	 * move it to the head of the list. The test of the back
1861 	 * pointer and the reference count of zero is because
1862 	 * it will be removed from the free list by getnewvnode,
1863 	 * but will not have its reference count incremented until
1864 	 * after calling vgone. If the reference count were
1865 	 * incremented first, vgone would (incorrectly) try to
1866 	 * close the previous instance of the underlying object.
1867 	 */
1868 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1869 		s = splbio();
1870 		simple_lock(&vnode_free_list_slock);
1871 		if (vp->v_flag & VFREE) {
1872 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1873 		} else if (vp->v_flag & VTBFREE) {
1874 			TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
1875 			vp->v_flag &= ~VTBFREE;
1876 			freevnodes++;
1877 		} else
1878 			freevnodes++;
1879 		vp->v_flag |= VFREE;
1880 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1881 		simple_unlock(&vnode_free_list_slock);
1882 		splx(s);
1883 	}
1884 
1885 	vp->v_type = VBAD;
1886 	simple_unlock(&vp->v_interlock);
1887 }
1888 
1889 /*
1890  * Lookup a vnode by device number.
1891  */
1892 int
1893 vfinddev(dev, type, vpp)
1894 	dev_t dev;
1895 	enum vtype type;
1896 	struct vnode **vpp;
1897 {
1898 	struct vnode *vp;
1899 
1900 	simple_lock(&spechash_slock);
1901 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1902 		if (type == vp->v_type) {
1903 			*vpp = vp;
1904 			simple_unlock(&spechash_slock);
1905 			return (1);
1906 		}
1907 	}
1908 	simple_unlock(&spechash_slock);
1909 	return (0);
1910 }
1911 
1912 /*
1913  * Calculate the total number of references to a special device.
1914  */
1915 int
1916 vcount(vp)
1917 	struct vnode *vp;
1918 {
1919 	struct vnode *vq;
1920 	int count;
1921 
1922 	count = 0;
1923 	simple_lock(&spechash_slock);
1924 	SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
1925 		count += vq->v_usecount;
1926 	simple_unlock(&spechash_slock);
1927 	return (count);
1928 }
1929 
1930 /*
1931  * Same as above, but using the dev_t as argument
1932  */
1933 
1934 int
1935 count_dev(dev)
1936 	dev_t dev;
1937 {
1938 	struct vnode *vp;
1939 
1940 	vp = SLIST_FIRST(&dev->si_hlist);
1941 	if (vp == NULL)
1942 		return (0);
1943 	return(vcount(vp));
1944 }
1945 
1946 /*
1947  * Print out a description of a vnode.
1948  */
1949 static char *typename[] =
1950 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1951 
1952 void
1953 vprint(label, vp)
1954 	char *label;
1955 	struct vnode *vp;
1956 {
1957 	char buf[96];
1958 
1959 	if (label != NULL)
1960 		printf("%s: %p: ", label, (void *)vp);
1961 	else
1962 		printf("%p: ", (void *)vp);
1963 	printf("type %s, usecount %d, writecount %d, refcount %d,",
1964 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1965 	    vp->v_holdcnt);
1966 	buf[0] = '\0';
1967 	if (vp->v_flag & VROOT)
1968 		strcat(buf, "|VROOT");
1969 	if (vp->v_flag & VTEXT)
1970 		strcat(buf, "|VTEXT");
1971 	if (vp->v_flag & VSYSTEM)
1972 		strcat(buf, "|VSYSTEM");
1973 	if (vp->v_flag & VXLOCK)
1974 		strcat(buf, "|VXLOCK");
1975 	if (vp->v_flag & VXWANT)
1976 		strcat(buf, "|VXWANT");
1977 	if (vp->v_flag & VBWAIT)
1978 		strcat(buf, "|VBWAIT");
1979 	if (vp->v_flag & VDOOMED)
1980 		strcat(buf, "|VDOOMED");
1981 	if (vp->v_flag & VFREE)
1982 		strcat(buf, "|VFREE");
1983 	if (vp->v_flag & VOBJBUF)
1984 		strcat(buf, "|VOBJBUF");
1985 	if (buf[0] != '\0')
1986 		printf(" flags (%s)", &buf[1]);
1987 	if (vp->v_data == NULL) {
1988 		printf("\n");
1989 	} else {
1990 		printf("\n\t");
1991 		VOP_PRINT(vp);
1992 	}
1993 }
1994 
1995 #ifdef DDB
1996 #include <ddb/ddb.h>
1997 /*
1998  * List all of the locked vnodes in the system.
1999  * Called when debugging the kernel.
2000  */
2001 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2002 {
2003 	struct proc *p = curproc;	/* XXX */
2004 	struct mount *mp, *nmp;
2005 	struct vnode *vp;
2006 
2007 	printf("Locked vnodes\n");
2008 	simple_lock(&mountlist_slock);
2009 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2010 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2011 			nmp = TAILQ_NEXT(mp, mnt_list);
2012 			continue;
2013 		}
2014 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2015 			if (VOP_ISLOCKED(vp, NULL))
2016 				vprint((char *)0, vp);
2017 		}
2018 		simple_lock(&mountlist_slock);
2019 		nmp = TAILQ_NEXT(mp, mnt_list);
2020 		vfs_unbusy(mp, p);
2021 	}
2022 	simple_unlock(&mountlist_slock);
2023 }
2024 #endif
2025 
2026 /*
2027  * Top level filesystem related information gathering.
2028  */
2029 static int	sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS);
2030 
2031 static int
2032 vfs_sysctl SYSCTL_HANDLER_ARGS
2033 {
2034 	int *name = (int *)arg1 - 1;	/* XXX */
2035 	u_int namelen = arg2 + 1;	/* XXX */
2036 	struct vfsconf *vfsp;
2037 
2038 #if 1 || defined(COMPAT_PRELITE2)
2039 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2040 	if (namelen == 1)
2041 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2042 #endif
2043 
2044 #ifdef notyet
2045 	/* all sysctl names at this level are at least name and field */
2046 	if (namelen < 2)
2047 		return (ENOTDIR);		/* overloaded */
2048 	if (name[0] != VFS_GENERIC) {
2049 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2050 			if (vfsp->vfc_typenum == name[0])
2051 				break;
2052 		if (vfsp == NULL)
2053 			return (EOPNOTSUPP);
2054 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2055 		    oldp, oldlenp, newp, newlen, p));
2056 	}
2057 #endif
2058 	switch (name[1]) {
2059 	case VFS_MAXTYPENUM:
2060 		if (namelen != 2)
2061 			return (ENOTDIR);
2062 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2063 	case VFS_CONF:
2064 		if (namelen != 3)
2065 			return (ENOTDIR);	/* overloaded */
2066 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2067 			if (vfsp->vfc_typenum == name[2])
2068 				break;
2069 		if (vfsp == NULL)
2070 			return (EOPNOTSUPP);
2071 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2072 	}
2073 	return (EOPNOTSUPP);
2074 }
2075 
2076 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2077 	"Generic filesystem");
2078 
2079 #if 1 || defined(COMPAT_PRELITE2)
2080 
2081 static int
2082 sysctl_ovfs_conf SYSCTL_HANDLER_ARGS
2083 {
2084 	int error;
2085 	struct vfsconf *vfsp;
2086 	struct ovfsconf ovfs;
2087 
2088 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2089 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2090 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2091 		ovfs.vfc_index = vfsp->vfc_typenum;
2092 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2093 		ovfs.vfc_flags = vfsp->vfc_flags;
2094 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2095 		if (error)
2096 			return error;
2097 	}
2098 	return 0;
2099 }
2100 
2101 #endif /* 1 || COMPAT_PRELITE2 */
2102 
2103 #if 0
2104 #define KINFO_VNODESLOP	10
2105 /*
2106  * Dump vnode list (via sysctl).
2107  * Copyout address of vnode followed by vnode.
2108  */
2109 /* ARGSUSED */
2110 static int
2111 sysctl_vnode SYSCTL_HANDLER_ARGS
2112 {
2113 	struct proc *p = curproc;	/* XXX */
2114 	struct mount *mp, *nmp;
2115 	struct vnode *nvp, *vp;
2116 	int error;
2117 
2118 #define VPTRSZ	sizeof (struct vnode *)
2119 #define VNODESZ	sizeof (struct vnode)
2120 
2121 	req->lock = 0;
2122 	if (!req->oldptr) /* Make an estimate */
2123 		return (SYSCTL_OUT(req, 0,
2124 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2125 
2126 	simple_lock(&mountlist_slock);
2127 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2128 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2129 			nmp = TAILQ_NEXT(mp, mnt_list);
2130 			continue;
2131 		}
2132 again:
2133 		simple_lock(&mntvnode_slock);
2134 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2135 		     vp != NULL;
2136 		     vp = nvp) {
2137 			/*
2138 			 * Check that the vp is still associated with
2139 			 * this filesystem.  RACE: could have been
2140 			 * recycled onto the same filesystem.
2141 			 */
2142 			if (vp->v_mount != mp) {
2143 				simple_unlock(&mntvnode_slock);
2144 				goto again;
2145 			}
2146 			nvp = LIST_NEXT(vp, v_mntvnodes);
2147 			simple_unlock(&mntvnode_slock);
2148 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2149 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2150 				return (error);
2151 			simple_lock(&mntvnode_slock);
2152 		}
2153 		simple_unlock(&mntvnode_slock);
2154 		simple_lock(&mountlist_slock);
2155 		nmp = TAILQ_NEXT(mp, mnt_list);
2156 		vfs_unbusy(mp, p);
2157 	}
2158 	simple_unlock(&mountlist_slock);
2159 
2160 	return (0);
2161 }
2162 #endif
2163 
2164 /*
2165  * XXX
2166  * Exporting the vnode list on large systems causes them to crash.
2167  * Exporting the vnode list on medium systems causes sysctl to coredump.
2168  */
2169 #if 0
2170 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2171 	0, 0, sysctl_vnode, "S,vnode", "");
2172 #endif
2173 
2174 /*
2175  * Check to see if a filesystem is mounted on a block device.
2176  */
2177 int
2178 vfs_mountedon(vp)
2179 	struct vnode *vp;
2180 {
2181 
2182 	if (vp->v_specmountpoint != NULL)
2183 		return (EBUSY);
2184 	return (0);
2185 }
2186 
2187 /*
2188  * Unmount all filesystems. The list is traversed in reverse order
2189  * of mounting to avoid dependencies.
2190  */
2191 void
2192 vfs_unmountall()
2193 {
2194 	struct mount *mp;
2195 	struct proc *p;
2196 	int error;
2197 
2198 	if (curproc != NULL)
2199 		p = curproc;
2200 	else
2201 		p = initproc;	/* XXX XXX should this be proc0? */
2202 	/*
2203 	 * Since this only runs when rebooting, it is not interlocked.
2204 	 */
2205 	while(!TAILQ_EMPTY(&mountlist)) {
2206 		mp = TAILQ_LAST(&mountlist, mntlist);
2207 		error = dounmount(mp, MNT_FORCE, p);
2208 		if (error) {
2209 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2210 			printf("unmount of %s failed (",
2211 			    mp->mnt_stat.f_mntonname);
2212 			if (error == EBUSY)
2213 				printf("BUSY)\n");
2214 			else
2215 				printf("%d)\n", error);
2216 		} else {
2217 			/* The unmount has removed mp from the mountlist */
2218 		}
2219 	}
2220 }
2221 
2222 /*
2223  * Build hash lists of net addresses and hang them off the mount point.
2224  * Called by ufs_mount() to set up the lists of export addresses.
2225  */
2226 static int
2227 vfs_hang_addrlist(mp, nep, argp)
2228 	struct mount *mp;
2229 	struct netexport *nep;
2230 	struct export_args *argp;
2231 {
2232 	register struct netcred *np;
2233 	register struct radix_node_head *rnh;
2234 	register int i;
2235 	struct radix_node *rn;
2236 	struct sockaddr *saddr, *smask = 0;
2237 	struct domain *dom;
2238 	int error;
2239 
2240 	if (argp->ex_addrlen == 0) {
2241 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2242 			return (EPERM);
2243 		np = &nep->ne_defexported;
2244 		np->netc_exflags = argp->ex_flags;
2245 		np->netc_anon = argp->ex_anon;
2246 		np->netc_anon.cr_ref = 1;
2247 		mp->mnt_flag |= MNT_DEFEXPORTED;
2248 		return (0);
2249 	}
2250 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2251 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2252 	bzero((caddr_t) np, i);
2253 	saddr = (struct sockaddr *) (np + 1);
2254 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2255 		goto out;
2256 	if (saddr->sa_len > argp->ex_addrlen)
2257 		saddr->sa_len = argp->ex_addrlen;
2258 	if (argp->ex_masklen) {
2259 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2260 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2261 		if (error)
2262 			goto out;
2263 		if (smask->sa_len > argp->ex_masklen)
2264 			smask->sa_len = argp->ex_masklen;
2265 	}
2266 	i = saddr->sa_family;
2267 	if ((rnh = nep->ne_rtable[i]) == 0) {
2268 		/*
2269 		 * Seems silly to initialize every AF when most are not used,
2270 		 * do so on demand here
2271 		 */
2272 		for (dom = domains; dom; dom = dom->dom_next)
2273 			if (dom->dom_family == i && dom->dom_rtattach) {
2274 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2275 				    dom->dom_rtoffset);
2276 				break;
2277 			}
2278 		if ((rnh = nep->ne_rtable[i]) == 0) {
2279 			error = ENOBUFS;
2280 			goto out;
2281 		}
2282 	}
2283 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2284 	    np->netc_rnodes);
2285 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2286 		error = EPERM;
2287 		goto out;
2288 	}
2289 	np->netc_exflags = argp->ex_flags;
2290 	np->netc_anon = argp->ex_anon;
2291 	np->netc_anon.cr_ref = 1;
2292 	return (0);
2293 out:
2294 	free(np, M_NETADDR);
2295 	return (error);
2296 }
2297 
2298 /* ARGSUSED */
2299 static int
2300 vfs_free_netcred(rn, w)
2301 	struct radix_node *rn;
2302 	void *w;
2303 {
2304 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
2305 
2306 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2307 	free((caddr_t) rn, M_NETADDR);
2308 	return (0);
2309 }
2310 
2311 /*
2312  * Free the net address hash lists that are hanging off the mount points.
2313  */
2314 static void
2315 vfs_free_addrlist(nep)
2316 	struct netexport *nep;
2317 {
2318 	register int i;
2319 	register struct radix_node_head *rnh;
2320 
2321 	for (i = 0; i <= AF_MAX; i++)
2322 		if ((rnh = nep->ne_rtable[i])) {
2323 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2324 			    (caddr_t) rnh);
2325 			free((caddr_t) rnh, M_RTABLE);
2326 			nep->ne_rtable[i] = 0;
2327 		}
2328 }
2329 
2330 int
2331 vfs_export(mp, nep, argp)
2332 	struct mount *mp;
2333 	struct netexport *nep;
2334 	struct export_args *argp;
2335 {
2336 	int error;
2337 
2338 	if (argp->ex_flags & MNT_DELEXPORT) {
2339 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2340 			vfs_setpublicfs(NULL, NULL, NULL);
2341 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2342 		}
2343 		vfs_free_addrlist(nep);
2344 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2345 	}
2346 	if (argp->ex_flags & MNT_EXPORTED) {
2347 		if (argp->ex_flags & MNT_EXPUBLIC) {
2348 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2349 				return (error);
2350 			mp->mnt_flag |= MNT_EXPUBLIC;
2351 		}
2352 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2353 			return (error);
2354 		mp->mnt_flag |= MNT_EXPORTED;
2355 	}
2356 	return (0);
2357 }
2358 
2359 
2360 /*
2361  * Set the publicly exported filesystem (WebNFS). Currently, only
2362  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2363  */
2364 int
2365 vfs_setpublicfs(mp, nep, argp)
2366 	struct mount *mp;
2367 	struct netexport *nep;
2368 	struct export_args *argp;
2369 {
2370 	int error;
2371 	struct vnode *rvp;
2372 	char *cp;
2373 
2374 	/*
2375 	 * mp == NULL -> invalidate the current info, the FS is
2376 	 * no longer exported. May be called from either vfs_export
2377 	 * or unmount, so check if it hasn't already been done.
2378 	 */
2379 	if (mp == NULL) {
2380 		if (nfs_pub.np_valid) {
2381 			nfs_pub.np_valid = 0;
2382 			if (nfs_pub.np_index != NULL) {
2383 				FREE(nfs_pub.np_index, M_TEMP);
2384 				nfs_pub.np_index = NULL;
2385 			}
2386 		}
2387 		return (0);
2388 	}
2389 
2390 	/*
2391 	 * Only one allowed at a time.
2392 	 */
2393 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2394 		return (EBUSY);
2395 
2396 	/*
2397 	 * Get real filehandle for root of exported FS.
2398 	 */
2399 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2400 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2401 
2402 	if ((error = VFS_ROOT(mp, &rvp)))
2403 		return (error);
2404 
2405 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2406 		return (error);
2407 
2408 	vput(rvp);
2409 
2410 	/*
2411 	 * If an indexfile was specified, pull it in.
2412 	 */
2413 	if (argp->ex_indexfile != NULL) {
2414 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2415 		    M_WAITOK);
2416 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2417 		    MAXNAMLEN, (size_t *)0);
2418 		if (!error) {
2419 			/*
2420 			 * Check for illegal filenames.
2421 			 */
2422 			for (cp = nfs_pub.np_index; *cp; cp++) {
2423 				if (*cp == '/') {
2424 					error = EINVAL;
2425 					break;
2426 				}
2427 			}
2428 		}
2429 		if (error) {
2430 			FREE(nfs_pub.np_index, M_TEMP);
2431 			return (error);
2432 		}
2433 	}
2434 
2435 	nfs_pub.np_mount = mp;
2436 	nfs_pub.np_valid = 1;
2437 	return (0);
2438 }
2439 
2440 struct netcred *
2441 vfs_export_lookup(mp, nep, nam)
2442 	register struct mount *mp;
2443 	struct netexport *nep;
2444 	struct sockaddr *nam;
2445 {
2446 	register struct netcred *np;
2447 	register struct radix_node_head *rnh;
2448 	struct sockaddr *saddr;
2449 
2450 	np = NULL;
2451 	if (mp->mnt_flag & MNT_EXPORTED) {
2452 		/*
2453 		 * Lookup in the export list first.
2454 		 */
2455 		if (nam != NULL) {
2456 			saddr = nam;
2457 			rnh = nep->ne_rtable[saddr->sa_family];
2458 			if (rnh != NULL) {
2459 				np = (struct netcred *)
2460 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2461 							      rnh);
2462 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2463 					np = NULL;
2464 			}
2465 		}
2466 		/*
2467 		 * If no address match, use the default if it exists.
2468 		 */
2469 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2470 			np = &nep->ne_defexported;
2471 	}
2472 	return (np);
2473 }
2474 
2475 /*
2476  * perform msync on all vnodes under a mount point
2477  * the mount point must be locked.
2478  */
2479 void
2480 vfs_msync(struct mount *mp, int flags) {
2481 	struct vnode *vp, *nvp;
2482 	struct vm_object *obj;
2483 	int anyio, tries;
2484 
2485 	tries = 5;
2486 loop:
2487 	anyio = 0;
2488 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
2489 
2490 		nvp = LIST_NEXT(vp, v_mntvnodes);
2491 
2492 		if (vp->v_mount != mp) {
2493 			goto loop;
2494 		}
2495 
2496 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2497 			continue;
2498 
2499 		if (flags != MNT_WAIT) {
2500 			obj = vp->v_object;
2501 			if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2502 				continue;
2503 			if (VOP_ISLOCKED(vp, NULL))
2504 				continue;
2505 		}
2506 
2507 		simple_lock(&vp->v_interlock);
2508 		if (vp->v_object &&
2509 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
2510 			if (!vget(vp,
2511 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2512 				if (vp->v_object) {
2513 					vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2514 					anyio = 1;
2515 				}
2516 				vput(vp);
2517 			}
2518 		} else {
2519 			simple_unlock(&vp->v_interlock);
2520 		}
2521 	}
2522 	if (anyio && (--tries > 0))
2523 		goto loop;
2524 }
2525 
2526 /*
2527  * Create the VM object needed for VMIO and mmap support.  This
2528  * is done for all VREG files in the system.  Some filesystems might
2529  * afford the additional metadata buffering capability of the
2530  * VMIO code by making the device node be VMIO mode also.
2531  *
2532  * vp must be locked when vfs_object_create is called.
2533  */
2534 int
2535 vfs_object_create(vp, p, cred)
2536 	struct vnode *vp;
2537 	struct proc *p;
2538 	struct ucred *cred;
2539 {
2540 	struct vattr vat;
2541 	vm_object_t object;
2542 	int error = 0;
2543 
2544 	if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE)
2545 		return 0;
2546 
2547 retry:
2548 	if ((object = vp->v_object) == NULL) {
2549 		if (vp->v_type == VREG || vp->v_type == VDIR) {
2550 			if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
2551 				goto retn;
2552 			object = vnode_pager_alloc(vp, vat.va_size, 0, 0);
2553 		} else if (devsw(vp->v_rdev) != NULL) {
2554 			/*
2555 			 * This simply allocates the biggest object possible
2556 			 * for a disk vnode.  This should be fixed, but doesn't
2557 			 * cause any problems (yet).
2558 			 */
2559 			object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0);
2560 		} else {
2561 			goto retn;
2562 		}
2563 		/*
2564 		 * Dereference the reference we just created.  This assumes
2565 		 * that the object is associated with the vp.
2566 		 */
2567 		object->ref_count--;
2568 		vp->v_usecount--;
2569 	} else {
2570 		if (object->flags & OBJ_DEAD) {
2571 			VOP_UNLOCK(vp, 0, p);
2572 			tsleep(object, PVM, "vodead", 0);
2573 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
2574 			goto retry;
2575 		}
2576 	}
2577 
2578 	KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object"));
2579 	vp->v_flag |= VOBJBUF;
2580 
2581 retn:
2582 	return error;
2583 }
2584 
2585 static void
2586 vfree(vp)
2587 	struct vnode *vp;
2588 {
2589 	int s;
2590 
2591 	s = splbio();
2592 	simple_lock(&vnode_free_list_slock);
2593 	if (vp->v_flag & VTBFREE) {
2594 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2595 		vp->v_flag &= ~VTBFREE;
2596 	}
2597 	if (vp->v_flag & VAGE) {
2598 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2599 	} else {
2600 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2601 	}
2602 	freevnodes++;
2603 	simple_unlock(&vnode_free_list_slock);
2604 	vp->v_flag &= ~VAGE;
2605 	vp->v_flag |= VFREE;
2606 	splx(s);
2607 }
2608 
2609 void
2610 vbusy(vp)
2611 	struct vnode *vp;
2612 {
2613 	int s;
2614 
2615 	s = splbio();
2616 	simple_lock(&vnode_free_list_slock);
2617 	if (vp->v_flag & VTBFREE) {
2618 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2619 		vp->v_flag &= ~VTBFREE;
2620 	} else {
2621 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2622 		freevnodes--;
2623 	}
2624 	simple_unlock(&vnode_free_list_slock);
2625 	vp->v_flag &= ~(VFREE|VAGE);
2626 	splx(s);
2627 }
2628 
2629 /*
2630  * Record a process's interest in events which might happen to
2631  * a vnode.  Because poll uses the historic select-style interface
2632  * internally, this routine serves as both the ``check for any
2633  * pending events'' and the ``record my interest in future events''
2634  * functions.  (These are done together, while the lock is held,
2635  * to avoid race conditions.)
2636  */
2637 int
2638 vn_pollrecord(vp, p, events)
2639 	struct vnode *vp;
2640 	struct proc *p;
2641 	short events;
2642 {
2643 	simple_lock(&vp->v_pollinfo.vpi_lock);
2644 	if (vp->v_pollinfo.vpi_revents & events) {
2645 		/*
2646 		 * This leaves events we are not interested
2647 		 * in available for the other process which
2648 		 * which presumably had requested them
2649 		 * (otherwise they would never have been
2650 		 * recorded).
2651 		 */
2652 		events &= vp->v_pollinfo.vpi_revents;
2653 		vp->v_pollinfo.vpi_revents &= ~events;
2654 
2655 		simple_unlock(&vp->v_pollinfo.vpi_lock);
2656 		return events;
2657 	}
2658 	vp->v_pollinfo.vpi_events |= events;
2659 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2660 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2661 	return 0;
2662 }
2663 
2664 /*
2665  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2666  * it is possible for us to miss an event due to race conditions, but
2667  * that condition is expected to be rare, so for the moment it is the
2668  * preferred interface.
2669  */
2670 void
2671 vn_pollevent(vp, events)
2672 	struct vnode *vp;
2673 	short events;
2674 {
2675 	simple_lock(&vp->v_pollinfo.vpi_lock);
2676 	if (vp->v_pollinfo.vpi_events & events) {
2677 		/*
2678 		 * We clear vpi_events so that we don't
2679 		 * call selwakeup() twice if two events are
2680 		 * posted before the polling process(es) is
2681 		 * awakened.  This also ensures that we take at
2682 		 * most one selwakeup() if the polling process
2683 		 * is no longer interested.  However, it does
2684 		 * mean that only one event can be noticed at
2685 		 * a time.  (Perhaps we should only clear those
2686 		 * event bits which we note?) XXX
2687 		 */
2688 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2689 		vp->v_pollinfo.vpi_revents |= events;
2690 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2691 	}
2692 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2693 }
2694 
2695 /*
2696  * Wake up anyone polling on vp because it is being revoked.
2697  * This depends on dead_poll() returning POLLHUP for correct
2698  * behavior.
2699  */
2700 void
2701 vn_pollgone(vp)
2702 	struct vnode *vp;
2703 {
2704 	simple_lock(&vp->v_pollinfo.vpi_lock);
2705 	if (vp->v_pollinfo.vpi_events) {
2706 		vp->v_pollinfo.vpi_events = 0;
2707 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2708 	}
2709 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2710 }
2711 
2712 
2713 
2714 /*
2715  * Routine to create and manage a filesystem syncer vnode.
2716  */
2717 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2718 static int	sync_fsync __P((struct  vop_fsync_args *));
2719 static int	sync_inactive __P((struct  vop_inactive_args *));
2720 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2721 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2722 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2723 static int	sync_print __P((struct vop_print_args *));
2724 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2725 
2726 static vop_t **sync_vnodeop_p;
2727 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2728 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2729 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2730 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2731 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2732 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2733 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2734 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2735 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2736 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2737 	{ NULL, NULL }
2738 };
2739 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2740 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2741 
2742 VNODEOP_SET(sync_vnodeop_opv_desc);
2743 
2744 /*
2745  * Create a new filesystem syncer vnode for the specified mount point.
2746  */
2747 int
2748 vfs_allocate_syncvnode(mp)
2749 	struct mount *mp;
2750 {
2751 	struct vnode *vp;
2752 	static long start, incr, next;
2753 	int error;
2754 
2755 	/* Allocate a new vnode */
2756 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2757 		mp->mnt_syncer = NULL;
2758 		return (error);
2759 	}
2760 	vp->v_type = VNON;
2761 	/*
2762 	 * Place the vnode onto the syncer worklist. We attempt to
2763 	 * scatter them about on the list so that they will go off
2764 	 * at evenly distributed times even if all the filesystems
2765 	 * are mounted at once.
2766 	 */
2767 	next += incr;
2768 	if (next == 0 || next > syncer_maxdelay) {
2769 		start /= 2;
2770 		incr /= 2;
2771 		if (start == 0) {
2772 			start = syncer_maxdelay / 2;
2773 			incr = syncer_maxdelay;
2774 		}
2775 		next = start;
2776 	}
2777 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2778 	mp->mnt_syncer = vp;
2779 	return (0);
2780 }
2781 
2782 /*
2783  * Do a lazy sync of the filesystem.
2784  */
2785 static int
2786 sync_fsync(ap)
2787 	struct vop_fsync_args /* {
2788 		struct vnode *a_vp;
2789 		struct ucred *a_cred;
2790 		int a_waitfor;
2791 		struct proc *a_p;
2792 	} */ *ap;
2793 {
2794 	struct vnode *syncvp = ap->a_vp;
2795 	struct mount *mp = syncvp->v_mount;
2796 	struct proc *p = ap->a_p;
2797 	int asyncflag;
2798 
2799 	/*
2800 	 * We only need to do something if this is a lazy evaluation.
2801 	 */
2802 	if (ap->a_waitfor != MNT_LAZY)
2803 		return (0);
2804 
2805 	/*
2806 	 * Move ourselves to the back of the sync list.
2807 	 */
2808 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2809 
2810 	/*
2811 	 * Walk the list of vnodes pushing all that are dirty and
2812 	 * not already on the sync list.
2813 	 */
2814 	simple_lock(&mountlist_slock);
2815 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) {
2816 		simple_unlock(&mountlist_slock);
2817 		return (0);
2818 	}
2819 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2820 	mp->mnt_flag &= ~MNT_ASYNC;
2821 	vfs_msync(mp, MNT_NOWAIT);
2822 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2823 	if (asyncflag)
2824 		mp->mnt_flag |= MNT_ASYNC;
2825 	vfs_unbusy(mp, p);
2826 	return (0);
2827 }
2828 
2829 /*
2830  * The syncer vnode is no referenced.
2831  */
2832 static int
2833 sync_inactive(ap)
2834 	struct vop_inactive_args /* {
2835 		struct vnode *a_vp;
2836 		struct proc *a_p;
2837 	} */ *ap;
2838 {
2839 
2840 	vgone(ap->a_vp);
2841 	return (0);
2842 }
2843 
2844 /*
2845  * The syncer vnode is no longer needed and is being decommissioned.
2846  *
2847  * Modifications to the worklist must be protected at splbio().
2848  */
2849 static int
2850 sync_reclaim(ap)
2851 	struct vop_reclaim_args /* {
2852 		struct vnode *a_vp;
2853 	} */ *ap;
2854 {
2855 	struct vnode *vp = ap->a_vp;
2856 	int s;
2857 
2858 	s = splbio();
2859 	vp->v_mount->mnt_syncer = NULL;
2860 	if (vp->v_flag & VONWORKLST) {
2861 		LIST_REMOVE(vp, v_synclist);
2862 		vp->v_flag &= ~VONWORKLST;
2863 	}
2864 	splx(s);
2865 
2866 	return (0);
2867 }
2868 
2869 /*
2870  * Print out a syncer vnode.
2871  */
2872 static int
2873 sync_print(ap)
2874 	struct vop_print_args /* {
2875 		struct vnode *a_vp;
2876 	} */ *ap;
2877 {
2878 	struct vnode *vp = ap->a_vp;
2879 
2880 	printf("syncer vnode");
2881 	if (vp->v_vnlock != NULL)
2882 		lockmgr_printinfo(vp->v_vnlock);
2883 	printf("\n");
2884 	return (0);
2885 }
2886 
2887 /*
2888  * extract the dev_t from a VBLK or VCHR
2889  */
2890 dev_t
2891 vn_todev(vp)
2892 	struct vnode *vp;
2893 {
2894 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2895 		return (NODEV);
2896 	return (vp->v_rdev);
2897 }
2898 
2899 /*
2900  * Check if vnode represents a disk device
2901  */
2902 int
2903 vn_isdisk(vp, errp)
2904 	struct vnode *vp;
2905 	int *errp;
2906 {
2907 	if (vp->v_type != VBLK && vp->v_type != VCHR) {
2908 		if (errp != NULL)
2909 			*errp = ENOTBLK;
2910 		return (0);
2911 	}
2912 	if (vp->v_rdev == NULL) {
2913 		if (errp != NULL)
2914 			*errp = ENXIO;
2915 		return (0);
2916 	}
2917 	if (!devsw(vp->v_rdev)) {
2918 		if (errp != NULL)
2919 			*errp = ENXIO;
2920 		return (0);
2921 	}
2922 	if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) {
2923 		if (errp != NULL)
2924 			*errp = ENOTBLK;
2925 		return (0);
2926 	}
2927 	if (errp != NULL)
2928 		*errp = 0;
2929 	return (1);
2930 }
2931 
2932 void
2933 NDFREE(ndp, flags)
2934      struct nameidata *ndp;
2935      const uint flags;
2936 {
2937 	if (!(flags & NDF_NO_FREE_PNBUF) &&
2938 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
2939 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
2940 		ndp->ni_cnd.cn_flags &= ~HASBUF;
2941 	}
2942 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
2943 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
2944 	    ndp->ni_dvp != ndp->ni_vp)
2945 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_proc);
2946 	if (!(flags & NDF_NO_DVP_RELE) &&
2947 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
2948 		vrele(ndp->ni_dvp);
2949 		ndp->ni_dvp = NULL;
2950 	}
2951 	if (!(flags & NDF_NO_VP_UNLOCK) &&
2952 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
2953 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_proc);
2954 	if (!(flags & NDF_NO_VP_RELE) &&
2955 	    ndp->ni_vp) {
2956 		vrele(ndp->ni_vp);
2957 		ndp->ni_vp = NULL;
2958 	}
2959 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
2960 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
2961 		vrele(ndp->ni_startdir);
2962 		ndp->ni_startdir = NULL;
2963 	}
2964 }
2965