xref: /freebsd/sys/kern/vfs_subr.c (revision a1a4f1a0d87b594d3f17a97dc0127eec1417e6f6)
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_param.h>
68 #include <vm/vm_prot.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_extern.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_pager.h>
75 #include <vm/vnode_pager.h>
76 #include <vm/vm_zone.h>
77 #include <sys/sysctl.h>
78 
79 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
80 
81 static void	insmntque __P((struct vnode *vp, struct mount *mp));
82 static void	vclean __P((struct vnode *vp, int flags, struct proc *p));
83 static void	vfree __P((struct vnode *));
84 static void	vgonel __P((struct vnode *vp, struct proc *p));
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;	/* mounted filesystem list */
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 	CIRCLEQ_INIT(&mountlist);
178 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
179 	/*
180 	 * Initialize the filesystem syncer.
181 	 */
182 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
183 		&syncer_mask);
184 	syncer_maxdelay = syncer_mask + 1;
185 }
186 
187 /*
188  * Mark a mount point as busy. Used to synchronize access and to delay
189  * unmounting. Interlock is not released on failure.
190  */
191 int
192 vfs_busy(mp, flags, interlkp, p)
193 	struct mount *mp;
194 	int flags;
195 	struct simplelock *interlkp;
196 	struct proc *p;
197 {
198 	int lkflags;
199 
200 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
201 		if (flags & LK_NOWAIT)
202 			return (ENOENT);
203 		mp->mnt_kern_flag |= MNTK_MWAIT;
204 		if (interlkp) {
205 			simple_unlock(interlkp);
206 		}
207 		/*
208 		 * Since all busy locks are shared except the exclusive
209 		 * lock granted when unmounting, the only place that a
210 		 * wakeup needs to be done is at the release of the
211 		 * exclusive lock at the end of dounmount.
212 		 */
213 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
214 		if (interlkp) {
215 			simple_lock(interlkp);
216 		}
217 		return (ENOENT);
218 	}
219 	lkflags = LK_SHARED | LK_NOPAUSE;
220 	if (interlkp)
221 		lkflags |= LK_INTERLOCK;
222 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
223 		panic("vfs_busy: unexpected lock failure");
224 	return (0);
225 }
226 
227 /*
228  * Free a busy filesystem.
229  */
230 void
231 vfs_unbusy(mp, p)
232 	struct mount *mp;
233 	struct proc *p;
234 {
235 
236 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
237 }
238 
239 /*
240  * Lookup a filesystem type, and if found allocate and initialize
241  * a mount structure for it.
242  *
243  * Devname is usually updated by mount(8) after booting.
244  */
245 int
246 vfs_rootmountalloc(fstypename, devname, mpp)
247 	char *fstypename;
248 	char *devname;
249 	struct mount **mpp;
250 {
251 	struct proc *p = curproc;	/* XXX */
252 	struct vfsconf *vfsp;
253 	struct mount *mp;
254 
255 	if (fstypename == NULL)
256 		return (ENODEV);
257 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
258 		if (!strcmp(vfsp->vfc_name, fstypename))
259 			break;
260 	if (vfsp == NULL)
261 		return (ENODEV);
262 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
263 	bzero((char *)mp, (u_long)sizeof(struct mount));
264 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
265 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
266 	LIST_INIT(&mp->mnt_vnodelist);
267 	mp->mnt_vfc = vfsp;
268 	mp->mnt_op = vfsp->vfc_vfsops;
269 	mp->mnt_flag = MNT_RDONLY;
270 	mp->mnt_vnodecovered = NULLVP;
271 	vfsp->vfc_refcount++;
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 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
320 	    mp = mp->mnt_list.cqe_next) {
321 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
322 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
323 			simple_unlock(&mountlist_slock);
324 			return (mp);
325 	    }
326 	}
327 	simple_unlock(&mountlist_slock);
328 	return ((struct mount *) 0);
329 }
330 
331 /*
332  * Get a new unique fsid
333  */
334 void
335 vfs_getnewfsid(mp)
336 	struct mount *mp;
337 {
338 	static u_short xxxfs_mntid;
339 
340 	fsid_t tfsid;
341 	int mtype;
342 
343 	simple_lock(&mntid_slock);
344 	mtype = mp->mnt_vfc->vfc_typenum;
345 	mp->mnt_stat.f_fsid.val[0] = makeudev(255, mtype);
346 	mp->mnt_stat.f_fsid.val[1] = mtype;
347 	if (xxxfs_mntid == 0)
348 		++xxxfs_mntid;
349 	tfsid.val[0] = makeudev(255, mtype + (xxxfs_mntid << 16));
350 	tfsid.val[1] = mtype;
351 	if (mountlist.cqh_first != (void *)&mountlist) {
352 		while (vfs_getvfs(&tfsid)) {
353 			xxxfs_mntid++;
354 			tfsid.val[0] = makeudev(255,
355 			    mtype + (xxxfs_mntid << 16));
356 		}
357 	}
358 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
359 	simple_unlock(&mntid_slock);
360 }
361 
362 /*
363  * Knob to control the precision of file timestamps:
364  *
365  *   0 = seconds only; nanoseconds zeroed.
366  *   1 = seconds and nanoseconds, accurate within 1/HZ.
367  *   2 = seconds and nanoseconds, truncated to microseconds.
368  * >=3 = seconds and nanoseconds, maximum precision.
369  */
370 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
371 
372 static int timestamp_precision = TSP_SEC;
373 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
374     &timestamp_precision, 0, "");
375 
376 /*
377  * Get a current timestamp.
378  */
379 void
380 vfs_timestamp(tsp)
381 	struct timespec *tsp;
382 {
383 	struct timeval tv;
384 
385 	switch (timestamp_precision) {
386 	case TSP_SEC:
387 		tsp->tv_sec = time_second;
388 		tsp->tv_nsec = 0;
389 		break;
390 	case TSP_HZ:
391 		getnanotime(tsp);
392 		break;
393 	case TSP_USEC:
394 		microtime(&tv);
395 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
396 		break;
397 	case TSP_NSEC:
398 	default:
399 		nanotime(tsp);
400 		break;
401 	}
402 }
403 
404 /*
405  * Set vnode attributes to VNOVAL
406  */
407 void
408 vattr_null(vap)
409 	register struct vattr *vap;
410 {
411 
412 	vap->va_type = VNON;
413 	vap->va_size = VNOVAL;
414 	vap->va_bytes = VNOVAL;
415 	vap->va_mode = VNOVAL;
416 	vap->va_nlink = VNOVAL;
417 	vap->va_uid = VNOVAL;
418 	vap->va_gid = VNOVAL;
419 	vap->va_fsid = VNOVAL;
420 	vap->va_fileid = VNOVAL;
421 	vap->va_blocksize = VNOVAL;
422 	vap->va_rdev = VNOVAL;
423 	vap->va_atime.tv_sec = VNOVAL;
424 	vap->va_atime.tv_nsec = VNOVAL;
425 	vap->va_mtime.tv_sec = VNOVAL;
426 	vap->va_mtime.tv_nsec = VNOVAL;
427 	vap->va_ctime.tv_sec = VNOVAL;
428 	vap->va_ctime.tv_nsec = VNOVAL;
429 	vap->va_flags = VNOVAL;
430 	vap->va_gen = VNOVAL;
431 	vap->va_vaflags = 0;
432 }
433 
434 /*
435  * Routines having to do with the management of the vnode table.
436  */
437 extern vop_t **dead_vnodeop_p;
438 
439 /*
440  * Return the next vnode from the free list.
441  */
442 int
443 getnewvnode(tag, mp, vops, vpp)
444 	enum vtagtype tag;
445 	struct mount *mp;
446 	vop_t **vops;
447 	struct vnode **vpp;
448 {
449 	int s;
450 	struct proc *p = curproc;	/* XXX */
451 	struct vnode *vp, *tvp, *nvp;
452 	vm_object_t object;
453 	TAILQ_HEAD(freelst, vnode) vnode_tmp_list;
454 
455 	/*
456 	 * We take the least recently used vnode from the freelist
457 	 * if we can get it and it has no cached pages, and no
458 	 * namecache entries are relative to it.
459 	 * Otherwise we allocate a new vnode
460 	 */
461 
462 	s = splbio();
463 	simple_lock(&vnode_free_list_slock);
464 	TAILQ_INIT(&vnode_tmp_list);
465 
466 	for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) {
467 		nvp = TAILQ_NEXT(vp, v_freelist);
468 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
469 		if (vp->v_flag & VAGE) {
470 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
471 		} else {
472 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
473 		}
474 		vp->v_flag &= ~(VTBFREE|VAGE);
475 		vp->v_flag |= VFREE;
476 		if (vp->v_usecount)
477 			panic("tobe free vnode isn't");
478 		freevnodes++;
479 	}
480 
481 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
482 		vp = NULL;
483 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
484 		/*
485 		 * XXX: this is only here to be backwards compatible
486 		 */
487 		vp = NULL;
488 	} else {
489 		for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) {
490 			nvp = TAILQ_NEXT(vp, v_freelist);
491 			if (!simple_lock_try(&vp->v_interlock))
492 				continue;
493 			if (vp->v_usecount)
494 				panic("free vnode isn't");
495 
496 			object = vp->v_object;
497 			if (object && (object->resident_page_count || object->ref_count)) {
498 				printf("object inconsistant state: RPC: %d, RC: %d\n",
499 					object->resident_page_count, object->ref_count);
500 				/* Don't recycle if it's caching some pages */
501 				TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
502 				TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist);
503 				continue;
504 			} else if (LIST_FIRST(&vp->v_cache_src)) {
505 				/* Don't recycle if active in the namecache */
506 				simple_unlock(&vp->v_interlock);
507 				continue;
508 			} else {
509 				break;
510 			}
511 		}
512 	}
513 
514 	for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) {
515 		nvp = TAILQ_NEXT(tvp, v_freelist);
516 		TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist);
517 		TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist);
518 		simple_unlock(&tvp->v_interlock);
519 	}
520 
521 	if (vp) {
522 		vp->v_flag |= VDOOMED;
523 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
524 		freevnodes--;
525 		simple_unlock(&vnode_free_list_slock);
526 		cache_purge(vp);
527 		vp->v_lease = NULL;
528 		if (vp->v_type != VBAD) {
529 			vgonel(vp, p);
530 		} else {
531 			simple_unlock(&vp->v_interlock);
532 		}
533 
534 #ifdef INVARIANTS
535 		{
536 			int s;
537 
538 			if (vp->v_data)
539 				panic("cleaned vnode isn't");
540 			s = splbio();
541 			if (vp->v_numoutput)
542 				panic("Clean vnode has pending I/O's");
543 			splx(s);
544 		}
545 #endif
546 		vp->v_flag = 0;
547 		vp->v_lastr = 0;
548 		vp->v_lastw = 0;
549 		vp->v_lasta = 0;
550 		vp->v_cstart = 0;
551 		vp->v_clen = 0;
552 		vp->v_socket = 0;
553 		vp->v_writecount = 0;	/* XXX */
554 		vp->v_maxio = 0;
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 						VOP_BWRITE(bp->b_vp, bp);
706 					}
707 				} else {
708 					bremfree(bp);
709 					(void) VOP_BWRITE(bp->b_vp, 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 && (((nbp->b_xflags & B_VNCLEAN) == 0)||
785 					 (nbp->b_vp != vp) ||
786 					 (nbp->b_flags & B_DELWRI))) {
787 					goto restart;
788 				}
789 			}
790 		}
791 
792 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
793 			nbp = TAILQ_NEXT(bp, b_vnbufs);
794 			if (bp->b_lblkno >= trunclbn) {
795 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
796 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
797 					goto restart;
798 				} else {
799 					bremfree(bp);
800 					bp->b_flags |= (B_INVAL | B_RELBUF);
801 					bp->b_flags &= ~B_ASYNC;
802 					brelse(bp);
803 					anyfreed = 1;
804 				}
805 				if (nbp && (((nbp->b_xflags & B_VNDIRTY) == 0)||
806 					 (nbp->b_vp != vp) ||
807 					 (nbp->b_flags & B_DELWRI) == 0)) {
808 					goto restart;
809 				}
810 			}
811 		}
812 	}
813 
814 	if (length > 0) {
815 restartsync:
816 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
817 			nbp = TAILQ_NEXT(bp, b_vnbufs);
818 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
819 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
820 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
821 					goto restart;
822 				} else {
823 					bremfree(bp);
824 					if (bp->b_vp == vp) {
825 						bp->b_flags |= B_ASYNC;
826 					} else {
827 						bp->b_flags &= ~B_ASYNC;
828 					}
829 					VOP_BWRITE(bp->b_vp, bp);
830 				}
831 				goto restartsync;
832 			}
833 
834 		}
835 	}
836 
837 	while (vp->v_numoutput > 0) {
838 		vp->v_flag |= VBWAIT;
839 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
840 	}
841 
842 	splx(s);
843 
844 	vnode_pager_setsize(vp, length);
845 
846 	return (0);
847 }
848 
849 /*
850  * Associate a buffer with a vnode.
851  */
852 void
853 bgetvp(vp, bp)
854 	register struct vnode *vp;
855 	register struct buf *bp;
856 {
857 	int s;
858 
859 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
860 
861 	vhold(vp);
862 	bp->b_vp = vp;
863 	bp->b_dev = vn_todev(vp);
864 	/*
865 	 * Insert onto list for new vnode.
866 	 */
867 	s = splbio();
868 	bp->b_xflags |= B_VNCLEAN;
869 	bp->b_xflags &= ~B_VNDIRTY;
870 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
871 	splx(s);
872 }
873 
874 /*
875  * Disassociate a buffer from a vnode.
876  */
877 void
878 brelvp(bp)
879 	register struct buf *bp;
880 {
881 	struct vnode *vp;
882 	struct buflists *listheadp;
883 	int s;
884 
885 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
886 
887 	/*
888 	 * Delete from old vnode list, if on one.
889 	 */
890 	vp = bp->b_vp;
891 	s = splbio();
892 	if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
893 		if (bp->b_xflags & B_VNDIRTY)
894 			listheadp = &vp->v_dirtyblkhd;
895 		else
896 			listheadp = &vp->v_cleanblkhd;
897 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
898 		bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
899 	}
900 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
901 		vp->v_flag &= ~VONWORKLST;
902 		LIST_REMOVE(vp, v_synclist);
903 	}
904 	splx(s);
905 	bp->b_vp = (struct vnode *) 0;
906 	vdrop(vp);
907 }
908 
909 /*
910  * The workitem queue.
911  *
912  * It is useful to delay writes of file data and filesystem metadata
913  * for tens of seconds so that quickly created and deleted files need
914  * not waste disk bandwidth being created and removed. To realize this,
915  * we append vnodes to a "workitem" queue. When running with a soft
916  * updates implementation, most pending metadata dependencies should
917  * not wait for more than a few seconds. Thus, mounted on block devices
918  * are delayed only about a half the time that file data is delayed.
919  * Similarly, directory updates are more critical, so are only delayed
920  * about a third the time that file data is delayed. Thus, there are
921  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
922  * one each second (driven off the filesystem syncer process). The
923  * syncer_delayno variable indicates the next queue that is to be processed.
924  * Items that need to be processed soon are placed in this queue:
925  *
926  *	syncer_workitem_pending[syncer_delayno]
927  *
928  * A delay of fifteen seconds is done by placing the request fifteen
929  * entries later in the queue:
930  *
931  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
932  *
933  */
934 
935 /*
936  * Add an item to the syncer work queue.
937  */
938 static void
939 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
940 {
941 	int s, slot;
942 
943 	s = splbio();
944 
945 	if (vp->v_flag & VONWORKLST) {
946 		LIST_REMOVE(vp, v_synclist);
947 	}
948 
949 	if (delay > syncer_maxdelay - 2)
950 		delay = syncer_maxdelay - 2;
951 	slot = (syncer_delayno + delay) & syncer_mask;
952 
953 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
954 	vp->v_flag |= VONWORKLST;
955 	splx(s);
956 }
957 
958 struct  proc *updateproc;
959 static void sched_sync __P((void));
960 static struct kproc_desc up_kp = {
961 	"syncer",
962 	sched_sync,
963 	&updateproc
964 };
965 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
966 
967 /*
968  * System filesystem synchronizer daemon.
969  */
970 void
971 sched_sync(void)
972 {
973 	struct synclist *slp;
974 	struct vnode *vp;
975 	long starttime;
976 	int s;
977 	struct proc *p = updateproc;
978 
979 	p->p_flag |= P_BUFEXHAUST;
980 
981 	for (;;) {
982 		starttime = time_second;
983 
984 		/*
985 		 * Push files whose dirty time has expired.  Be careful
986 		 * of interrupt race on slp queue.
987 		 */
988 		s = splbio();
989 		slp = &syncer_workitem_pending[syncer_delayno];
990 		syncer_delayno += 1;
991 		if (syncer_delayno == syncer_maxdelay)
992 			syncer_delayno = 0;
993 		splx(s);
994 
995 		while ((vp = LIST_FIRST(slp)) != NULL) {
996 			if (VOP_ISLOCKED(vp) == 0) {
997 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
998 				(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
999 				VOP_UNLOCK(vp, 0, p);
1000 			}
1001 			s = splbio();
1002 			if (LIST_FIRST(slp) == vp) {
1003 				/*
1004 				 * Note: v_tag VT_VFS vps can remain on the
1005 				 * worklist too with no dirty blocks, but
1006 				 * since sync_fsync() moves it to a different
1007 				 * slot we are safe.
1008 				 */
1009 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1010 				    vp->v_type != VBLK)
1011 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1012 				/*
1013 				 * Put us back on the worklist.  The worklist
1014 				 * routine will remove us from our current
1015 				 * position and then add us back in at a later
1016 				 * position.
1017 				 */
1018 				vn_syncer_add_to_worklist(vp, syncdelay);
1019 			}
1020 			splx(s);
1021 		}
1022 
1023 		/*
1024 		 * Do soft update processing.
1025 		 */
1026 		if (bioops.io_sync)
1027 			(*bioops.io_sync)(NULL);
1028 
1029 		/*
1030 		 * The variable rushjob allows the kernel to speed up the
1031 		 * processing of the filesystem syncer process. A rushjob
1032 		 * value of N tells the filesystem syncer to process the next
1033 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1034 		 * is used by the soft update code to speed up the filesystem
1035 		 * syncer process when the incore state is getting so far
1036 		 * ahead of the disk that the kernel memory pool is being
1037 		 * threatened with exhaustion.
1038 		 */
1039 		if (rushjob > 0) {
1040 			rushjob -= 1;
1041 			continue;
1042 		}
1043 		/*
1044 		 * If it has taken us less than a second to process the
1045 		 * current work, then wait. Otherwise start right over
1046 		 * again. We can still lose time if any single round
1047 		 * takes more than two seconds, but it does not really
1048 		 * matter as we are just trying to generally pace the
1049 		 * filesystem activity.
1050 		 */
1051 		if (time_second == starttime)
1052 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1053 	}
1054 }
1055 
1056 /*
1057  * Request the syncer daemon to speed up its work.
1058  * We never push it to speed up more than half of its
1059  * normal turn time, otherwise it could take over the cpu.
1060  */
1061 int
1062 speedup_syncer()
1063 {
1064 	int s;
1065 
1066 	s = splhigh();
1067 	if (updateproc->p_wchan == &lbolt)
1068 		setrunnable(updateproc);
1069 	splx(s);
1070 	if (rushjob < syncdelay / 2) {
1071 		rushjob += 1;
1072 		stat_rush_requests += 1;
1073 		return (1);
1074 	}
1075 	return(0);
1076 }
1077 
1078 /*
1079  * Associate a p-buffer with a vnode.
1080  *
1081  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1082  * with the buffer.  i.e. the bp has not been linked into the vnode or
1083  * ref-counted.
1084  */
1085 void
1086 pbgetvp(vp, bp)
1087 	register struct vnode *vp;
1088 	register struct buf *bp;
1089 {
1090 
1091 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1092 
1093 	bp->b_vp = vp;
1094 	bp->b_flags |= B_PAGING;
1095 	bp->b_dev = vn_todev(vp);
1096 }
1097 
1098 /*
1099  * Disassociate a p-buffer from a vnode.
1100  */
1101 void
1102 pbrelvp(bp)
1103 	register struct buf *bp;
1104 {
1105 
1106 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1107 
1108 #if !defined(MAX_PERF)
1109 	/* XXX REMOVE ME */
1110 	if (bp->b_vnbufs.tqe_next != NULL) {
1111 		panic(
1112 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1113 		    bp,
1114 		    (int)bp->b_flags
1115 		);
1116 	}
1117 #endif
1118 	bp->b_vp = (struct vnode *) 0;
1119 	bp->b_flags &= ~B_PAGING;
1120 }
1121 
1122 void
1123 pbreassignbuf(bp, newvp)
1124 	struct buf *bp;
1125 	struct vnode *newvp;
1126 {
1127 #if !defined(MAX_PERF)
1128 	if ((bp->b_flags & B_PAGING) == 0) {
1129 		panic(
1130 		    "pbreassignbuf() on non phys bp %p",
1131 		    bp
1132 		);
1133 	}
1134 #endif
1135 	bp->b_vp = newvp;
1136 }
1137 
1138 /*
1139  * Reassign a buffer from one vnode to another.
1140  * Used to assign file specific control information
1141  * (indirect blocks) to the vnode to which they belong.
1142  */
1143 void
1144 reassignbuf(bp, newvp)
1145 	register struct buf *bp;
1146 	register struct vnode *newvp;
1147 {
1148 	struct buflists *listheadp;
1149 	int delay;
1150 	int s;
1151 
1152 	if (newvp == NULL) {
1153 		printf("reassignbuf: NULL");
1154 		return;
1155 	}
1156 	++reassignbufcalls;
1157 
1158 #if !defined(MAX_PERF)
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 #endif
1166 
1167 	s = splbio();
1168 	/*
1169 	 * Delete from old vnode list, if on one.
1170 	 */
1171 	if (bp->b_xflags & (B_VNDIRTY|B_VNCLEAN)) {
1172 		if (bp->b_xflags & B_VNDIRTY)
1173 			listheadp = &bp->b_vp->v_dirtyblkhd;
1174 		else
1175 			listheadp = &bp->b_vp->v_cleanblkhd;
1176 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1177 		bp->b_xflags &= ~(B_VNDIRTY|B_VNCLEAN);
1178 		if (bp->b_vp != newvp) {
1179 			vdrop(bp->b_vp);
1180 			bp->b_vp = NULL;	/* for clarification */
1181 		}
1182 	}
1183 	/*
1184 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1185 	 * of clean buffers.
1186 	 */
1187 	if (bp->b_flags & B_DELWRI) {
1188 		struct buf *tbp;
1189 
1190 		listheadp = &newvp->v_dirtyblkhd;
1191 		if ((newvp->v_flag & VONWORKLST) == 0) {
1192 			switch (newvp->v_type) {
1193 			case VDIR:
1194 				delay = dirdelay;
1195 				break;
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 |= B_VNDIRTY;
1208 		tbp = TAILQ_FIRST(listheadp);
1209 		if (tbp == NULL ||
1210 		    bp->b_lblkno == 0 ||
1211 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1212 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1213 			++reassignbufsortgood;
1214 		} else if (bp->b_lblkno < 0) {
1215 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1216 			++reassignbufsortgood;
1217 		} else if (reassignbufmethod == 1) {
1218 			/*
1219 			 * New sorting algorithm, only handle sequential case,
1220 			 * otherwise guess.
1221 			 */
1222 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1223 			    (tbp->b_xflags & B_VNDIRTY)) {
1224 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1225 				++reassignbufsortgood;
1226 			} else {
1227 				TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1228 				++reassignbufsortbad;
1229 			}
1230 		} else {
1231 			/*
1232 			 * Old sorting algorithm, scan queue and insert
1233 			 */
1234 			struct buf *ttbp;
1235 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1236 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1237 				++reassignbufloops;
1238 				tbp = ttbp;
1239 			}
1240 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1241 		}
1242 	} else {
1243 		bp->b_xflags |= B_VNCLEAN;
1244 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1245 		if ((newvp->v_flag & VONWORKLST) &&
1246 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1247 			newvp->v_flag &= ~VONWORKLST;
1248 			LIST_REMOVE(newvp, v_synclist);
1249 		}
1250 	}
1251 	if (bp->b_vp != newvp) {
1252 		bp->b_vp = newvp;
1253 		vhold(bp->b_vp);
1254 	}
1255 	splx(s);
1256 }
1257 
1258 /*
1259  * Create a vnode for a block device.
1260  * Used for mounting the root file system.
1261  */
1262 int
1263 bdevvp(dev, vpp)
1264 	dev_t dev;
1265 	struct vnode **vpp;
1266 {
1267 	register struct vnode *vp;
1268 	struct vnode *nvp;
1269 	int error;
1270 
1271 	if (dev == NODEV) {
1272 		*vpp = NULLVP;
1273 		return (ENXIO);
1274 	}
1275 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1276 	if (error) {
1277 		*vpp = NULLVP;
1278 		return (error);
1279 	}
1280 	vp = nvp;
1281 	vp->v_type = VBLK;
1282 	addalias(vp, dev);
1283 	*vpp = vp;
1284 	return (0);
1285 }
1286 
1287 /*
1288  * Add vnode to the alias list hung off the dev_t.
1289  *
1290  * The reason for this gunk is that multiple vnodes can reference
1291  * the same physical device, so checking vp->v_usecount to see
1292  * how many users there are is inadequate; the v_usecount for
1293  * the vnodes need to be accumulated.  vcount() does that.
1294  */
1295 void
1296 addaliasu(nvp, nvp_rdev)
1297 	struct vnode *nvp;
1298 	udev_t nvp_rdev;
1299 {
1300 
1301 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1302 		panic("addaliasu on non-special vnode");
1303 
1304 	nvp->v_rdev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
1305 	simple_lock(&spechash_slock);
1306 	SLIST_INSERT_HEAD(&nvp->v_rdev->si_hlist, nvp, v_specnext);
1307 	simple_unlock(&spechash_slock);
1308 }
1309 
1310 void
1311 addalias(nvp, dev)
1312 	struct vnode *nvp;
1313 	dev_t dev;
1314 {
1315 
1316 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1317 		panic("addalias on non-special vnode");
1318 
1319 	nvp->v_rdev = dev;
1320 	simple_lock(&spechash_slock);
1321 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1322 	simple_unlock(&spechash_slock);
1323 }
1324 
1325 /*
1326  * Grab a particular vnode from the free list, increment its
1327  * reference count and lock it. The vnode lock bit is set if the
1328  * vnode is being eliminated in vgone. The process is awakened
1329  * when the transition is completed, and an error returned to
1330  * indicate that the vnode is no longer usable (possibly having
1331  * been changed to a new file system type).
1332  */
1333 int
1334 vget(vp, flags, p)
1335 	register struct vnode *vp;
1336 	int flags;
1337 	struct proc *p;
1338 {
1339 	int error;
1340 
1341 	/*
1342 	 * If the vnode is in the process of being cleaned out for
1343 	 * another use, we wait for the cleaning to finish and then
1344 	 * return failure. Cleaning is determined by checking that
1345 	 * the VXLOCK flag is set.
1346 	 */
1347 	if ((flags & LK_INTERLOCK) == 0) {
1348 		simple_lock(&vp->v_interlock);
1349 	}
1350 	if (vp->v_flag & VXLOCK) {
1351 		vp->v_flag |= VXWANT;
1352 		simple_unlock(&vp->v_interlock);
1353 		tsleep((caddr_t)vp, PINOD, "vget", 0);
1354 		return (ENOENT);
1355 	}
1356 
1357 	vp->v_usecount++;
1358 
1359 	if (VSHOULDBUSY(vp))
1360 		vbusy(vp);
1361 	if (flags & LK_TYPE_MASK) {
1362 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1363 			/*
1364 			 * must expand vrele here because we do not want
1365 			 * to call VOP_INACTIVE if the reference count
1366 			 * drops back to zero since it was never really
1367 			 * active. We must remove it from the free list
1368 			 * before sleeping so that multiple processes do
1369 			 * not try to recycle it.
1370 			 */
1371 			simple_lock(&vp->v_interlock);
1372 			vp->v_usecount--;
1373 			if (VSHOULDFREE(vp))
1374 				vfree(vp);
1375 			simple_unlock(&vp->v_interlock);
1376 		}
1377 		return (error);
1378 	}
1379 	simple_unlock(&vp->v_interlock);
1380 	return (0);
1381 }
1382 
1383 void
1384 vref(struct vnode *vp)
1385 {
1386 	simple_lock(&vp->v_interlock);
1387 	vp->v_usecount++;
1388 	simple_unlock(&vp->v_interlock);
1389 }
1390 
1391 /*
1392  * Vnode put/release.
1393  * If count drops to zero, call inactive routine and return to freelist.
1394  */
1395 void
1396 vrele(vp)
1397 	struct vnode *vp;
1398 {
1399 	struct proc *p = curproc;	/* XXX */
1400 
1401 	KASSERT(vp != NULL, ("vrele: null vp"));
1402 
1403 	simple_lock(&vp->v_interlock);
1404 
1405 	if (vp->v_usecount > 1) {
1406 
1407 		vp->v_usecount--;
1408 		simple_unlock(&vp->v_interlock);
1409 
1410 		return;
1411 	}
1412 
1413 	if (vp->v_usecount == 1) {
1414 
1415 		vp->v_usecount--;
1416 		if (VSHOULDFREE(vp))
1417 			vfree(vp);
1418 	/*
1419 	 * If we are doing a vput, the node is already locked, and we must
1420 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1421 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1422 	 */
1423 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1424 			VOP_INACTIVE(vp, p);
1425 		}
1426 
1427 	} else {
1428 #ifdef DIAGNOSTIC
1429 		vprint("vrele: negative ref count", vp);
1430 		simple_unlock(&vp->v_interlock);
1431 #endif
1432 		panic("vrele: negative ref cnt");
1433 	}
1434 }
1435 
1436 void
1437 vput(vp)
1438 	struct vnode *vp;
1439 {
1440 	struct proc *p = curproc;	/* XXX */
1441 
1442 	KASSERT(vp != NULL, ("vput: null vp"));
1443 
1444 	simple_lock(&vp->v_interlock);
1445 
1446 	if (vp->v_usecount > 1) {
1447 
1448 		vp->v_usecount--;
1449 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1450 		return;
1451 
1452 	}
1453 
1454 	if (vp->v_usecount == 1) {
1455 
1456 		vp->v_usecount--;
1457 		if (VSHOULDFREE(vp))
1458 			vfree(vp);
1459 	/*
1460 	 * If we are doing a vput, the node is already locked, and we must
1461 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1462 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1463 	 */
1464 		simple_unlock(&vp->v_interlock);
1465 		VOP_INACTIVE(vp, p);
1466 
1467 	} else {
1468 #ifdef DIAGNOSTIC
1469 		vprint("vput: negative ref count", vp);
1470 #endif
1471 		panic("vput: negative ref cnt");
1472 	}
1473 }
1474 
1475 /*
1476  * Somebody doesn't want the vnode recycled.
1477  */
1478 void
1479 vhold(vp)
1480 	register struct vnode *vp;
1481 {
1482 	int s;
1483 
1484   	s = splbio();
1485 	vp->v_holdcnt++;
1486 	if (VSHOULDBUSY(vp))
1487 		vbusy(vp);
1488 	splx(s);
1489 }
1490 
1491 /*
1492  * One less who cares about this vnode.
1493  */
1494 void
1495 vdrop(vp)
1496 	register struct vnode *vp;
1497 {
1498 	int s;
1499 
1500 	s = splbio();
1501 	if (vp->v_holdcnt <= 0)
1502 		panic("vdrop: holdcnt");
1503 	vp->v_holdcnt--;
1504 	if (VSHOULDFREE(vp))
1505 		vfree(vp);
1506 	splx(s);
1507 }
1508 
1509 /*
1510  * Remove any vnodes in the vnode table belonging to mount point mp.
1511  *
1512  * If MNT_NOFORCE is specified, there should not be any active ones,
1513  * return error if any are found (nb: this is a user error, not a
1514  * system error). If MNT_FORCE is specified, detach any active vnodes
1515  * that are found.
1516  */
1517 #ifdef DIAGNOSTIC
1518 static int busyprt = 0;		/* print out busy vnodes */
1519 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1520 #endif
1521 
1522 int
1523 vflush(mp, skipvp, flags)
1524 	struct mount *mp;
1525 	struct vnode *skipvp;
1526 	int flags;
1527 {
1528 	struct proc *p = curproc;	/* XXX */
1529 	struct vnode *vp, *nvp;
1530 	int busy = 0;
1531 
1532 	simple_lock(&mntvnode_slock);
1533 loop:
1534 	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
1535 		/*
1536 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1537 		 * Start over if it has (it won't be on the list anymore).
1538 		 */
1539 		if (vp->v_mount != mp)
1540 			goto loop;
1541 		nvp = vp->v_mntvnodes.le_next;
1542 		/*
1543 		 * Skip over a selected vnode.
1544 		 */
1545 		if (vp == skipvp)
1546 			continue;
1547 
1548 		simple_lock(&vp->v_interlock);
1549 		/*
1550 		 * Skip over a vnodes marked VSYSTEM.
1551 		 */
1552 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1553 			simple_unlock(&vp->v_interlock);
1554 			continue;
1555 		}
1556 		/*
1557 		 * If WRITECLOSE is set, only flush out regular file vnodes
1558 		 * open for writing.
1559 		 */
1560 		if ((flags & WRITECLOSE) &&
1561 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1562 			simple_unlock(&vp->v_interlock);
1563 			continue;
1564 		}
1565 
1566 		/*
1567 		 * With v_usecount == 0, all we need to do is clear out the
1568 		 * vnode data structures and we are done.
1569 		 */
1570 		if (vp->v_usecount == 0) {
1571 			simple_unlock(&mntvnode_slock);
1572 			vgonel(vp, p);
1573 			simple_lock(&mntvnode_slock);
1574 			continue;
1575 		}
1576 
1577 		/*
1578 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1579 		 * or character devices, revert to an anonymous device. For
1580 		 * all other files, just kill them.
1581 		 */
1582 		if (flags & FORCECLOSE) {
1583 			simple_unlock(&mntvnode_slock);
1584 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1585 				vgonel(vp, p);
1586 			} else {
1587 				vclean(vp, 0, p);
1588 				vp->v_op = spec_vnodeop_p;
1589 				insmntque(vp, (struct mount *) 0);
1590 			}
1591 			simple_lock(&mntvnode_slock);
1592 			continue;
1593 		}
1594 #ifdef DIAGNOSTIC
1595 		if (busyprt)
1596 			vprint("vflush: busy vnode", vp);
1597 #endif
1598 		simple_unlock(&vp->v_interlock);
1599 		busy++;
1600 	}
1601 	simple_unlock(&mntvnode_slock);
1602 	if (busy)
1603 		return (EBUSY);
1604 	return (0);
1605 }
1606 
1607 /*
1608  * Disassociate the underlying file system from a vnode.
1609  */
1610 static void
1611 vclean(vp, flags, p)
1612 	struct vnode *vp;
1613 	int flags;
1614 	struct proc *p;
1615 {
1616 	int active;
1617 	vm_object_t obj;
1618 
1619 	/*
1620 	 * Check to see if the vnode is in use. If so we have to reference it
1621 	 * before we clean it out so that its count cannot fall to zero and
1622 	 * generate a race against ourselves to recycle it.
1623 	 */
1624 	if ((active = vp->v_usecount))
1625 		vp->v_usecount++;
1626 
1627 	/*
1628 	 * Prevent the vnode from being recycled or brought into use while we
1629 	 * clean it out.
1630 	 */
1631 	if (vp->v_flag & VXLOCK)
1632 		panic("vclean: deadlock");
1633 	vp->v_flag |= VXLOCK;
1634 	/*
1635 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1636 	 * have the object locked while it cleans it out. The VOP_LOCK
1637 	 * ensures that the VOP_INACTIVE routine is done with its work.
1638 	 * For active vnodes, it ensures that no other activity can
1639 	 * occur while the underlying object is being cleaned out.
1640 	 */
1641 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1642 
1643 	/*
1644 	 * Clean out any buffers associated with the vnode.
1645 	 */
1646 	vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1647 	if ((obj = vp->v_object) != NULL) {
1648 		if (obj->ref_count == 0) {
1649 			/*
1650 			 * This is a normal way of shutting down the object/vnode
1651 			 * association.
1652 			 */
1653 			vm_object_terminate(obj);
1654 		} else {
1655 			/*
1656 			 * Woe to the process that tries to page now :-).
1657 			 */
1658 			vm_pager_deallocate(obj);
1659 		}
1660 	}
1661 
1662 	/*
1663 	 * If purging an active vnode, it must be closed and
1664 	 * deactivated before being reclaimed. Note that the
1665 	 * VOP_INACTIVE will unlock the vnode.
1666 	 */
1667 	if (active) {
1668 		if (flags & DOCLOSE)
1669 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1670 		VOP_INACTIVE(vp, p);
1671 	} else {
1672 		/*
1673 		 * Any other processes trying to obtain this lock must first
1674 		 * wait for VXLOCK to clear, then call the new lock operation.
1675 		 */
1676 		VOP_UNLOCK(vp, 0, p);
1677 	}
1678 	/*
1679 	 * Reclaim the vnode.
1680 	 */
1681 	if (VOP_RECLAIM(vp, p))
1682 		panic("vclean: cannot reclaim");
1683 
1684 	if (active)
1685 		vrele(vp);
1686 
1687 	cache_purge(vp);
1688 	if (vp->v_vnlock) {
1689 		FREE(vp->v_vnlock, M_VNODE);
1690 		vp->v_vnlock = NULL;
1691 	}
1692 
1693 	if (VSHOULDFREE(vp))
1694 		vfree(vp);
1695 
1696 	/*
1697 	 * Done with purge, notify sleepers of the grim news.
1698 	 */
1699 	vp->v_op = dead_vnodeop_p;
1700 	vn_pollgone(vp);
1701 	vp->v_tag = VT_NON;
1702 	vp->v_flag &= ~VXLOCK;
1703 	if (vp->v_flag & VXWANT) {
1704 		vp->v_flag &= ~VXWANT;
1705 		wakeup((caddr_t) vp);
1706 	}
1707 }
1708 
1709 /*
1710  * Eliminate all activity associated with the requested vnode
1711  * and with all vnodes aliased to the requested vnode.
1712  */
1713 int
1714 vop_revoke(ap)
1715 	struct vop_revoke_args /* {
1716 		struct vnode *a_vp;
1717 		int a_flags;
1718 	} */ *ap;
1719 {
1720 	struct vnode *vp, *vq;
1721 	dev_t dev;
1722 
1723 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1724 
1725 	vp = ap->a_vp;
1726 	/*
1727 	 * If a vgone (or vclean) is already in progress,
1728 	 * wait until it is done and return.
1729 	 */
1730 	if (vp->v_flag & VXLOCK) {
1731 		vp->v_flag |= VXWANT;
1732 		simple_unlock(&vp->v_interlock);
1733 		tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0);
1734 		return (0);
1735 	}
1736 	dev = vp->v_rdev;
1737 	for (;;) {
1738 		simple_lock(&spechash_slock);
1739 		vq = SLIST_FIRST(&dev->si_hlist);
1740 		simple_unlock(&spechash_slock);
1741 		if (!vq)
1742 			break;
1743 		vgone(vq);
1744 	}
1745 	return (0);
1746 }
1747 
1748 /*
1749  * Recycle an unused vnode to the front of the free list.
1750  * Release the passed interlock if the vnode will be recycled.
1751  */
1752 int
1753 vrecycle(vp, inter_lkp, p)
1754 	struct vnode *vp;
1755 	struct simplelock *inter_lkp;
1756 	struct proc *p;
1757 {
1758 
1759 	simple_lock(&vp->v_interlock);
1760 	if (vp->v_usecount == 0) {
1761 		if (inter_lkp) {
1762 			simple_unlock(inter_lkp);
1763 		}
1764 		vgonel(vp, p);
1765 		return (1);
1766 	}
1767 	simple_unlock(&vp->v_interlock);
1768 	return (0);
1769 }
1770 
1771 /*
1772  * Eliminate all activity associated with a vnode
1773  * in preparation for reuse.
1774  */
1775 void
1776 vgone(vp)
1777 	register struct vnode *vp;
1778 {
1779 	struct proc *p = curproc;	/* XXX */
1780 
1781 	simple_lock(&vp->v_interlock);
1782 	vgonel(vp, p);
1783 }
1784 
1785 /*
1786  * vgone, with the vp interlock held.
1787  */
1788 static void
1789 vgonel(vp, p)
1790 	struct vnode *vp;
1791 	struct proc *p;
1792 {
1793 	int s;
1794 
1795 	/*
1796 	 * If a vgone (or vclean) is already in progress,
1797 	 * wait until it is done and return.
1798 	 */
1799 	if (vp->v_flag & VXLOCK) {
1800 		vp->v_flag |= VXWANT;
1801 		simple_unlock(&vp->v_interlock);
1802 		tsleep((caddr_t)vp, PINOD, "vgone", 0);
1803 		return;
1804 	}
1805 
1806 	/*
1807 	 * Clean out the filesystem specific data.
1808 	 */
1809 	vclean(vp, DOCLOSE, p);
1810 	simple_lock(&vp->v_interlock);
1811 
1812 	/*
1813 	 * Delete from old mount point vnode list, if on one.
1814 	 */
1815 	if (vp->v_mount != NULL)
1816 		insmntque(vp, (struct mount *)0);
1817 	/*
1818 	 * If special device, remove it from special device alias list
1819 	 * if it is on one.
1820 	 */
1821 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1822 		simple_lock(&spechash_slock);
1823 		SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
1824 		simple_unlock(&spechash_slock);
1825 		vp->v_rdev = NULL;
1826 	}
1827 
1828 	/*
1829 	 * If it is on the freelist and not already at the head,
1830 	 * move it to the head of the list. The test of the back
1831 	 * pointer and the reference count of zero is because
1832 	 * it will be removed from the free list by getnewvnode,
1833 	 * but will not have its reference count incremented until
1834 	 * after calling vgone. If the reference count were
1835 	 * incremented first, vgone would (incorrectly) try to
1836 	 * close the previous instance of the underlying object.
1837 	 */
1838 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1839 		s = splbio();
1840 		simple_lock(&vnode_free_list_slock);
1841 		if (vp->v_flag & VFREE) {
1842 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1843 		} else if (vp->v_flag & VTBFREE) {
1844 			TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
1845 			vp->v_flag &= ~VTBFREE;
1846 			freevnodes++;
1847 		} else
1848 			freevnodes++;
1849 		vp->v_flag |= VFREE;
1850 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1851 		simple_unlock(&vnode_free_list_slock);
1852 		splx(s);
1853 	}
1854 
1855 	vp->v_type = VBAD;
1856 	simple_unlock(&vp->v_interlock);
1857 }
1858 
1859 /*
1860  * Lookup a vnode by device number.
1861  */
1862 int
1863 vfinddev(dev, type, vpp)
1864 	dev_t dev;
1865 	enum vtype type;
1866 	struct vnode **vpp;
1867 {
1868 	struct vnode *vp;
1869 
1870 	simple_lock(&spechash_slock);
1871 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1872 		if (type == vp->v_type) {
1873 			*vpp = vp;
1874 			simple_unlock(&spechash_slock);
1875 			return (1);
1876 		}
1877 	}
1878 	simple_unlock(&spechash_slock);
1879 	return (0);
1880 }
1881 
1882 /*
1883  * Calculate the total number of references to a special device.
1884  */
1885 int
1886 vcount(vp)
1887 	struct vnode *vp;
1888 {
1889 	struct vnode *vq, *vnext;
1890 	int count;
1891 
1892 	count = 0;
1893 	simple_lock(&spechash_slock);
1894 	SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
1895 		count += vq->v_usecount;
1896 	simple_unlock(&spechash_slock);
1897 	return (count);
1898 }
1899 
1900 /*
1901  * Print out a description of a vnode.
1902  */
1903 static char *typename[] =
1904 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1905 
1906 void
1907 vprint(label, vp)
1908 	char *label;
1909 	struct vnode *vp;
1910 {
1911 	char buf[96];
1912 
1913 	if (label != NULL)
1914 		printf("%s: %p: ", label, (void *)vp);
1915 	else
1916 		printf("%p: ", (void *)vp);
1917 	printf("type %s, usecount %d, writecount %d, refcount %d,",
1918 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1919 	    vp->v_holdcnt);
1920 	buf[0] = '\0';
1921 	if (vp->v_flag & VROOT)
1922 		strcat(buf, "|VROOT");
1923 	if (vp->v_flag & VTEXT)
1924 		strcat(buf, "|VTEXT");
1925 	if (vp->v_flag & VSYSTEM)
1926 		strcat(buf, "|VSYSTEM");
1927 	if (vp->v_flag & VXLOCK)
1928 		strcat(buf, "|VXLOCK");
1929 	if (vp->v_flag & VXWANT)
1930 		strcat(buf, "|VXWANT");
1931 	if (vp->v_flag & VBWAIT)
1932 		strcat(buf, "|VBWAIT");
1933 	if (vp->v_flag & VDOOMED)
1934 		strcat(buf, "|VDOOMED");
1935 	if (vp->v_flag & VFREE)
1936 		strcat(buf, "|VFREE");
1937 	if (vp->v_flag & VOBJBUF)
1938 		strcat(buf, "|VOBJBUF");
1939 	if (buf[0] != '\0')
1940 		printf(" flags (%s)", &buf[1]);
1941 	if (vp->v_data == NULL) {
1942 		printf("\n");
1943 	} else {
1944 		printf("\n\t");
1945 		VOP_PRINT(vp);
1946 	}
1947 }
1948 
1949 #ifdef DDB
1950 #include <ddb/ddb.h>
1951 /*
1952  * List all of the locked vnodes in the system.
1953  * Called when debugging the kernel.
1954  */
1955 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1956 {
1957 	struct proc *p = curproc;	/* XXX */
1958 	struct mount *mp, *nmp;
1959 	struct vnode *vp;
1960 
1961 	printf("Locked vnodes\n");
1962 	simple_lock(&mountlist_slock);
1963 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
1964 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
1965 			nmp = mp->mnt_list.cqe_next;
1966 			continue;
1967 		}
1968 		for (vp = mp->mnt_vnodelist.lh_first;
1969 		     vp != NULL;
1970 		     vp = vp->v_mntvnodes.le_next) {
1971 			if (VOP_ISLOCKED(vp))
1972 				vprint((char *)0, vp);
1973 		}
1974 		simple_lock(&mountlist_slock);
1975 		nmp = mp->mnt_list.cqe_next;
1976 		vfs_unbusy(mp, p);
1977 	}
1978 	simple_unlock(&mountlist_slock);
1979 }
1980 #endif
1981 
1982 /*
1983  * Top level filesystem related information gathering.
1984  */
1985 static int	sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS);
1986 
1987 static int
1988 vfs_sysctl SYSCTL_HANDLER_ARGS
1989 {
1990 	int *name = (int *)arg1 - 1;	/* XXX */
1991 	u_int namelen = arg2 + 1;	/* XXX */
1992 	struct vfsconf *vfsp;
1993 
1994 #if 1 || defined(COMPAT_PRELITE2)
1995 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1996 	if (namelen == 1)
1997 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1998 #endif
1999 
2000 #ifdef notyet
2001 	/* all sysctl names at this level are at least name and field */
2002 	if (namelen < 2)
2003 		return (ENOTDIR);		/* overloaded */
2004 	if (name[0] != VFS_GENERIC) {
2005 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2006 			if (vfsp->vfc_typenum == name[0])
2007 				break;
2008 		if (vfsp == NULL)
2009 			return (EOPNOTSUPP);
2010 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2011 		    oldp, oldlenp, newp, newlen, p));
2012 	}
2013 #endif
2014 	switch (name[1]) {
2015 	case VFS_MAXTYPENUM:
2016 		if (namelen != 2)
2017 			return (ENOTDIR);
2018 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2019 	case VFS_CONF:
2020 		if (namelen != 3)
2021 			return (ENOTDIR);	/* overloaded */
2022 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2023 			if (vfsp->vfc_typenum == name[2])
2024 				break;
2025 		if (vfsp == NULL)
2026 			return (EOPNOTSUPP);
2027 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2028 	}
2029 	return (EOPNOTSUPP);
2030 }
2031 
2032 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2033 	"Generic filesystem");
2034 
2035 #if 1 || defined(COMPAT_PRELITE2)
2036 
2037 static int
2038 sysctl_ovfs_conf SYSCTL_HANDLER_ARGS
2039 {
2040 	int error;
2041 	struct vfsconf *vfsp;
2042 	struct ovfsconf ovfs;
2043 
2044 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2045 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2046 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2047 		ovfs.vfc_index = vfsp->vfc_typenum;
2048 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2049 		ovfs.vfc_flags = vfsp->vfc_flags;
2050 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2051 		if (error)
2052 			return error;
2053 	}
2054 	return 0;
2055 }
2056 
2057 #endif /* 1 || COMPAT_PRELITE2 */
2058 
2059 #if 0
2060 #define KINFO_VNODESLOP	10
2061 /*
2062  * Dump vnode list (via sysctl).
2063  * Copyout address of vnode followed by vnode.
2064  */
2065 /* ARGSUSED */
2066 static int
2067 sysctl_vnode SYSCTL_HANDLER_ARGS
2068 {
2069 	struct proc *p = curproc;	/* XXX */
2070 	struct mount *mp, *nmp;
2071 	struct vnode *nvp, *vp;
2072 	int error;
2073 
2074 #define VPTRSZ	sizeof (struct vnode *)
2075 #define VNODESZ	sizeof (struct vnode)
2076 
2077 	req->lock = 0;
2078 	if (!req->oldptr) /* Make an estimate */
2079 		return (SYSCTL_OUT(req, 0,
2080 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2081 
2082 	simple_lock(&mountlist_slock);
2083 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
2084 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) {
2085 			nmp = mp->mnt_list.cqe_next;
2086 			continue;
2087 		}
2088 again:
2089 		simple_lock(&mntvnode_slock);
2090 		for (vp = mp->mnt_vnodelist.lh_first;
2091 		     vp != NULL;
2092 		     vp = nvp) {
2093 			/*
2094 			 * Check that the vp is still associated with
2095 			 * this filesystem.  RACE: could have been
2096 			 * recycled onto the same filesystem.
2097 			 */
2098 			if (vp->v_mount != mp) {
2099 				simple_unlock(&mntvnode_slock);
2100 				goto again;
2101 			}
2102 			nvp = vp->v_mntvnodes.le_next;
2103 			simple_unlock(&mntvnode_slock);
2104 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2105 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2106 				return (error);
2107 			simple_lock(&mntvnode_slock);
2108 		}
2109 		simple_unlock(&mntvnode_slock);
2110 		simple_lock(&mountlist_slock);
2111 		nmp = mp->mnt_list.cqe_next;
2112 		vfs_unbusy(mp, p);
2113 	}
2114 	simple_unlock(&mountlist_slock);
2115 
2116 	return (0);
2117 }
2118 #endif
2119 
2120 /*
2121  * XXX
2122  * Exporting the vnode list on large systems causes them to crash.
2123  * Exporting the vnode list on medium systems causes sysctl to coredump.
2124  */
2125 #if 0
2126 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2127 	0, 0, sysctl_vnode, "S,vnode", "");
2128 #endif
2129 
2130 /*
2131  * Check to see if a filesystem is mounted on a block device.
2132  */
2133 int
2134 vfs_mountedon(vp)
2135 	struct vnode *vp;
2136 {
2137 
2138 	if (vp->v_specmountpoint != NULL)
2139 		return (EBUSY);
2140 	return (0);
2141 }
2142 
2143 /*
2144  * Unmount all filesystems. The list is traversed in reverse order
2145  * of mounting to avoid dependencies.
2146  */
2147 void
2148 vfs_unmountall()
2149 {
2150 	struct mount *mp, *nmp;
2151 	struct proc *p;
2152 	int error;
2153 
2154 	if (curproc != NULL)
2155 		p = curproc;
2156 	else
2157 		p = initproc;	/* XXX XXX should this be proc0? */
2158 	/*
2159 	 * Since this only runs when rebooting, it is not interlocked.
2160 	 */
2161 	for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2162 		nmp = mp->mnt_list.cqe_prev;
2163 		error = dounmount(mp, MNT_FORCE, p);
2164 		if (error) {
2165 			printf("unmount of %s failed (",
2166 			    mp->mnt_stat.f_mntonname);
2167 			if (error == EBUSY)
2168 				printf("BUSY)\n");
2169 			else
2170 				printf("%d)\n", error);
2171 		}
2172 	}
2173 }
2174 
2175 /*
2176  * Build hash lists of net addresses and hang them off the mount point.
2177  * Called by ufs_mount() to set up the lists of export addresses.
2178  */
2179 static int
2180 vfs_hang_addrlist(mp, nep, argp)
2181 	struct mount *mp;
2182 	struct netexport *nep;
2183 	struct export_args *argp;
2184 {
2185 	register struct netcred *np;
2186 	register struct radix_node_head *rnh;
2187 	register int i;
2188 	struct radix_node *rn;
2189 	struct sockaddr *saddr, *smask = 0;
2190 	struct domain *dom;
2191 	int error;
2192 
2193 	if (argp->ex_addrlen == 0) {
2194 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2195 			return (EPERM);
2196 		np = &nep->ne_defexported;
2197 		np->netc_exflags = argp->ex_flags;
2198 		np->netc_anon = argp->ex_anon;
2199 		np->netc_anon.cr_ref = 1;
2200 		mp->mnt_flag |= MNT_DEFEXPORTED;
2201 		return (0);
2202 	}
2203 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2204 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2205 	bzero((caddr_t) np, i);
2206 	saddr = (struct sockaddr *) (np + 1);
2207 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2208 		goto out;
2209 	if (saddr->sa_len > argp->ex_addrlen)
2210 		saddr->sa_len = argp->ex_addrlen;
2211 	if (argp->ex_masklen) {
2212 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2213 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2214 		if (error)
2215 			goto out;
2216 		if (smask->sa_len > argp->ex_masklen)
2217 			smask->sa_len = argp->ex_masklen;
2218 	}
2219 	i = saddr->sa_family;
2220 	if ((rnh = nep->ne_rtable[i]) == 0) {
2221 		/*
2222 		 * Seems silly to initialize every AF when most are not used,
2223 		 * do so on demand here
2224 		 */
2225 		for (dom = domains; dom; dom = dom->dom_next)
2226 			if (dom->dom_family == i && dom->dom_rtattach) {
2227 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2228 				    dom->dom_rtoffset);
2229 				break;
2230 			}
2231 		if ((rnh = nep->ne_rtable[i]) == 0) {
2232 			error = ENOBUFS;
2233 			goto out;
2234 		}
2235 	}
2236 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2237 	    np->netc_rnodes);
2238 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2239 		error = EPERM;
2240 		goto out;
2241 	}
2242 	np->netc_exflags = argp->ex_flags;
2243 	np->netc_anon = argp->ex_anon;
2244 	np->netc_anon.cr_ref = 1;
2245 	return (0);
2246 out:
2247 	free(np, M_NETADDR);
2248 	return (error);
2249 }
2250 
2251 /* ARGSUSED */
2252 static int
2253 vfs_free_netcred(rn, w)
2254 	struct radix_node *rn;
2255 	void *w;
2256 {
2257 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
2258 
2259 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2260 	free((caddr_t) rn, M_NETADDR);
2261 	return (0);
2262 }
2263 
2264 /*
2265  * Free the net address hash lists that are hanging off the mount points.
2266  */
2267 static void
2268 vfs_free_addrlist(nep)
2269 	struct netexport *nep;
2270 {
2271 	register int i;
2272 	register struct radix_node_head *rnh;
2273 
2274 	for (i = 0; i <= AF_MAX; i++)
2275 		if ((rnh = nep->ne_rtable[i])) {
2276 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2277 			    (caddr_t) rnh);
2278 			free((caddr_t) rnh, M_RTABLE);
2279 			nep->ne_rtable[i] = 0;
2280 		}
2281 }
2282 
2283 int
2284 vfs_export(mp, nep, argp)
2285 	struct mount *mp;
2286 	struct netexport *nep;
2287 	struct export_args *argp;
2288 {
2289 	int error;
2290 
2291 	if (argp->ex_flags & MNT_DELEXPORT) {
2292 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2293 			vfs_setpublicfs(NULL, NULL, NULL);
2294 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2295 		}
2296 		vfs_free_addrlist(nep);
2297 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2298 	}
2299 	if (argp->ex_flags & MNT_EXPORTED) {
2300 		if (argp->ex_flags & MNT_EXPUBLIC) {
2301 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2302 				return (error);
2303 			mp->mnt_flag |= MNT_EXPUBLIC;
2304 		}
2305 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2306 			return (error);
2307 		mp->mnt_flag |= MNT_EXPORTED;
2308 	}
2309 	return (0);
2310 }
2311 
2312 
2313 /*
2314  * Set the publicly exported filesystem (WebNFS). Currently, only
2315  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2316  */
2317 int
2318 vfs_setpublicfs(mp, nep, argp)
2319 	struct mount *mp;
2320 	struct netexport *nep;
2321 	struct export_args *argp;
2322 {
2323 	int error;
2324 	struct vnode *rvp;
2325 	char *cp;
2326 
2327 	/*
2328 	 * mp == NULL -> invalidate the current info, the FS is
2329 	 * no longer exported. May be called from either vfs_export
2330 	 * or unmount, so check if it hasn't already been done.
2331 	 */
2332 	if (mp == NULL) {
2333 		if (nfs_pub.np_valid) {
2334 			nfs_pub.np_valid = 0;
2335 			if (nfs_pub.np_index != NULL) {
2336 				FREE(nfs_pub.np_index, M_TEMP);
2337 				nfs_pub.np_index = NULL;
2338 			}
2339 		}
2340 		return (0);
2341 	}
2342 
2343 	/*
2344 	 * Only one allowed at a time.
2345 	 */
2346 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2347 		return (EBUSY);
2348 
2349 	/*
2350 	 * Get real filehandle for root of exported FS.
2351 	 */
2352 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2353 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2354 
2355 	if ((error = VFS_ROOT(mp, &rvp)))
2356 		return (error);
2357 
2358 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2359 		return (error);
2360 
2361 	vput(rvp);
2362 
2363 	/*
2364 	 * If an indexfile was specified, pull it in.
2365 	 */
2366 	if (argp->ex_indexfile != NULL) {
2367 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2368 		    M_WAITOK);
2369 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2370 		    MAXNAMLEN, (size_t *)0);
2371 		if (!error) {
2372 			/*
2373 			 * Check for illegal filenames.
2374 			 */
2375 			for (cp = nfs_pub.np_index; *cp; cp++) {
2376 				if (*cp == '/') {
2377 					error = EINVAL;
2378 					break;
2379 				}
2380 			}
2381 		}
2382 		if (error) {
2383 			FREE(nfs_pub.np_index, M_TEMP);
2384 			return (error);
2385 		}
2386 	}
2387 
2388 	nfs_pub.np_mount = mp;
2389 	nfs_pub.np_valid = 1;
2390 	return (0);
2391 }
2392 
2393 struct netcred *
2394 vfs_export_lookup(mp, nep, nam)
2395 	register struct mount *mp;
2396 	struct netexport *nep;
2397 	struct sockaddr *nam;
2398 {
2399 	register struct netcred *np;
2400 	register struct radix_node_head *rnh;
2401 	struct sockaddr *saddr;
2402 
2403 	np = NULL;
2404 	if (mp->mnt_flag & MNT_EXPORTED) {
2405 		/*
2406 		 * Lookup in the export list first.
2407 		 */
2408 		if (nam != NULL) {
2409 			saddr = nam;
2410 			rnh = nep->ne_rtable[saddr->sa_family];
2411 			if (rnh != NULL) {
2412 				np = (struct netcred *)
2413 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2414 							      rnh);
2415 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2416 					np = NULL;
2417 			}
2418 		}
2419 		/*
2420 		 * If no address match, use the default if it exists.
2421 		 */
2422 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2423 			np = &nep->ne_defexported;
2424 	}
2425 	return (np);
2426 }
2427 
2428 /*
2429  * perform msync on all vnodes under a mount point
2430  * the mount point must be locked.
2431  */
2432 void
2433 vfs_msync(struct mount *mp, int flags) {
2434 	struct vnode *vp, *nvp;
2435 	struct vm_object *obj;
2436 	int anyio, tries;
2437 
2438 	tries = 5;
2439 loop:
2440 	anyio = 0;
2441 	for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
2442 
2443 		nvp = vp->v_mntvnodes.le_next;
2444 
2445 		if (vp->v_mount != mp) {
2446 			goto loop;
2447 		}
2448 
2449 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2450 			continue;
2451 
2452 		if (flags != MNT_WAIT) {
2453 			obj = vp->v_object;
2454 			if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2455 				continue;
2456 			if (VOP_ISLOCKED(vp))
2457 				continue;
2458 		}
2459 
2460 		simple_lock(&vp->v_interlock);
2461 		if (vp->v_object &&
2462 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
2463 			if (!vget(vp,
2464 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2465 				if (vp->v_object) {
2466 					vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : 0);
2467 					anyio = 1;
2468 				}
2469 				vput(vp);
2470 			}
2471 		} else {
2472 			simple_unlock(&vp->v_interlock);
2473 		}
2474 	}
2475 	if (anyio && (--tries > 0))
2476 		goto loop;
2477 }
2478 
2479 /*
2480  * Create the VM object needed for VMIO and mmap support.  This
2481  * is done for all VREG files in the system.  Some filesystems might
2482  * afford the additional metadata buffering capability of the
2483  * VMIO code by making the device node be VMIO mode also.
2484  *
2485  * vp must be locked when vfs_object_create is called.
2486  */
2487 int
2488 vfs_object_create(vp, p, cred)
2489 	struct vnode *vp;
2490 	struct proc *p;
2491 	struct ucred *cred;
2492 {
2493 	struct vattr vat;
2494 	vm_object_t object;
2495 	int error = 0;
2496 
2497 	if (vp->v_type != VBLK && vn_canvmio(vp) == FALSE)
2498 		return 0;
2499 
2500 retry:
2501 	if ((object = vp->v_object) == NULL) {
2502 		if (vp->v_type == VREG || vp->v_type == VDIR) {
2503 			if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
2504 				goto retn;
2505 			object = vnode_pager_alloc(vp, vat.va_size, 0, 0);
2506 		} else if (devsw(vp->v_rdev) != NULL) {
2507 			/*
2508 			 * This simply allocates the biggest object possible
2509 			 * for a VBLK vnode.  This should be fixed, but doesn't
2510 			 * cause any problems (yet).
2511 			 */
2512 			object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0);
2513 		} else {
2514 			goto retn;
2515 		}
2516 		/*
2517 		 * Dereference the reference we just created.  This assumes
2518 		 * that the object is associated with the vp.
2519 		 */
2520 		object->ref_count--;
2521 		vp->v_usecount--;
2522 	} else {
2523 		if (object->flags & OBJ_DEAD) {
2524 			VOP_UNLOCK(vp, 0, p);
2525 			tsleep(object, PVM, "vodead", 0);
2526 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
2527 			goto retry;
2528 		}
2529 	}
2530 
2531 	KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object"));
2532 	vp->v_flag |= VOBJBUF;
2533 
2534 retn:
2535 	return error;
2536 }
2537 
2538 static void
2539 vfree(vp)
2540 	struct vnode *vp;
2541 {
2542 	int s;
2543 
2544 	s = splbio();
2545 	simple_lock(&vnode_free_list_slock);
2546 	if (vp->v_flag & VTBFREE) {
2547 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2548 		vp->v_flag &= ~VTBFREE;
2549 	}
2550 	if (vp->v_flag & VAGE) {
2551 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2552 	} else {
2553 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2554 	}
2555 	freevnodes++;
2556 	simple_unlock(&vnode_free_list_slock);
2557 	vp->v_flag &= ~VAGE;
2558 	vp->v_flag |= VFREE;
2559 	splx(s);
2560 }
2561 
2562 void
2563 vbusy(vp)
2564 	struct vnode *vp;
2565 {
2566 	int s;
2567 
2568 	s = splbio();
2569 	simple_lock(&vnode_free_list_slock);
2570 	if (vp->v_flag & VTBFREE) {
2571 		TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist);
2572 		vp->v_flag &= ~VTBFREE;
2573 	} else {
2574 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2575 		freevnodes--;
2576 	}
2577 	simple_unlock(&vnode_free_list_slock);
2578 	vp->v_flag &= ~(VFREE|VAGE);
2579 	splx(s);
2580 }
2581 
2582 /*
2583  * Record a process's interest in events which might happen to
2584  * a vnode.  Because poll uses the historic select-style interface
2585  * internally, this routine serves as both the ``check for any
2586  * pending events'' and the ``record my interest in future events''
2587  * functions.  (These are done together, while the lock is held,
2588  * to avoid race conditions.)
2589  */
2590 int
2591 vn_pollrecord(vp, p, events)
2592 	struct vnode *vp;
2593 	struct proc *p;
2594 	short events;
2595 {
2596 	simple_lock(&vp->v_pollinfo.vpi_lock);
2597 	if (vp->v_pollinfo.vpi_revents & events) {
2598 		/*
2599 		 * This leaves events we are not interested
2600 		 * in available for the other process which
2601 		 * which presumably had requested them
2602 		 * (otherwise they would never have been
2603 		 * recorded).
2604 		 */
2605 		events &= vp->v_pollinfo.vpi_revents;
2606 		vp->v_pollinfo.vpi_revents &= ~events;
2607 
2608 		simple_unlock(&vp->v_pollinfo.vpi_lock);
2609 		return events;
2610 	}
2611 	vp->v_pollinfo.vpi_events |= events;
2612 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2613 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2614 	return 0;
2615 }
2616 
2617 /*
2618  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2619  * it is possible for us to miss an event due to race conditions, but
2620  * that condition is expected to be rare, so for the moment it is the
2621  * preferred interface.
2622  */
2623 void
2624 vn_pollevent(vp, events)
2625 	struct vnode *vp;
2626 	short events;
2627 {
2628 	simple_lock(&vp->v_pollinfo.vpi_lock);
2629 	if (vp->v_pollinfo.vpi_events & events) {
2630 		/*
2631 		 * We clear vpi_events so that we don't
2632 		 * call selwakeup() twice if two events are
2633 		 * posted before the polling process(es) is
2634 		 * awakened.  This also ensures that we take at
2635 		 * most one selwakeup() if the polling process
2636 		 * is no longer interested.  However, it does
2637 		 * mean that only one event can be noticed at
2638 		 * a time.  (Perhaps we should only clear those
2639 		 * event bits which we note?) XXX
2640 		 */
2641 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2642 		vp->v_pollinfo.vpi_revents |= events;
2643 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2644 	}
2645 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2646 }
2647 
2648 /*
2649  * Wake up anyone polling on vp because it is being revoked.
2650  * This depends on dead_poll() returning POLLHUP for correct
2651  * behavior.
2652  */
2653 void
2654 vn_pollgone(vp)
2655 	struct vnode *vp;
2656 {
2657 	simple_lock(&vp->v_pollinfo.vpi_lock);
2658 	if (vp->v_pollinfo.vpi_events) {
2659 		vp->v_pollinfo.vpi_events = 0;
2660 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2661 	}
2662 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2663 }
2664 
2665 
2666 
2667 /*
2668  * Routine to create and manage a filesystem syncer vnode.
2669  */
2670 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2671 static int	sync_fsync __P((struct  vop_fsync_args *));
2672 static int	sync_inactive __P((struct  vop_inactive_args *));
2673 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2674 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2675 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2676 static int	sync_print __P((struct vop_print_args *));
2677 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2678 
2679 static vop_t **sync_vnodeop_p;
2680 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2681 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2682 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2683 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2684 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2685 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2686 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2687 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2688 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2689 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2690 	{ NULL, NULL }
2691 };
2692 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2693 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2694 
2695 VNODEOP_SET(sync_vnodeop_opv_desc);
2696 
2697 /*
2698  * Create a new filesystem syncer vnode for the specified mount point.
2699  */
2700 int
2701 vfs_allocate_syncvnode(mp)
2702 	struct mount *mp;
2703 {
2704 	struct vnode *vp;
2705 	static long start, incr, next;
2706 	int error;
2707 
2708 	/* Allocate a new vnode */
2709 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2710 		mp->mnt_syncer = NULL;
2711 		return (error);
2712 	}
2713 	vp->v_type = VNON;
2714 	/*
2715 	 * Place the vnode onto the syncer worklist. We attempt to
2716 	 * scatter them about on the list so that they will go off
2717 	 * at evenly distributed times even if all the filesystems
2718 	 * are mounted at once.
2719 	 */
2720 	next += incr;
2721 	if (next == 0 || next > syncer_maxdelay) {
2722 		start /= 2;
2723 		incr /= 2;
2724 		if (start == 0) {
2725 			start = syncer_maxdelay / 2;
2726 			incr = syncer_maxdelay;
2727 		}
2728 		next = start;
2729 	}
2730 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2731 	mp->mnt_syncer = vp;
2732 	return (0);
2733 }
2734 
2735 /*
2736  * Do a lazy sync of the filesystem.
2737  */
2738 static int
2739 sync_fsync(ap)
2740 	struct vop_fsync_args /* {
2741 		struct vnode *a_vp;
2742 		struct ucred *a_cred;
2743 		int a_waitfor;
2744 		struct proc *a_p;
2745 	} */ *ap;
2746 {
2747 	struct vnode *syncvp = ap->a_vp;
2748 	struct mount *mp = syncvp->v_mount;
2749 	struct proc *p = ap->a_p;
2750 	int asyncflag;
2751 
2752 	/*
2753 	 * We only need to do something if this is a lazy evaluation.
2754 	 */
2755 	if (ap->a_waitfor != MNT_LAZY)
2756 		return (0);
2757 
2758 	/*
2759 	 * Move ourselves to the back of the sync list.
2760 	 */
2761 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2762 
2763 	/*
2764 	 * Walk the list of vnodes pushing all that are dirty and
2765 	 * not already on the sync list.
2766 	 */
2767 	simple_lock(&mountlist_slock);
2768 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) {
2769 		simple_unlock(&mountlist_slock);
2770 		return (0);
2771 	}
2772 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2773 	mp->mnt_flag &= ~MNT_ASYNC;
2774 	vfs_msync(mp, MNT_NOWAIT);
2775 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2776 	if (asyncflag)
2777 		mp->mnt_flag |= MNT_ASYNC;
2778 	vfs_unbusy(mp, p);
2779 	return (0);
2780 }
2781 
2782 /*
2783  * The syncer vnode is no referenced.
2784  */
2785 static int
2786 sync_inactive(ap)
2787 	struct vop_inactive_args /* {
2788 		struct vnode *a_vp;
2789 		struct proc *a_p;
2790 	} */ *ap;
2791 {
2792 
2793 	vgone(ap->a_vp);
2794 	return (0);
2795 }
2796 
2797 /*
2798  * The syncer vnode is no longer needed and is being decommissioned.
2799  *
2800  * Modifications to the worklist must be protected at splbio().
2801  */
2802 static int
2803 sync_reclaim(ap)
2804 	struct vop_reclaim_args /* {
2805 		struct vnode *a_vp;
2806 	} */ *ap;
2807 {
2808 	struct vnode *vp = ap->a_vp;
2809 	int s;
2810 
2811 	s = splbio();
2812 	vp->v_mount->mnt_syncer = NULL;
2813 	if (vp->v_flag & VONWORKLST) {
2814 		LIST_REMOVE(vp, v_synclist);
2815 		vp->v_flag &= ~VONWORKLST;
2816 	}
2817 	splx(s);
2818 
2819 	return (0);
2820 }
2821 
2822 /*
2823  * Print out a syncer vnode.
2824  */
2825 static int
2826 sync_print(ap)
2827 	struct vop_print_args /* {
2828 		struct vnode *a_vp;
2829 	} */ *ap;
2830 {
2831 	struct vnode *vp = ap->a_vp;
2832 
2833 	printf("syncer vnode");
2834 	if (vp->v_vnlock != NULL)
2835 		lockmgr_printinfo(vp->v_vnlock);
2836 	printf("\n");
2837 	return (0);
2838 }
2839 
2840 /*
2841  * extract the dev_t from a VBLK or VCHR
2842  */
2843 dev_t
2844 vn_todev(vp)
2845 	struct vnode *vp;
2846 {
2847 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2848 		return (NODEV);
2849 	return (vp->v_rdev);
2850 }
2851 
2852 /*
2853  * Check if vnode represents a disk device
2854  */
2855 int
2856 vn_isdisk(vp)
2857 	struct vnode *vp;
2858 {
2859 	if (vp->v_type != VBLK)
2860 		return (0);
2861 	if (!devsw(vp->v_rdev))
2862 		return (0);
2863 	if (!(devsw(vp->v_rdev)->d_flags & D_DISK))
2864 		return (0);
2865 	return (1);
2866 }
2867 
2868