xref: /freebsd/sys/kern/vfs_subr.c (revision 1a2cdef4962b47be5057809ce730a733b7f3c27c)
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 #include "opt_ffs.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/dirent.h>
54 #include <sys/domain.h>
55 #include <sys/eventhandler.h>
56 #include <sys/event.h>
57 #include <sys/fcntl.h>
58 #include <sys/kernel.h>
59 #include <sys/kthread.h>
60 #include <sys/ktr.h>
61 #include <sys/malloc.h>
62 #include <sys/mount.h>
63 #include <sys/mutex.h>
64 #include <sys/namei.h>
65 #include <sys/proc.h>
66 #include <sys/reboot.h>
67 #include <sys/socket.h>
68 #include <sys/stat.h>
69 #include <sys/sysctl.h>
70 #include <sys/vmmeter.h>
71 #include <sys/vnode.h>
72 
73 #include <machine/limits.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_extern.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_pager.h>
82 #include <vm/vnode_pager.h>
83 #include <vm/vm_zone.h>
84 
85 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
86 
87 static void	addalias __P((struct vnode *vp, dev_t nvp_rdev));
88 static void	insmntque __P((struct vnode *vp, struct mount *mp));
89 static void	vclean __P((struct vnode *vp, int flags, struct proc *p));
90 
91 /*
92  * Number of vnodes in existence.  Increased whenever getnewvnode()
93  * allocates a new vnode, never decreased.
94  */
95 static unsigned long	numvnodes;
96 SYSCTL_LONG(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
97 
98 /*
99  * Conversion tables for conversion from vnode types to inode formats
100  * and back.
101  */
102 enum vtype iftovt_tab[16] = {
103 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
104 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
105 };
106 int vttoif_tab[9] = {
107 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
108 	S_IFSOCK, S_IFIFO, S_IFMT,
109 };
110 
111 /*
112  * List of vnodes that are ready for recycling.
113  */
114 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
115 
116 /*
117  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
118  * getnewvnode() will return a newly allocated vnode.
119  */
120 static u_long wantfreevnodes = 25;
121 SYSCTL_LONG(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
122 /* Number of vnodes in the free list. */
123 static u_long freevnodes = 0;
124 SYSCTL_LONG(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
125 
126 /*
127  * Various variables used for debugging the new implementation of
128  * reassignbuf().
129  * XXX these are probably of (very) limited utility now.
130  */
131 static int reassignbufcalls;
132 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
133 static int reassignbufloops;
134 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
135 static int reassignbufsortgood;
136 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
137 static int reassignbufsortbad;
138 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
139 /* Set to 0 for old insertion-sort based reassignbuf, 1 for modern method. */
140 static int reassignbufmethod = 1;
141 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
142 
143 #ifdef ENABLE_VFS_IOOPT
144 /* See NOTES for a description of this setting. */
145 int vfs_ioopt = 0;
146 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
147 #endif
148 
149 /* List of mounted filesystems. */
150 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
151 
152 /* For any iteration/modification of mountlist */
153 struct mtx mountlist_mtx;
154 
155 /* For any iteration/modification of mnt_vnodelist */
156 struct mtx mntvnode_mtx;
157 
158 /*
159  * Cache for the mount type id assigned to NFS.  This is used for
160  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
161  */
162 int	nfs_mount_type = -1;
163 
164 /* To keep more than one thread at a time from running vfs_getnewfsid */
165 static struct mtx mntid_mtx;
166 
167 /* For any iteration/modification of vnode_free_list */
168 static struct mtx vnode_free_list_mtx;
169 
170 /*
171  * For any iteration/modification of dev->si_hlist (linked through
172  * v_specnext)
173  */
174 static struct mtx spechash_mtx;
175 
176 /* Publicly exported FS */
177 struct nfs_public nfs_pub;
178 
179 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
180 static vm_zone_t vnode_zone;
181 
182 /* Set to 1 to print out reclaim of active vnodes */
183 int	prtactive = 0;
184 
185 /*
186  * The workitem queue.
187  *
188  * It is useful to delay writes of file data and filesystem metadata
189  * for tens of seconds so that quickly created and deleted files need
190  * not waste disk bandwidth being created and removed. To realize this,
191  * we append vnodes to a "workitem" queue. When running with a soft
192  * updates implementation, most pending metadata dependencies should
193  * not wait for more than a few seconds. Thus, mounted on block devices
194  * are delayed only about a half the time that file data is delayed.
195  * Similarly, directory updates are more critical, so are only delayed
196  * about a third the time that file data is delayed. Thus, there are
197  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
198  * one each second (driven off the filesystem syncer process). The
199  * syncer_delayno variable indicates the next queue that is to be processed.
200  * Items that need to be processed soon are placed in this queue:
201  *
202  *	syncer_workitem_pending[syncer_delayno]
203  *
204  * A delay of fifteen seconds is done by placing the request fifteen
205  * entries later in the queue:
206  *
207  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
208  *
209  */
210 static int syncer_delayno = 0;
211 static long syncer_mask;
212 LIST_HEAD(synclist, vnode);
213 static struct synclist *syncer_workitem_pending;
214 
215 #define SYNCER_MAXDELAY		32
216 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
217 time_t syncdelay = 30;		/* max time to delay syncing data */
218 time_t filedelay = 30;		/* time to delay syncing files */
219 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
220 time_t dirdelay = 29;		/* time to delay syncing directories */
221 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
222 time_t metadelay = 28;		/* time to delay syncing metadata */
223 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
224 static int rushjob;		/* number of slots to run ASAP */
225 static int stat_rush_requests;	/* number of times I/O speeded up */
226 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
227 
228 /*
229  * Number of vnodes we want to exist at any one time.  This is mostly used
230  * to size hash tables in vnode-related code.  It is normally not used in
231  * getnewvnode(), as wantfreevnodes is normally nonzero.)
232  *
233  * XXX desiredvnodes is historical cruft and should not exist.
234  */
235 int desiredvnodes;
236 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
237     &desiredvnodes, 0, "Maximum number of vnodes");
238 
239 static void	vfs_free_addrlist __P((struct netexport *nep));
240 static int	vfs_free_netcred __P((struct radix_node *rn, void *w));
241 static int	vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
242 				       struct export_args *argp));
243 
244 /*
245  * Initialize the vnode management data structures.
246  */
247 static void
248 vntblinit(void *dummy __unused)
249 {
250 
251 	desiredvnodes = maxproc + cnt.v_page_count / 4;
252 	mtx_init(&mountlist_mtx, "mountlist", MTX_DEF);
253 	mtx_init(&mntvnode_mtx, "mntvnode", MTX_DEF);
254 	mtx_init(&mntid_mtx, "mntid", MTX_DEF);
255 	mtx_init(&spechash_mtx, "spechash", MTX_DEF);
256 	TAILQ_INIT(&vnode_free_list);
257 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", MTX_DEF);
258 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
259 	/*
260 	 * Initialize the filesystem syncer.
261 	 */
262 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
263 		&syncer_mask);
264 	syncer_maxdelay = syncer_mask + 1;
265 }
266 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
267 
268 
269 /*
270  * Mark a mount point as busy. Used to synchronize access and to delay
271  * unmounting. Interlock is not released on failure.
272  */
273 int
274 vfs_busy(mp, flags, interlkp, p)
275 	struct mount *mp;
276 	int flags;
277 	struct mtx *interlkp;
278 	struct proc *p;
279 {
280 	int lkflags;
281 
282 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
283 		if (flags & LK_NOWAIT)
284 			return (ENOENT);
285 		mp->mnt_kern_flag |= MNTK_MWAIT;
286 		/*
287 		 * Since all busy locks are shared except the exclusive
288 		 * lock granted when unmounting, the only place that a
289 		 * wakeup needs to be done is at the release of the
290 		 * exclusive lock at the end of dounmount.
291 		 */
292 		msleep((caddr_t)mp, interlkp, PVFS, "vfs_busy", 0);
293 		return (ENOENT);
294 	}
295 	lkflags = LK_SHARED | LK_NOPAUSE;
296 	if (interlkp)
297 		lkflags |= LK_INTERLOCK;
298 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
299 		panic("vfs_busy: unexpected lock failure");
300 	return (0);
301 }
302 
303 /*
304  * Free a busy filesystem.
305  */
306 void
307 vfs_unbusy(mp, p)
308 	struct mount *mp;
309 	struct proc *p;
310 {
311 
312 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
313 }
314 
315 /*
316  * Lookup a filesystem type, and if found allocate and initialize
317  * a mount structure for it.
318  *
319  * Devname is usually updated by mount(8) after booting.
320  */
321 int
322 vfs_rootmountalloc(fstypename, devname, mpp)
323 	char *fstypename;
324 	char *devname;
325 	struct mount **mpp;
326 {
327 	struct proc *p = curproc;	/* XXX */
328 	struct vfsconf *vfsp;
329 	struct mount *mp;
330 
331 	if (fstypename == NULL)
332 		return (ENODEV);
333 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
334 		if (!strcmp(vfsp->vfc_name, fstypename))
335 			break;
336 	if (vfsp == NULL)
337 		return (ENODEV);
338 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
339 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
340 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
341 	LIST_INIT(&mp->mnt_vnodelist);
342 	mp->mnt_vfc = vfsp;
343 	mp->mnt_op = vfsp->vfc_vfsops;
344 	mp->mnt_flag = MNT_RDONLY;
345 	mp->mnt_vnodecovered = NULLVP;
346 	vfsp->vfc_refcount++;
347 	mp->mnt_iosize_max = DFLTPHYS;
348 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
349 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
350 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
351 	mp->mnt_stat.f_mntonname[0] = '/';
352 	mp->mnt_stat.f_mntonname[1] = 0;
353 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
354 	*mpp = mp;
355 	return (0);
356 }
357 
358 /*
359  * Find an appropriate filesystem to use for the root. If a filesystem
360  * has not been preselected, walk through the list of known filesystems
361  * trying those that have mountroot routines, and try them until one
362  * works or we have tried them all.
363  */
364 #ifdef notdef	/* XXX JH */
365 int
366 lite2_vfs_mountroot()
367 {
368 	struct vfsconf *vfsp;
369 	extern int (*lite2_mountroot) __P((void));
370 	int error;
371 
372 	if (lite2_mountroot != NULL)
373 		return ((*lite2_mountroot)());
374 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
375 		if (vfsp->vfc_mountroot == NULL)
376 			continue;
377 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
378 			return (0);
379 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
380 	}
381 	return (ENODEV);
382 }
383 #endif
384 
385 /*
386  * Lookup a mount point by filesystem identifier.
387  */
388 struct mount *
389 vfs_getvfs(fsid)
390 	fsid_t *fsid;
391 {
392 	register struct mount *mp;
393 
394 	mtx_lock(&mountlist_mtx);
395 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
396 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
397 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
398 			mtx_unlock(&mountlist_mtx);
399 			return (mp);
400 	    }
401 	}
402 	mtx_unlock(&mountlist_mtx);
403 	return ((struct mount *) 0);
404 }
405 
406 /*
407  * Get a new unique fsid.  Try to make its val[0] unique, since this value
408  * will be used to create fake device numbers for stat().  Also try (but
409  * not so hard) make its val[0] unique mod 2^16, since some emulators only
410  * support 16-bit device numbers.  We end up with unique val[0]'s for the
411  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
412  *
413  * Keep in mind that several mounts may be running in parallel.  Starting
414  * the search one past where the previous search terminated is both a
415  * micro-optimization and a defense against returning the same fsid to
416  * different mounts.
417  */
418 void
419 vfs_getnewfsid(mp)
420 	struct mount *mp;
421 {
422 	static u_int16_t mntid_base;
423 	fsid_t tfsid;
424 	int mtype;
425 
426 	mtx_lock(&mntid_mtx);
427 	mtype = mp->mnt_vfc->vfc_typenum;
428 	tfsid.val[1] = mtype;
429 	mtype = (mtype & 0xFF) << 24;
430 	for (;;) {
431 		tfsid.val[0] = makeudev(255,
432 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
433 		mntid_base++;
434 		if (vfs_getvfs(&tfsid) == NULL)
435 			break;
436 	}
437 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
438 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
439 	mtx_unlock(&mntid_mtx);
440 }
441 
442 /*
443  * Knob to control the precision of file timestamps:
444  *
445  *   0 = seconds only; nanoseconds zeroed.
446  *   1 = seconds and nanoseconds, accurate within 1/HZ.
447  *   2 = seconds and nanoseconds, truncated to microseconds.
448  * >=3 = seconds and nanoseconds, maximum precision.
449  */
450 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
451 
452 static int timestamp_precision = TSP_SEC;
453 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
454     &timestamp_precision, 0, "");
455 
456 /*
457  * Get a current timestamp.
458  */
459 void
460 vfs_timestamp(tsp)
461 	struct timespec *tsp;
462 {
463 	struct timeval tv;
464 
465 	switch (timestamp_precision) {
466 	case TSP_SEC:
467 		tsp->tv_sec = time_second;
468 		tsp->tv_nsec = 0;
469 		break;
470 	case TSP_HZ:
471 		getnanotime(tsp);
472 		break;
473 	case TSP_USEC:
474 		microtime(&tv);
475 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
476 		break;
477 	case TSP_NSEC:
478 	default:
479 		nanotime(tsp);
480 		break;
481 	}
482 }
483 
484 /*
485  * Set vnode attributes to VNOVAL
486  */
487 void
488 vattr_null(vap)
489 	register struct vattr *vap;
490 {
491 
492 	vap->va_type = VNON;
493 	vap->va_size = VNOVAL;
494 	vap->va_bytes = VNOVAL;
495 	vap->va_mode = VNOVAL;
496 	vap->va_nlink = VNOVAL;
497 	vap->va_uid = VNOVAL;
498 	vap->va_gid = VNOVAL;
499 	vap->va_fsid = VNOVAL;
500 	vap->va_fileid = VNOVAL;
501 	vap->va_blocksize = VNOVAL;
502 	vap->va_rdev = VNOVAL;
503 	vap->va_atime.tv_sec = VNOVAL;
504 	vap->va_atime.tv_nsec = VNOVAL;
505 	vap->va_mtime.tv_sec = VNOVAL;
506 	vap->va_mtime.tv_nsec = VNOVAL;
507 	vap->va_ctime.tv_sec = VNOVAL;
508 	vap->va_ctime.tv_nsec = VNOVAL;
509 	vap->va_flags = VNOVAL;
510 	vap->va_gen = VNOVAL;
511 	vap->va_vaflags = 0;
512 }
513 
514 /*
515  * Routines having to do with the management of the vnode table.
516  */
517 
518 /*
519  * Return the next vnode from the free list.
520  */
521 int
522 getnewvnode(tag, mp, vops, vpp)
523 	enum vtagtype tag;
524 	struct mount *mp;
525 	vop_t **vops;
526 	struct vnode **vpp;
527 {
528 	int s, count;
529 	struct proc *p = curproc;	/* XXX */
530 	struct vnode *vp = NULL;
531 	struct mount *vnmp;
532 	vm_object_t object;
533 
534 	/*
535 	 * We take the least recently used vnode from the freelist
536 	 * if we can get it and it has no cached pages, and no
537 	 * namecache entries are relative to it.
538 	 * Otherwise we allocate a new vnode
539 	 */
540 
541 	s = splbio();
542 	mtx_lock(&vnode_free_list_mtx);
543 
544 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
545 		vp = NULL;
546 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
547 		/*
548 		 * XXX: this is only here to be backwards compatible
549 		 */
550 		vp = NULL;
551 	} else for (count = 0; count < freevnodes; count++) {
552 		vp = TAILQ_FIRST(&vnode_free_list);
553 		if (vp == NULL || vp->v_usecount)
554 			panic("getnewvnode: free vnode isn't");
555 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
556 		/*
557 		 * Don't recycle if active in the namecache or
558 		 * if it still has cached pages or we cannot get
559 		 * its interlock.
560 		 */
561 		if (LIST_FIRST(&vp->v_cache_src) != NULL ||
562 		    (VOP_GETVOBJECT(vp, &object) == 0 &&
563 		     (object->resident_page_count || object->ref_count)) ||
564 		    !mtx_trylock(&vp->v_interlock)) {
565 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
566 			vp = NULL;
567 			continue;
568 		}
569 		/*
570 		 * Skip over it if its filesystem is being suspended.
571 		 */
572 		if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0)
573 			break;
574 		mtx_unlock(&vp->v_interlock);
575 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
576 		vp = NULL;
577 	}
578 	if (vp) {
579 		vp->v_flag |= VDOOMED;
580 		vp->v_flag &= ~VFREE;
581 		freevnodes--;
582 		mtx_unlock(&vnode_free_list_mtx);
583 		cache_purge(vp);
584 		vp->v_lease = NULL;
585 		if (vp->v_type != VBAD) {
586 			vgonel(vp, p);
587 		} else {
588 			mtx_unlock(&vp->v_interlock);
589 		}
590 		vn_finished_write(vnmp);
591 
592 #ifdef INVARIANTS
593 		{
594 			int s;
595 
596 			if (vp->v_data)
597 				panic("cleaned vnode isn't");
598 			s = splbio();
599 			if (vp->v_numoutput)
600 				panic("Clean vnode has pending I/O's");
601 			splx(s);
602 			if (vp->v_writecount != 0)
603 				panic("Non-zero write count");
604 		}
605 #endif
606 		vp->v_flag = 0;
607 		vp->v_lastw = 0;
608 		vp->v_lasta = 0;
609 		vp->v_cstart = 0;
610 		vp->v_clen = 0;
611 		vp->v_socket = 0;
612 	} else {
613 		mtx_unlock(&vnode_free_list_mtx);
614 		vp = (struct vnode *) zalloc(vnode_zone);
615 		bzero((char *) vp, sizeof *vp);
616 		mtx_init(&vp->v_interlock, "vnode interlock", MTX_DEF);
617 		vp->v_dd = vp;
618 		mtx_init(&vp->v_pollinfo.vpi_lock, "vnode pollinfo", MTX_DEF);
619 		cache_purge(vp);
620 		LIST_INIT(&vp->v_cache_src);
621 		TAILQ_INIT(&vp->v_cache_dst);
622 		numvnodes++;
623 	}
624 
625 	TAILQ_INIT(&vp->v_cleanblkhd);
626 	TAILQ_INIT(&vp->v_dirtyblkhd);
627 	vp->v_type = VNON;
628 	vp->v_tag = tag;
629 	vp->v_op = vops;
630 	lockinit(&vp->v_lock, PVFS, "vnlock", 0, LK_NOPAUSE);
631 	insmntque(vp, mp);
632 	*vpp = vp;
633 	vp->v_usecount = 1;
634 	vp->v_data = 0;
635 	splx(s);
636 
637 	vfs_object_create(vp, p, p->p_ucred);
638 	return (0);
639 }
640 
641 /*
642  * Move a vnode from one mount queue to another.
643  */
644 static void
645 insmntque(vp, mp)
646 	register struct vnode *vp;
647 	register struct mount *mp;
648 {
649 
650 	mtx_lock(&mntvnode_mtx);
651 	/*
652 	 * Delete from old mount point vnode list, if on one.
653 	 */
654 	if (vp->v_mount != NULL)
655 		LIST_REMOVE(vp, v_mntvnodes);
656 	/*
657 	 * Insert into list of vnodes for the new mount point, if available.
658 	 */
659 	if ((vp->v_mount = mp) == NULL) {
660 		mtx_unlock(&mntvnode_mtx);
661 		return;
662 	}
663 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
664 	mtx_unlock(&mntvnode_mtx);
665 }
666 
667 /*
668  * Update outstanding I/O count and do wakeup if requested.
669  */
670 void
671 vwakeup(bp)
672 	register struct buf *bp;
673 {
674 	register struct vnode *vp;
675 
676 	bp->b_flags &= ~B_WRITEINPROG;
677 	if ((vp = bp->b_vp)) {
678 		vp->v_numoutput--;
679 		if (vp->v_numoutput < 0)
680 			panic("vwakeup: neg numoutput");
681 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
682 			vp->v_flag &= ~VBWAIT;
683 			wakeup((caddr_t) &vp->v_numoutput);
684 		}
685 	}
686 }
687 
688 /*
689  * Flush out and invalidate all buffers associated with a vnode.
690  * Called with the underlying object locked.
691  */
692 int
693 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
694 	register struct vnode *vp;
695 	int flags;
696 	struct ucred *cred;
697 	struct proc *p;
698 	int slpflag, slptimeo;
699 {
700 	register struct buf *bp;
701 	struct buf *nbp, *blist;
702 	int s, error;
703 	vm_object_t object;
704 
705 	if (flags & V_SAVE) {
706 		s = splbio();
707 		while (vp->v_numoutput) {
708 			vp->v_flag |= VBWAIT;
709 			error = tsleep((caddr_t)&vp->v_numoutput,
710 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
711 			if (error) {
712 				splx(s);
713 				return (error);
714 			}
715 		}
716 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
717 			splx(s);
718 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
719 				return (error);
720 			s = splbio();
721 			if (vp->v_numoutput > 0 ||
722 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
723 				panic("vinvalbuf: dirty bufs");
724 		}
725 		splx(s);
726   	}
727 	s = splbio();
728 	for (;;) {
729 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
730 		if (!blist)
731 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
732 		if (!blist)
733 			break;
734 
735 		for (bp = blist; bp; bp = nbp) {
736 			nbp = TAILQ_NEXT(bp, b_vnbufs);
737 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
738 				error = BUF_TIMELOCK(bp,
739 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
740 				    "vinvalbuf", slpflag, slptimeo);
741 				if (error == ENOLCK)
742 					break;
743 				splx(s);
744 				return (error);
745 			}
746 			/*
747 			 * XXX Since there are no node locks for NFS, I
748 			 * believe there is a slight chance that a delayed
749 			 * write will occur while sleeping just above, so
750 			 * check for it.  Note that vfs_bio_awrite expects
751 			 * buffers to reside on a queue, while BUF_WRITE and
752 			 * brelse do not.
753 			 */
754 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
755 				(flags & V_SAVE)) {
756 
757 				if (bp->b_vp == vp) {
758 					if (bp->b_flags & B_CLUSTEROK) {
759 						BUF_UNLOCK(bp);
760 						vfs_bio_awrite(bp);
761 					} else {
762 						bremfree(bp);
763 						bp->b_flags |= B_ASYNC;
764 						BUF_WRITE(bp);
765 					}
766 				} else {
767 					bremfree(bp);
768 					(void) BUF_WRITE(bp);
769 				}
770 				break;
771 			}
772 			bremfree(bp);
773 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
774 			bp->b_flags &= ~B_ASYNC;
775 			brelse(bp);
776 		}
777 	}
778 
779 	while (vp->v_numoutput > 0) {
780 		vp->v_flag |= VBWAIT;
781 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
782 	}
783 
784 	splx(s);
785 
786 	/*
787 	 * Destroy the copy in the VM cache, too.
788 	 */
789 	mtx_lock(&vp->v_interlock);
790 	if (VOP_GETVOBJECT(vp, &object) == 0) {
791 		vm_object_page_remove(object, 0, 0,
792 			(flags & V_SAVE) ? TRUE : FALSE);
793 	}
794 	mtx_unlock(&vp->v_interlock);
795 
796 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
797 		panic("vinvalbuf: flush failed");
798 	return (0);
799 }
800 
801 /*
802  * Truncate a file's buffer and pages to a specified length.  This
803  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
804  * sync activity.
805  */
806 int
807 vtruncbuf(vp, cred, p, length, blksize)
808 	register struct vnode *vp;
809 	struct ucred *cred;
810 	struct proc *p;
811 	off_t length;
812 	int blksize;
813 {
814 	register struct buf *bp;
815 	struct buf *nbp;
816 	int s, anyfreed;
817 	int trunclbn;
818 
819 	/*
820 	 * Round up to the *next* lbn.
821 	 */
822 	trunclbn = (length + blksize - 1) / blksize;
823 
824 	s = splbio();
825 restart:
826 	anyfreed = 1;
827 	for (;anyfreed;) {
828 		anyfreed = 0;
829 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
830 			nbp = TAILQ_NEXT(bp, b_vnbufs);
831 			if (bp->b_lblkno >= trunclbn) {
832 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
833 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
834 					goto restart;
835 				} else {
836 					bremfree(bp);
837 					bp->b_flags |= (B_INVAL | B_RELBUF);
838 					bp->b_flags &= ~B_ASYNC;
839 					brelse(bp);
840 					anyfreed = 1;
841 				}
842 				if (nbp &&
843 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
844 				    (nbp->b_vp != vp) ||
845 				    (nbp->b_flags & B_DELWRI))) {
846 					goto restart;
847 				}
848 			}
849 		}
850 
851 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
852 			nbp = TAILQ_NEXT(bp, b_vnbufs);
853 			if (bp->b_lblkno >= trunclbn) {
854 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
855 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
856 					goto restart;
857 				} else {
858 					bremfree(bp);
859 					bp->b_flags |= (B_INVAL | B_RELBUF);
860 					bp->b_flags &= ~B_ASYNC;
861 					brelse(bp);
862 					anyfreed = 1;
863 				}
864 				if (nbp &&
865 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
866 				    (nbp->b_vp != vp) ||
867 				    (nbp->b_flags & B_DELWRI) == 0)) {
868 					goto restart;
869 				}
870 			}
871 		}
872 	}
873 
874 	if (length > 0) {
875 restartsync:
876 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
877 			nbp = TAILQ_NEXT(bp, b_vnbufs);
878 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
879 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
880 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
881 					goto restart;
882 				} else {
883 					bremfree(bp);
884 					if (bp->b_vp == vp) {
885 						bp->b_flags |= B_ASYNC;
886 					} else {
887 						bp->b_flags &= ~B_ASYNC;
888 					}
889 					BUF_WRITE(bp);
890 				}
891 				goto restartsync;
892 			}
893 
894 		}
895 	}
896 
897 	while (vp->v_numoutput > 0) {
898 		vp->v_flag |= VBWAIT;
899 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
900 	}
901 
902 	splx(s);
903 
904 	vnode_pager_setsize(vp, length);
905 
906 	return (0);
907 }
908 
909 /*
910  * Associate a buffer with a vnode.
911  */
912 void
913 bgetvp(vp, bp)
914 	register struct vnode *vp;
915 	register struct buf *bp;
916 {
917 	int s;
918 
919 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
920 
921 	vhold(vp);
922 	bp->b_vp = vp;
923 	bp->b_dev = vn_todev(vp);
924 	/*
925 	 * Insert onto list for new vnode.
926 	 */
927 	s = splbio();
928 	bp->b_xflags |= BX_VNCLEAN;
929 	bp->b_xflags &= ~BX_VNDIRTY;
930 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
931 	splx(s);
932 }
933 
934 /*
935  * Disassociate a buffer from a vnode.
936  */
937 void
938 brelvp(bp)
939 	register struct buf *bp;
940 {
941 	struct vnode *vp;
942 	struct buflists *listheadp;
943 	int s;
944 
945 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
946 
947 	/*
948 	 * Delete from old vnode list, if on one.
949 	 */
950 	vp = bp->b_vp;
951 	s = splbio();
952 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
953 		if (bp->b_xflags & BX_VNDIRTY)
954 			listheadp = &vp->v_dirtyblkhd;
955 		else
956 			listheadp = &vp->v_cleanblkhd;
957 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
958 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
959 	}
960 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
961 		vp->v_flag &= ~VONWORKLST;
962 		LIST_REMOVE(vp, v_synclist);
963 	}
964 	splx(s);
965 	bp->b_vp = (struct vnode *) 0;
966 	vdrop(vp);
967 }
968 
969 /*
970  * Add an item to the syncer work queue.
971  */
972 static void
973 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
974 {
975 	int s, slot;
976 
977 	s = splbio();
978 
979 	if (vp->v_flag & VONWORKLST) {
980 		LIST_REMOVE(vp, v_synclist);
981 	}
982 
983 	if (delay > syncer_maxdelay - 2)
984 		delay = syncer_maxdelay - 2;
985 	slot = (syncer_delayno + delay) & syncer_mask;
986 
987 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
988 	vp->v_flag |= VONWORKLST;
989 	splx(s);
990 }
991 
992 struct  proc *updateproc;
993 static void sched_sync __P((void));
994 static struct kproc_desc up_kp = {
995 	"syncer",
996 	sched_sync,
997 	&updateproc
998 };
999 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1000 
1001 /*
1002  * System filesystem synchronizer daemon.
1003  */
1004 void
1005 sched_sync(void)
1006 {
1007 	struct synclist *slp;
1008 	struct vnode *vp;
1009 	struct mount *mp;
1010 	long starttime;
1011 	int s;
1012 	struct proc *p = updateproc;
1013 
1014 	mtx_lock(&Giant);
1015 
1016 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
1017 	    SHUTDOWN_PRI_LAST);
1018 
1019 	for (;;) {
1020 		kthread_suspend_check(p);
1021 
1022 		starttime = time_second;
1023 
1024 		/*
1025 		 * Push files whose dirty time has expired.  Be careful
1026 		 * of interrupt race on slp queue.
1027 		 */
1028 		s = splbio();
1029 		slp = &syncer_workitem_pending[syncer_delayno];
1030 		syncer_delayno += 1;
1031 		if (syncer_delayno == syncer_maxdelay)
1032 			syncer_delayno = 0;
1033 		splx(s);
1034 
1035 		while ((vp = LIST_FIRST(slp)) != NULL) {
1036 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1037 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1038 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1039 				(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
1040 				VOP_UNLOCK(vp, 0, p);
1041 				vn_finished_write(mp);
1042 			}
1043 			s = splbio();
1044 			if (LIST_FIRST(slp) == vp) {
1045 				/*
1046 				 * Note: v_tag VT_VFS vps can remain on the
1047 				 * worklist too with no dirty blocks, but
1048 				 * since sync_fsync() moves it to a different
1049 				 * slot we are safe.
1050 				 */
1051 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1052 				    !vn_isdisk(vp, NULL))
1053 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1054 				/*
1055 				 * Put us back on the worklist.  The worklist
1056 				 * routine will remove us from our current
1057 				 * position and then add us back in at a later
1058 				 * position.
1059 				 */
1060 				vn_syncer_add_to_worklist(vp, syncdelay);
1061 			}
1062 			splx(s);
1063 		}
1064 
1065 		/*
1066 		 * Do soft update processing.
1067 		 */
1068 #ifdef SOFTUPDATES
1069 		softdep_process_worklist(NULL);
1070 #endif
1071 
1072 		/*
1073 		 * The variable rushjob allows the kernel to speed up the
1074 		 * processing of the filesystem syncer process. A rushjob
1075 		 * value of N tells the filesystem syncer to process the next
1076 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1077 		 * is used by the soft update code to speed up the filesystem
1078 		 * syncer process when the incore state is getting so far
1079 		 * ahead of the disk that the kernel memory pool is being
1080 		 * threatened with exhaustion.
1081 		 */
1082 		if (rushjob > 0) {
1083 			rushjob -= 1;
1084 			continue;
1085 		}
1086 		/*
1087 		 * If it has taken us less than a second to process the
1088 		 * current work, then wait. Otherwise start right over
1089 		 * again. We can still lose time if any single round
1090 		 * takes more than two seconds, but it does not really
1091 		 * matter as we are just trying to generally pace the
1092 		 * filesystem activity.
1093 		 */
1094 		if (time_second == starttime)
1095 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1096 	}
1097 }
1098 
1099 /*
1100  * Request the syncer daemon to speed up its work.
1101  * We never push it to speed up more than half of its
1102  * normal turn time, otherwise it could take over the cpu.
1103  */
1104 int
1105 speedup_syncer()
1106 {
1107 
1108 	mtx_lock_spin(&sched_lock);
1109 	if (updateproc->p_wchan == &lbolt)
1110 		setrunnable(updateproc);
1111 	mtx_unlock_spin(&sched_lock);
1112 	if (rushjob < syncdelay / 2) {
1113 		rushjob += 1;
1114 		stat_rush_requests += 1;
1115 		return (1);
1116 	}
1117 	return(0);
1118 }
1119 
1120 /*
1121  * Associate a p-buffer with a vnode.
1122  *
1123  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1124  * with the buffer.  i.e. the bp has not been linked into the vnode or
1125  * ref-counted.
1126  */
1127 void
1128 pbgetvp(vp, bp)
1129 	register struct vnode *vp;
1130 	register struct buf *bp;
1131 {
1132 
1133 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1134 
1135 	bp->b_vp = vp;
1136 	bp->b_flags |= B_PAGING;
1137 	bp->b_dev = vn_todev(vp);
1138 }
1139 
1140 /*
1141  * Disassociate a p-buffer from a vnode.
1142  */
1143 void
1144 pbrelvp(bp)
1145 	register struct buf *bp;
1146 {
1147 
1148 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1149 
1150 	/* XXX REMOVE ME */
1151 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1152 		panic(
1153 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1154 		    bp,
1155 		    (int)bp->b_flags
1156 		);
1157 	}
1158 	bp->b_vp = (struct vnode *) 0;
1159 	bp->b_flags &= ~B_PAGING;
1160 }
1161 
1162 /*
1163  * Change the vnode a pager buffer is associated with.
1164  */
1165 void
1166 pbreassignbuf(bp, newvp)
1167 	struct buf *bp;
1168 	struct vnode *newvp;
1169 {
1170 
1171 	KASSERT(bp->b_flags & B_PAGING,
1172 	    ("pbreassignbuf() on non phys bp %p", bp));
1173 	bp->b_vp = newvp;
1174 }
1175 
1176 /*
1177  * Reassign a buffer from one vnode to another.
1178  * Used to assign file specific control information
1179  * (indirect blocks) to the vnode to which they belong.
1180  */
1181 void
1182 reassignbuf(bp, newvp)
1183 	register struct buf *bp;
1184 	register struct vnode *newvp;
1185 {
1186 	struct buflists *listheadp;
1187 	int delay;
1188 	int s;
1189 
1190 	if (newvp == NULL) {
1191 		printf("reassignbuf: NULL");
1192 		return;
1193 	}
1194 	++reassignbufcalls;
1195 
1196 	/*
1197 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1198 	 * is not fully linked in.
1199 	 */
1200 	if (bp->b_flags & B_PAGING)
1201 		panic("cannot reassign paging buffer");
1202 
1203 	s = splbio();
1204 	/*
1205 	 * Delete from old vnode list, if on one.
1206 	 */
1207 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1208 		if (bp->b_xflags & BX_VNDIRTY)
1209 			listheadp = &bp->b_vp->v_dirtyblkhd;
1210 		else
1211 			listheadp = &bp->b_vp->v_cleanblkhd;
1212 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1213 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1214 		if (bp->b_vp != newvp) {
1215 			vdrop(bp->b_vp);
1216 			bp->b_vp = NULL;	/* for clarification */
1217 		}
1218 	}
1219 	/*
1220 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1221 	 * of clean buffers.
1222 	 */
1223 	if (bp->b_flags & B_DELWRI) {
1224 		struct buf *tbp;
1225 
1226 		listheadp = &newvp->v_dirtyblkhd;
1227 		if ((newvp->v_flag & VONWORKLST) == 0) {
1228 			switch (newvp->v_type) {
1229 			case VDIR:
1230 				delay = dirdelay;
1231 				break;
1232 			case VCHR:
1233 				if (newvp->v_rdev->si_mountpoint != NULL) {
1234 					delay = metadelay;
1235 					break;
1236 				}
1237 				/* fall through */
1238 			default:
1239 				delay = filedelay;
1240 			}
1241 			vn_syncer_add_to_worklist(newvp, delay);
1242 		}
1243 		bp->b_xflags |= BX_VNDIRTY;
1244 		tbp = TAILQ_FIRST(listheadp);
1245 		if (tbp == NULL ||
1246 		    bp->b_lblkno == 0 ||
1247 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1248 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1249 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1250 			++reassignbufsortgood;
1251 		} else if (bp->b_lblkno < 0) {
1252 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1253 			++reassignbufsortgood;
1254 		} else if (reassignbufmethod == 1) {
1255 			/*
1256 			 * New sorting algorithm, only handle sequential case,
1257 			 * otherwise append to end (but before metadata)
1258 			 */
1259 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1260 			    (tbp->b_xflags & BX_VNDIRTY)) {
1261 				/*
1262 				 * Found the best place to insert the buffer
1263 				 */
1264 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1265 				++reassignbufsortgood;
1266 			} else {
1267 				/*
1268 				 * Missed, append to end, but before meta-data.
1269 				 * We know that the head buffer in the list is
1270 				 * not meta-data due to prior conditionals.
1271 				 *
1272 				 * Indirect effects:  NFS second stage write
1273 				 * tends to wind up here, giving maximum
1274 				 * distance between the unstable write and the
1275 				 * commit rpc.
1276 				 */
1277 				tbp = TAILQ_LAST(listheadp, buflists);
1278 				while (tbp && tbp->b_lblkno < 0)
1279 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1280 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1281 				++reassignbufsortbad;
1282 			}
1283 		} else {
1284 			/*
1285 			 * Old sorting algorithm, scan queue and insert
1286 			 */
1287 			struct buf *ttbp;
1288 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1289 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1290 				++reassignbufloops;
1291 				tbp = ttbp;
1292 			}
1293 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1294 		}
1295 	} else {
1296 		bp->b_xflags |= BX_VNCLEAN;
1297 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1298 		if ((newvp->v_flag & VONWORKLST) &&
1299 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1300 			newvp->v_flag &= ~VONWORKLST;
1301 			LIST_REMOVE(newvp, v_synclist);
1302 		}
1303 	}
1304 	if (bp->b_vp != newvp) {
1305 		bp->b_vp = newvp;
1306 		vhold(bp->b_vp);
1307 	}
1308 	splx(s);
1309 }
1310 
1311 /*
1312  * Create a vnode for a device.
1313  * Used for mounting the root file system.
1314  */
1315 int
1316 bdevvp(dev, vpp)
1317 	dev_t dev;
1318 	struct vnode **vpp;
1319 {
1320 	register struct vnode *vp;
1321 	struct vnode *nvp;
1322 	int error;
1323 
1324 	if (dev == NODEV) {
1325 		*vpp = NULLVP;
1326 		return (ENXIO);
1327 	}
1328 	if (vfinddev(dev, VCHR, vpp))
1329 		return (0);
1330 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1331 	if (error) {
1332 		*vpp = NULLVP;
1333 		return (error);
1334 	}
1335 	vp = nvp;
1336 	vp->v_type = VCHR;
1337 	addalias(vp, dev);
1338 	*vpp = vp;
1339 	return (0);
1340 }
1341 
1342 /*
1343  * Add vnode to the alias list hung off the dev_t.
1344  *
1345  * The reason for this gunk is that multiple vnodes can reference
1346  * the same physical device, so checking vp->v_usecount to see
1347  * how many users there are is inadequate; the v_usecount for
1348  * the vnodes need to be accumulated.  vcount() does that.
1349  */
1350 struct vnode *
1351 addaliasu(nvp, nvp_rdev)
1352 	struct vnode *nvp;
1353 	udev_t nvp_rdev;
1354 {
1355 	struct vnode *ovp;
1356 	vop_t **ops;
1357 	dev_t dev;
1358 
1359 	if (nvp->v_type == VBLK)
1360 		return (nvp);
1361 	if (nvp->v_type != VCHR)
1362 		panic("addaliasu on non-special vnode");
1363 	dev = udev2dev(nvp_rdev, 0);
1364 	/*
1365 	 * Check to see if we have a bdevvp vnode with no associated
1366 	 * filesystem. If so, we want to associate the filesystem of
1367 	 * the new newly instigated vnode with the bdevvp vnode and
1368 	 * discard the newly created vnode rather than leaving the
1369 	 * bdevvp vnode lying around with no associated filesystem.
1370 	 */
1371 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1372 		addalias(nvp, dev);
1373 		return (nvp);
1374 	}
1375 	/*
1376 	 * Discard unneeded vnode, but save its node specific data.
1377 	 * Note that if there is a lock, it is carried over in the
1378 	 * node specific data to the replacement vnode.
1379 	 */
1380 	vref(ovp);
1381 	ovp->v_data = nvp->v_data;
1382 	ovp->v_tag = nvp->v_tag;
1383 	nvp->v_data = NULL;
1384 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1385 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1386 	if (nvp->v_vnlock)
1387 		ovp->v_vnlock = &ovp->v_lock;
1388 	ops = ovp->v_op;
1389 	ovp->v_op = nvp->v_op;
1390 	if (VOP_ISLOCKED(nvp, curproc)) {
1391 		VOP_UNLOCK(nvp, 0, curproc);
1392 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curproc);
1393 	}
1394 	nvp->v_op = ops;
1395 	insmntque(ovp, nvp->v_mount);
1396 	vrele(nvp);
1397 	vgone(nvp);
1398 	return (ovp);
1399 }
1400 
1401 /* This is a local helper function that do the same as addaliasu, but for a
1402  * dev_t instead of an udev_t. */
1403 static void
1404 addalias(nvp, dev)
1405 	struct vnode *nvp;
1406 	dev_t dev;
1407 {
1408 
1409 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1410 	nvp->v_rdev = dev;
1411 	mtx_lock(&spechash_mtx);
1412 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1413 	mtx_unlock(&spechash_mtx);
1414 }
1415 
1416 /*
1417  * Grab a particular vnode from the free list, increment its
1418  * reference count and lock it. The vnode lock bit is set if the
1419  * vnode is being eliminated in vgone. The process is awakened
1420  * when the transition is completed, and an error returned to
1421  * indicate that the vnode is no longer usable (possibly having
1422  * been changed to a new file system type).
1423  */
1424 int
1425 vget(vp, flags, p)
1426 	register struct vnode *vp;
1427 	int flags;
1428 	struct proc *p;
1429 {
1430 	int error;
1431 
1432 	/*
1433 	 * If the vnode is in the process of being cleaned out for
1434 	 * another use, we wait for the cleaning to finish and then
1435 	 * return failure. Cleaning is determined by checking that
1436 	 * the VXLOCK flag is set.
1437 	 */
1438 	if ((flags & LK_INTERLOCK) == 0)
1439 		mtx_lock(&vp->v_interlock);
1440 	if (vp->v_flag & VXLOCK) {
1441 		if (vp->v_vxproc == curproc) {
1442 			printf("VXLOCK interlock avoided\n");
1443 		} else {
1444 			vp->v_flag |= VXWANT;
1445 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1446 			    "vget", 0);
1447 			return (ENOENT);
1448 		}
1449 	}
1450 
1451 	vp->v_usecount++;
1452 
1453 	if (VSHOULDBUSY(vp))
1454 		vbusy(vp);
1455 	if (flags & LK_TYPE_MASK) {
1456 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1457 			/*
1458 			 * must expand vrele here because we do not want
1459 			 * to call VOP_INACTIVE if the reference count
1460 			 * drops back to zero since it was never really
1461 			 * active. We must remove it from the free list
1462 			 * before sleeping so that multiple processes do
1463 			 * not try to recycle it.
1464 			 */
1465 			mtx_lock(&vp->v_interlock);
1466 			vp->v_usecount--;
1467 			if (VSHOULDFREE(vp))
1468 				vfree(vp);
1469 			mtx_unlock(&vp->v_interlock);
1470 		}
1471 		return (error);
1472 	}
1473 	mtx_unlock(&vp->v_interlock);
1474 	return (0);
1475 }
1476 
1477 /*
1478  * Increase the reference count of a vnode.
1479  */
1480 void
1481 vref(struct vnode *vp)
1482 {
1483 	mtx_lock(&vp->v_interlock);
1484 	vp->v_usecount++;
1485 	mtx_unlock(&vp->v_interlock);
1486 }
1487 
1488 /*
1489  * Vnode put/release.
1490  * If count drops to zero, call inactive routine and return to freelist.
1491  */
1492 void
1493 vrele(vp)
1494 	struct vnode *vp;
1495 {
1496 	struct proc *p = curproc;	/* XXX */
1497 
1498 	KASSERT(vp != NULL, ("vrele: null vp"));
1499 
1500 	mtx_lock(&vp->v_interlock);
1501 
1502 	KASSERT(vp->v_writecount < vp->v_usecount, ("vrele: missed vn_close"));
1503 
1504 	if (vp->v_usecount > 1) {
1505 
1506 		vp->v_usecount--;
1507 		mtx_unlock(&vp->v_interlock);
1508 
1509 		return;
1510 	}
1511 
1512 	if (vp->v_usecount == 1) {
1513 
1514 		vp->v_usecount--;
1515 		if (VSHOULDFREE(vp))
1516 			vfree(vp);
1517 	/*
1518 	 * If we are doing a vput, the node is already locked, and we must
1519 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1520 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1521 	 */
1522 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1523 			VOP_INACTIVE(vp, p);
1524 		}
1525 
1526 	} else {
1527 #ifdef DIAGNOSTIC
1528 		vprint("vrele: negative ref count", vp);
1529 		mtx_unlock(&vp->v_interlock);
1530 #endif
1531 		panic("vrele: negative ref cnt");
1532 	}
1533 }
1534 
1535 /*
1536  * Release an already locked vnode.  This give the same effects as
1537  * unlock+vrele(), but takes less time and avoids releasing and
1538  * re-aquiring the lock (as vrele() aquires the lock internally.)
1539  */
1540 void
1541 vput(vp)
1542 	struct vnode *vp;
1543 {
1544 	struct proc *p = curproc;	/* XXX */
1545 
1546 	KASSERT(vp != NULL, ("vput: null vp"));
1547 	mtx_lock(&vp->v_interlock);
1548 	KASSERT(vp->v_writecount < vp->v_usecount, ("vput: missed vn_close"));
1549 
1550 	if (vp->v_usecount > 1) {
1551 
1552 		vp->v_usecount--;
1553 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1554 		return;
1555 
1556 	}
1557 
1558 	if (vp->v_usecount == 1) {
1559 
1560 		vp->v_usecount--;
1561 		if (VSHOULDFREE(vp))
1562 			vfree(vp);
1563 	/*
1564 	 * If we are doing a vput, the node is already locked, and we must
1565 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1566 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1567 	 */
1568 		mtx_unlock(&vp->v_interlock);
1569 		VOP_INACTIVE(vp, p);
1570 
1571 	} else {
1572 #ifdef DIAGNOSTIC
1573 		vprint("vput: negative ref count", vp);
1574 #endif
1575 		panic("vput: negative ref cnt");
1576 	}
1577 }
1578 
1579 /*
1580  * Somebody doesn't want the vnode recycled.
1581  */
1582 void
1583 vhold(vp)
1584 	register struct vnode *vp;
1585 {
1586 	int s;
1587 
1588   	s = splbio();
1589 	vp->v_holdcnt++;
1590 	if (VSHOULDBUSY(vp))
1591 		vbusy(vp);
1592 	splx(s);
1593 }
1594 
1595 /*
1596  * Note that there is one less who cares about this vnode.  vdrop() is the
1597  * opposite of vhold().
1598  */
1599 void
1600 vdrop(vp)
1601 	register struct vnode *vp;
1602 {
1603 	int s;
1604 
1605 	s = splbio();
1606 	if (vp->v_holdcnt <= 0)
1607 		panic("vdrop: holdcnt");
1608 	vp->v_holdcnt--;
1609 	if (VSHOULDFREE(vp))
1610 		vfree(vp);
1611 	splx(s);
1612 }
1613 
1614 /*
1615  * Remove any vnodes in the vnode table belonging to mount point mp.
1616  *
1617  * If MNT_NOFORCE is specified, there should not be any active ones,
1618  * return error if any are found (nb: this is a user error, not a
1619  * system error). If MNT_FORCE is specified, detach any active vnodes
1620  * that are found.
1621  */
1622 #ifdef DIAGNOSTIC
1623 static int busyprt = 0;		/* print out busy vnodes */
1624 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1625 #endif
1626 
1627 int
1628 vflush(mp, skipvp, flags)
1629 	struct mount *mp;
1630 	struct vnode *skipvp;
1631 	int flags;
1632 {
1633 	struct proc *p = curproc;	/* XXX */
1634 	struct vnode *vp, *nvp;
1635 	int busy = 0;
1636 
1637 	mtx_lock(&mntvnode_mtx);
1638 loop:
1639 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1640 		/*
1641 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1642 		 * Start over if it has (it won't be on the list anymore).
1643 		 */
1644 		if (vp->v_mount != mp)
1645 			goto loop;
1646 		nvp = LIST_NEXT(vp, v_mntvnodes);
1647 		/*
1648 		 * Skip over a selected vnode.
1649 		 */
1650 		if (vp == skipvp)
1651 			continue;
1652 
1653 		mtx_lock(&vp->v_interlock);
1654 		/*
1655 		 * Skip over a vnodes marked VSYSTEM.
1656 		 */
1657 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1658 			mtx_unlock(&vp->v_interlock);
1659 			continue;
1660 		}
1661 		/*
1662 		 * If WRITECLOSE is set, only flush out regular file vnodes
1663 		 * open for writing.
1664 		 */
1665 		if ((flags & WRITECLOSE) &&
1666 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1667 			mtx_unlock(&vp->v_interlock);
1668 			continue;
1669 		}
1670 
1671 		/*
1672 		 * With v_usecount == 0, all we need to do is clear out the
1673 		 * vnode data structures and we are done.
1674 		 */
1675 		if (vp->v_usecount == 0) {
1676 			mtx_unlock(&mntvnode_mtx);
1677 			vgonel(vp, p);
1678 			mtx_lock(&mntvnode_mtx);
1679 			continue;
1680 		}
1681 
1682 		/*
1683 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1684 		 * or character devices, revert to an anonymous device. For
1685 		 * all other files, just kill them.
1686 		 */
1687 		if (flags & FORCECLOSE) {
1688 			mtx_unlock(&mntvnode_mtx);
1689 			if (vp->v_type != VCHR) {
1690 				vgonel(vp, p);
1691 			} else {
1692 				vclean(vp, 0, p);
1693 				vp->v_op = spec_vnodeop_p;
1694 				insmntque(vp, (struct mount *) 0);
1695 			}
1696 			mtx_lock(&mntvnode_mtx);
1697 			continue;
1698 		}
1699 #ifdef DIAGNOSTIC
1700 		if (busyprt)
1701 			vprint("vflush: busy vnode", vp);
1702 #endif
1703 		mtx_unlock(&vp->v_interlock);
1704 		busy++;
1705 	}
1706 	mtx_unlock(&mntvnode_mtx);
1707 	if (busy)
1708 		return (EBUSY);
1709 	return (0);
1710 }
1711 
1712 /*
1713  * Disassociate the underlying file system from a vnode.
1714  */
1715 static void
1716 vclean(vp, flags, p)
1717 	struct vnode *vp;
1718 	int flags;
1719 	struct proc *p;
1720 {
1721 	int active;
1722 
1723 	/*
1724 	 * Check to see if the vnode is in use. If so we have to reference it
1725 	 * before we clean it out so that its count cannot fall to zero and
1726 	 * generate a race against ourselves to recycle it.
1727 	 */
1728 	if ((active = vp->v_usecount))
1729 		vp->v_usecount++;
1730 
1731 	/*
1732 	 * Prevent the vnode from being recycled or brought into use while we
1733 	 * clean it out.
1734 	 */
1735 	if (vp->v_flag & VXLOCK)
1736 		panic("vclean: deadlock");
1737 	vp->v_flag |= VXLOCK;
1738 	vp->v_vxproc = curproc;
1739 	/*
1740 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1741 	 * have the object locked while it cleans it out. The VOP_LOCK
1742 	 * ensures that the VOP_INACTIVE routine is done with its work.
1743 	 * For active vnodes, it ensures that no other activity can
1744 	 * occur while the underlying object is being cleaned out.
1745 	 */
1746 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1747 
1748 	/*
1749 	 * Clean out any buffers associated with the vnode.
1750 	 * If the flush fails, just toss the buffers.
1751 	 */
1752 	if (flags & DOCLOSE) {
1753 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
1754 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
1755 		if (vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0) != 0)
1756 			vinvalbuf(vp, 0, NOCRED, p, 0, 0);
1757 	}
1758 
1759 	VOP_DESTROYVOBJECT(vp);
1760 
1761 	/*
1762 	 * If purging an active vnode, it must be closed and
1763 	 * deactivated before being reclaimed. Note that the
1764 	 * VOP_INACTIVE will unlock the vnode.
1765 	 */
1766 	if (active) {
1767 		if (flags & DOCLOSE)
1768 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1769 		VOP_INACTIVE(vp, p);
1770 	} else {
1771 		/*
1772 		 * Any other processes trying to obtain this lock must first
1773 		 * wait for VXLOCK to clear, then call the new lock operation.
1774 		 */
1775 		VOP_UNLOCK(vp, 0, p);
1776 	}
1777 	/*
1778 	 * Reclaim the vnode.
1779 	 */
1780 	if (VOP_RECLAIM(vp, p))
1781 		panic("vclean: cannot reclaim");
1782 
1783 	if (active) {
1784 		/*
1785 		 * Inline copy of vrele() since VOP_INACTIVE
1786 		 * has already been called.
1787 		 */
1788 		mtx_lock(&vp->v_interlock);
1789 		if (--vp->v_usecount <= 0) {
1790 #ifdef DIAGNOSTIC
1791 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1792 				vprint("vclean: bad ref count", vp);
1793 				panic("vclean: ref cnt");
1794 			}
1795 #endif
1796 			vfree(vp);
1797 		}
1798 		mtx_unlock(&vp->v_interlock);
1799 	}
1800 
1801 	cache_purge(vp);
1802 	vp->v_vnlock = NULL;
1803 	lockdestroy(&vp->v_lock);
1804 
1805 	if (VSHOULDFREE(vp))
1806 		vfree(vp);
1807 
1808 	/*
1809 	 * Done with purge, notify sleepers of the grim news.
1810 	 */
1811 	vp->v_op = dead_vnodeop_p;
1812 	vn_pollgone(vp);
1813 	vp->v_tag = VT_NON;
1814 	vp->v_flag &= ~VXLOCK;
1815 	vp->v_vxproc = NULL;
1816 	if (vp->v_flag & VXWANT) {
1817 		vp->v_flag &= ~VXWANT;
1818 		wakeup((caddr_t) vp);
1819 	}
1820 }
1821 
1822 /*
1823  * Eliminate all activity associated with the requested vnode
1824  * and with all vnodes aliased to the requested vnode.
1825  */
1826 int
1827 vop_revoke(ap)
1828 	struct vop_revoke_args /* {
1829 		struct vnode *a_vp;
1830 		int a_flags;
1831 	} */ *ap;
1832 {
1833 	struct vnode *vp, *vq;
1834 	dev_t dev;
1835 
1836 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1837 
1838 	vp = ap->a_vp;
1839 	/*
1840 	 * If a vgone (or vclean) is already in progress,
1841 	 * wait until it is done and return.
1842 	 */
1843 	if (vp->v_flag & VXLOCK) {
1844 		vp->v_flag |= VXWANT;
1845 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1846 		    "vop_revokeall", 0);
1847 		return (0);
1848 	}
1849 	dev = vp->v_rdev;
1850 	for (;;) {
1851 		mtx_lock(&spechash_mtx);
1852 		vq = SLIST_FIRST(&dev->si_hlist);
1853 		mtx_unlock(&spechash_mtx);
1854 		if (!vq)
1855 			break;
1856 		vgone(vq);
1857 	}
1858 	return (0);
1859 }
1860 
1861 /*
1862  * Recycle an unused vnode to the front of the free list.
1863  * Release the passed interlock if the vnode will be recycled.
1864  */
1865 int
1866 vrecycle(vp, inter_lkp, p)
1867 	struct vnode *vp;
1868 	struct mtx *inter_lkp;
1869 	struct proc *p;
1870 {
1871 
1872 	mtx_lock(&vp->v_interlock);
1873 	if (vp->v_usecount == 0) {
1874 		if (inter_lkp) {
1875 			mtx_unlock(inter_lkp);
1876 		}
1877 		vgonel(vp, p);
1878 		return (1);
1879 	}
1880 	mtx_unlock(&vp->v_interlock);
1881 	return (0);
1882 }
1883 
1884 /*
1885  * Eliminate all activity associated with a vnode
1886  * in preparation for reuse.
1887  */
1888 void
1889 vgone(vp)
1890 	register struct vnode *vp;
1891 {
1892 	struct proc *p = curproc;	/* XXX */
1893 
1894 	mtx_lock(&vp->v_interlock);
1895 	vgonel(vp, p);
1896 }
1897 
1898 /*
1899  * vgone, with the vp interlock held.
1900  */
1901 void
1902 vgonel(vp, p)
1903 	struct vnode *vp;
1904 	struct proc *p;
1905 {
1906 	int s;
1907 
1908 	/*
1909 	 * If a vgone (or vclean) is already in progress,
1910 	 * wait until it is done and return.
1911 	 */
1912 	if (vp->v_flag & VXLOCK) {
1913 		vp->v_flag |= VXWANT;
1914 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1915 		    "vgone", 0);
1916 		return;
1917 	}
1918 
1919 	/*
1920 	 * Clean out the filesystem specific data.
1921 	 */
1922 	vclean(vp, DOCLOSE, p);
1923 	mtx_lock(&vp->v_interlock);
1924 
1925 	/*
1926 	 * Delete from old mount point vnode list, if on one.
1927 	 */
1928 	if (vp->v_mount != NULL)
1929 		insmntque(vp, (struct mount *)0);
1930 	/*
1931 	 * If special device, remove it from special device alias list
1932 	 * if it is on one.
1933 	 */
1934 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
1935 		mtx_lock(&spechash_mtx);
1936 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
1937 		freedev(vp->v_rdev);
1938 		mtx_unlock(&spechash_mtx);
1939 		vp->v_rdev = NULL;
1940 	}
1941 
1942 	/*
1943 	 * If it is on the freelist and not already at the head,
1944 	 * move it to the head of the list. The test of the
1945 	 * VDOOMED flag and the reference count of zero is because
1946 	 * it will be removed from the free list by getnewvnode,
1947 	 * but will not have its reference count incremented until
1948 	 * after calling vgone. If the reference count were
1949 	 * incremented first, vgone would (incorrectly) try to
1950 	 * close the previous instance of the underlying object.
1951 	 */
1952 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1953 		s = splbio();
1954 		mtx_lock(&vnode_free_list_mtx);
1955 		if (vp->v_flag & VFREE)
1956 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1957 		else
1958 			freevnodes++;
1959 		vp->v_flag |= VFREE;
1960 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1961 		mtx_unlock(&vnode_free_list_mtx);
1962 		splx(s);
1963 	}
1964 
1965 	vp->v_type = VBAD;
1966 	mtx_unlock(&vp->v_interlock);
1967 }
1968 
1969 /*
1970  * Lookup a vnode by device number.
1971  */
1972 int
1973 vfinddev(dev, type, vpp)
1974 	dev_t dev;
1975 	enum vtype type;
1976 	struct vnode **vpp;
1977 {
1978 	struct vnode *vp;
1979 
1980 	mtx_lock(&spechash_mtx);
1981 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1982 		if (type == vp->v_type) {
1983 			*vpp = vp;
1984 			mtx_unlock(&spechash_mtx);
1985 			return (1);
1986 		}
1987 	}
1988 	mtx_unlock(&spechash_mtx);
1989 	return (0);
1990 }
1991 
1992 /*
1993  * Calculate the total number of references to a special device.
1994  */
1995 int
1996 vcount(vp)
1997 	struct vnode *vp;
1998 {
1999 	struct vnode *vq;
2000 	int count;
2001 
2002 	count = 0;
2003 	mtx_lock(&spechash_mtx);
2004 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2005 		count += vq->v_usecount;
2006 	mtx_unlock(&spechash_mtx);
2007 	return (count);
2008 }
2009 
2010 /*
2011  * Same as above, but using the dev_t as argument
2012  */
2013 int
2014 count_dev(dev)
2015 	dev_t dev;
2016 {
2017 	struct vnode *vp;
2018 
2019 	vp = SLIST_FIRST(&dev->si_hlist);
2020 	if (vp == NULL)
2021 		return (0);
2022 	return(vcount(vp));
2023 }
2024 
2025 /*
2026  * Print out a description of a vnode.
2027  */
2028 static char *typename[] =
2029 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2030 
2031 void
2032 vprint(label, vp)
2033 	char *label;
2034 	struct vnode *vp;
2035 {
2036 	char buf[96];
2037 
2038 	if (label != NULL)
2039 		printf("%s: %p: ", label, (void *)vp);
2040 	else
2041 		printf("%p: ", (void *)vp);
2042 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2043 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2044 	    vp->v_holdcnt);
2045 	buf[0] = '\0';
2046 	if (vp->v_flag & VROOT)
2047 		strcat(buf, "|VROOT");
2048 	if (vp->v_flag & VTEXT)
2049 		strcat(buf, "|VTEXT");
2050 	if (vp->v_flag & VSYSTEM)
2051 		strcat(buf, "|VSYSTEM");
2052 	if (vp->v_flag & VXLOCK)
2053 		strcat(buf, "|VXLOCK");
2054 	if (vp->v_flag & VXWANT)
2055 		strcat(buf, "|VXWANT");
2056 	if (vp->v_flag & VBWAIT)
2057 		strcat(buf, "|VBWAIT");
2058 	if (vp->v_flag & VDOOMED)
2059 		strcat(buf, "|VDOOMED");
2060 	if (vp->v_flag & VFREE)
2061 		strcat(buf, "|VFREE");
2062 	if (vp->v_flag & VOBJBUF)
2063 		strcat(buf, "|VOBJBUF");
2064 	if (buf[0] != '\0')
2065 		printf(" flags (%s)", &buf[1]);
2066 	if (vp->v_data == NULL) {
2067 		printf("\n");
2068 	} else {
2069 		printf("\n\t");
2070 		VOP_PRINT(vp);
2071 	}
2072 }
2073 
2074 #ifdef DDB
2075 #include <ddb/ddb.h>
2076 /*
2077  * List all of the locked vnodes in the system.
2078  * Called when debugging the kernel.
2079  */
2080 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2081 {
2082 	struct proc *p = curproc;	/* XXX */
2083 	struct mount *mp, *nmp;
2084 	struct vnode *vp;
2085 
2086 	printf("Locked vnodes\n");
2087 	mtx_lock(&mountlist_mtx);
2088 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2089 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, p)) {
2090 			nmp = TAILQ_NEXT(mp, mnt_list);
2091 			continue;
2092 		}
2093 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2094 			if (VOP_ISLOCKED(vp, NULL))
2095 				vprint((char *)0, vp);
2096 		}
2097 		mtx_lock(&mountlist_mtx);
2098 		nmp = TAILQ_NEXT(mp, mnt_list);
2099 		vfs_unbusy(mp, p);
2100 	}
2101 	mtx_unlock(&mountlist_mtx);
2102 }
2103 #endif
2104 
2105 /*
2106  * Top level filesystem related information gathering.
2107  */
2108 static int	sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2109 
2110 static int
2111 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2112 {
2113 	int *name = (int *)arg1 - 1;	/* XXX */
2114 	u_int namelen = arg2 + 1;	/* XXX */
2115 	struct vfsconf *vfsp;
2116 
2117 #if 1 || defined(COMPAT_PRELITE2)
2118 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2119 	if (namelen == 1)
2120 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2121 #endif
2122 
2123 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2124 #ifdef notyet
2125 	/* all sysctl names at this level are at least name and field */
2126 	if (namelen < 2)
2127 		return (ENOTDIR);		/* overloaded */
2128 	if (name[0] != VFS_GENERIC) {
2129 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2130 			if (vfsp->vfc_typenum == name[0])
2131 				break;
2132 		if (vfsp == NULL)
2133 			return (EOPNOTSUPP);
2134 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2135 		    oldp, oldlenp, newp, newlen, p));
2136 	}
2137 #endif
2138 	switch (name[1]) {
2139 	case VFS_MAXTYPENUM:
2140 		if (namelen != 2)
2141 			return (ENOTDIR);
2142 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2143 	case VFS_CONF:
2144 		if (namelen != 3)
2145 			return (ENOTDIR);	/* overloaded */
2146 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2147 			if (vfsp->vfc_typenum == name[2])
2148 				break;
2149 		if (vfsp == NULL)
2150 			return (EOPNOTSUPP);
2151 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2152 	}
2153 	return (EOPNOTSUPP);
2154 }
2155 
2156 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2157 	"Generic filesystem");
2158 
2159 #if 1 || defined(COMPAT_PRELITE2)
2160 
2161 static int
2162 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2163 {
2164 	int error;
2165 	struct vfsconf *vfsp;
2166 	struct ovfsconf ovfs;
2167 
2168 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2169 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2170 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2171 		ovfs.vfc_index = vfsp->vfc_typenum;
2172 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2173 		ovfs.vfc_flags = vfsp->vfc_flags;
2174 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2175 		if (error)
2176 			return error;
2177 	}
2178 	return 0;
2179 }
2180 
2181 #endif /* 1 || COMPAT_PRELITE2 */
2182 
2183 #if COMPILING_LINT
2184 #define KINFO_VNODESLOP	10
2185 /*
2186  * Dump vnode list (via sysctl).
2187  * Copyout address of vnode followed by vnode.
2188  */
2189 /* ARGSUSED */
2190 static int
2191 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2192 {
2193 	struct proc *p = curproc;	/* XXX */
2194 	struct mount *mp, *nmp;
2195 	struct vnode *nvp, *vp;
2196 	int error;
2197 
2198 #define VPTRSZ	sizeof (struct vnode *)
2199 #define VNODESZ	sizeof (struct vnode)
2200 
2201 	req->lock = 0;
2202 	if (!req->oldptr) /* Make an estimate */
2203 		return (SYSCTL_OUT(req, 0,
2204 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2205 
2206 	mtx_lock(&mountlist_mtx);
2207 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2208 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, p)) {
2209 			nmp = TAILQ_NEXT(mp, mnt_list);
2210 			continue;
2211 		}
2212 again:
2213 		mtx_lock(&mntvnode_mtx);
2214 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2215 		     vp != NULL;
2216 		     vp = nvp) {
2217 			/*
2218 			 * Check that the vp is still associated with
2219 			 * this filesystem.  RACE: could have been
2220 			 * recycled onto the same filesystem.
2221 			 */
2222 			if (vp->v_mount != mp) {
2223 				mtx_unlock(&mntvnode_mtx);
2224 				goto again;
2225 			}
2226 			nvp = LIST_NEXT(vp, v_mntvnodes);
2227 			mtx_unlock(&mntvnode_mtx);
2228 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2229 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2230 				return (error);
2231 			mtx_lock(&mntvnode_mtx);
2232 		}
2233 		mtx_unlock(&mntvnode_mtx);
2234 		mtx_lock(&mountlist_mtx);
2235 		nmp = TAILQ_NEXT(mp, mnt_list);
2236 		vfs_unbusy(mp, p);
2237 	}
2238 	mtx_unlock(&mountlist_mtx);
2239 
2240 	return (0);
2241 }
2242 
2243 /*
2244  * XXX
2245  * Exporting the vnode list on large systems causes them to crash.
2246  * Exporting the vnode list on medium systems causes sysctl to coredump.
2247  */
2248 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2249 	0, 0, sysctl_vnode, "S,vnode", "");
2250 #endif
2251 
2252 /*
2253  * Check to see if a filesystem is mounted on a block device.
2254  */
2255 int
2256 vfs_mountedon(vp)
2257 	struct vnode *vp;
2258 {
2259 
2260 	if (vp->v_rdev->si_mountpoint != NULL)
2261 		return (EBUSY);
2262 	return (0);
2263 }
2264 
2265 /*
2266  * Unmount all filesystems. The list is traversed in reverse order
2267  * of mounting to avoid dependencies.
2268  */
2269 void
2270 vfs_unmountall()
2271 {
2272 	struct mount *mp;
2273 	struct proc *p;
2274 	int error;
2275 
2276 	if (curproc != NULL)
2277 		p = curproc;
2278 	else
2279 		p = initproc;	/* XXX XXX should this be proc0? */
2280 	/*
2281 	 * Since this only runs when rebooting, it is not interlocked.
2282 	 */
2283 	while(!TAILQ_EMPTY(&mountlist)) {
2284 		mp = TAILQ_LAST(&mountlist, mntlist);
2285 		error = dounmount(mp, MNT_FORCE, p);
2286 		if (error) {
2287 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2288 			printf("unmount of %s failed (",
2289 			    mp->mnt_stat.f_mntonname);
2290 			if (error == EBUSY)
2291 				printf("BUSY)\n");
2292 			else
2293 				printf("%d)\n", error);
2294 		} else {
2295 			/* The unmount has removed mp from the mountlist */
2296 		}
2297 	}
2298 }
2299 
2300 /*
2301  * Build hash lists of net addresses and hang them off the mount point.
2302  * Called by ufs_mount() to set up the lists of export addresses.
2303  */
2304 static int
2305 vfs_hang_addrlist(mp, nep, argp)
2306 	struct mount *mp;
2307 	struct netexport *nep;
2308 	struct export_args *argp;
2309 {
2310 	register struct netcred *np;
2311 	register struct radix_node_head *rnh;
2312 	register int i;
2313 	struct radix_node *rn;
2314 	struct sockaddr *saddr, *smask = 0;
2315 	struct domain *dom;
2316 	int error;
2317 
2318 	if (argp->ex_addrlen == 0) {
2319 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2320 			return (EPERM);
2321 		np = &nep->ne_defexported;
2322 		np->netc_exflags = argp->ex_flags;
2323 		bzero(&np->netc_anon, sizeof(np->netc_anon));
2324 		np->netc_anon.cr_uid = argp->ex_anon.cr_uid;
2325 		np->netc_anon.cr_ngroups = argp->ex_anon.cr_ngroups;
2326 		bcopy(argp->ex_anon.cr_groups, np->netc_anon.cr_groups,
2327 		    sizeof(np->netc_anon.cr_groups));
2328 		np->netc_anon.cr_ref = 1;
2329 		mp->mnt_flag |= MNT_DEFEXPORTED;
2330 		return (0);
2331 	}
2332 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2333 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK | M_ZERO);
2334 	saddr = (struct sockaddr *) (np + 1);
2335 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2336 		goto out;
2337 	if (saddr->sa_len > argp->ex_addrlen)
2338 		saddr->sa_len = argp->ex_addrlen;
2339 	if (argp->ex_masklen) {
2340 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2341 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2342 		if (error)
2343 			goto out;
2344 		if (smask->sa_len > argp->ex_masklen)
2345 			smask->sa_len = argp->ex_masklen;
2346 	}
2347 	i = saddr->sa_family;
2348 	if ((rnh = nep->ne_rtable[i]) == 0) {
2349 		/*
2350 		 * Seems silly to initialize every AF when most are not used,
2351 		 * do so on demand here
2352 		 */
2353 		for (dom = domains; dom; dom = dom->dom_next)
2354 			if (dom->dom_family == i && dom->dom_rtattach) {
2355 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2356 				    dom->dom_rtoffset);
2357 				break;
2358 			}
2359 		if ((rnh = nep->ne_rtable[i]) == 0) {
2360 			error = ENOBUFS;
2361 			goto out;
2362 		}
2363 	}
2364 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2365 	    np->netc_rnodes);
2366 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2367 		error = EPERM;
2368 		goto out;
2369 	}
2370 	np->netc_exflags = argp->ex_flags;
2371 	bzero(&np->netc_anon, sizeof(np->netc_anon));
2372 	np->netc_anon.cr_uid = argp->ex_anon.cr_uid;
2373 	np->netc_anon.cr_ngroups = argp->ex_anon.cr_ngroups;
2374 	bcopy(argp->ex_anon.cr_groups, np->netc_anon.cr_groups,
2375 	    sizeof(np->netc_anon.cr_groups));
2376 	np->netc_anon.cr_ref = 1;
2377 	return (0);
2378 out:
2379 	free(np, M_NETADDR);
2380 	return (error);
2381 }
2382 
2383 /* Helper for vfs_free_addrlist. */
2384 /* ARGSUSED */
2385 static int
2386 vfs_free_netcred(rn, w)
2387 	struct radix_node *rn;
2388 	void *w;
2389 {
2390 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
2391 
2392 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2393 	free((caddr_t) rn, M_NETADDR);
2394 	return (0);
2395 }
2396 
2397 /*
2398  * Free the net address hash lists that are hanging off the mount points.
2399  */
2400 static void
2401 vfs_free_addrlist(nep)
2402 	struct netexport *nep;
2403 {
2404 	register int i;
2405 	register struct radix_node_head *rnh;
2406 
2407 	for (i = 0; i <= AF_MAX; i++)
2408 		if ((rnh = nep->ne_rtable[i])) {
2409 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2410 			    (caddr_t) rnh);
2411 			free((caddr_t) rnh, M_RTABLE);
2412 			nep->ne_rtable[i] = 0;
2413 		}
2414 }
2415 
2416 /*
2417  * High level function to manipulate export options on a mount point
2418  * and the passed in netexport.
2419  * Struct export_args *argp is the variable used to twiddle options,
2420  * the structure is described in sys/mount.h
2421  */
2422 int
2423 vfs_export(mp, nep, argp)
2424 	struct mount *mp;
2425 	struct netexport *nep;
2426 	struct export_args *argp;
2427 {
2428 	int error;
2429 
2430 	if (argp->ex_flags & MNT_DELEXPORT) {
2431 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2432 			vfs_setpublicfs(NULL, NULL, NULL);
2433 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2434 		}
2435 		vfs_free_addrlist(nep);
2436 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2437 	}
2438 	if (argp->ex_flags & MNT_EXPORTED) {
2439 		if (argp->ex_flags & MNT_EXPUBLIC) {
2440 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2441 				return (error);
2442 			mp->mnt_flag |= MNT_EXPUBLIC;
2443 		}
2444 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2445 			return (error);
2446 		mp->mnt_flag |= MNT_EXPORTED;
2447 	}
2448 	return (0);
2449 }
2450 
2451 /*
2452  * Set the publicly exported filesystem (WebNFS). Currently, only
2453  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2454  */
2455 int
2456 vfs_setpublicfs(mp, nep, argp)
2457 	struct mount *mp;
2458 	struct netexport *nep;
2459 	struct export_args *argp;
2460 {
2461 	int error;
2462 	struct vnode *rvp;
2463 	char *cp;
2464 
2465 	/*
2466 	 * mp == NULL -> invalidate the current info, the FS is
2467 	 * no longer exported. May be called from either vfs_export
2468 	 * or unmount, so check if it hasn't already been done.
2469 	 */
2470 	if (mp == NULL) {
2471 		if (nfs_pub.np_valid) {
2472 			nfs_pub.np_valid = 0;
2473 			if (nfs_pub.np_index != NULL) {
2474 				FREE(nfs_pub.np_index, M_TEMP);
2475 				nfs_pub.np_index = NULL;
2476 			}
2477 		}
2478 		return (0);
2479 	}
2480 
2481 	/*
2482 	 * Only one allowed at a time.
2483 	 */
2484 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2485 		return (EBUSY);
2486 
2487 	/*
2488 	 * Get real filehandle for root of exported FS.
2489 	 */
2490 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2491 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2492 
2493 	if ((error = VFS_ROOT(mp, &rvp)))
2494 		return (error);
2495 
2496 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2497 		return (error);
2498 
2499 	vput(rvp);
2500 
2501 	/*
2502 	 * If an indexfile was specified, pull it in.
2503 	 */
2504 	if (argp->ex_indexfile != NULL) {
2505 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2506 		    M_WAITOK);
2507 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2508 		    MAXNAMLEN, (size_t *)0);
2509 		if (!error) {
2510 			/*
2511 			 * Check for illegal filenames.
2512 			 */
2513 			for (cp = nfs_pub.np_index; *cp; cp++) {
2514 				if (*cp == '/') {
2515 					error = EINVAL;
2516 					break;
2517 				}
2518 			}
2519 		}
2520 		if (error) {
2521 			FREE(nfs_pub.np_index, M_TEMP);
2522 			return (error);
2523 		}
2524 	}
2525 
2526 	nfs_pub.np_mount = mp;
2527 	nfs_pub.np_valid = 1;
2528 	return (0);
2529 }
2530 
2531 /*
2532  * Used by the filesystems to determine if a given network address
2533  * (passed in 'nam') is present in thier exports list, returns a pointer
2534  * to struct netcred so that the filesystem can examine it for
2535  * access rights (read/write/etc).
2536  */
2537 struct netcred *
2538 vfs_export_lookup(mp, nep, nam)
2539 	register struct mount *mp;
2540 	struct netexport *nep;
2541 	struct sockaddr *nam;
2542 {
2543 	register struct netcred *np;
2544 	register struct radix_node_head *rnh;
2545 	struct sockaddr *saddr;
2546 
2547 	np = NULL;
2548 	if (mp->mnt_flag & MNT_EXPORTED) {
2549 		/*
2550 		 * Lookup in the export list first.
2551 		 */
2552 		if (nam != NULL) {
2553 			saddr = nam;
2554 			rnh = nep->ne_rtable[saddr->sa_family];
2555 			if (rnh != NULL) {
2556 				np = (struct netcred *)
2557 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2558 							      rnh);
2559 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2560 					np = NULL;
2561 			}
2562 		}
2563 		/*
2564 		 * If no address match, use the default if it exists.
2565 		 */
2566 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2567 			np = &nep->ne_defexported;
2568 	}
2569 	return (np);
2570 }
2571 
2572 /*
2573  * perform msync on all vnodes under a mount point
2574  * the mount point must be locked.
2575  */
2576 void
2577 vfs_msync(struct mount *mp, int flags) {
2578 	struct vnode *vp, *nvp;
2579 	struct vm_object *obj;
2580 	int anyio, tries;
2581 
2582 	tries = 5;
2583 loop:
2584 	anyio = 0;
2585 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
2586 
2587 		nvp = LIST_NEXT(vp, v_mntvnodes);
2588 
2589 		if (vp->v_mount != mp) {
2590 			goto loop;
2591 		}
2592 
2593 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2594 			continue;
2595 
2596 		if (flags != MNT_WAIT) {
2597 			if (VOP_GETVOBJECT(vp, &obj) != 0 ||
2598 			    (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2599 				continue;
2600 			if (VOP_ISLOCKED(vp, NULL))
2601 				continue;
2602 		}
2603 
2604 		mtx_lock(&vp->v_interlock);
2605 		if (VOP_GETVOBJECT(vp, &obj) == 0 &&
2606 		    (obj->flags & OBJ_MIGHTBEDIRTY)) {
2607 			if (!vget(vp,
2608 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2609 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2610 					vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2611 					anyio = 1;
2612 				}
2613 				vput(vp);
2614 			}
2615 		} else {
2616 			mtx_unlock(&vp->v_interlock);
2617 		}
2618 	}
2619 	if (anyio && (--tries > 0))
2620 		goto loop;
2621 }
2622 
2623 /*
2624  * Create the VM object needed for VMIO and mmap support.  This
2625  * is done for all VREG files in the system.  Some filesystems might
2626  * afford the additional metadata buffering capability of the
2627  * VMIO code by making the device node be VMIO mode also.
2628  *
2629  * vp must be locked when vfs_object_create is called.
2630  */
2631 int
2632 vfs_object_create(vp, p, cred)
2633 	struct vnode *vp;
2634 	struct proc *p;
2635 	struct ucred *cred;
2636 {
2637 	return (VOP_CREATEVOBJECT(vp, cred, p));
2638 }
2639 
2640 /*
2641  * Mark a vnode as free, putting it up for recycling.
2642  */
2643 void
2644 vfree(vp)
2645 	struct vnode *vp;
2646 {
2647 	int s;
2648 
2649 	s = splbio();
2650 	mtx_lock(&vnode_free_list_mtx);
2651 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2652 	if (vp->v_flag & VAGE) {
2653 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2654 	} else {
2655 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2656 	}
2657 	freevnodes++;
2658 	mtx_unlock(&vnode_free_list_mtx);
2659 	vp->v_flag &= ~VAGE;
2660 	vp->v_flag |= VFREE;
2661 	splx(s);
2662 }
2663 
2664 /*
2665  * Opposite of vfree() - mark a vnode as in use.
2666  */
2667 void
2668 vbusy(vp)
2669 	struct vnode *vp;
2670 {
2671 	int s;
2672 
2673 	s = splbio();
2674 	mtx_lock(&vnode_free_list_mtx);
2675 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2676 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2677 	freevnodes--;
2678 	mtx_unlock(&vnode_free_list_mtx);
2679 	vp->v_flag &= ~(VFREE|VAGE);
2680 	splx(s);
2681 }
2682 
2683 /*
2684  * Record a process's interest in events which might happen to
2685  * a vnode.  Because poll uses the historic select-style interface
2686  * internally, this routine serves as both the ``check for any
2687  * pending events'' and the ``record my interest in future events''
2688  * functions.  (These are done together, while the lock is held,
2689  * to avoid race conditions.)
2690  */
2691 int
2692 vn_pollrecord(vp, p, events)
2693 	struct vnode *vp;
2694 	struct proc *p;
2695 	short events;
2696 {
2697 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2698 	if (vp->v_pollinfo.vpi_revents & events) {
2699 		/*
2700 		 * This leaves events we are not interested
2701 		 * in available for the other process which
2702 		 * which presumably had requested them
2703 		 * (otherwise they would never have been
2704 		 * recorded).
2705 		 */
2706 		events &= vp->v_pollinfo.vpi_revents;
2707 		vp->v_pollinfo.vpi_revents &= ~events;
2708 
2709 		mtx_unlock(&vp->v_pollinfo.vpi_lock);
2710 		return events;
2711 	}
2712 	vp->v_pollinfo.vpi_events |= events;
2713 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2714 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2715 	return 0;
2716 }
2717 
2718 /*
2719  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2720  * it is possible for us to miss an event due to race conditions, but
2721  * that condition is expected to be rare, so for the moment it is the
2722  * preferred interface.
2723  */
2724 void
2725 vn_pollevent(vp, events)
2726 	struct vnode *vp;
2727 	short events;
2728 {
2729 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2730 	if (vp->v_pollinfo.vpi_events & events) {
2731 		/*
2732 		 * We clear vpi_events so that we don't
2733 		 * call selwakeup() twice if two events are
2734 		 * posted before the polling process(es) is
2735 		 * awakened.  This also ensures that we take at
2736 		 * most one selwakeup() if the polling process
2737 		 * is no longer interested.  However, it does
2738 		 * mean that only one event can be noticed at
2739 		 * a time.  (Perhaps we should only clear those
2740 		 * event bits which we note?) XXX
2741 		 */
2742 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2743 		vp->v_pollinfo.vpi_revents |= events;
2744 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2745 	}
2746 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2747 }
2748 
2749 #define VN_KNOTE(vp, b) \
2750 	KNOTE((struct klist *)&vp->v_pollinfo.vpi_selinfo.si_note, (b))
2751 
2752 /*
2753  * Wake up anyone polling on vp because it is being revoked.
2754  * This depends on dead_poll() returning POLLHUP for correct
2755  * behavior.
2756  */
2757 void
2758 vn_pollgone(vp)
2759 	struct vnode *vp;
2760 {
2761 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2762         VN_KNOTE(vp, NOTE_REVOKE);
2763 	if (vp->v_pollinfo.vpi_events) {
2764 		vp->v_pollinfo.vpi_events = 0;
2765 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2766 	}
2767 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2768 }
2769 
2770 
2771 
2772 /*
2773  * Routine to create and manage a filesystem syncer vnode.
2774  */
2775 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2776 static int	sync_fsync __P((struct  vop_fsync_args *));
2777 static int	sync_inactive __P((struct  vop_inactive_args *));
2778 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2779 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2780 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2781 static int	sync_print __P((struct vop_print_args *));
2782 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2783 
2784 static vop_t **sync_vnodeop_p;
2785 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2786 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2787 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2788 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2789 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2790 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2791 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2792 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2793 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2794 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2795 	{ NULL, NULL }
2796 };
2797 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2798 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2799 
2800 VNODEOP_SET(sync_vnodeop_opv_desc);
2801 
2802 /*
2803  * Create a new filesystem syncer vnode for the specified mount point.
2804  */
2805 int
2806 vfs_allocate_syncvnode(mp)
2807 	struct mount *mp;
2808 {
2809 	struct vnode *vp;
2810 	static long start, incr, next;
2811 	int error;
2812 
2813 	/* Allocate a new vnode */
2814 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2815 		mp->mnt_syncer = NULL;
2816 		return (error);
2817 	}
2818 	vp->v_type = VNON;
2819 	/*
2820 	 * Place the vnode onto the syncer worklist. We attempt to
2821 	 * scatter them about on the list so that they will go off
2822 	 * at evenly distributed times even if all the filesystems
2823 	 * are mounted at once.
2824 	 */
2825 	next += incr;
2826 	if (next == 0 || next > syncer_maxdelay) {
2827 		start /= 2;
2828 		incr /= 2;
2829 		if (start == 0) {
2830 			start = syncer_maxdelay / 2;
2831 			incr = syncer_maxdelay;
2832 		}
2833 		next = start;
2834 	}
2835 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2836 	mp->mnt_syncer = vp;
2837 	return (0);
2838 }
2839 
2840 /*
2841  * Do a lazy sync of the filesystem.
2842  */
2843 static int
2844 sync_fsync(ap)
2845 	struct vop_fsync_args /* {
2846 		struct vnode *a_vp;
2847 		struct ucred *a_cred;
2848 		int a_waitfor;
2849 		struct proc *a_p;
2850 	} */ *ap;
2851 {
2852 	struct vnode *syncvp = ap->a_vp;
2853 	struct mount *mp = syncvp->v_mount;
2854 	struct proc *p = ap->a_p;
2855 	int asyncflag;
2856 
2857 	/*
2858 	 * We only need to do something if this is a lazy evaluation.
2859 	 */
2860 	if (ap->a_waitfor != MNT_LAZY)
2861 		return (0);
2862 
2863 	/*
2864 	 * Move ourselves to the back of the sync list.
2865 	 */
2866 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2867 
2868 	/*
2869 	 * Walk the list of vnodes pushing all that are dirty and
2870 	 * not already on the sync list.
2871 	 */
2872 	mtx_lock(&mountlist_mtx);
2873 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, p) != 0) {
2874 		mtx_unlock(&mountlist_mtx);
2875 		return (0);
2876 	}
2877 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2878 		vfs_unbusy(mp, p);
2879 		return (0);
2880 	}
2881 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2882 	mp->mnt_flag &= ~MNT_ASYNC;
2883 	vfs_msync(mp, MNT_NOWAIT);
2884 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2885 	if (asyncflag)
2886 		mp->mnt_flag |= MNT_ASYNC;
2887 	vn_finished_write(mp);
2888 	vfs_unbusy(mp, p);
2889 	return (0);
2890 }
2891 
2892 /*
2893  * The syncer vnode is no referenced.
2894  */
2895 static int
2896 sync_inactive(ap)
2897 	struct vop_inactive_args /* {
2898 		struct vnode *a_vp;
2899 		struct proc *a_p;
2900 	} */ *ap;
2901 {
2902 
2903 	vgone(ap->a_vp);
2904 	return (0);
2905 }
2906 
2907 /*
2908  * The syncer vnode is no longer needed and is being decommissioned.
2909  *
2910  * Modifications to the worklist must be protected at splbio().
2911  */
2912 static int
2913 sync_reclaim(ap)
2914 	struct vop_reclaim_args /* {
2915 		struct vnode *a_vp;
2916 	} */ *ap;
2917 {
2918 	struct vnode *vp = ap->a_vp;
2919 	int s;
2920 
2921 	s = splbio();
2922 	vp->v_mount->mnt_syncer = NULL;
2923 	if (vp->v_flag & VONWORKLST) {
2924 		LIST_REMOVE(vp, v_synclist);
2925 		vp->v_flag &= ~VONWORKLST;
2926 	}
2927 	splx(s);
2928 
2929 	return (0);
2930 }
2931 
2932 /*
2933  * Print out a syncer vnode.
2934  */
2935 static int
2936 sync_print(ap)
2937 	struct vop_print_args /* {
2938 		struct vnode *a_vp;
2939 	} */ *ap;
2940 {
2941 	struct vnode *vp = ap->a_vp;
2942 
2943 	printf("syncer vnode");
2944 	if (vp->v_vnlock != NULL)
2945 		lockmgr_printinfo(vp->v_vnlock);
2946 	printf("\n");
2947 	return (0);
2948 }
2949 
2950 /*
2951  * extract the dev_t from a VCHR
2952  */
2953 dev_t
2954 vn_todev(vp)
2955 	struct vnode *vp;
2956 {
2957 	if (vp->v_type != VCHR)
2958 		return (NODEV);
2959 	return (vp->v_rdev);
2960 }
2961 
2962 /*
2963  * Check if vnode represents a disk device
2964  */
2965 int
2966 vn_isdisk(vp, errp)
2967 	struct vnode *vp;
2968 	int *errp;
2969 {
2970 	struct cdevsw *cdevsw;
2971 
2972 	if (vp->v_type != VCHR) {
2973 		if (errp != NULL)
2974 			*errp = ENOTBLK;
2975 		return (0);
2976 	}
2977 	if (vp->v_rdev == NULL) {
2978 		if (errp != NULL)
2979 			*errp = ENXIO;
2980 		return (0);
2981 	}
2982 	cdevsw = devsw(vp->v_rdev);
2983 	if (cdevsw == NULL) {
2984 		if (errp != NULL)
2985 			*errp = ENXIO;
2986 		return (0);
2987 	}
2988 	if (!(cdevsw->d_flags & D_DISK)) {
2989 		if (errp != NULL)
2990 			*errp = ENOTBLK;
2991 		return (0);
2992 	}
2993 	if (errp != NULL)
2994 		*errp = 0;
2995 	return (1);
2996 }
2997 
2998 /*
2999  * Free data allocated by namei(); see namei(9) for details.
3000  */
3001 void
3002 NDFREE(ndp, flags)
3003      struct nameidata *ndp;
3004      const uint flags;
3005 {
3006 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3007 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3008 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3009 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3010 	}
3011 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3012 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3013 	    ndp->ni_dvp != ndp->ni_vp)
3014 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_proc);
3015 	if (!(flags & NDF_NO_DVP_RELE) &&
3016 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3017 		vrele(ndp->ni_dvp);
3018 		ndp->ni_dvp = NULL;
3019 	}
3020 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3021 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3022 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_proc);
3023 	if (!(flags & NDF_NO_VP_RELE) &&
3024 	    ndp->ni_vp) {
3025 		vrele(ndp->ni_vp);
3026 		ndp->ni_vp = NULL;
3027 	}
3028 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3029 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3030 		vrele(ndp->ni_startdir);
3031 		ndp->ni_startdir = NULL;
3032 	}
3033 }
3034 
3035 /*
3036  * Common file system object access control check routine.  Accepts a
3037  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3038  * and optional call-by-reference privused argument allowing vaccess()
3039  * to indicate to the caller whether privilege was used to satisfy the
3040  * request.  Returns 0 on success, or an errno on failure.
3041  */
3042 int
3043 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3044 	enum vtype type;
3045 	mode_t file_mode;
3046 	uid_t file_uid;
3047 	gid_t file_gid;
3048 	mode_t acc_mode;
3049 	struct ucred *cred;
3050 	int *privused;
3051 {
3052 	mode_t dac_granted;
3053 #ifdef CAPABILITIES
3054 	mode_t cap_granted;
3055 #endif
3056 
3057 	/*
3058 	 * Look for a normal, non-privileged way to access the file/directory
3059 	 * as requested.  If it exists, go with that.
3060 	 */
3061 
3062 	if (privused != NULL)
3063 		*privused = 0;
3064 
3065 	dac_granted = 0;
3066 
3067 	/* Check the owner. */
3068 	if (cred->cr_uid == file_uid) {
3069 		dac_granted |= VADMIN;
3070 		if (file_mode & S_IXUSR)
3071 			dac_granted |= VEXEC;
3072 		if (file_mode & S_IRUSR)
3073 			dac_granted |= VREAD;
3074 		if (file_mode & S_IWUSR)
3075 			dac_granted |= VWRITE;
3076 
3077 		if ((acc_mode & dac_granted) == acc_mode)
3078 			return (0);
3079 
3080 		goto privcheck;
3081 	}
3082 
3083 	/* Otherwise, check the groups (first match) */
3084 	if (groupmember(file_gid, cred)) {
3085 		if (file_mode & S_IXGRP)
3086 			dac_granted |= VEXEC;
3087 		if (file_mode & S_IRGRP)
3088 			dac_granted |= VREAD;
3089 		if (file_mode & S_IWGRP)
3090 			dac_granted |= VWRITE;
3091 
3092 		if ((acc_mode & dac_granted) == acc_mode)
3093 			return (0);
3094 
3095 		goto privcheck;
3096 	}
3097 
3098 	/* Otherwise, check everyone else. */
3099 	if (file_mode & S_IXOTH)
3100 		dac_granted |= VEXEC;
3101 	if (file_mode & S_IROTH)
3102 		dac_granted |= VREAD;
3103 	if (file_mode & S_IWOTH)
3104 		dac_granted |= VWRITE;
3105 	if ((acc_mode & dac_granted) == acc_mode)
3106 		return (0);
3107 
3108 privcheck:
3109 	if (!suser_xxx(cred, NULL, PRISON_ROOT)) {
3110 		/* XXX audit: privilege used */
3111 		if (privused != NULL)
3112 			*privused = 1;
3113 		return (0);
3114 	}
3115 
3116 #ifdef CAPABILITIES
3117 	/*
3118 	 * Build a capability mask to determine if the set of capabilities
3119 	 * satisfies the requirements when combined with the granted mask
3120 	 * from above.
3121 	 * For each capability, if the capability is required, bitwise
3122 	 * or the request type onto the cap_granted mask.
3123 	 */
3124 	cap_granted = 0;
3125 	if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3126 	    !cap_check_xxx(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3127 	    cap_granted |= VEXEC;
3128 
3129 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3130 	    !cap_check_xxx(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3131 		cap_granted |= VREAD;
3132 
3133 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3134 	    !cap_check_xxx(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3135 		cap_granted |= VWRITE;
3136 
3137 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3138 	    !cap_check_xxx(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3139 		cap_granted |= VADMIN;
3140 
3141 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3142 		/* XXX audit: privilege used */
3143 		if (privused != NULL)
3144 			*privused = 1;
3145 		return (0);
3146 	}
3147 #endif
3148 
3149 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3150 }
3151