xref: /dflybsd-src/sys/kern/vfs_subr.c (revision 9dbf638f09a3ee45768de45e3d72f7a2b87281b9)
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: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_subr.c,v 1.19 2003/09/01 00:35:29 hmp Exp $
41  */
42 
43 /*
44  * External virtual filesystem routines
45  */
46 #include "opt_ddb.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/mount.h>
61 #include <sys/proc.h>
62 #include <sys/namei.h>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
70 
71 #include <machine/limits.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vnode_pager.h>
81 #include <vm/vm_zone.h>
82 
83 #include <sys/buf2.h>
84 #include <sys/thread2.h>
85 
86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
87 
88 static void	insmntque (struct vnode *vp, struct mount *mp);
89 static void	vclean (struct vnode *vp, int flags, struct thread *td);
90 static unsigned long	numvnodes;
91 static void	vlruvp(struct vnode *vp);
92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
93 
94 enum vtype iftovt_tab[16] = {
95 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
96 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
97 };
98 int vttoif_tab[9] = {
99 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
100 	S_IFSOCK, S_IFIFO, S_IFMT,
101 };
102 
103 static TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
104 
105 static u_long wantfreevnodes = 25;
106 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
107 static u_long freevnodes = 0;
108 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
109 
110 static int reassignbufcalls;
111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
112 static int reassignbufloops;
113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
114 static int reassignbufsortgood;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
116 static int reassignbufsortbad;
117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
118 static int reassignbufmethod = 1;
119 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
120 static int nameileafonly = 0;
121 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
122 
123 #ifdef ENABLE_VFS_IOOPT
124 int vfs_ioopt = 0;
125 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
126 #endif
127 
128 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
129 struct lwkt_token mountlist_token;
130 struct lwkt_token mntvnode_token;
131 int	nfs_mount_type = -1;
132 static struct lwkt_token mntid_token;
133 static struct lwkt_token vnode_free_list_token;
134 static struct lwkt_token spechash_token;
135 struct nfs_public nfs_pub;	/* publicly exported FS */
136 static vm_zone_t vnode_zone;
137 
138 /*
139  * The workitem queue.
140  */
141 #define SYNCER_MAXDELAY		32
142 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
143 time_t syncdelay = 30;		/* max time to delay syncing data */
144 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0,
145 	"VFS data synchronization delay");
146 time_t filedelay = 30;		/* time to delay syncing files */
147 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
148 	"File synchronization delay");
149 time_t dirdelay = 29;		/* time to delay syncing directories */
150 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
151 	"Directory synchronization delay");
152 time_t metadelay = 28;		/* time to delay syncing metadata */
153 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
154 	"VFS metadata synchronization delay");
155 static int rushjob;			/* number of slots to run ASAP */
156 static int stat_rush_requests;	/* number of times I/O speeded up */
157 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
158 
159 static int syncer_delayno = 0;
160 static long syncer_mask;
161 LIST_HEAD(synclist, vnode);
162 static struct synclist *syncer_workitem_pending;
163 
164 int desiredvnodes;
165 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
166     &desiredvnodes, 0, "Maximum number of vnodes");
167 static int minvnodes;
168 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
169     &minvnodes, 0, "Minimum number of vnodes");
170 static int vnlru_nowhere = 0;
171 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
172     "Number of times the vnlru process ran without success");
173 
174 static void	vfs_free_addrlist (struct netexport *nep);
175 static int	vfs_free_netcred (struct radix_node *rn, void *w);
176 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
177 				       struct export_args *argp);
178 
179 /*
180  * Initialize the vnode management data structures.
181  */
182 void
183 vntblinit()
184 {
185 
186 	desiredvnodes = maxproc + vmstats.v_page_count / 4;
187 	minvnodes = desiredvnodes / 4;
188 	lwkt_inittoken(&mntvnode_token);
189 	lwkt_inittoken(&mntid_token);
190 	lwkt_inittoken(&spechash_token);
191 	TAILQ_INIT(&vnode_free_list);
192 	lwkt_inittoken(&vnode_free_list_token);
193 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
194 	/*
195 	 * Initialize the filesystem syncer.
196 	 */
197 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
198 		&syncer_mask);
199 	syncer_maxdelay = syncer_mask + 1;
200 }
201 
202 /*
203  * Mark a mount point as busy. Used to synchronize access and to delay
204  * unmounting. Interlock is not released on failure.
205  */
206 int
207 vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp,
208 	struct thread *td)
209 {
210 	int lkflags;
211 
212 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
213 		if (flags & LK_NOWAIT)
214 			return (ENOENT);
215 		mp->mnt_kern_flag |= MNTK_MWAIT;
216 		if (interlkp) {
217 			lwkt_reltoken(interlkp);
218 		}
219 		/*
220 		 * Since all busy locks are shared except the exclusive
221 		 * lock granted when unmounting, the only place that a
222 		 * wakeup needs to be done is at the release of the
223 		 * exclusive lock at the end of dounmount.
224 		 */
225 		tsleep((caddr_t)mp, 0, "vfs_busy", 0);
226 		if (interlkp) {
227 			lwkt_gettoken(interlkp);
228 		}
229 		return (ENOENT);
230 	}
231 	lkflags = LK_SHARED | LK_NOPAUSE;
232 	if (interlkp)
233 		lkflags |= LK_INTERLOCK;
234 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
235 		panic("vfs_busy: unexpected lock failure");
236 	return (0);
237 }
238 
239 /*
240  * Free a busy filesystem.
241  */
242 void
243 vfs_unbusy(struct mount *mp, struct thread *td)
244 {
245 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
246 }
247 
248 /*
249  * Lookup a filesystem type, and if found allocate and initialize
250  * a mount structure for it.
251  *
252  * Devname is usually updated by mount(8) after booting.
253  */
254 int
255 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
256 {
257 	struct thread *td = curthread;	/* XXX */
258 	struct vfsconf *vfsp;
259 	struct mount *mp;
260 
261 	if (fstypename == NULL)
262 		return (ENODEV);
263 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
264 		if (!strcmp(vfsp->vfc_name, fstypename))
265 			break;
266 	if (vfsp == NULL)
267 		return (ENODEV);
268 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
269 	bzero((char *)mp, (u_long)sizeof(struct mount));
270 	lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
271 	(void)vfs_busy(mp, LK_NOWAIT, 0, td);
272 	TAILQ_INIT(&mp->mnt_nvnodelist);
273 	TAILQ_INIT(&mp->mnt_reservedvnlist);
274 	mp->mnt_nvnodelistsize = 0;
275 	mp->mnt_vfc = vfsp;
276 	mp->mnt_op = vfsp->vfc_vfsops;
277 	mp->mnt_flag = MNT_RDONLY;
278 	mp->mnt_vnodecovered = NULLVP;
279 	vfsp->vfc_refcount++;
280 	mp->mnt_iosize_max = DFLTPHYS;
281 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
282 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
283 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
284 	mp->mnt_stat.f_mntonname[0] = '/';
285 	mp->mnt_stat.f_mntonname[1] = 0;
286 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
287 	*mpp = mp;
288 	return (0);
289 }
290 
291 /*
292  * Find an appropriate filesystem to use for the root. If a filesystem
293  * has not been preselected, walk through the list of known filesystems
294  * trying those that have mountroot routines, and try them until one
295  * works or we have tried them all.
296  */
297 #ifdef notdef	/* XXX JH */
298 int
299 lite2_vfs_mountroot()
300 {
301 	struct vfsconf *vfsp;
302 	extern int (*lite2_mountroot) (void);
303 	int error;
304 
305 	if (lite2_mountroot != NULL)
306 		return ((*lite2_mountroot)());
307 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
308 		if (vfsp->vfc_mountroot == NULL)
309 			continue;
310 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
311 			return (0);
312 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
313 	}
314 	return (ENODEV);
315 }
316 #endif
317 
318 /*
319  * Lookup a mount point by filesystem identifier.
320  */
321 struct mount *
322 vfs_getvfs(fsid)
323 	fsid_t *fsid;
324 {
325 	struct mount *mp;
326 
327 	lwkt_gettoken(&mountlist_token);
328 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
329 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
330 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
331 			lwkt_reltoken(&mountlist_token);
332 			return (mp);
333 	    }
334 	}
335 	lwkt_reltoken(&mountlist_token);
336 	return ((struct mount *) 0);
337 }
338 
339 /*
340  * Get a new unique fsid.  Try to make its val[0] unique, since this value
341  * will be used to create fake device numbers for stat().  Also try (but
342  * not so hard) make its val[0] unique mod 2^16, since some emulators only
343  * support 16-bit device numbers.  We end up with unique val[0]'s for the
344  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
345  *
346  * Keep in mind that several mounts may be running in parallel.  Starting
347  * the search one past where the previous search terminated is both a
348  * micro-optimization and a defense against returning the same fsid to
349  * different mounts.
350  */
351 void
352 vfs_getnewfsid(mp)
353 	struct mount *mp;
354 {
355 	static u_int16_t mntid_base;
356 	fsid_t tfsid;
357 	int mtype;
358 
359 	lwkt_gettoken(&mntid_token);
360 	mtype = mp->mnt_vfc->vfc_typenum;
361 	tfsid.val[1] = mtype;
362 	mtype = (mtype & 0xFF) << 24;
363 	for (;;) {
364 		tfsid.val[0] = makeudev(255,
365 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
366 		mntid_base++;
367 		if (vfs_getvfs(&tfsid) == NULL)
368 			break;
369 	}
370 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
371 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
372 	lwkt_reltoken(&mntid_token);
373 }
374 
375 /*
376  * Knob to control the precision of file timestamps:
377  *
378  *   0 = seconds only; nanoseconds zeroed.
379  *   1 = seconds and nanoseconds, accurate within 1/HZ.
380  *   2 = seconds and nanoseconds, truncated to microseconds.
381  * >=3 = seconds and nanoseconds, maximum precision.
382  */
383 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
384 
385 static int timestamp_precision = TSP_SEC;
386 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
387     &timestamp_precision, 0, "");
388 
389 /*
390  * Get a current timestamp.
391  */
392 void
393 vfs_timestamp(tsp)
394 	struct timespec *tsp;
395 {
396 	struct timeval tv;
397 
398 	switch (timestamp_precision) {
399 	case TSP_SEC:
400 		tsp->tv_sec = time_second;
401 		tsp->tv_nsec = 0;
402 		break;
403 	case TSP_HZ:
404 		getnanotime(tsp);
405 		break;
406 	case TSP_USEC:
407 		microtime(&tv);
408 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
409 		break;
410 	case TSP_NSEC:
411 	default:
412 		nanotime(tsp);
413 		break;
414 	}
415 }
416 
417 /*
418  * Set vnode attributes to VNOVAL
419  */
420 void
421 vattr_null(vap)
422 	struct vattr *vap;
423 {
424 
425 	vap->va_type = VNON;
426 	vap->va_size = VNOVAL;
427 	vap->va_bytes = VNOVAL;
428 	vap->va_mode = VNOVAL;
429 	vap->va_nlink = VNOVAL;
430 	vap->va_uid = VNOVAL;
431 	vap->va_gid = VNOVAL;
432 	vap->va_fsid = VNOVAL;
433 	vap->va_fileid = VNOVAL;
434 	vap->va_blocksize = VNOVAL;
435 	vap->va_rdev = VNOVAL;
436 	vap->va_atime.tv_sec = VNOVAL;
437 	vap->va_atime.tv_nsec = VNOVAL;
438 	vap->va_mtime.tv_sec = VNOVAL;
439 	vap->va_mtime.tv_nsec = VNOVAL;
440 	vap->va_ctime.tv_sec = VNOVAL;
441 	vap->va_ctime.tv_nsec = VNOVAL;
442 	vap->va_flags = VNOVAL;
443 	vap->va_gen = VNOVAL;
444 	vap->va_vaflags = 0;
445 }
446 
447 /*
448  * This routine is called when we have too many vnodes.  It attempts
449  * to free <count> vnodes and will potentially free vnodes that still
450  * have VM backing store (VM backing store is typically the cause
451  * of a vnode blowout so we want to do this).  Therefore, this operation
452  * is not considered cheap.
453  *
454  * A number of conditions may prevent a vnode from being reclaimed.
455  * the buffer cache may have references on the vnode, a directory
456  * vnode may still have references due to the namei cache representing
457  * underlying files, or the vnode may be in active use.   It is not
458  * desireable to reuse such vnodes.  These conditions may cause the
459  * number of vnodes to reach some minimum value regardless of what
460  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
461  */
462 static int
463 vlrureclaim(struct mount *mp)
464 {
465 	struct vnode *vp;
466 	int done;
467 	int trigger;
468 	int usevnodes;
469 	int count;
470 	int gen;
471 
472 	/*
473 	 * Calculate the trigger point, don't allow user
474 	 * screwups to blow us up.   This prevents us from
475 	 * recycling vnodes with lots of resident pages.  We
476 	 * aren't trying to free memory, we are trying to
477 	 * free vnodes.
478 	 */
479 	usevnodes = desiredvnodes;
480 	if (usevnodes <= 0)
481 		usevnodes = 1;
482 	trigger = vmstats.v_page_count * 2 / usevnodes;
483 
484 	done = 0;
485 	gen = lwkt_gettoken(&mntvnode_token);
486 	count = mp->mnt_nvnodelistsize / 10 + 1;
487 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
488 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
489 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
490 
491 		if (vp->v_type != VNON &&
492 		    vp->v_type != VBAD &&
493 		    VMIGHTFREE(vp) &&		/* critical path opt */
494 		    (vp->v_object == NULL || vp->v_object->resident_page_count < trigger)
495 		) {
496 			lwkt_gettoken(&vp->v_interlock);
497 			if (lwkt_gentoken(&mntvnode_token, &gen) == 0) {
498 				if (VMIGHTFREE(vp)) {
499 					vgonel(vp, curthread);
500 					done++;
501 				} else {
502 					lwkt_reltoken(&vp->v_interlock);
503 				}
504 			} else {
505 				lwkt_reltoken(&vp->v_interlock);
506 			}
507 		}
508 		--count;
509 	}
510 	lwkt_reltoken(&mntvnode_token);
511 	return done;
512 }
513 
514 /*
515  * Attempt to recycle vnodes in a context that is always safe to block.
516  * Calling vlrurecycle() from the bowels of file system code has some
517  * interesting deadlock problems.
518  */
519 static struct thread *vnlruthread;
520 static int vnlruproc_sig;
521 
522 static void
523 vnlru_proc(void)
524 {
525 	struct mount *mp, *nmp;
526 	int s;
527 	int done;
528 	struct thread *td = curthread;
529 
530 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
531 	    SHUTDOWN_PRI_FIRST);
532 
533 	s = splbio();
534 	for (;;) {
535 		kproc_suspend_loop();
536 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
537 			vnlruproc_sig = 0;
538 			wakeup(&vnlruproc_sig);
539 			tsleep(td, 0, "vlruwt", hz);
540 			continue;
541 		}
542 		done = 0;
543 		lwkt_gettoken(&mountlist_token);
544 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
545 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
546 				nmp = TAILQ_NEXT(mp, mnt_list);
547 				continue;
548 			}
549 			done += vlrureclaim(mp);
550 			lwkt_gettoken(&mountlist_token);
551 			nmp = TAILQ_NEXT(mp, mnt_list);
552 			vfs_unbusy(mp, td);
553 		}
554 		lwkt_reltoken(&mountlist_token);
555 		if (done == 0) {
556 			vnlru_nowhere++;
557 			tsleep(td, 0, "vlrup", hz * 3);
558 		}
559 	}
560 	splx(s);
561 }
562 
563 static struct kproc_desc vnlru_kp = {
564 	"vnlru",
565 	vnlru_proc,
566 	&vnlruthread
567 };
568 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
569 
570 /*
571  * Routines having to do with the management of the vnode table.
572  */
573 extern vop_t **dead_vnodeop_p;
574 
575 /*
576  * Return the next vnode from the free list.
577  */
578 int
579 getnewvnode(tag, mp, vops, vpp)
580 	enum vtagtype tag;
581 	struct mount *mp;
582 	vop_t **vops;
583 	struct vnode **vpp;
584 {
585 	int s;
586 	int gen;
587 	int vgen;
588 	struct thread *td = curthread;	/* XXX */
589 	struct vnode *vp = NULL;
590 	vm_object_t object;
591 
592 	s = splbio();
593 
594 	/*
595 	 * Try to reuse vnodes if we hit the max.  This situation only
596 	 * occurs in certain large-memory (2G+) situations.  We cannot
597 	 * attempt to directly reclaim vnodes due to nasty recursion
598 	 * problems.
599 	 */
600 	while (numvnodes - freevnodes > desiredvnodes) {
601 		if (vnlruproc_sig == 0) {
602 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
603 			wakeup(vnlruthread);
604 		}
605 		tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
606 	}
607 
608 
609 	/*
610 	 * Attempt to reuse a vnode already on the free list, allocating
611 	 * a new vnode if we can't find one or if we have not reached a
612 	 * good minimum for good LRU performance.
613 	 */
614 	gen = lwkt_gettoken(&vnode_free_list_token);
615 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
616 		int count;
617 
618 		for (count = 0; count < freevnodes; count++) {
619 			vp = TAILQ_FIRST(&vnode_free_list);
620 			if (vp == NULL || vp->v_usecount)
621 				panic("getnewvnode: free vnode isn't");
622 
623 			/*
624 			 * Get the vnode's interlock, then re-obtain
625 			 * vnode_free_list_token in case we lost it.  If we
626 			 * did lose it while getting the vnode interlock,
627 			 * even if we got it back again, then retry.
628 			 */
629 			vgen = lwkt_gettoken(&vp->v_interlock);
630 			if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
631 				--count;
632 				lwkt_reltoken(&vp->v_interlock);
633 				vp = NULL;
634 				continue;
635 			}
636 
637 			/*
638 			 * Whew!  We have both tokens.  Since we didn't lose
639 			 * the free list VFREE had better still be set.  But
640 			 * we aren't out of the woods yet.  We have to get
641 			 * the object (may block).  If the vnode is not
642 			 * suitable then move it to the end of the list
643 			 * if we can.  If we can't move it to the end of the
644 			 * list retry again.
645 			 */
646 			if ((VOP_GETVOBJECT(vp, &object) == 0 &&
647 			    (object->resident_page_count || object->ref_count))
648 			) {
649 				if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 &&
650 				   lwkt_gentoken(&vnode_free_list_token, &gen) == 0
651 				) {
652 					TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
653 					TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
654 				} else {
655 					--count;
656 				}
657 				lwkt_reltoken(&vp->v_interlock);
658 				vp = NULL;
659 				continue;
660 			}
661 
662 			/*
663 			 * Still not out of the woods.  VOBJECT might have
664 			 * blocked, if we did not retain our tokens we have
665 			 * to retry.
666 			 */
667 			if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 ||
668 			    lwkt_gentoken(&vnode_free_list_token, &gen) != 0) {
669 				--count;
670 				vp = NULL;
671 				continue;
672 			}
673 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
674 			KKASSERT(vp->v_flag & VFREE);
675 
676 			if (LIST_FIRST(&vp->v_cache_src)) {
677 				/*
678 				 * note: nameileafonly sysctl is temporary,
679 				 * for debugging only, and will eventually be
680 				 * removed.
681 				 */
682 				if (nameileafonly > 0) {
683 					/*
684 					 * Do not reuse namei-cached directory
685 					 * vnodes that have cached
686 					 * subdirectories.
687 					 */
688 					if (cache_leaf_test(vp) < 0) {
689 						lwkt_reltoken(&vp->v_interlock);
690 						TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
691 						vp = NULL;
692 						continue;
693 					}
694 				} else if (nameileafonly < 0 ||
695 					    vmiodirenable == 0) {
696 					/*
697 					 * Do not reuse namei-cached directory
698 					 * vnodes if nameileafonly is -1 or
699 					 * if VMIO backing for directories is
700 					 * turned off (otherwise we reuse them
701 					 * too quickly).
702 					 */
703 					lwkt_reltoken(&vp->v_interlock);
704 					TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
705 					vp = NULL;
706 					continue;
707 				}
708 			}
709 			break;
710 		}
711 	}
712 
713 	if (vp) {
714 		vp->v_flag |= VDOOMED;
715 		vp->v_flag &= ~VFREE;
716 		freevnodes--;
717 		lwkt_reltoken(&vnode_free_list_token);
718 		cache_purge(vp);	/* YYY may block */
719 		vp->v_lease = NULL;
720 		if (vp->v_type != VBAD) {
721 			vgonel(vp, td);
722 		} else {
723 			lwkt_reltoken(&vp->v_interlock);
724 		}
725 
726 #ifdef INVARIANTS
727 		{
728 			int s;
729 
730 			if (vp->v_data)
731 				panic("cleaned vnode isn't");
732 			s = splbio();
733 			if (vp->v_numoutput)
734 				panic("Clean vnode has pending I/O's");
735 			splx(s);
736 		}
737 #endif
738 		vp->v_flag = 0;
739 		vp->v_lastw = 0;
740 		vp->v_lasta = 0;
741 		vp->v_cstart = 0;
742 		vp->v_clen = 0;
743 		vp->v_socket = 0;
744 		vp->v_writecount = 0;	/* XXX */
745 	} else {
746 		lwkt_reltoken(&vnode_free_list_token);
747 		vp = (struct vnode *) zalloc(vnode_zone);
748 		bzero((char *) vp, sizeof *vp);
749 		lwkt_inittoken(&vp->v_interlock);
750 		vp->v_dd = vp;
751 		cache_purge(vp);
752 		LIST_INIT(&vp->v_cache_src);
753 		TAILQ_INIT(&vp->v_cache_dst);
754 		numvnodes++;
755 	}
756 
757 	TAILQ_INIT(&vp->v_cleanblkhd);
758 	TAILQ_INIT(&vp->v_dirtyblkhd);
759 	vp->v_type = VNON;
760 	vp->v_tag = tag;
761 	vp->v_op = vops;
762 	insmntque(vp, mp);
763 	*vpp = vp;
764 	vp->v_usecount = 1;
765 	vp->v_data = 0;
766 	splx(s);
767 
768 	vfs_object_create(vp, td);
769 	return (0);
770 }
771 
772 /*
773  * Move a vnode from one mount queue to another.
774  */
775 static void
776 insmntque(vp, mp)
777 	struct vnode *vp;
778 	struct mount *mp;
779 {
780 
781 	lwkt_gettoken(&mntvnode_token);
782 	/*
783 	 * Delete from old mount point vnode list, if on one.
784 	 */
785 	if (vp->v_mount != NULL) {
786 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
787 			("bad mount point vnode list size"));
788 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
789 		vp->v_mount->mnt_nvnodelistsize--;
790 	}
791 	/*
792 	 * Insert into list of vnodes for the new mount point, if available.
793 	 */
794 	if ((vp->v_mount = mp) == NULL) {
795 		lwkt_reltoken(&mntvnode_token);
796 		return;
797 	}
798 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
799 	mp->mnt_nvnodelistsize++;
800 	lwkt_reltoken(&mntvnode_token);
801 }
802 
803 /*
804  * Update outstanding I/O count and do wakeup if requested.
805  */
806 void
807 vwakeup(bp)
808 	struct buf *bp;
809 {
810 	struct vnode *vp;
811 
812 	bp->b_flags &= ~B_WRITEINPROG;
813 	if ((vp = bp->b_vp)) {
814 		vp->v_numoutput--;
815 		if (vp->v_numoutput < 0)
816 			panic("vwakeup: neg numoutput");
817 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
818 			vp->v_flag &= ~VBWAIT;
819 			wakeup((caddr_t) &vp->v_numoutput);
820 		}
821 	}
822 }
823 
824 /*
825  * Flush out and invalidate all buffers associated with a vnode.
826  * Called with the underlying object locked.
827  */
828 int
829 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
830 	int slpflag, int slptimeo)
831 {
832 	struct buf *bp;
833 	struct buf *nbp, *blist;
834 	int s, error;
835 	vm_object_t object;
836 
837 	if (flags & V_SAVE) {
838 		s = splbio();
839 		while (vp->v_numoutput) {
840 			vp->v_flag |= VBWAIT;
841 			error = tsleep((caddr_t)&vp->v_numoutput,
842 			    slpflag, "vinvlbuf", slptimeo);
843 			if (error) {
844 				splx(s);
845 				return (error);
846 			}
847 		}
848 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
849 			splx(s);
850 			if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
851 				return (error);
852 			s = splbio();
853 			if (vp->v_numoutput > 0 ||
854 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
855 				panic("vinvalbuf: dirty bufs");
856 		}
857 		splx(s);
858   	}
859 	s = splbio();
860 	for (;;) {
861 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
862 		if (!blist)
863 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
864 		if (!blist)
865 			break;
866 
867 		for (bp = blist; bp; bp = nbp) {
868 			nbp = TAILQ_NEXT(bp, b_vnbufs);
869 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
870 				error = BUF_TIMELOCK(bp,
871 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
872 				    "vinvalbuf", slpflag, slptimeo);
873 				if (error == ENOLCK)
874 					break;
875 				splx(s);
876 				return (error);
877 			}
878 			/*
879 			 * XXX Since there are no node locks for NFS, I
880 			 * believe there is a slight chance that a delayed
881 			 * write will occur while sleeping just above, so
882 			 * check for it.  Note that vfs_bio_awrite expects
883 			 * buffers to reside on a queue, while VOP_BWRITE and
884 			 * brelse do not.
885 			 */
886 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
887 				(flags & V_SAVE)) {
888 
889 				if (bp->b_vp == vp) {
890 					if (bp->b_flags & B_CLUSTEROK) {
891 						BUF_UNLOCK(bp);
892 						vfs_bio_awrite(bp);
893 					} else {
894 						bremfree(bp);
895 						bp->b_flags |= B_ASYNC;
896 						VOP_BWRITE(bp->b_vp, bp);
897 					}
898 				} else {
899 					bremfree(bp);
900 					(void) VOP_BWRITE(bp->b_vp, bp);
901 				}
902 				break;
903 			}
904 			bremfree(bp);
905 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
906 			bp->b_flags &= ~B_ASYNC;
907 			brelse(bp);
908 		}
909 	}
910 
911 	/*
912 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
913 	 * have write I/O in-progress but if there is a VM object then the
914 	 * VM object can also have read-I/O in-progress.
915 	 */
916 	do {
917 		while (vp->v_numoutput > 0) {
918 			vp->v_flag |= VBWAIT;
919 			tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
920 		}
921 		if (VOP_GETVOBJECT(vp, &object) == 0) {
922 			while (object->paging_in_progress)
923 				vm_object_pip_sleep(object, "vnvlbx");
924 		}
925 	} while (vp->v_numoutput > 0);
926 
927 	splx(s);
928 
929 	/*
930 	 * Destroy the copy in the VM cache, too.
931 	 */
932 	lwkt_gettoken(&vp->v_interlock);
933 	if (VOP_GETVOBJECT(vp, &object) == 0) {
934 		vm_object_page_remove(object, 0, 0,
935 			(flags & V_SAVE) ? TRUE : FALSE);
936 	}
937 	lwkt_reltoken(&vp->v_interlock);
938 
939 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
940 		panic("vinvalbuf: flush failed");
941 	return (0);
942 }
943 
944 /*
945  * Truncate a file's buffer and pages to a specified length.  This
946  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
947  * sync activity.
948  */
949 int
950 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
951 {
952 	struct buf *bp;
953 	struct buf *nbp;
954 	int s, anyfreed;
955 	int trunclbn;
956 
957 	/*
958 	 * Round up to the *next* lbn.
959 	 */
960 	trunclbn = (length + blksize - 1) / blksize;
961 
962 	s = splbio();
963 restart:
964 	anyfreed = 1;
965 	for (;anyfreed;) {
966 		anyfreed = 0;
967 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
968 			nbp = TAILQ_NEXT(bp, b_vnbufs);
969 			if (bp->b_lblkno >= trunclbn) {
970 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
971 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
972 					goto restart;
973 				} else {
974 					bremfree(bp);
975 					bp->b_flags |= (B_INVAL | B_RELBUF);
976 					bp->b_flags &= ~B_ASYNC;
977 					brelse(bp);
978 					anyfreed = 1;
979 				}
980 				if (nbp &&
981 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
982 				    (nbp->b_vp != vp) ||
983 				    (nbp->b_flags & B_DELWRI))) {
984 					goto restart;
985 				}
986 			}
987 		}
988 
989 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
990 			nbp = TAILQ_NEXT(bp, b_vnbufs);
991 			if (bp->b_lblkno >= trunclbn) {
992 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
993 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
994 					goto restart;
995 				} else {
996 					bremfree(bp);
997 					bp->b_flags |= (B_INVAL | B_RELBUF);
998 					bp->b_flags &= ~B_ASYNC;
999 					brelse(bp);
1000 					anyfreed = 1;
1001 				}
1002 				if (nbp &&
1003 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1004 				    (nbp->b_vp != vp) ||
1005 				    (nbp->b_flags & B_DELWRI) == 0)) {
1006 					goto restart;
1007 				}
1008 			}
1009 		}
1010 	}
1011 
1012 	if (length > 0) {
1013 restartsync:
1014 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1015 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1016 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1017 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1018 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1019 					goto restart;
1020 				} else {
1021 					bremfree(bp);
1022 					if (bp->b_vp == vp) {
1023 						bp->b_flags |= B_ASYNC;
1024 					} else {
1025 						bp->b_flags &= ~B_ASYNC;
1026 					}
1027 					VOP_BWRITE(bp->b_vp, bp);
1028 				}
1029 				goto restartsync;
1030 			}
1031 
1032 		}
1033 	}
1034 
1035 	while (vp->v_numoutput > 0) {
1036 		vp->v_flag |= VBWAIT;
1037 		tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1038 	}
1039 
1040 	splx(s);
1041 
1042 	vnode_pager_setsize(vp, length);
1043 
1044 	return (0);
1045 }
1046 
1047 /*
1048  * Associate a buffer with a vnode.
1049  */
1050 void
1051 bgetvp(vp, bp)
1052 	struct vnode *vp;
1053 	struct buf *bp;
1054 {
1055 	int s;
1056 
1057 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1058 
1059 	vhold(vp);
1060 	bp->b_vp = vp;
1061 	bp->b_dev = vn_todev(vp);
1062 	/*
1063 	 * Insert onto list for new vnode.
1064 	 */
1065 	s = splbio();
1066 	bp->b_xflags |= BX_VNCLEAN;
1067 	bp->b_xflags &= ~BX_VNDIRTY;
1068 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1069 	splx(s);
1070 }
1071 
1072 /*
1073  * Disassociate a buffer from a vnode.
1074  */
1075 void
1076 brelvp(bp)
1077 	struct buf *bp;
1078 {
1079 	struct vnode *vp;
1080 	struct buflists *listheadp;
1081 	int s;
1082 
1083 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1084 
1085 	/*
1086 	 * Delete from old vnode list, if on one.
1087 	 */
1088 	vp = bp->b_vp;
1089 	s = splbio();
1090 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1091 		if (bp->b_xflags & BX_VNDIRTY)
1092 			listheadp = &vp->v_dirtyblkhd;
1093 		else
1094 			listheadp = &vp->v_cleanblkhd;
1095 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1096 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1097 	}
1098 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1099 		vp->v_flag &= ~VONWORKLST;
1100 		LIST_REMOVE(vp, v_synclist);
1101 	}
1102 	splx(s);
1103 	bp->b_vp = (struct vnode *) 0;
1104 	vdrop(vp);
1105 }
1106 
1107 /*
1108  * The workitem queue.
1109  *
1110  * It is useful to delay writes of file data and filesystem metadata
1111  * for tens of seconds so that quickly created and deleted files need
1112  * not waste disk bandwidth being created and removed. To realize this,
1113  * we append vnodes to a "workitem" queue. When running with a soft
1114  * updates implementation, most pending metadata dependencies should
1115  * not wait for more than a few seconds. Thus, mounted on block devices
1116  * are delayed only about a half the time that file data is delayed.
1117  * Similarly, directory updates are more critical, so are only delayed
1118  * about a third the time that file data is delayed. Thus, there are
1119  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1120  * one each second (driven off the filesystem syncer process). The
1121  * syncer_delayno variable indicates the next queue that is to be processed.
1122  * Items that need to be processed soon are placed in this queue:
1123  *
1124  *	syncer_workitem_pending[syncer_delayno]
1125  *
1126  * A delay of fifteen seconds is done by placing the request fifteen
1127  * entries later in the queue:
1128  *
1129  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1130  *
1131  */
1132 
1133 /*
1134  * Add an item to the syncer work queue.
1135  */
1136 static void
1137 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1138 {
1139 	int s, slot;
1140 
1141 	s = splbio();
1142 
1143 	if (vp->v_flag & VONWORKLST) {
1144 		LIST_REMOVE(vp, v_synclist);
1145 	}
1146 
1147 	if (delay > syncer_maxdelay - 2)
1148 		delay = syncer_maxdelay - 2;
1149 	slot = (syncer_delayno + delay) & syncer_mask;
1150 
1151 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1152 	vp->v_flag |= VONWORKLST;
1153 	splx(s);
1154 }
1155 
1156 struct  thread *updatethread;
1157 static void sched_sync (void);
1158 static struct kproc_desc up_kp = {
1159 	"syncer",
1160 	sched_sync,
1161 	&updatethread
1162 };
1163 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1164 
1165 /*
1166  * System filesystem synchronizer daemon.
1167  */
1168 void
1169 sched_sync(void)
1170 {
1171 	struct synclist *slp;
1172 	struct vnode *vp;
1173 	long starttime;
1174 	int s;
1175 	struct thread *td = curthread;
1176 
1177 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1178 	    SHUTDOWN_PRI_LAST);
1179 
1180 	for (;;) {
1181 		kproc_suspend_loop();
1182 
1183 		starttime = time_second;
1184 
1185 		/*
1186 		 * Push files whose dirty time has expired.  Be careful
1187 		 * of interrupt race on slp queue.
1188 		 */
1189 		s = splbio();
1190 		slp = &syncer_workitem_pending[syncer_delayno];
1191 		syncer_delayno += 1;
1192 		if (syncer_delayno == syncer_maxdelay)
1193 			syncer_delayno = 0;
1194 		splx(s);
1195 
1196 		while ((vp = LIST_FIRST(slp)) != NULL) {
1197 			if (VOP_ISLOCKED(vp, NULL) == 0) {
1198 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1199 				(void) VOP_FSYNC(vp, MNT_LAZY, td);
1200 				VOP_UNLOCK(vp, 0, td);
1201 			}
1202 			s = splbio();
1203 			if (LIST_FIRST(slp) == vp) {
1204 				/*
1205 				 * Note: v_tag VT_VFS vps can remain on the
1206 				 * worklist too with no dirty blocks, but
1207 				 * since sync_fsync() moves it to a different
1208 				 * slot we are safe.
1209 				 */
1210 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1211 				    !vn_isdisk(vp, NULL))
1212 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1213 				/*
1214 				 * Put us back on the worklist.  The worklist
1215 				 * routine will remove us from our current
1216 				 * position and then add us back in at a later
1217 				 * position.
1218 				 */
1219 				vn_syncer_add_to_worklist(vp, syncdelay);
1220 			}
1221 			splx(s);
1222 		}
1223 
1224 		/*
1225 		 * Do soft update processing.
1226 		 */
1227 		if (bioops.io_sync)
1228 			(*bioops.io_sync)(NULL);
1229 
1230 		/*
1231 		 * The variable rushjob allows the kernel to speed up the
1232 		 * processing of the filesystem syncer process. A rushjob
1233 		 * value of N tells the filesystem syncer to process the next
1234 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1235 		 * is used by the soft update code to speed up the filesystem
1236 		 * syncer process when the incore state is getting so far
1237 		 * ahead of the disk that the kernel memory pool is being
1238 		 * threatened with exhaustion.
1239 		 */
1240 		if (rushjob > 0) {
1241 			rushjob -= 1;
1242 			continue;
1243 		}
1244 		/*
1245 		 * If it has taken us less than a second to process the
1246 		 * current work, then wait. Otherwise start right over
1247 		 * again. We can still lose time if any single round
1248 		 * takes more than two seconds, but it does not really
1249 		 * matter as we are just trying to generally pace the
1250 		 * filesystem activity.
1251 		 */
1252 		if (time_second == starttime)
1253 			tsleep(&lbolt, 0, "syncer", 0);
1254 	}
1255 }
1256 
1257 /*
1258  * Request the syncer daemon to speed up its work.
1259  * We never push it to speed up more than half of its
1260  * normal turn time, otherwise it could take over the cpu.
1261  *
1262  * YYY wchan field protected by the BGL.
1263  */
1264 int
1265 speedup_syncer()
1266 {
1267 	crit_enter();
1268 	if (updatethread->td_wchan == &lbolt) { /* YYY */
1269 		unsleep(updatethread);
1270 		lwkt_schedule(updatethread);
1271 	}
1272 	crit_exit();
1273 	if (rushjob < syncdelay / 2) {
1274 		rushjob += 1;
1275 		stat_rush_requests += 1;
1276 		return (1);
1277 	}
1278 	return(0);
1279 }
1280 
1281 /*
1282  * Associate a p-buffer with a vnode.
1283  *
1284  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1285  * with the buffer.  i.e. the bp has not been linked into the vnode or
1286  * ref-counted.
1287  */
1288 void
1289 pbgetvp(vp, bp)
1290 	struct vnode *vp;
1291 	struct buf *bp;
1292 {
1293 
1294 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1295 
1296 	bp->b_vp = vp;
1297 	bp->b_flags |= B_PAGING;
1298 	bp->b_dev = vn_todev(vp);
1299 }
1300 
1301 /*
1302  * Disassociate a p-buffer from a vnode.
1303  */
1304 void
1305 pbrelvp(bp)
1306 	struct buf *bp;
1307 {
1308 
1309 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1310 
1311 	/* XXX REMOVE ME */
1312 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1313 		panic(
1314 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1315 		    bp,
1316 		    (int)bp->b_flags
1317 		);
1318 	}
1319 	bp->b_vp = (struct vnode *) 0;
1320 	bp->b_flags &= ~B_PAGING;
1321 }
1322 
1323 void
1324 pbreassignbuf(bp, newvp)
1325 	struct buf *bp;
1326 	struct vnode *newvp;
1327 {
1328 	if ((bp->b_flags & B_PAGING) == 0) {
1329 		panic(
1330 		    "pbreassignbuf() on non phys bp %p",
1331 		    bp
1332 		);
1333 	}
1334 	bp->b_vp = newvp;
1335 }
1336 
1337 /*
1338  * Reassign a buffer from one vnode to another.
1339  * Used to assign file specific control information
1340  * (indirect blocks) to the vnode to which they belong.
1341  */
1342 void
1343 reassignbuf(bp, newvp)
1344 	struct buf *bp;
1345 	struct vnode *newvp;
1346 {
1347 	struct buflists *listheadp;
1348 	int delay;
1349 	int s;
1350 
1351 	if (newvp == NULL) {
1352 		printf("reassignbuf: NULL");
1353 		return;
1354 	}
1355 	++reassignbufcalls;
1356 
1357 	/*
1358 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1359 	 * is not fully linked in.
1360 	 */
1361 	if (bp->b_flags & B_PAGING)
1362 		panic("cannot reassign paging buffer");
1363 
1364 	s = splbio();
1365 	/*
1366 	 * Delete from old vnode list, if on one.
1367 	 */
1368 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1369 		if (bp->b_xflags & BX_VNDIRTY)
1370 			listheadp = &bp->b_vp->v_dirtyblkhd;
1371 		else
1372 			listheadp = &bp->b_vp->v_cleanblkhd;
1373 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1374 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1375 		if (bp->b_vp != newvp) {
1376 			vdrop(bp->b_vp);
1377 			bp->b_vp = NULL;	/* for clarification */
1378 		}
1379 	}
1380 	/*
1381 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1382 	 * of clean buffers.
1383 	 */
1384 	if (bp->b_flags & B_DELWRI) {
1385 		struct buf *tbp;
1386 
1387 		listheadp = &newvp->v_dirtyblkhd;
1388 		if ((newvp->v_flag & VONWORKLST) == 0) {
1389 			switch (newvp->v_type) {
1390 			case VDIR:
1391 				delay = dirdelay;
1392 				break;
1393 			case VCHR:
1394 			case VBLK:
1395 				if (newvp->v_specmountpoint != NULL) {
1396 					delay = metadelay;
1397 					break;
1398 				}
1399 				/* fall through */
1400 			default:
1401 				delay = filedelay;
1402 			}
1403 			vn_syncer_add_to_worklist(newvp, delay);
1404 		}
1405 		bp->b_xflags |= BX_VNDIRTY;
1406 		tbp = TAILQ_FIRST(listheadp);
1407 		if (tbp == NULL ||
1408 		    bp->b_lblkno == 0 ||
1409 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1410 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1411 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1412 			++reassignbufsortgood;
1413 		} else if (bp->b_lblkno < 0) {
1414 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1415 			++reassignbufsortgood;
1416 		} else if (reassignbufmethod == 1) {
1417 			/*
1418 			 * New sorting algorithm, only handle sequential case,
1419 			 * otherwise append to end (but before metadata)
1420 			 */
1421 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1422 			    (tbp->b_xflags & BX_VNDIRTY)) {
1423 				/*
1424 				 * Found the best place to insert the buffer
1425 				 */
1426 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1427 				++reassignbufsortgood;
1428 			} else {
1429 				/*
1430 				 * Missed, append to end, but before meta-data.
1431 				 * We know that the head buffer in the list is
1432 				 * not meta-data due to prior conditionals.
1433 				 *
1434 				 * Indirect effects:  NFS second stage write
1435 				 * tends to wind up here, giving maximum
1436 				 * distance between the unstable write and the
1437 				 * commit rpc.
1438 				 */
1439 				tbp = TAILQ_LAST(listheadp, buflists);
1440 				while (tbp && tbp->b_lblkno < 0)
1441 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1442 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1443 				++reassignbufsortbad;
1444 			}
1445 		} else {
1446 			/*
1447 			 * Old sorting algorithm, scan queue and insert
1448 			 */
1449 			struct buf *ttbp;
1450 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1451 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1452 				++reassignbufloops;
1453 				tbp = ttbp;
1454 			}
1455 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1456 		}
1457 	} else {
1458 		bp->b_xflags |= BX_VNCLEAN;
1459 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1460 		if ((newvp->v_flag & VONWORKLST) &&
1461 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1462 			newvp->v_flag &= ~VONWORKLST;
1463 			LIST_REMOVE(newvp, v_synclist);
1464 		}
1465 	}
1466 	if (bp->b_vp != newvp) {
1467 		bp->b_vp = newvp;
1468 		vhold(bp->b_vp);
1469 	}
1470 	splx(s);
1471 }
1472 
1473 /*
1474  * Create a vnode for a block device.
1475  * Used for mounting the root file system.
1476  */
1477 int
1478 bdevvp(dev, vpp)
1479 	dev_t dev;
1480 	struct vnode **vpp;
1481 {
1482 	struct vnode *vp;
1483 	struct vnode *nvp;
1484 	int error;
1485 
1486 	if (dev == NODEV) {
1487 		*vpp = NULLVP;
1488 		return (ENXIO);
1489 	}
1490 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1491 	if (error) {
1492 		*vpp = NULLVP;
1493 		return (error);
1494 	}
1495 	vp = nvp;
1496 	vp->v_type = VBLK;
1497 	addalias(vp, dev);
1498 	*vpp = vp;
1499 	return (0);
1500 }
1501 
1502 /*
1503  * Add a vnode to the alias list hung off the dev_t.
1504  *
1505  * The reason for this gunk is that multiple vnodes can reference
1506  * the same physical device, so checking vp->v_usecount to see
1507  * how many users there are is inadequate; the v_usecount for
1508  * the vnodes need to be accumulated.  vcount() does that.
1509  */
1510 void
1511 addaliasu(struct vnode *nvp, udev_t nvp_rdev)
1512 {
1513 	dev_t dev;
1514 
1515 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1516 		panic("addaliasu on non-special vnode");
1517 	dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
1518 	if (dev != NODEV) {
1519 		nvp->v_rdev = dev;
1520 		addalias(nvp, dev);
1521 	} else
1522 		nvp->v_rdev = NULL;
1523 }
1524 
1525 void
1526 addalias(struct vnode *nvp, dev_t dev)
1527 {
1528 
1529 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1530 		panic("addalias on non-special vnode");
1531 
1532 	nvp->v_rdev = dev;
1533 	lwkt_gettoken(&spechash_token);
1534 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1535 	lwkt_reltoken(&spechash_token);
1536 }
1537 
1538 /*
1539  * Grab a particular vnode from the free list, increment its
1540  * reference count and lock it. The vnode lock bit is set if the
1541  * vnode is being eliminated in vgone. The process is awakened
1542  * when the transition is completed, and an error returned to
1543  * indicate that the vnode is no longer usable (possibly having
1544  * been changed to a new file system type).
1545  */
1546 int
1547 vget(vp, flags, td)
1548 	struct vnode *vp;
1549 	int flags;
1550 	struct thread *td;
1551 {
1552 	int error;
1553 
1554 	/*
1555 	 * If the vnode is in the process of being cleaned out for
1556 	 * another use, we wait for the cleaning to finish and then
1557 	 * return failure. Cleaning is determined by checking that
1558 	 * the VXLOCK flag is set.
1559 	 */
1560 	if ((flags & LK_INTERLOCK) == 0) {
1561 		lwkt_gettoken(&vp->v_interlock);
1562 	}
1563 	if (vp->v_flag & VXLOCK) {
1564 		if (vp->v_vxproc == curproc) {
1565 #if 0
1566 			/* this can now occur in normal operation */
1567 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1568 #endif
1569 		} else {
1570 			vp->v_flag |= VXWANT;
1571 			lwkt_reltoken(&vp->v_interlock);
1572 			tsleep((caddr_t)vp, 0, "vget", 0);
1573 			return (ENOENT);
1574 		}
1575 	}
1576 
1577 	vp->v_usecount++;
1578 
1579 	if (VSHOULDBUSY(vp))
1580 		vbusy(vp);
1581 	if (flags & LK_TYPE_MASK) {
1582 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1583 			/*
1584 			 * must expand vrele here because we do not want
1585 			 * to call VOP_INACTIVE if the reference count
1586 			 * drops back to zero since it was never really
1587 			 * active. We must remove it from the free list
1588 			 * before sleeping so that multiple processes do
1589 			 * not try to recycle it.
1590 			 */
1591 			lwkt_gettoken(&vp->v_interlock);
1592 			vp->v_usecount--;
1593 			if (VSHOULDFREE(vp))
1594 				vfree(vp);
1595 			else
1596 				vlruvp(vp);
1597 			lwkt_reltoken(&vp->v_interlock);
1598 		}
1599 		return (error);
1600 	}
1601 	lwkt_reltoken(&vp->v_interlock);
1602 	return (0);
1603 }
1604 
1605 void
1606 vref(struct vnode *vp)
1607 {
1608 	lwkt_gettoken(&vp->v_interlock);
1609 	vp->v_usecount++;
1610 	lwkt_reltoken(&vp->v_interlock);
1611 }
1612 
1613 /*
1614  * Vnode put/release.
1615  * If count drops to zero, call inactive routine and return to freelist.
1616  */
1617 void
1618 vrele(struct vnode *vp)
1619 {
1620 	struct thread *td = curthread;	/* XXX */
1621 
1622 	KASSERT(vp != NULL, ("vrele: null vp"));
1623 
1624 	lwkt_gettoken(&vp->v_interlock);
1625 
1626 	if (vp->v_usecount > 1) {
1627 
1628 		vp->v_usecount--;
1629 		lwkt_reltoken(&vp->v_interlock);
1630 
1631 		return;
1632 	}
1633 
1634 	if (vp->v_usecount == 1) {
1635 		vp->v_usecount--;
1636 		/*
1637 		 * We must call VOP_INACTIVE with the node locked.
1638 		 * If we are doing a vpu, the node is already locked,
1639 		 * but, in the case of vrele, we must explicitly lock
1640 		 * the vnode before calling VOP_INACTIVE
1641 		 */
1642 
1643 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1644 			VOP_INACTIVE(vp, td);
1645 		if (VSHOULDFREE(vp))
1646 			vfree(vp);
1647 		else
1648 			vlruvp(vp);
1649 	} else {
1650 #ifdef DIAGNOSTIC
1651 		vprint("vrele: negative ref count", vp);
1652 		lwkt_reltoken(&vp->v_interlock);
1653 #endif
1654 		panic("vrele: negative ref cnt");
1655 	}
1656 }
1657 
1658 void
1659 vput(struct vnode *vp)
1660 {
1661 	struct thread *td = curthread;	/* XXX */
1662 
1663 	KASSERT(vp != NULL, ("vput: null vp"));
1664 
1665 	lwkt_gettoken(&vp->v_interlock);
1666 
1667 	if (vp->v_usecount > 1) {
1668 		vp->v_usecount--;
1669 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
1670 		return;
1671 	}
1672 
1673 	if (vp->v_usecount == 1) {
1674 		vp->v_usecount--;
1675 		/*
1676 		 * We must call VOP_INACTIVE with the node locked.
1677 		 * If we are doing a vpu, the node is already locked,
1678 		 * so we just need to release the vnode mutex.
1679 		 */
1680 		lwkt_reltoken(&vp->v_interlock);
1681 		VOP_INACTIVE(vp, td);
1682 		if (VSHOULDFREE(vp))
1683 			vfree(vp);
1684 		else
1685 			vlruvp(vp);
1686 	} else {
1687 #ifdef DIAGNOSTIC
1688 		vprint("vput: negative ref count", vp);
1689 #endif
1690 		panic("vput: negative ref cnt");
1691 	}
1692 }
1693 
1694 /*
1695  * Somebody doesn't want the vnode recycled.
1696  */
1697 void
1698 vhold(vp)
1699 	struct vnode *vp;
1700 {
1701 	int s;
1702 
1703   	s = splbio();
1704 	vp->v_holdcnt++;
1705 	if (VSHOULDBUSY(vp))
1706 		vbusy(vp);
1707 	splx(s);
1708 }
1709 
1710 /*
1711  * One less who cares about this vnode.
1712  */
1713 void
1714 vdrop(vp)
1715 	struct vnode *vp;
1716 {
1717 	int s;
1718 
1719 	s = splbio();
1720 	if (vp->v_holdcnt <= 0)
1721 		panic("vdrop: holdcnt");
1722 	vp->v_holdcnt--;
1723 	if (VSHOULDFREE(vp))
1724 		vfree(vp);
1725 	splx(s);
1726 }
1727 
1728 /*
1729  * Remove any vnodes in the vnode table belonging to mount point mp.
1730  *
1731  * If FORCECLOSE is not specified, there should not be any active ones,
1732  * return error if any are found (nb: this is a user error, not a
1733  * system error). If FORCECLOSE is specified, detach any active vnodes
1734  * that are found.
1735  *
1736  * If WRITECLOSE is set, only flush out regular file vnodes open for
1737  * writing.
1738  *
1739  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1740  *
1741  * `rootrefs' specifies the base reference count for the root vnode
1742  * of this filesystem. The root vnode is considered busy if its
1743  * v_usecount exceeds this value. On a successful return, vflush()
1744  * will call vrele() on the root vnode exactly rootrefs times.
1745  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1746  * be zero.
1747  */
1748 #ifdef DIAGNOSTIC
1749 static int busyprt = 0;		/* print out busy vnodes */
1750 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1751 #endif
1752 
1753 int
1754 vflush(mp, rootrefs, flags)
1755 	struct mount *mp;
1756 	int rootrefs;
1757 	int flags;
1758 {
1759 	struct thread *td = curthread;	/* XXX */
1760 	struct vnode *vp, *nvp, *rootvp = NULL;
1761 	struct vattr vattr;
1762 	int busy = 0, error;
1763 
1764 	if (rootrefs > 0) {
1765 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1766 		    ("vflush: bad args"));
1767 		/*
1768 		 * Get the filesystem root vnode. We can vput() it
1769 		 * immediately, since with rootrefs > 0, it won't go away.
1770 		 */
1771 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1772 			return (error);
1773 		vput(rootvp);
1774 	}
1775 	lwkt_gettoken(&mntvnode_token);
1776 loop:
1777 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1778 		/*
1779 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1780 		 * Start over if it has (it won't be on the list anymore).
1781 		 */
1782 		if (vp->v_mount != mp)
1783 			goto loop;
1784 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1785 
1786 		lwkt_gettoken(&vp->v_interlock);
1787 		/*
1788 		 * Skip over a vnodes marked VSYSTEM.
1789 		 */
1790 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1791 			lwkt_reltoken(&vp->v_interlock);
1792 			continue;
1793 		}
1794 		/*
1795 		 * If WRITECLOSE is set, flush out unlinked but still open
1796 		 * files (even if open only for reading) and regular file
1797 		 * vnodes open for writing.
1798 		 */
1799 		if ((flags & WRITECLOSE) &&
1800 		    (vp->v_type == VNON ||
1801 		    (VOP_GETATTR(vp, &vattr, td) == 0 &&
1802 		    vattr.va_nlink > 0)) &&
1803 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1804 			lwkt_reltoken(&vp->v_interlock);
1805 			continue;
1806 		}
1807 
1808 		/*
1809 		 * With v_usecount == 0, all we need to do is clear out the
1810 		 * vnode data structures and we are done.
1811 		 */
1812 		if (vp->v_usecount == 0) {
1813 			lwkt_reltoken(&mntvnode_token);
1814 			vgonel(vp, td);
1815 			lwkt_gettoken(&mntvnode_token);
1816 			continue;
1817 		}
1818 
1819 		/*
1820 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1821 		 * or character devices, revert to an anonymous device. For
1822 		 * all other files, just kill them.
1823 		 */
1824 		if (flags & FORCECLOSE) {
1825 			lwkt_reltoken(&mntvnode_token);
1826 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1827 				vgonel(vp, td);
1828 			} else {
1829 				vclean(vp, 0, td);
1830 				vp->v_op = spec_vnodeop_p;
1831 				insmntque(vp, (struct mount *) 0);
1832 			}
1833 			lwkt_gettoken(&mntvnode_token);
1834 			continue;
1835 		}
1836 #ifdef DIAGNOSTIC
1837 		if (busyprt)
1838 			vprint("vflush: busy vnode", vp);
1839 #endif
1840 		lwkt_reltoken(&vp->v_interlock);
1841 		busy++;
1842 	}
1843 	lwkt_reltoken(&mntvnode_token);
1844 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1845 		/*
1846 		 * If just the root vnode is busy, and if its refcount
1847 		 * is equal to `rootrefs', then go ahead and kill it.
1848 		 */
1849 		lwkt_gettoken(&rootvp->v_interlock);
1850 		KASSERT(busy > 0, ("vflush: not busy"));
1851 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1852 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
1853 			vgonel(rootvp, td);
1854 			busy = 0;
1855 		} else
1856 			lwkt_reltoken(&rootvp->v_interlock);
1857 	}
1858 	if (busy)
1859 		return (EBUSY);
1860 	for (; rootrefs > 0; rootrefs--)
1861 		vrele(rootvp);
1862 	return (0);
1863 }
1864 
1865 /*
1866  * We do not want to recycle the vnode too quickly.
1867  *
1868  * XXX we can't move vp's around the nvnodelist without really screwing
1869  * up the efficiency of filesystem SYNC and friends.  This code is
1870  * disabled until we fix the syncing code's scanning algorithm.
1871  */
1872 static void
1873 vlruvp(struct vnode *vp)
1874 {
1875 #if 0
1876 	struct mount *mp;
1877 
1878 	if ((mp = vp->v_mount) != NULL) {
1879 		lwkt_gettoken(&mntvnode_token);
1880 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1881 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1882 		lwkt_reltoken(&mntvnode_token);
1883 	}
1884 #endif
1885 }
1886 
1887 /*
1888  * Disassociate the underlying file system from a vnode.
1889  */
1890 static void
1891 vclean(struct vnode *vp, int flags, struct thread *td)
1892 {
1893 	int active;
1894 
1895 	/*
1896 	 * Check to see if the vnode is in use. If so we have to reference it
1897 	 * before we clean it out so that its count cannot fall to zero and
1898 	 * generate a race against ourselves to recycle it.
1899 	 */
1900 	if ((active = vp->v_usecount))
1901 		vp->v_usecount++;
1902 
1903 	/*
1904 	 * Prevent the vnode from being recycled or brought into use while we
1905 	 * clean it out.
1906 	 */
1907 	if (vp->v_flag & VXLOCK)
1908 		panic("vclean: deadlock");
1909 	vp->v_flag |= VXLOCK;
1910 	vp->v_vxproc = curproc;
1911 	/*
1912 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1913 	 * have the object locked while it cleans it out. The VOP_LOCK
1914 	 * ensures that the VOP_INACTIVE routine is done with its work.
1915 	 * For active vnodes, it ensures that no other activity can
1916 	 * occur while the underlying object is being cleaned out.
1917 	 */
1918 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
1919 
1920 	/*
1921 	 * Clean out any buffers associated with the vnode.
1922 	 */
1923 	vinvalbuf(vp, V_SAVE, td, 0, 0);
1924 
1925 	VOP_DESTROYVOBJECT(vp);
1926 
1927 	/*
1928 	 * If purging an active vnode, it must be closed and
1929 	 * deactivated before being reclaimed. Note that the
1930 	 * VOP_INACTIVE will unlock the vnode.
1931 	 */
1932 	if (active) {
1933 		if (flags & DOCLOSE)
1934 			VOP_CLOSE(vp, FNONBLOCK, td);
1935 		VOP_INACTIVE(vp, td);
1936 	} else {
1937 		/*
1938 		 * Any other processes trying to obtain this lock must first
1939 		 * wait for VXLOCK to clear, then call the new lock operation.
1940 		 */
1941 		VOP_UNLOCK(vp, 0, td);
1942 	}
1943 	/*
1944 	 * Reclaim the vnode.
1945 	 */
1946 	if (VOP_RECLAIM(vp, td))
1947 		panic("vclean: cannot reclaim");
1948 
1949 	if (active) {
1950 		/*
1951 		 * Inline copy of vrele() since VOP_INACTIVE
1952 		 * has already been called.
1953 		 */
1954 		lwkt_gettoken(&vp->v_interlock);
1955 		if (--vp->v_usecount <= 0) {
1956 #ifdef DIAGNOSTIC
1957 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1958 				vprint("vclean: bad ref count", vp);
1959 				panic("vclean: ref cnt");
1960 			}
1961 #endif
1962 			vfree(vp);
1963 		}
1964 		lwkt_reltoken(&vp->v_interlock);
1965 	}
1966 
1967 	cache_purge(vp);
1968 	vp->v_vnlock = NULL;
1969 
1970 	if (VSHOULDFREE(vp))
1971 		vfree(vp);
1972 
1973 	/*
1974 	 * Done with purge, notify sleepers of the grim news.
1975 	 */
1976 	vp->v_op = dead_vnodeop_p;
1977 	vn_pollgone(vp);
1978 	vp->v_tag = VT_NON;
1979 	vp->v_flag &= ~VXLOCK;
1980 	vp->v_vxproc = NULL;
1981 	if (vp->v_flag & VXWANT) {
1982 		vp->v_flag &= ~VXWANT;
1983 		wakeup((caddr_t) vp);
1984 	}
1985 }
1986 
1987 /*
1988  * Eliminate all activity associated with the requested vnode
1989  * and with all vnodes aliased to the requested vnode.
1990  */
1991 int
1992 vop_revoke(ap)
1993 	struct vop_revoke_args /* {
1994 		struct vnode *a_vp;
1995 		int a_flags;
1996 	} */ *ap;
1997 {
1998 	struct vnode *vp, *vq;
1999 	dev_t dev;
2000 
2001 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2002 
2003 	vp = ap->a_vp;
2004 	/*
2005 	 * If a vgone (or vclean) is already in progress,
2006 	 * wait until it is done and return.
2007 	 */
2008 	if (vp->v_flag & VXLOCK) {
2009 		vp->v_flag |= VXWANT;
2010 		lwkt_reltoken(&vp->v_interlock);
2011 		tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2012 		return (0);
2013 	}
2014 	dev = vp->v_rdev;
2015 	for (;;) {
2016 		lwkt_gettoken(&spechash_token);
2017 		vq = SLIST_FIRST(&dev->si_hlist);
2018 		lwkt_reltoken(&spechash_token);
2019 		if (!vq)
2020 			break;
2021 		vgone(vq);
2022 	}
2023 	return (0);
2024 }
2025 
2026 /*
2027  * Recycle an unused vnode to the front of the free list.
2028  * Release the passed interlock if the vnode will be recycled.
2029  */
2030 int
2031 vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td)
2032 {
2033 	lwkt_gettoken(&vp->v_interlock);
2034 	if (vp->v_usecount == 0) {
2035 		if (inter_lkp) {
2036 			lwkt_reltoken(inter_lkp);
2037 		}
2038 		vgonel(vp, td);
2039 		return (1);
2040 	}
2041 	lwkt_reltoken(&vp->v_interlock);
2042 	return (0);
2043 }
2044 
2045 /*
2046  * Eliminate all activity associated with a vnode
2047  * in preparation for reuse.
2048  */
2049 void
2050 vgone(struct vnode *vp)
2051 {
2052 	struct thread *td = curthread;	/* XXX */
2053 
2054 	lwkt_gettoken(&vp->v_interlock);
2055 	vgonel(vp, td);
2056 }
2057 
2058 /*
2059  * vgone, with the vp interlock held.
2060  */
2061 void
2062 vgonel(struct vnode *vp, struct thread *td)
2063 {
2064 	int s;
2065 
2066 	/*
2067 	 * If a vgone (or vclean) is already in progress,
2068 	 * wait until it is done and return.
2069 	 */
2070 	if (vp->v_flag & VXLOCK) {
2071 		vp->v_flag |= VXWANT;
2072 		lwkt_reltoken(&vp->v_interlock);
2073 		tsleep((caddr_t)vp, 0, "vgone", 0);
2074 		return;
2075 	}
2076 
2077 	/*
2078 	 * Clean out the filesystem specific data.
2079 	 */
2080 	vclean(vp, DOCLOSE, td);
2081 	lwkt_gettoken(&vp->v_interlock);
2082 
2083 	/*
2084 	 * Delete from old mount point vnode list, if on one.
2085 	 */
2086 	if (vp->v_mount != NULL)
2087 		insmntque(vp, (struct mount *)0);
2088 	/*
2089 	 * If special device, remove it from special device alias list
2090 	 * if it is on one.
2091 	 */
2092 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2093 		lwkt_gettoken(&spechash_token);
2094 		SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2095 		freedev(vp->v_rdev);
2096 		lwkt_reltoken(&spechash_token);
2097 		vp->v_rdev = NULL;
2098 	}
2099 
2100 	/*
2101 	 * If it is on the freelist and not already at the head,
2102 	 * move it to the head of the list. The test of the
2103 	 * VDOOMED flag and the reference count of zero is because
2104 	 * it will be removed from the free list by getnewvnode,
2105 	 * but will not have its reference count incremented until
2106 	 * after calling vgone. If the reference count were
2107 	 * incremented first, vgone would (incorrectly) try to
2108 	 * close the previous instance of the underlying object.
2109 	 */
2110 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2111 		s = splbio();
2112 		lwkt_gettoken(&vnode_free_list_token);
2113 		if (vp->v_flag & VFREE)
2114 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2115 		else
2116 			freevnodes++;
2117 		vp->v_flag |= VFREE;
2118 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2119 		lwkt_reltoken(&vnode_free_list_token);
2120 		splx(s);
2121 	}
2122 
2123 	vp->v_type = VBAD;
2124 	lwkt_reltoken(&vp->v_interlock);
2125 }
2126 
2127 /*
2128  * Lookup a vnode by device number.
2129  */
2130 int
2131 vfinddev(dev, type, vpp)
2132 	dev_t dev;
2133 	enum vtype type;
2134 	struct vnode **vpp;
2135 {
2136 	struct vnode *vp;
2137 
2138 	lwkt_gettoken(&spechash_token);
2139 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2140 		if (type == vp->v_type) {
2141 			*vpp = vp;
2142 			lwkt_reltoken(&spechash_token);
2143 			return (1);
2144 		}
2145 	}
2146 	lwkt_reltoken(&spechash_token);
2147 	return (0);
2148 }
2149 
2150 /*
2151  * Calculate the total number of references to a special device.
2152  */
2153 int
2154 vcount(vp)
2155 	struct vnode *vp;
2156 {
2157 	struct vnode *vq;
2158 	int count;
2159 
2160 	count = 0;
2161 	lwkt_gettoken(&spechash_token);
2162 	SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2163 		count += vq->v_usecount;
2164 	lwkt_reltoken(&spechash_token);
2165 	return (count);
2166 }
2167 
2168 /*
2169  * Same as above, but using the dev_t as argument
2170  */
2171 
2172 int
2173 count_dev(dev)
2174 	dev_t dev;
2175 {
2176 	struct vnode *vp;
2177 
2178 	vp = SLIST_FIRST(&dev->si_hlist);
2179 	if (vp == NULL)
2180 		return (0);
2181 	return(vcount(vp));
2182 }
2183 
2184 /*
2185  * Print out a description of a vnode.
2186  */
2187 static char *typename[] =
2188 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2189 
2190 void
2191 vprint(label, vp)
2192 	char *label;
2193 	struct vnode *vp;
2194 {
2195 	char buf[96];
2196 
2197 	if (label != NULL)
2198 		printf("%s: %p: ", label, (void *)vp);
2199 	else
2200 		printf("%p: ", (void *)vp);
2201 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2202 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2203 	    vp->v_holdcnt);
2204 	buf[0] = '\0';
2205 	if (vp->v_flag & VROOT)
2206 		strcat(buf, "|VROOT");
2207 	if (vp->v_flag & VTEXT)
2208 		strcat(buf, "|VTEXT");
2209 	if (vp->v_flag & VSYSTEM)
2210 		strcat(buf, "|VSYSTEM");
2211 	if (vp->v_flag & VXLOCK)
2212 		strcat(buf, "|VXLOCK");
2213 	if (vp->v_flag & VXWANT)
2214 		strcat(buf, "|VXWANT");
2215 	if (vp->v_flag & VBWAIT)
2216 		strcat(buf, "|VBWAIT");
2217 	if (vp->v_flag & VDOOMED)
2218 		strcat(buf, "|VDOOMED");
2219 	if (vp->v_flag & VFREE)
2220 		strcat(buf, "|VFREE");
2221 	if (vp->v_flag & VOBJBUF)
2222 		strcat(buf, "|VOBJBUF");
2223 	if (buf[0] != '\0')
2224 		printf(" flags (%s)", &buf[1]);
2225 	if (vp->v_data == NULL) {
2226 		printf("\n");
2227 	} else {
2228 		printf("\n\t");
2229 		VOP_PRINT(vp);
2230 	}
2231 }
2232 
2233 #ifdef DDB
2234 #include <ddb/ddb.h>
2235 /*
2236  * List all of the locked vnodes in the system.
2237  * Called when debugging the kernel.
2238  */
2239 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2240 {
2241 	struct thread *td = curthread;	/* XXX */
2242 	struct mount *mp, *nmp;
2243 	struct vnode *vp;
2244 
2245 	printf("Locked vnodes\n");
2246 	lwkt_gettoken(&mountlist_token);
2247 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2248 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) {
2249 			nmp = TAILQ_NEXT(mp, mnt_list);
2250 			continue;
2251 		}
2252 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2253 			if (VOP_ISLOCKED(vp, NULL))
2254 				vprint((char *)0, vp);
2255 		}
2256 		lwkt_gettoken(&mountlist_token);
2257 		nmp = TAILQ_NEXT(mp, mnt_list);
2258 		vfs_unbusy(mp, td);
2259 	}
2260 	lwkt_reltoken(&mountlist_token);
2261 }
2262 #endif
2263 
2264 /*
2265  * Top level filesystem related information gathering.
2266  */
2267 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2268 
2269 static int
2270 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2271 {
2272 	int *name = (int *)arg1 - 1;	/* XXX */
2273 	u_int namelen = arg2 + 1;	/* XXX */
2274 	struct vfsconf *vfsp;
2275 
2276 #if 1 || defined(COMPAT_PRELITE2)
2277 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2278 	if (namelen == 1)
2279 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2280 #endif
2281 
2282 #ifdef notyet
2283 	/* all sysctl names at this level are at least name and field */
2284 	if (namelen < 2)
2285 		return (ENOTDIR);		/* overloaded */
2286 	if (name[0] != VFS_GENERIC) {
2287 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2288 			if (vfsp->vfc_typenum == name[0])
2289 				break;
2290 		if (vfsp == NULL)
2291 			return (EOPNOTSUPP);
2292 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2293 		    oldp, oldlenp, newp, newlen, p));
2294 	}
2295 #endif
2296 	switch (name[1]) {
2297 	case VFS_MAXTYPENUM:
2298 		if (namelen != 2)
2299 			return (ENOTDIR);
2300 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2301 	case VFS_CONF:
2302 		if (namelen != 3)
2303 			return (ENOTDIR);	/* overloaded */
2304 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2305 			if (vfsp->vfc_typenum == name[2])
2306 				break;
2307 		if (vfsp == NULL)
2308 			return (EOPNOTSUPP);
2309 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2310 	}
2311 	return (EOPNOTSUPP);
2312 }
2313 
2314 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2315 	"Generic filesystem");
2316 
2317 #if 1 || defined(COMPAT_PRELITE2)
2318 
2319 static int
2320 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2321 {
2322 	int error;
2323 	struct vfsconf *vfsp;
2324 	struct ovfsconf ovfs;
2325 
2326 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2327 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2328 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2329 		ovfs.vfc_index = vfsp->vfc_typenum;
2330 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2331 		ovfs.vfc_flags = vfsp->vfc_flags;
2332 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2333 		if (error)
2334 			return error;
2335 	}
2336 	return 0;
2337 }
2338 
2339 #endif /* 1 || COMPAT_PRELITE2 */
2340 
2341 #if 0
2342 #define KINFO_VNODESLOP	10
2343 /*
2344  * Dump vnode list (via sysctl).
2345  * Copyout address of vnode followed by vnode.
2346  */
2347 /* ARGSUSED */
2348 static int
2349 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2350 {
2351 	struct proc *p = curproc;	/* XXX */
2352 	struct mount *mp, *nmp;
2353 	struct vnode *nvp, *vp;
2354 	int error;
2355 
2356 #define VPTRSZ	sizeof (struct vnode *)
2357 #define VNODESZ	sizeof (struct vnode)
2358 
2359 	req->lock = 0;
2360 	if (!req->oldptr) /* Make an estimate */
2361 		return (SYSCTL_OUT(req, 0,
2362 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2363 
2364 	lwkt_gettoken(&mountlist_token);
2365 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2366 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) {
2367 			nmp = TAILQ_NEXT(mp, mnt_list);
2368 			continue;
2369 		}
2370 again:
2371 		lwkt_gettoken(&mntvnode_token);
2372 		for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2373 		     vp != NULL;
2374 		     vp = nvp) {
2375 			/*
2376 			 * Check that the vp is still associated with
2377 			 * this filesystem.  RACE: could have been
2378 			 * recycled onto the same filesystem.
2379 			 */
2380 			if (vp->v_mount != mp) {
2381 				lwkt_reltoken(&mntvnode_token);
2382 				goto again;
2383 			}
2384 			nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2385 			lwkt_reltoken(&mntvnode_token);
2386 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2387 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2388 				return (error);
2389 			lwkt_gettoken(&mntvnode_token);
2390 		}
2391 		lwkt_reltoken(&mntvnode_token);
2392 		lwkt_gettoken(&mountlist_token);
2393 		nmp = TAILQ_NEXT(mp, mnt_list);
2394 		vfs_unbusy(mp, p);
2395 	}
2396 	lwkt_reltoken(&mountlist_token);
2397 
2398 	return (0);
2399 }
2400 #endif
2401 
2402 /*
2403  * XXX
2404  * Exporting the vnode list on large systems causes them to crash.
2405  * Exporting the vnode list on medium systems causes sysctl to coredump.
2406  */
2407 #if 0
2408 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2409 	0, 0, sysctl_vnode, "S,vnode", "");
2410 #endif
2411 
2412 /*
2413  * Check to see if a filesystem is mounted on a block device.
2414  */
2415 int
2416 vfs_mountedon(vp)
2417 	struct vnode *vp;
2418 {
2419 
2420 	if (vp->v_specmountpoint != NULL)
2421 		return (EBUSY);
2422 	return (0);
2423 }
2424 
2425 /*
2426  * Unmount all filesystems. The list is traversed in reverse order
2427  * of mounting to avoid dependencies.
2428  */
2429 void
2430 vfs_unmountall()
2431 {
2432 	struct mount *mp;
2433 	struct thread *td = curthread;
2434 	int error;
2435 
2436 	if (td->td_proc == NULL)
2437 		td = initproc->p_thread;	/* XXX XXX use proc0 instead? */
2438 
2439 	/*
2440 	 * Since this only runs when rebooting, it is not interlocked.
2441 	 */
2442 	while(!TAILQ_EMPTY(&mountlist)) {
2443 		mp = TAILQ_LAST(&mountlist, mntlist);
2444 		error = dounmount(mp, MNT_FORCE, td);
2445 		if (error) {
2446 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2447 			printf("unmount of %s failed (",
2448 			    mp->mnt_stat.f_mntonname);
2449 			if (error == EBUSY)
2450 				printf("BUSY)\n");
2451 			else
2452 				printf("%d)\n", error);
2453 		} else {
2454 			/* The unmount has removed mp from the mountlist */
2455 		}
2456 	}
2457 }
2458 
2459 /*
2460  * Build hash lists of net addresses and hang them off the mount point.
2461  * Called by ufs_mount() to set up the lists of export addresses.
2462  */
2463 static int
2464 vfs_hang_addrlist(mp, nep, argp)
2465 	struct mount *mp;
2466 	struct netexport *nep;
2467 	struct export_args *argp;
2468 {
2469 	struct netcred *np;
2470 	struct radix_node_head *rnh;
2471 	int i;
2472 	struct radix_node *rn;
2473 	struct sockaddr *saddr, *smask = 0;
2474 	struct domain *dom;
2475 	int error;
2476 
2477 	if (argp->ex_addrlen == 0) {
2478 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2479 			return (EPERM);
2480 		np = &nep->ne_defexported;
2481 		np->netc_exflags = argp->ex_flags;
2482 		np->netc_anon = argp->ex_anon;
2483 		np->netc_anon.cr_ref = 1;
2484 		mp->mnt_flag |= MNT_DEFEXPORTED;
2485 		return (0);
2486 	}
2487 
2488 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2489 		return (EINVAL);
2490 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2491 		return (EINVAL);
2492 
2493 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2494 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2495 	bzero((caddr_t) np, i);
2496 	saddr = (struct sockaddr *) (np + 1);
2497 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2498 		goto out;
2499 	if (saddr->sa_len > argp->ex_addrlen)
2500 		saddr->sa_len = argp->ex_addrlen;
2501 	if (argp->ex_masklen) {
2502 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2503 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2504 		if (error)
2505 			goto out;
2506 		if (smask->sa_len > argp->ex_masklen)
2507 			smask->sa_len = argp->ex_masklen;
2508 	}
2509 	i = saddr->sa_family;
2510 	if ((rnh = nep->ne_rtable[i]) == 0) {
2511 		/*
2512 		 * Seems silly to initialize every AF when most are not used,
2513 		 * do so on demand here
2514 		 */
2515 		for (dom = domains; dom; dom = dom->dom_next)
2516 			if (dom->dom_family == i && dom->dom_rtattach) {
2517 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2518 				    dom->dom_rtoffset);
2519 				break;
2520 			}
2521 		if ((rnh = nep->ne_rtable[i]) == 0) {
2522 			error = ENOBUFS;
2523 			goto out;
2524 		}
2525 	}
2526 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2527 	    np->netc_rnodes);
2528 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2529 		error = EPERM;
2530 		goto out;
2531 	}
2532 	np->netc_exflags = argp->ex_flags;
2533 	np->netc_anon = argp->ex_anon;
2534 	np->netc_anon.cr_ref = 1;
2535 	return (0);
2536 out:
2537 	free(np, M_NETADDR);
2538 	return (error);
2539 }
2540 
2541 /* ARGSUSED */
2542 static int
2543 vfs_free_netcred(rn, w)
2544 	struct radix_node *rn;
2545 	void *w;
2546 {
2547 	struct radix_node_head *rnh = (struct radix_node_head *) w;
2548 
2549 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2550 	free((caddr_t) rn, M_NETADDR);
2551 	return (0);
2552 }
2553 
2554 /*
2555  * Free the net address hash lists that are hanging off the mount points.
2556  */
2557 static void
2558 vfs_free_addrlist(nep)
2559 	struct netexport *nep;
2560 {
2561 	int i;
2562 	struct radix_node_head *rnh;
2563 
2564 	for (i = 0; i <= AF_MAX; i++)
2565 		if ((rnh = nep->ne_rtable[i])) {
2566 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2567 			    (caddr_t) rnh);
2568 			free((caddr_t) rnh, M_RTABLE);
2569 			nep->ne_rtable[i] = 0;
2570 		}
2571 }
2572 
2573 int
2574 vfs_export(mp, nep, argp)
2575 	struct mount *mp;
2576 	struct netexport *nep;
2577 	struct export_args *argp;
2578 {
2579 	int error;
2580 
2581 	if (argp->ex_flags & MNT_DELEXPORT) {
2582 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2583 			vfs_setpublicfs(NULL, NULL, NULL);
2584 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2585 		}
2586 		vfs_free_addrlist(nep);
2587 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2588 	}
2589 	if (argp->ex_flags & MNT_EXPORTED) {
2590 		if (argp->ex_flags & MNT_EXPUBLIC) {
2591 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2592 				return (error);
2593 			mp->mnt_flag |= MNT_EXPUBLIC;
2594 		}
2595 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2596 			return (error);
2597 		mp->mnt_flag |= MNT_EXPORTED;
2598 	}
2599 	return (0);
2600 }
2601 
2602 
2603 /*
2604  * Set the publicly exported filesystem (WebNFS). Currently, only
2605  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2606  */
2607 int
2608 vfs_setpublicfs(mp, nep, argp)
2609 	struct mount *mp;
2610 	struct netexport *nep;
2611 	struct export_args *argp;
2612 {
2613 	int error;
2614 	struct vnode *rvp;
2615 	char *cp;
2616 
2617 	/*
2618 	 * mp == NULL -> invalidate the current info, the FS is
2619 	 * no longer exported. May be called from either vfs_export
2620 	 * or unmount, so check if it hasn't already been done.
2621 	 */
2622 	if (mp == NULL) {
2623 		if (nfs_pub.np_valid) {
2624 			nfs_pub.np_valid = 0;
2625 			if (nfs_pub.np_index != NULL) {
2626 				FREE(nfs_pub.np_index, M_TEMP);
2627 				nfs_pub.np_index = NULL;
2628 			}
2629 		}
2630 		return (0);
2631 	}
2632 
2633 	/*
2634 	 * Only one allowed at a time.
2635 	 */
2636 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2637 		return (EBUSY);
2638 
2639 	/*
2640 	 * Get real filehandle for root of exported FS.
2641 	 */
2642 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2643 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2644 
2645 	if ((error = VFS_ROOT(mp, &rvp)))
2646 		return (error);
2647 
2648 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2649 		return (error);
2650 
2651 	vput(rvp);
2652 
2653 	/*
2654 	 * If an indexfile was specified, pull it in.
2655 	 */
2656 	if (argp->ex_indexfile != NULL) {
2657 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2658 		    M_WAITOK);
2659 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2660 		    MAXNAMLEN, (size_t *)0);
2661 		if (!error) {
2662 			/*
2663 			 * Check for illegal filenames.
2664 			 */
2665 			for (cp = nfs_pub.np_index; *cp; cp++) {
2666 				if (*cp == '/') {
2667 					error = EINVAL;
2668 					break;
2669 				}
2670 			}
2671 		}
2672 		if (error) {
2673 			FREE(nfs_pub.np_index, M_TEMP);
2674 			return (error);
2675 		}
2676 	}
2677 
2678 	nfs_pub.np_mount = mp;
2679 	nfs_pub.np_valid = 1;
2680 	return (0);
2681 }
2682 
2683 struct netcred *
2684 vfs_export_lookup(mp, nep, nam)
2685 	struct mount *mp;
2686 	struct netexport *nep;
2687 	struct sockaddr *nam;
2688 {
2689 	struct netcred *np;
2690 	struct radix_node_head *rnh;
2691 	struct sockaddr *saddr;
2692 
2693 	np = NULL;
2694 	if (mp->mnt_flag & MNT_EXPORTED) {
2695 		/*
2696 		 * Lookup in the export list first.
2697 		 */
2698 		if (nam != NULL) {
2699 			saddr = nam;
2700 			rnh = nep->ne_rtable[saddr->sa_family];
2701 			if (rnh != NULL) {
2702 				np = (struct netcred *)
2703 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2704 							      rnh);
2705 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2706 					np = NULL;
2707 			}
2708 		}
2709 		/*
2710 		 * If no address match, use the default if it exists.
2711 		 */
2712 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2713 			np = &nep->ne_defexported;
2714 	}
2715 	return (np);
2716 }
2717 
2718 /*
2719  * perform msync on all vnodes under a mount point
2720  * the mount point must be locked.
2721  */
2722 void
2723 vfs_msync(struct mount *mp, int flags)
2724 {
2725 	struct thread *td = curthread;	/* XXX */
2726 	struct vnode *vp, *nvp;
2727 	struct vm_object *obj;
2728 	int tries;
2729 
2730 	tries = 5;
2731 	lwkt_gettoken(&mntvnode_token);
2732 loop:
2733 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2734 		if (vp->v_mount != mp) {
2735 			if (--tries > 0)
2736 				goto loop;
2737 			break;
2738 		}
2739 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2740 
2741 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2742 			continue;
2743 
2744 		/*
2745 		 * There could be hundreds of thousands of vnodes, we cannot
2746 		 * afford to do anything heavy-weight until we have a fairly
2747 		 * good indication that there is something to do.
2748 		 */
2749 		if ((vp->v_flag & VOBJDIRTY) &&
2750 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2751 			lwkt_reltoken(&mntvnode_token);
2752 			if (!vget(vp,
2753 			    LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) {
2754 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2755 					vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2756 				}
2757 				vput(vp);
2758 			}
2759 			lwkt_gettoken(&mntvnode_token);
2760 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2761 				if (--tries > 0)
2762 					goto loop;
2763 				break;
2764 			}
2765 		}
2766 	}
2767 	lwkt_reltoken(&mntvnode_token);
2768 }
2769 
2770 /*
2771  * Create the VM object needed for VMIO and mmap support.  This
2772  * is done for all VREG files in the system.  Some filesystems might
2773  * afford the additional metadata buffering capability of the
2774  * VMIO code by making the device node be VMIO mode also.
2775  *
2776  * vp must be locked when vfs_object_create is called.
2777  */
2778 int
2779 vfs_object_create(struct vnode *vp, struct thread *td)
2780 {
2781 	return (VOP_CREATEVOBJECT(vp, td));
2782 }
2783 
2784 void
2785 vfree(vp)
2786 	struct vnode *vp;
2787 {
2788 	int s;
2789 
2790 	s = splbio();
2791 	lwkt_gettoken(&vnode_free_list_token);
2792 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2793 	if (vp->v_flag & VAGE) {
2794 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2795 	} else {
2796 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2797 	}
2798 	freevnodes++;
2799 	lwkt_reltoken(&vnode_free_list_token);
2800 	vp->v_flag &= ~VAGE;
2801 	vp->v_flag |= VFREE;
2802 	splx(s);
2803 }
2804 
2805 void
2806 vbusy(vp)
2807 	struct vnode *vp;
2808 {
2809 	int s;
2810 
2811 	s = splbio();
2812 	lwkt_gettoken(&vnode_free_list_token);
2813 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2814 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2815 	freevnodes--;
2816 	lwkt_reltoken(&vnode_free_list_token);
2817 	vp->v_flag &= ~(VFREE|VAGE);
2818 	splx(s);
2819 }
2820 
2821 /*
2822  * Record a process's interest in events which might happen to
2823  * a vnode.  Because poll uses the historic select-style interface
2824  * internally, this routine serves as both the ``check for any
2825  * pending events'' and the ``record my interest in future events''
2826  * functions.  (These are done together, while the lock is held,
2827  * to avoid race conditions.)
2828  */
2829 int
2830 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
2831 {
2832 	lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2833 	if (vp->v_pollinfo.vpi_revents & events) {
2834 		/*
2835 		 * This leaves events we are not interested
2836 		 * in available for the other process which
2837 		 * which presumably had requested them
2838 		 * (otherwise they would never have been
2839 		 * recorded).
2840 		 */
2841 		events &= vp->v_pollinfo.vpi_revents;
2842 		vp->v_pollinfo.vpi_revents &= ~events;
2843 
2844 		lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2845 		return events;
2846 	}
2847 	vp->v_pollinfo.vpi_events |= events;
2848 	selrecord(td, &vp->v_pollinfo.vpi_selinfo);
2849 	lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2850 	return 0;
2851 }
2852 
2853 /*
2854  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2855  * it is possible for us to miss an event due to race conditions, but
2856  * that condition is expected to be rare, so for the moment it is the
2857  * preferred interface.
2858  */
2859 void
2860 vn_pollevent(vp, events)
2861 	struct vnode *vp;
2862 	short events;
2863 {
2864 	lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2865 	if (vp->v_pollinfo.vpi_events & events) {
2866 		/*
2867 		 * We clear vpi_events so that we don't
2868 		 * call selwakeup() twice if two events are
2869 		 * posted before the polling process(es) is
2870 		 * awakened.  This also ensures that we take at
2871 		 * most one selwakeup() if the polling process
2872 		 * is no longer interested.  However, it does
2873 		 * mean that only one event can be noticed at
2874 		 * a time.  (Perhaps we should only clear those
2875 		 * event bits which we note?) XXX
2876 		 */
2877 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2878 		vp->v_pollinfo.vpi_revents |= events;
2879 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2880 	}
2881 	lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2882 }
2883 
2884 /*
2885  * Wake up anyone polling on vp because it is being revoked.
2886  * This depends on dead_poll() returning POLLHUP for correct
2887  * behavior.
2888  */
2889 void
2890 vn_pollgone(vp)
2891 	struct vnode *vp;
2892 {
2893 	lwkt_gettoken(&vp->v_pollinfo.vpi_token);
2894 	if (vp->v_pollinfo.vpi_events) {
2895 		vp->v_pollinfo.vpi_events = 0;
2896 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2897 	}
2898 	lwkt_reltoken(&vp->v_pollinfo.vpi_token);
2899 }
2900 
2901 
2902 
2903 /*
2904  * Routine to create and manage a filesystem syncer vnode.
2905  */
2906 #define sync_close ((int (*) (struct  vop_close_args *))nullop)
2907 static int	sync_fsync (struct  vop_fsync_args *);
2908 static int	sync_inactive (struct  vop_inactive_args *);
2909 static int	sync_reclaim  (struct  vop_reclaim_args *);
2910 #define sync_lock ((int (*) (struct  vop_lock_args *))vop_nolock)
2911 #define sync_unlock ((int (*) (struct  vop_unlock_args *))vop_nounlock)
2912 static int	sync_print (struct vop_print_args *);
2913 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
2914 
2915 static vop_t **sync_vnodeop_p;
2916 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2917 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2918 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2919 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2920 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2921 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2922 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2923 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2924 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2925 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2926 	{ NULL, NULL }
2927 };
2928 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2929 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2930 
2931 VNODEOP_SET(sync_vnodeop_opv_desc);
2932 
2933 /*
2934  * Create a new filesystem syncer vnode for the specified mount point.
2935  */
2936 int
2937 vfs_allocate_syncvnode(mp)
2938 	struct mount *mp;
2939 {
2940 	struct vnode *vp;
2941 	static long start, incr, next;
2942 	int error;
2943 
2944 	/* Allocate a new vnode */
2945 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2946 		mp->mnt_syncer = NULL;
2947 		return (error);
2948 	}
2949 	vp->v_type = VNON;
2950 	/*
2951 	 * Place the vnode onto the syncer worklist. We attempt to
2952 	 * scatter them about on the list so that they will go off
2953 	 * at evenly distributed times even if all the filesystems
2954 	 * are mounted at once.
2955 	 */
2956 	next += incr;
2957 	if (next == 0 || next > syncer_maxdelay) {
2958 		start /= 2;
2959 		incr /= 2;
2960 		if (start == 0) {
2961 			start = syncer_maxdelay / 2;
2962 			incr = syncer_maxdelay;
2963 		}
2964 		next = start;
2965 	}
2966 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2967 	mp->mnt_syncer = vp;
2968 	return (0);
2969 }
2970 
2971 /*
2972  * Do a lazy sync of the filesystem.
2973  */
2974 static int
2975 sync_fsync(ap)
2976 	struct vop_fsync_args /* {
2977 		struct vnode *a_vp;
2978 		struct ucred *a_cred;
2979 		int a_waitfor;
2980 		struct thread *a_td;
2981 	} */ *ap;
2982 {
2983 	struct vnode *syncvp = ap->a_vp;
2984 	struct mount *mp = syncvp->v_mount;
2985 	struct thread *td = ap->a_td;
2986 	int asyncflag;
2987 
2988 	/*
2989 	 * We only need to do something if this is a lazy evaluation.
2990 	 */
2991 	if (ap->a_waitfor != MNT_LAZY)
2992 		return (0);
2993 
2994 	/*
2995 	 * Move ourselves to the back of the sync list.
2996 	 */
2997 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2998 
2999 	/*
3000 	 * Walk the list of vnodes pushing all that are dirty and
3001 	 * not already on the sync list.
3002 	 */
3003 	lwkt_gettoken(&mountlist_token);
3004 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) {
3005 		lwkt_reltoken(&mountlist_token);
3006 		return (0);
3007 	}
3008 	asyncflag = mp->mnt_flag & MNT_ASYNC;
3009 	mp->mnt_flag &= ~MNT_ASYNC;
3010 	vfs_msync(mp, MNT_NOWAIT);
3011 	VFS_SYNC(mp, MNT_LAZY, td);
3012 	if (asyncflag)
3013 		mp->mnt_flag |= MNT_ASYNC;
3014 	vfs_unbusy(mp, td);
3015 	return (0);
3016 }
3017 
3018 /*
3019  * The syncer vnode is no referenced.
3020  */
3021 static int
3022 sync_inactive(ap)
3023 	struct vop_inactive_args /* {
3024 		struct vnode *a_vp;
3025 		struct proc *a_p;
3026 	} */ *ap;
3027 {
3028 
3029 	vgone(ap->a_vp);
3030 	return (0);
3031 }
3032 
3033 /*
3034  * The syncer vnode is no longer needed and is being decommissioned.
3035  *
3036  * Modifications to the worklist must be protected at splbio().
3037  */
3038 static int
3039 sync_reclaim(ap)
3040 	struct vop_reclaim_args /* {
3041 		struct vnode *a_vp;
3042 	} */ *ap;
3043 {
3044 	struct vnode *vp = ap->a_vp;
3045 	int s;
3046 
3047 	s = splbio();
3048 	vp->v_mount->mnt_syncer = NULL;
3049 	if (vp->v_flag & VONWORKLST) {
3050 		LIST_REMOVE(vp, v_synclist);
3051 		vp->v_flag &= ~VONWORKLST;
3052 	}
3053 	splx(s);
3054 
3055 	return (0);
3056 }
3057 
3058 /*
3059  * Print out a syncer vnode.
3060  */
3061 static int
3062 sync_print(ap)
3063 	struct vop_print_args /* {
3064 		struct vnode *a_vp;
3065 	} */ *ap;
3066 {
3067 	struct vnode *vp = ap->a_vp;
3068 
3069 	printf("syncer vnode");
3070 	if (vp->v_vnlock != NULL)
3071 		lockmgr_printinfo(vp->v_vnlock);
3072 	printf("\n");
3073 	return (0);
3074 }
3075 
3076 /*
3077  * extract the dev_t from a VBLK or VCHR
3078  */
3079 dev_t
3080 vn_todev(vp)
3081 	struct vnode *vp;
3082 {
3083 	if (vp->v_type != VBLK && vp->v_type != VCHR)
3084 		return (NODEV);
3085 	return (vp->v_rdev);
3086 }
3087 
3088 /*
3089  * Check if vnode represents a disk device
3090  */
3091 int
3092 vn_isdisk(vp, errp)
3093 	struct vnode *vp;
3094 	int *errp;
3095 {
3096 	if (vp->v_type != VBLK && vp->v_type != VCHR) {
3097 		if (errp != NULL)
3098 			*errp = ENOTBLK;
3099 		return (0);
3100 	}
3101 	if (vp->v_rdev == NULL) {
3102 		if (errp != NULL)
3103 			*errp = ENXIO;
3104 		return (0);
3105 	}
3106 	if (!dev_dport(vp->v_rdev)) {
3107 		if (errp != NULL)
3108 			*errp = ENXIO;
3109 		return (0);
3110 	}
3111 	if (!(dev_dflags(vp->v_rdev) & D_DISK)) {
3112 		if (errp != NULL)
3113 			*errp = ENOTBLK;
3114 		return (0);
3115 	}
3116 	if (errp != NULL)
3117 		*errp = 0;
3118 	return (1);
3119 }
3120 
3121 void
3122 NDFREE(ndp, flags)
3123      struct nameidata *ndp;
3124      const uint flags;
3125 {
3126 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3127 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3128 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3129 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3130 	}
3131 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3132 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3133 	    ndp->ni_dvp != ndp->ni_vp)
3134 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td);
3135 	if (!(flags & NDF_NO_DVP_RELE) &&
3136 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3137 		vrele(ndp->ni_dvp);
3138 		ndp->ni_dvp = NULL;
3139 	}
3140 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3141 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3142 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td);
3143 	if (!(flags & NDF_NO_VP_RELE) &&
3144 	    ndp->ni_vp) {
3145 		vrele(ndp->ni_vp);
3146 		ndp->ni_vp = NULL;
3147 	}
3148 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3149 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3150 		vrele(ndp->ni_startdir);
3151 		ndp->ni_startdir = NULL;
3152 	}
3153 }
3154