xref: /dflybsd-src/sys/kern/vfs_subr.c (revision bc76a771df54af7e361532b257cecc26227736b4)
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 1.29 2004/04/08 17:56:48 dillon 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/vm_kern.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
83 
84 #include <sys/buf2.h>
85 #include <sys/thread2.h>
86 
87 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
88 
89 static void	insmntque (struct vnode *vp, struct mount *mp);
90 static void	vclean (struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td);
91 static unsigned long	numvnodes;
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 
121 #ifdef ENABLE_VFS_IOOPT
122 int vfs_ioopt = 0;
123 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
124 #endif
125 
126 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
127 struct lwkt_token mountlist_token;
128 struct lwkt_token mntvnode_token;
129 int	nfs_mount_type = -1;
130 static struct lwkt_token mntid_token;
131 static struct lwkt_token vnode_free_list_token;
132 static struct lwkt_token spechash_token;
133 struct nfs_public nfs_pub;	/* publicly exported FS */
134 static vm_zone_t vnode_zone;
135 
136 /*
137  * The workitem queue.
138  */
139 #define SYNCER_MAXDELAY		32
140 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
141 time_t syncdelay = 30;		/* max time to delay syncing data */
142 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0,
143 	"VFS data synchronization delay");
144 time_t filedelay = 30;		/* time to delay syncing files */
145 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
146 	"File synchronization delay");
147 time_t dirdelay = 29;		/* time to delay syncing directories */
148 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
149 	"Directory synchronization delay");
150 time_t metadelay = 28;		/* time to delay syncing metadata */
151 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
152 	"VFS metadata synchronization delay");
153 static int rushjob;			/* number of slots to run ASAP */
154 static int stat_rush_requests;	/* number of times I/O speeded up */
155 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
156 
157 static int syncer_delayno = 0;
158 static long syncer_mask;
159 LIST_HEAD(synclist, vnode);
160 static struct synclist *syncer_workitem_pending;
161 
162 int desiredvnodes;
163 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
164     &desiredvnodes, 0, "Maximum number of vnodes");
165 static int minvnodes;
166 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
167     &minvnodes, 0, "Minimum number of vnodes");
168 static int vnlru_nowhere = 0;
169 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
170     "Number of times the vnlru process ran without success");
171 
172 static void	vfs_free_addrlist (struct netexport *nep);
173 static int	vfs_free_netcred (struct radix_node *rn, void *w);
174 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
175 				       struct export_args *argp);
176 
177 #define VSHOULDFREE(vp) \
178 	(!((vp)->v_flag & (VFREE|VDOOMED)) && \
179 	 !(vp)->v_holdcnt && !(vp)->v_usecount && \
180 	 (!(vp)->v_object || \
181 	  !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
182 
183 #define VMIGHTFREE(vp) \
184 	(((vp)->v_flag & (VFREE|VDOOMED|VXLOCK)) == 0 &&   \
185 	 cache_leaf_test(vp) == 0 && (vp)->v_usecount == 0)
186 
187 #define VSHOULDBUSY(vp) \
188 	(((vp)->v_flag & VFREE) && \
189 	 ((vp)->v_holdcnt || (vp)->v_usecount))
190 
191 static void vbusy(struct vnode *vp);
192 static void vfree(struct vnode *vp);
193 static void vmaybefree(struct vnode *vp);
194 
195 /*
196  * NOTE: the vnode interlock must be held on call.
197  */
198 static __inline void
199 vmaybefree(struct vnode *vp)
200 {
201 	if (VSHOULDFREE(vp))
202 		vfree(vp);
203 }
204 
205 /*
206  * Initialize the vnode management data structures.
207  */
208 void
209 vntblinit()
210 {
211 
212 	/*
213 	 * Desired vnodes is a result of the physical page count
214 	 * and the size of kernel's heap.  It scales in proportion
215 	 * to the amount of available physical memory.  This can
216 	 * cause trouble on 64-bit and large memory platforms.
217 	 */
218 	/* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
219 	desiredvnodes =
220 		min(maxproc + vmstats.v_page_count /4,
221 		    2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
222 		    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
223 
224 	minvnodes = desiredvnodes / 4;
225 	lwkt_token_init(&mountlist_token);
226 	lwkt_token_init(&mntvnode_token);
227 	lwkt_token_init(&mntid_token);
228 	lwkt_token_init(&spechash_token);
229 	TAILQ_INIT(&vnode_free_list);
230 	lwkt_token_init(&vnode_free_list_token);
231 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
232 	/*
233 	 * Initialize the filesystem syncer.
234 	 */
235 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
236 		&syncer_mask);
237 	syncer_maxdelay = syncer_mask + 1;
238 }
239 
240 /*
241  * Mark a mount point as busy. Used to synchronize access and to delay
242  * unmounting. Interlock is not released on failure.
243  */
244 int
245 vfs_busy(struct mount *mp, int flags, lwkt_tokref_t interlkp, struct thread *td)
246 {
247 	int lkflags;
248 
249 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
250 		if (flags & LK_NOWAIT)
251 			return (ENOENT);
252 		mp->mnt_kern_flag |= MNTK_MWAIT;
253 		/*
254 		 * Since all busy locks are shared except the exclusive
255 		 * lock granted when unmounting, the only place that a
256 		 * wakeup needs to be done is at the release of the
257 		 * exclusive lock at the end of dounmount.
258 		 *
259 		 * note: interlkp is a serializer and thus can be safely
260 		 * held through any sleep
261 		 */
262 		tsleep((caddr_t)mp, 0, "vfs_busy", 0);
263 		return (ENOENT);
264 	}
265 	lkflags = LK_SHARED | LK_NOPAUSE;
266 	if (interlkp)
267 		lkflags |= LK_INTERLOCK;
268 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
269 		panic("vfs_busy: unexpected lock failure");
270 	return (0);
271 }
272 
273 /*
274  * Free a busy filesystem.
275  */
276 void
277 vfs_unbusy(struct mount *mp, struct thread *td)
278 {
279 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
280 }
281 
282 /*
283  * Lookup a filesystem type, and if found allocate and initialize
284  * a mount structure for it.
285  *
286  * Devname is usually updated by mount(8) after booting.
287  */
288 int
289 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
290 {
291 	struct thread *td = curthread;	/* XXX */
292 	struct vfsconf *vfsp;
293 	struct mount *mp;
294 
295 	if (fstypename == NULL)
296 		return (ENODEV);
297 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
298 		if (!strcmp(vfsp->vfc_name, fstypename))
299 			break;
300 	if (vfsp == NULL)
301 		return (ENODEV);
302 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
303 	bzero((char *)mp, (u_long)sizeof(struct mount));
304 	lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
305 	vfs_busy(mp, LK_NOWAIT, NULL, td);
306 	TAILQ_INIT(&mp->mnt_nvnodelist);
307 	TAILQ_INIT(&mp->mnt_reservedvnlist);
308 	mp->mnt_nvnodelistsize = 0;
309 	mp->mnt_vfc = vfsp;
310 	mp->mnt_op = vfsp->vfc_vfsops;
311 	mp->mnt_flag = MNT_RDONLY;
312 	mp->mnt_vnodecovered = NULLVP;
313 	vfsp->vfc_refcount++;
314 	mp->mnt_iosize_max = DFLTPHYS;
315 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
316 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
317 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
318 	mp->mnt_stat.f_mntonname[0] = '/';
319 	mp->mnt_stat.f_mntonname[1] = 0;
320 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
321 	*mpp = mp;
322 	return (0);
323 }
324 
325 /*
326  * Find an appropriate filesystem to use for the root. If a filesystem
327  * has not been preselected, walk through the list of known filesystems
328  * trying those that have mountroot routines, and try them until one
329  * works or we have tried them all.
330  */
331 #ifdef notdef	/* XXX JH */
332 int
333 lite2_vfs_mountroot()
334 {
335 	struct vfsconf *vfsp;
336 	extern int (*lite2_mountroot) (void);
337 	int error;
338 
339 	if (lite2_mountroot != NULL)
340 		return ((*lite2_mountroot)());
341 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
342 		if (vfsp->vfc_mountroot == NULL)
343 			continue;
344 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
345 			return (0);
346 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
347 	}
348 	return (ENODEV);
349 }
350 #endif
351 
352 /*
353  * Lookup a mount point by filesystem identifier.
354  */
355 struct mount *
356 vfs_getvfs(fsid)
357 	fsid_t *fsid;
358 {
359 	struct mount *mp;
360 	lwkt_tokref ilock;
361 
362 	lwkt_gettoken(&ilock, &mountlist_token);
363 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
364 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
365 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
366 			break;
367 	    }
368 	}
369 	lwkt_reltoken(&ilock);
370 	return (mp);
371 }
372 
373 /*
374  * Get a new unique fsid.  Try to make its val[0] unique, since this value
375  * will be used to create fake device numbers for stat().  Also try (but
376  * not so hard) make its val[0] unique mod 2^16, since some emulators only
377  * support 16-bit device numbers.  We end up with unique val[0]'s for the
378  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
379  *
380  * Keep in mind that several mounts may be running in parallel.  Starting
381  * the search one past where the previous search terminated is both a
382  * micro-optimization and a defense against returning the same fsid to
383  * different mounts.
384  */
385 void
386 vfs_getnewfsid(mp)
387 	struct mount *mp;
388 {
389 	static u_int16_t mntid_base;
390 	lwkt_tokref ilock;
391 	fsid_t tfsid;
392 	int mtype;
393 
394 	lwkt_gettoken(&ilock, &mntid_token);
395 	mtype = mp->mnt_vfc->vfc_typenum;
396 	tfsid.val[1] = mtype;
397 	mtype = (mtype & 0xFF) << 24;
398 	for (;;) {
399 		tfsid.val[0] = makeudev(255,
400 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
401 		mntid_base++;
402 		if (vfs_getvfs(&tfsid) == NULL)
403 			break;
404 	}
405 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
406 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
407 	lwkt_reltoken(&ilock);
408 }
409 
410 /*
411  * Knob to control the precision of file timestamps:
412  *
413  *   0 = seconds only; nanoseconds zeroed.
414  *   1 = seconds and nanoseconds, accurate within 1/HZ.
415  *   2 = seconds and nanoseconds, truncated to microseconds.
416  * >=3 = seconds and nanoseconds, maximum precision.
417  */
418 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
419 
420 static int timestamp_precision = TSP_SEC;
421 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
422     &timestamp_precision, 0, "");
423 
424 /*
425  * Get a current timestamp.
426  */
427 void
428 vfs_timestamp(tsp)
429 	struct timespec *tsp;
430 {
431 	struct timeval tv;
432 
433 	switch (timestamp_precision) {
434 	case TSP_SEC:
435 		tsp->tv_sec = time_second;
436 		tsp->tv_nsec = 0;
437 		break;
438 	case TSP_HZ:
439 		getnanotime(tsp);
440 		break;
441 	case TSP_USEC:
442 		microtime(&tv);
443 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
444 		break;
445 	case TSP_NSEC:
446 	default:
447 		nanotime(tsp);
448 		break;
449 	}
450 }
451 
452 /*
453  * Set vnode attributes to VNOVAL
454  */
455 void
456 vattr_null(vap)
457 	struct vattr *vap;
458 {
459 
460 	vap->va_type = VNON;
461 	vap->va_size = VNOVAL;
462 	vap->va_bytes = VNOVAL;
463 	vap->va_mode = VNOVAL;
464 	vap->va_nlink = VNOVAL;
465 	vap->va_uid = VNOVAL;
466 	vap->va_gid = VNOVAL;
467 	vap->va_fsid = VNOVAL;
468 	vap->va_fileid = VNOVAL;
469 	vap->va_blocksize = VNOVAL;
470 	vap->va_rdev = VNOVAL;
471 	vap->va_atime.tv_sec = VNOVAL;
472 	vap->va_atime.tv_nsec = VNOVAL;
473 	vap->va_mtime.tv_sec = VNOVAL;
474 	vap->va_mtime.tv_nsec = VNOVAL;
475 	vap->va_ctime.tv_sec = VNOVAL;
476 	vap->va_ctime.tv_nsec = VNOVAL;
477 	vap->va_flags = VNOVAL;
478 	vap->va_gen = VNOVAL;
479 	vap->va_vaflags = 0;
480 }
481 
482 /*
483  * This routine is called when we have too many vnodes.  It attempts
484  * to free <count> vnodes and will potentially free vnodes that still
485  * have VM backing store (VM backing store is typically the cause
486  * of a vnode blowout so we want to do this).  Therefore, this operation
487  * is not considered cheap.
488  *
489  * A number of conditions may prevent a vnode from being reclaimed.
490  * the buffer cache may have references on the vnode, a directory
491  * vnode may still have references due to the namei cache representing
492  * underlying files, or the vnode may be in active use.   It is not
493  * desireable to reuse such vnodes.  These conditions may cause the
494  * number of vnodes to reach some minimum value regardless of what
495  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
496  */
497 static int
498 vlrureclaim(struct mount *mp)
499 {
500 	struct vnode *vp;
501 	lwkt_tokref ilock;
502 	lwkt_tokref vlock;
503 	int done;
504 	int trigger;
505 	int usevnodes;
506 	int count;
507 
508 	/*
509 	 * Calculate the trigger point, don't allow user
510 	 * screwups to blow us up.   This prevents us from
511 	 * recycling vnodes with lots of resident pages.  We
512 	 * aren't trying to free memory, we are trying to
513 	 * free vnodes.
514 	 */
515 	usevnodes = desiredvnodes;
516 	if (usevnodes <= 0)
517 		usevnodes = 1;
518 	trigger = vmstats.v_page_count * 2 / usevnodes;
519 
520 	done = 0;
521 	lwkt_gettoken(&ilock, &mntvnode_token);
522 	count = mp->mnt_nvnodelistsize / 10 + 1;
523 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
524 		/*
525 		 * __VNODESCAN__
526 		 *
527 		 * The VP will stick around while we hold mntvnode_token,
528 		 * at least until we block, so we can safely do an initial
529 		 * check.  But we have to check again after obtaining
530 		 * the vnode interlock.  vp->v_interlock points to stable
531 		 * storage so it's ok if the vp gets ripped out from
532 		 * under us while we are blocked.
533 		 */
534 		if (vp->v_type == VNON ||
535 		    vp->v_type == VBAD ||
536 		    !VMIGHTFREE(vp) ||		/* critical path opt */
537 		    (vp->v_object &&
538 		     vp->v_object->resident_page_count >= trigger)
539 		) {
540 			TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
541 			TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
542 			--count;
543 			continue;
544 		}
545 
546 		/*
547 		 * Get the interlock, delay moving the node to the tail so
548 		 * we don't race against new additions to the mountlist.
549 		 */
550 		lwkt_gettoken(&vlock, vp->v_interlock);
551 		if (TAILQ_FIRST(&mp->mnt_nvnodelist) != vp) {
552 			lwkt_reltoken(&vlock);
553 			continue;
554 		}
555 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
556 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
557 
558 		/*
559 		 * Must check again
560 		 */
561 		if (vp->v_type == VNON ||
562 		    vp->v_type == VBAD ||
563 		    !VMIGHTFREE(vp) ||		/* critical path opt */
564 		    (vp->v_object &&
565 		     vp->v_object->resident_page_count >= trigger)
566 		) {
567 			lwkt_reltoken(&vlock);
568 			--count;
569 			continue;
570 		}
571 		vgonel(vp, &vlock, curthread);
572 		++done;
573 		--count;
574 	}
575 	lwkt_reltoken(&ilock);
576 	return done;
577 }
578 
579 /*
580  * Attempt to recycle vnodes in a context that is always safe to block.
581  * Calling vlrurecycle() from the bowels of file system code has some
582  * interesting deadlock problems.
583  */
584 static struct thread *vnlruthread;
585 static int vnlruproc_sig;
586 
587 static void
588 vnlru_proc(void)
589 {
590 	struct mount *mp, *nmp;
591 	lwkt_tokref ilock;
592 	int s;
593 	int done;
594 	struct thread *td = curthread;
595 
596 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
597 	    SHUTDOWN_PRI_FIRST);
598 
599 	s = splbio();
600 	for (;;) {
601 		kproc_suspend_loop();
602 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
603 			vnlruproc_sig = 0;
604 			wakeup(&vnlruproc_sig);
605 			tsleep(td, 0, "vlruwt", hz);
606 			continue;
607 		}
608 		done = 0;
609 		lwkt_gettoken(&ilock, &mountlist_token);
610 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
611 			if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
612 				nmp = TAILQ_NEXT(mp, mnt_list);
613 				continue;
614 			}
615 			done += vlrureclaim(mp);
616 			lwkt_gettokref(&ilock);
617 			nmp = TAILQ_NEXT(mp, mnt_list);
618 			vfs_unbusy(mp, td);
619 		}
620 		lwkt_reltoken(&ilock);
621 		if (done == 0) {
622 			vnlru_nowhere++;
623 			tsleep(td, 0, "vlrup", hz * 3);
624 		}
625 	}
626 	splx(s);
627 }
628 
629 static struct kproc_desc vnlru_kp = {
630 	"vnlru",
631 	vnlru_proc,
632 	&vnlruthread
633 };
634 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
635 
636 /*
637  * Routines having to do with the management of the vnode table.
638  */
639 extern vop_t **dead_vnodeop_p;
640 
641 /*
642  * Return the next vnode from the free list.
643  */
644 int
645 getnewvnode(tag, mp, vops, vpp)
646 	enum vtagtype tag;
647 	struct mount *mp;
648 	vop_t **vops;
649 	struct vnode **vpp;
650 {
651 	int s;
652 	struct thread *td = curthread;	/* XXX */
653 	struct vnode *vp = NULL;
654 	struct vnode *xvp;
655 	vm_object_t object;
656 	lwkt_tokref ilock;
657 	lwkt_tokref vlock;
658 
659 	s = splbio();
660 
661 	/*
662 	 * Try to reuse vnodes if we hit the max.  This situation only
663 	 * occurs in certain large-memory (2G+) situations.  We cannot
664 	 * attempt to directly reclaim vnodes due to nasty recursion
665 	 * problems.
666 	 */
667 	while (numvnodes - freevnodes > desiredvnodes) {
668 		if (vnlruproc_sig == 0) {
669 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
670 			wakeup(vnlruthread);
671 		}
672 		tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
673 	}
674 
675 
676 	/*
677 	 * Attempt to reuse a vnode already on the free list, allocating
678 	 * a new vnode if we can't find one or if we have not reached a
679 	 * good minimum for good LRU performance.
680 	 */
681 	lwkt_gettoken(&ilock, &vnode_free_list_token);
682 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
683 		int count;
684 
685 		for (count = 0; count < freevnodes; count++) {
686 			/*
687 			 * __VNODESCAN__
688 			 *
689 			 * Pull the next vnode off the free list and do some
690 			 * sanity checks.  Note that regardless of how we
691 			 * block, if freevnodes is non-zero there had better
692 			 * be something on the list.
693 			 */
694 			vp = TAILQ_FIRST(&vnode_free_list);
695 			if (vp == NULL)
696 				panic("getnewvnode: free vnode isn't");
697 
698 			/*
699 			 * Move the vnode to the end of the list so other
700 			 * processes do not double-block trying to recycle
701 			 * the same vnode (as an optimization), then get
702 			 * the interlock.
703 			 */
704 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
705 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
706 
707 			/*
708 			 * Skip vnodes that are in the process of being
709 			 * held or referenced.  Since the act of adding or
710 			 * removing a vnode on the freelist requires a token
711 			 * and may block, the ref count may be adjusted
712 			 * prior to its addition or removal.
713 			 */
714 			if (VSHOULDBUSY(vp)) {
715 				vp = NULL;
716 				continue;
717 			}
718 
719 
720 			/*
721 			 * Obtain the vnode interlock and check that the
722 			 * vnode is still on the free list.
723 			 *
724 			 * This normally devolves into a degenerate case so
725 			 * it is optimal.   Loop up if it isn't.  Note that
726 			 * the vnode could be in the middle of being moved
727 			 * off the free list (the VSHOULDBUSY() check) and
728 			 * must be skipped if so.
729 			 */
730 			lwkt_gettoken(&vlock, vp->v_interlock);
731 			TAILQ_FOREACH_REVERSE(xvp, &vnode_free_list,
732 			    freelst, v_freelist) {
733 				if (vp == xvp)
734 					break;
735 			}
736 			if (vp != xvp || VSHOULDBUSY(vp)) {
737 				vp = NULL;
738 				continue;
739 			}
740 
741 			/*
742 			 * We now safely own the vnode.  If the vnode has
743 			 * an object do not recycle it if its VM object
744 			 * has resident pages or references.
745 			 */
746 			if ((VOP_GETVOBJECT(vp, &object) == 0 &&
747 			    (object->resident_page_count || object->ref_count))
748 			) {
749 				lwkt_reltoken(&vlock);
750 				vp = NULL;
751 				continue;
752 			}
753 
754 			/*
755 			 * We can almost reuse this vnode.  But we don't want
756 			 * to recycle it if the vnode has children in the
757 			 * namecache because that breaks the namecache's
758 			 * path element chain.  (YYY use nc_refs for the
759 			 * check?)
760 			 */
761 			KKASSERT(vp->v_flag & VFREE);
762 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
763 
764 			if (TAILQ_FIRST(&vp->v_namecache) == NULL ||
765 			    cache_leaf_test(vp) >= 0) {
766 				/* ok, we can reuse this vnode */
767 				break;
768 			}
769 			lwkt_reltoken(&vlock);
770 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
771 			vp = NULL;
772 		}
773 	}
774 
775 	/*
776 	 * If vp is non-NULL we hold it's interlock.
777 	 */
778 	if (vp) {
779 		vp->v_flag |= VDOOMED;
780 		vp->v_flag &= ~VFREE;
781 		freevnodes--;
782 		lwkt_reltoken(&ilock);
783 		cache_purge(vp);	/* YYY may block */
784 		vp->v_lease = NULL;
785 		if (vp->v_type != VBAD) {
786 			vgonel(vp, &vlock, td);
787 		} else {
788 			lwkt_reltoken(&vlock);
789 		}
790 
791 #ifdef INVARIANTS
792 		{
793 			int s;
794 
795 			if (vp->v_data)
796 				panic("cleaned vnode isn't");
797 			s = splbio();
798 			if (vp->v_numoutput)
799 				panic("Clean vnode has pending I/O's");
800 			splx(s);
801 		}
802 #endif
803 		vp->v_flag = 0;
804 		vp->v_lastw = 0;
805 		vp->v_lasta = 0;
806 		vp->v_cstart = 0;
807 		vp->v_clen = 0;
808 		vp->v_socket = 0;
809 		vp->v_writecount = 0;	/* XXX */
810 	} else {
811 		lwkt_reltoken(&ilock);
812 		vp = zalloc(vnode_zone);
813 		bzero(vp, sizeof(*vp));
814 		vp->v_interlock = lwkt_token_pool_get(vp);
815 		lwkt_token_init(&vp->v_pollinfo.vpi_token);
816 		cache_purge(vp);
817 		TAILQ_INIT(&vp->v_namecache);
818 		numvnodes++;
819 	}
820 
821 	TAILQ_INIT(&vp->v_cleanblkhd);
822 	TAILQ_INIT(&vp->v_dirtyblkhd);
823 	vp->v_type = VNON;
824 	vp->v_tag = tag;
825 	vp->v_op = vops;
826 	insmntque(vp, mp);
827 	*vpp = vp;
828 	vp->v_usecount = 1;
829 	vp->v_data = 0;
830 	splx(s);
831 
832 	vfs_object_create(vp, td);
833 	return (0);
834 }
835 
836 /*
837  * Move a vnode from one mount queue to another.
838  */
839 static void
840 insmntque(vp, mp)
841 	struct vnode *vp;
842 	struct mount *mp;
843 {
844 	lwkt_tokref ilock;
845 
846 	lwkt_gettoken(&ilock, &mntvnode_token);
847 	/*
848 	 * Delete from old mount point vnode list, if on one.
849 	 */
850 	if (vp->v_mount != NULL) {
851 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
852 			("bad mount point vnode list size"));
853 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
854 		vp->v_mount->mnt_nvnodelistsize--;
855 	}
856 	/*
857 	 * Insert into list of vnodes for the new mount point, if available.
858 	 */
859 	if ((vp->v_mount = mp) == NULL) {
860 		lwkt_reltoken(&ilock);
861 		return;
862 	}
863 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
864 	mp->mnt_nvnodelistsize++;
865 	lwkt_reltoken(&ilock);
866 }
867 
868 /*
869  * Update outstanding I/O count and do wakeup if requested.
870  */
871 void
872 vwakeup(bp)
873 	struct buf *bp;
874 {
875 	struct vnode *vp;
876 
877 	bp->b_flags &= ~B_WRITEINPROG;
878 	if ((vp = bp->b_vp)) {
879 		vp->v_numoutput--;
880 		if (vp->v_numoutput < 0)
881 			panic("vwakeup: neg numoutput");
882 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
883 			vp->v_flag &= ~VBWAIT;
884 			wakeup((caddr_t) &vp->v_numoutput);
885 		}
886 	}
887 }
888 
889 /*
890  * Flush out and invalidate all buffers associated with a vnode.
891  * Called with the underlying object locked.
892  */
893 int
894 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
895 	int slpflag, int slptimeo)
896 {
897 	struct buf *bp;
898 	struct buf *nbp, *blist;
899 	int s, error;
900 	vm_object_t object;
901 	lwkt_tokref vlock;
902 
903 	if (flags & V_SAVE) {
904 		s = splbio();
905 		while (vp->v_numoutput) {
906 			vp->v_flag |= VBWAIT;
907 			error = tsleep((caddr_t)&vp->v_numoutput,
908 			    slpflag, "vinvlbuf", slptimeo);
909 			if (error) {
910 				splx(s);
911 				return (error);
912 			}
913 		}
914 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
915 			splx(s);
916 			if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
917 				return (error);
918 			s = splbio();
919 			if (vp->v_numoutput > 0 ||
920 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
921 				panic("vinvalbuf: dirty bufs");
922 		}
923 		splx(s);
924   	}
925 	s = splbio();
926 	for (;;) {
927 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
928 		if (!blist)
929 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
930 		if (!blist)
931 			break;
932 
933 		for (bp = blist; bp; bp = nbp) {
934 			nbp = TAILQ_NEXT(bp, b_vnbufs);
935 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
936 				error = BUF_TIMELOCK(bp,
937 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
938 				    "vinvalbuf", slpflag, slptimeo);
939 				if (error == ENOLCK)
940 					break;
941 				splx(s);
942 				return (error);
943 			}
944 			/*
945 			 * XXX Since there are no node locks for NFS, I
946 			 * believe there is a slight chance that a delayed
947 			 * write will occur while sleeping just above, so
948 			 * check for it.  Note that vfs_bio_awrite expects
949 			 * buffers to reside on a queue, while VOP_BWRITE and
950 			 * brelse do not.
951 			 */
952 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
953 				(flags & V_SAVE)) {
954 
955 				if (bp->b_vp == vp) {
956 					if (bp->b_flags & B_CLUSTEROK) {
957 						BUF_UNLOCK(bp);
958 						vfs_bio_awrite(bp);
959 					} else {
960 						bremfree(bp);
961 						bp->b_flags |= B_ASYNC;
962 						VOP_BWRITE(bp->b_vp, bp);
963 					}
964 				} else {
965 					bremfree(bp);
966 					(void) VOP_BWRITE(bp->b_vp, bp);
967 				}
968 				break;
969 			}
970 			bremfree(bp);
971 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
972 			bp->b_flags &= ~B_ASYNC;
973 			brelse(bp);
974 		}
975 	}
976 
977 	/*
978 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
979 	 * have write I/O in-progress but if there is a VM object then the
980 	 * VM object can also have read-I/O in-progress.
981 	 */
982 	do {
983 		while (vp->v_numoutput > 0) {
984 			vp->v_flag |= VBWAIT;
985 			tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
986 		}
987 		if (VOP_GETVOBJECT(vp, &object) == 0) {
988 			while (object->paging_in_progress)
989 				vm_object_pip_sleep(object, "vnvlbx");
990 		}
991 	} while (vp->v_numoutput > 0);
992 
993 	splx(s);
994 
995 	/*
996 	 * Destroy the copy in the VM cache, too.
997 	 */
998 	lwkt_gettoken(&vlock, vp->v_interlock);
999 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1000 		vm_object_page_remove(object, 0, 0,
1001 			(flags & V_SAVE) ? TRUE : FALSE);
1002 	}
1003 	lwkt_reltoken(&vlock);
1004 
1005 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1006 		panic("vinvalbuf: flush failed");
1007 	return (0);
1008 }
1009 
1010 /*
1011  * Truncate a file's buffer and pages to a specified length.  This
1012  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1013  * sync activity.
1014  */
1015 int
1016 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
1017 {
1018 	struct buf *bp;
1019 	struct buf *nbp;
1020 	int s, anyfreed;
1021 	int trunclbn;
1022 
1023 	/*
1024 	 * Round up to the *next* lbn.
1025 	 */
1026 	trunclbn = (length + blksize - 1) / blksize;
1027 
1028 	s = splbio();
1029 restart:
1030 	anyfreed = 1;
1031 	for (;anyfreed;) {
1032 		anyfreed = 0;
1033 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1034 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1035 			if (bp->b_lblkno >= trunclbn) {
1036 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1037 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1038 					goto restart;
1039 				} else {
1040 					bremfree(bp);
1041 					bp->b_flags |= (B_INVAL | B_RELBUF);
1042 					bp->b_flags &= ~B_ASYNC;
1043 					brelse(bp);
1044 					anyfreed = 1;
1045 				}
1046 				if (nbp &&
1047 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1048 				    (nbp->b_vp != vp) ||
1049 				    (nbp->b_flags & B_DELWRI))) {
1050 					goto restart;
1051 				}
1052 			}
1053 		}
1054 
1055 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1056 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1057 			if (bp->b_lblkno >= trunclbn) {
1058 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1059 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1060 					goto restart;
1061 				} else {
1062 					bremfree(bp);
1063 					bp->b_flags |= (B_INVAL | B_RELBUF);
1064 					bp->b_flags &= ~B_ASYNC;
1065 					brelse(bp);
1066 					anyfreed = 1;
1067 				}
1068 				if (nbp &&
1069 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1070 				    (nbp->b_vp != vp) ||
1071 				    (nbp->b_flags & B_DELWRI) == 0)) {
1072 					goto restart;
1073 				}
1074 			}
1075 		}
1076 	}
1077 
1078 	if (length > 0) {
1079 restartsync:
1080 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1081 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1082 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1083 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1084 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1085 					goto restart;
1086 				} else {
1087 					bremfree(bp);
1088 					if (bp->b_vp == vp) {
1089 						bp->b_flags |= B_ASYNC;
1090 					} else {
1091 						bp->b_flags &= ~B_ASYNC;
1092 					}
1093 					VOP_BWRITE(bp->b_vp, bp);
1094 				}
1095 				goto restartsync;
1096 			}
1097 
1098 		}
1099 	}
1100 
1101 	while (vp->v_numoutput > 0) {
1102 		vp->v_flag |= VBWAIT;
1103 		tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1104 	}
1105 
1106 	splx(s);
1107 
1108 	vnode_pager_setsize(vp, length);
1109 
1110 	return (0);
1111 }
1112 
1113 /*
1114  * Associate a buffer with a vnode.
1115  */
1116 void
1117 bgetvp(vp, bp)
1118 	struct vnode *vp;
1119 	struct buf *bp;
1120 {
1121 	int s;
1122 
1123 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1124 
1125 	vhold(vp);
1126 	bp->b_vp = vp;
1127 	bp->b_dev = vn_todev(vp);
1128 	/*
1129 	 * Insert onto list for new vnode.
1130 	 */
1131 	s = splbio();
1132 	bp->b_xflags |= BX_VNCLEAN;
1133 	bp->b_xflags &= ~BX_VNDIRTY;
1134 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1135 	splx(s);
1136 }
1137 
1138 /*
1139  * Disassociate a buffer from a vnode.
1140  */
1141 void
1142 brelvp(bp)
1143 	struct buf *bp;
1144 {
1145 	struct vnode *vp;
1146 	struct buflists *listheadp;
1147 	int s;
1148 
1149 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1150 
1151 	/*
1152 	 * Delete from old vnode list, if on one.
1153 	 */
1154 	vp = bp->b_vp;
1155 	s = splbio();
1156 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1157 		if (bp->b_xflags & BX_VNDIRTY)
1158 			listheadp = &vp->v_dirtyblkhd;
1159 		else
1160 			listheadp = &vp->v_cleanblkhd;
1161 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1162 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1163 	}
1164 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1165 		vp->v_flag &= ~VONWORKLST;
1166 		LIST_REMOVE(vp, v_synclist);
1167 	}
1168 	splx(s);
1169 	bp->b_vp = (struct vnode *) 0;
1170 	vdrop(vp);
1171 }
1172 
1173 /*
1174  * The workitem queue.
1175  *
1176  * It is useful to delay writes of file data and filesystem metadata
1177  * for tens of seconds so that quickly created and deleted files need
1178  * not waste disk bandwidth being created and removed. To realize this,
1179  * we append vnodes to a "workitem" queue. When running with a soft
1180  * updates implementation, most pending metadata dependencies should
1181  * not wait for more than a few seconds. Thus, mounted on block devices
1182  * are delayed only about a half the time that file data is delayed.
1183  * Similarly, directory updates are more critical, so are only delayed
1184  * about a third the time that file data is delayed. Thus, there are
1185  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1186  * one each second (driven off the filesystem syncer process). The
1187  * syncer_delayno variable indicates the next queue that is to be processed.
1188  * Items that need to be processed soon are placed in this queue:
1189  *
1190  *	syncer_workitem_pending[syncer_delayno]
1191  *
1192  * A delay of fifteen seconds is done by placing the request fifteen
1193  * entries later in the queue:
1194  *
1195  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1196  *
1197  */
1198 
1199 /*
1200  * Add an item to the syncer work queue.
1201  */
1202 static void
1203 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1204 {
1205 	int s, slot;
1206 
1207 	s = splbio();
1208 
1209 	if (vp->v_flag & VONWORKLST) {
1210 		LIST_REMOVE(vp, v_synclist);
1211 	}
1212 
1213 	if (delay > syncer_maxdelay - 2)
1214 		delay = syncer_maxdelay - 2;
1215 	slot = (syncer_delayno + delay) & syncer_mask;
1216 
1217 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1218 	vp->v_flag |= VONWORKLST;
1219 	splx(s);
1220 }
1221 
1222 struct  thread *updatethread;
1223 static void sched_sync (void);
1224 static struct kproc_desc up_kp = {
1225 	"syncer",
1226 	sched_sync,
1227 	&updatethread
1228 };
1229 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1230 
1231 /*
1232  * System filesystem synchronizer daemon.
1233  */
1234 void
1235 sched_sync(void)
1236 {
1237 	struct synclist *slp;
1238 	struct vnode *vp;
1239 	long starttime;
1240 	int s;
1241 	struct thread *td = curthread;
1242 
1243 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1244 	    SHUTDOWN_PRI_LAST);
1245 
1246 	for (;;) {
1247 		kproc_suspend_loop();
1248 
1249 		starttime = time_second;
1250 
1251 		/*
1252 		 * Push files whose dirty time has expired.  Be careful
1253 		 * of interrupt race on slp queue.
1254 		 */
1255 		s = splbio();
1256 		slp = &syncer_workitem_pending[syncer_delayno];
1257 		syncer_delayno += 1;
1258 		if (syncer_delayno == syncer_maxdelay)
1259 			syncer_delayno = 0;
1260 		splx(s);
1261 
1262 		while ((vp = LIST_FIRST(slp)) != NULL) {
1263 			if (VOP_ISLOCKED(vp, NULL) == 0) {
1264 				vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
1265 				(void) VOP_FSYNC(vp, MNT_LAZY, td);
1266 				VOP_UNLOCK(vp, NULL, 0, td);
1267 			}
1268 			s = splbio();
1269 			if (LIST_FIRST(slp) == vp) {
1270 				/*
1271 				 * Note: v_tag VT_VFS vps can remain on the
1272 				 * worklist too with no dirty blocks, but
1273 				 * since sync_fsync() moves it to a different
1274 				 * slot we are safe.
1275 				 */
1276 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1277 				    !vn_isdisk(vp, NULL))
1278 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1279 				/*
1280 				 * Put us back on the worklist.  The worklist
1281 				 * routine will remove us from our current
1282 				 * position and then add us back in at a later
1283 				 * position.
1284 				 */
1285 				vn_syncer_add_to_worklist(vp, syncdelay);
1286 			}
1287 			splx(s);
1288 		}
1289 
1290 		/*
1291 		 * Do soft update processing.
1292 		 */
1293 		if (bioops.io_sync)
1294 			(*bioops.io_sync)(NULL);
1295 
1296 		/*
1297 		 * The variable rushjob allows the kernel to speed up the
1298 		 * processing of the filesystem syncer process. A rushjob
1299 		 * value of N tells the filesystem syncer to process the next
1300 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1301 		 * is used by the soft update code to speed up the filesystem
1302 		 * syncer process when the incore state is getting so far
1303 		 * ahead of the disk that the kernel memory pool is being
1304 		 * threatened with exhaustion.
1305 		 */
1306 		if (rushjob > 0) {
1307 			rushjob -= 1;
1308 			continue;
1309 		}
1310 		/*
1311 		 * If it has taken us less than a second to process the
1312 		 * current work, then wait. Otherwise start right over
1313 		 * again. We can still lose time if any single round
1314 		 * takes more than two seconds, but it does not really
1315 		 * matter as we are just trying to generally pace the
1316 		 * filesystem activity.
1317 		 */
1318 		if (time_second == starttime)
1319 			tsleep(&lbolt, 0, "syncer", 0);
1320 	}
1321 }
1322 
1323 /*
1324  * Request the syncer daemon to speed up its work.
1325  * We never push it to speed up more than half of its
1326  * normal turn time, otherwise it could take over the cpu.
1327  *
1328  * YYY wchan field protected by the BGL.
1329  */
1330 int
1331 speedup_syncer()
1332 {
1333 	crit_enter();
1334 	if (updatethread->td_wchan == &lbolt) { /* YYY */
1335 		unsleep(updatethread);
1336 		lwkt_schedule(updatethread);
1337 	}
1338 	crit_exit();
1339 	if (rushjob < syncdelay / 2) {
1340 		rushjob += 1;
1341 		stat_rush_requests += 1;
1342 		return (1);
1343 	}
1344 	return(0);
1345 }
1346 
1347 /*
1348  * Associate a p-buffer with a vnode.
1349  *
1350  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1351  * with the buffer.  i.e. the bp has not been linked into the vnode or
1352  * ref-counted.
1353  */
1354 void
1355 pbgetvp(vp, bp)
1356 	struct vnode *vp;
1357 	struct buf *bp;
1358 {
1359 
1360 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1361 
1362 	bp->b_vp = vp;
1363 	bp->b_flags |= B_PAGING;
1364 	bp->b_dev = vn_todev(vp);
1365 }
1366 
1367 /*
1368  * Disassociate a p-buffer from a vnode.
1369  */
1370 void
1371 pbrelvp(bp)
1372 	struct buf *bp;
1373 {
1374 
1375 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1376 
1377 	/* XXX REMOVE ME */
1378 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1379 		panic(
1380 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1381 		    bp,
1382 		    (int)bp->b_flags
1383 		);
1384 	}
1385 	bp->b_vp = (struct vnode *) 0;
1386 	bp->b_flags &= ~B_PAGING;
1387 }
1388 
1389 void
1390 pbreassignbuf(bp, newvp)
1391 	struct buf *bp;
1392 	struct vnode *newvp;
1393 {
1394 	if ((bp->b_flags & B_PAGING) == 0) {
1395 		panic(
1396 		    "pbreassignbuf() on non phys bp %p",
1397 		    bp
1398 		);
1399 	}
1400 	bp->b_vp = newvp;
1401 }
1402 
1403 /*
1404  * Reassign a buffer from one vnode to another.
1405  * Used to assign file specific control information
1406  * (indirect blocks) to the vnode to which they belong.
1407  */
1408 void
1409 reassignbuf(bp, newvp)
1410 	struct buf *bp;
1411 	struct vnode *newvp;
1412 {
1413 	struct buflists *listheadp;
1414 	int delay;
1415 	int s;
1416 
1417 	if (newvp == NULL) {
1418 		printf("reassignbuf: NULL");
1419 		return;
1420 	}
1421 	++reassignbufcalls;
1422 
1423 	/*
1424 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1425 	 * is not fully linked in.
1426 	 */
1427 	if (bp->b_flags & B_PAGING)
1428 		panic("cannot reassign paging buffer");
1429 
1430 	s = splbio();
1431 	/*
1432 	 * Delete from old vnode list, if on one.
1433 	 */
1434 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1435 		if (bp->b_xflags & BX_VNDIRTY)
1436 			listheadp = &bp->b_vp->v_dirtyblkhd;
1437 		else
1438 			listheadp = &bp->b_vp->v_cleanblkhd;
1439 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1440 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1441 		if (bp->b_vp != newvp) {
1442 			vdrop(bp->b_vp);
1443 			bp->b_vp = NULL;	/* for clarification */
1444 		}
1445 	}
1446 	/*
1447 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1448 	 * of clean buffers.
1449 	 */
1450 	if (bp->b_flags & B_DELWRI) {
1451 		struct buf *tbp;
1452 
1453 		listheadp = &newvp->v_dirtyblkhd;
1454 		if ((newvp->v_flag & VONWORKLST) == 0) {
1455 			switch (newvp->v_type) {
1456 			case VDIR:
1457 				delay = dirdelay;
1458 				break;
1459 			case VCHR:
1460 			case VBLK:
1461 				if (newvp->v_specmountpoint != NULL) {
1462 					delay = metadelay;
1463 					break;
1464 				}
1465 				/* fall through */
1466 			default:
1467 				delay = filedelay;
1468 			}
1469 			vn_syncer_add_to_worklist(newvp, delay);
1470 		}
1471 		bp->b_xflags |= BX_VNDIRTY;
1472 		tbp = TAILQ_FIRST(listheadp);
1473 		if (tbp == NULL ||
1474 		    bp->b_lblkno == 0 ||
1475 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1476 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1477 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1478 			++reassignbufsortgood;
1479 		} else if (bp->b_lblkno < 0) {
1480 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1481 			++reassignbufsortgood;
1482 		} else if (reassignbufmethod == 1) {
1483 			/*
1484 			 * New sorting algorithm, only handle sequential case,
1485 			 * otherwise append to end (but before metadata)
1486 			 */
1487 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1488 			    (tbp->b_xflags & BX_VNDIRTY)) {
1489 				/*
1490 				 * Found the best place to insert the buffer
1491 				 */
1492 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1493 				++reassignbufsortgood;
1494 			} else {
1495 				/*
1496 				 * Missed, append to end, but before meta-data.
1497 				 * We know that the head buffer in the list is
1498 				 * not meta-data due to prior conditionals.
1499 				 *
1500 				 * Indirect effects:  NFS second stage write
1501 				 * tends to wind up here, giving maximum
1502 				 * distance between the unstable write and the
1503 				 * commit rpc.
1504 				 */
1505 				tbp = TAILQ_LAST(listheadp, buflists);
1506 				while (tbp && tbp->b_lblkno < 0)
1507 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1508 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1509 				++reassignbufsortbad;
1510 			}
1511 		} else {
1512 			/*
1513 			 * Old sorting algorithm, scan queue and insert
1514 			 */
1515 			struct buf *ttbp;
1516 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1517 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1518 				++reassignbufloops;
1519 				tbp = ttbp;
1520 			}
1521 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1522 		}
1523 	} else {
1524 		bp->b_xflags |= BX_VNCLEAN;
1525 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1526 		if ((newvp->v_flag & VONWORKLST) &&
1527 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1528 			newvp->v_flag &= ~VONWORKLST;
1529 			LIST_REMOVE(newvp, v_synclist);
1530 		}
1531 	}
1532 	if (bp->b_vp != newvp) {
1533 		bp->b_vp = newvp;
1534 		vhold(bp->b_vp);
1535 	}
1536 	splx(s);
1537 }
1538 
1539 /*
1540  * Create a vnode for a block device.
1541  * Used for mounting the root file system.
1542  */
1543 int
1544 bdevvp(dev, vpp)
1545 	dev_t dev;
1546 	struct vnode **vpp;
1547 {
1548 	struct vnode *vp;
1549 	struct vnode *nvp;
1550 	int error;
1551 
1552 	if (dev == NODEV) {
1553 		*vpp = NULLVP;
1554 		return (ENXIO);
1555 	}
1556 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1557 	if (error) {
1558 		*vpp = NULLVP;
1559 		return (error);
1560 	}
1561 	vp = nvp;
1562 	vp->v_type = VBLK;
1563 	addalias(vp, dev);
1564 	*vpp = vp;
1565 	return (0);
1566 }
1567 
1568 /*
1569  * Add a vnode to the alias list hung off the dev_t.
1570  *
1571  * The reason for this gunk is that multiple vnodes can reference
1572  * the same physical device, so checking vp->v_usecount to see
1573  * how many users there are is inadequate; the v_usecount for
1574  * the vnodes need to be accumulated.  vcount() does that.
1575  */
1576 void
1577 addaliasu(struct vnode *nvp, udev_t nvp_rdev)
1578 {
1579 	dev_t dev;
1580 
1581 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1582 		panic("addaliasu on non-special vnode");
1583 	dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
1584 	if (dev != NODEV) {
1585 		nvp->v_rdev = dev;
1586 		addalias(nvp, dev);
1587 	} else
1588 		nvp->v_rdev = NULL;
1589 }
1590 
1591 void
1592 addalias(struct vnode *nvp, dev_t dev)
1593 {
1594 	lwkt_tokref ilock;
1595 
1596 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1597 		panic("addalias on non-special vnode");
1598 
1599 	nvp->v_rdev = dev;
1600 	lwkt_gettoken(&ilock, &spechash_token);
1601 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1602 	lwkt_reltoken(&ilock);
1603 }
1604 
1605 /*
1606  * Grab a particular vnode from the free list, increment its
1607  * reference count and lock it. The vnode lock bit is set if the
1608  * vnode is being eliminated in vgone. The process is awakened
1609  * when the transition is completed, and an error returned to
1610  * indicate that the vnode is no longer usable (possibly having
1611  * been changed to a new file system type).
1612  *
1613  * This code is very sensitive.  We are depending on the vnode interlock
1614  * to be maintained through to the vn_lock() call, which means that we
1615  * cannot block which means that we cannot call vbusy() until after vn_lock().
1616  * If the interlock is not maintained, the VXLOCK check will not properly
1617  * interlock against a vclean()'s LK_DRAIN operation on the lock.
1618  */
1619 int
1620 vget(struct vnode *vp, lwkt_tokref_t vlock, int flags, thread_t td)
1621 {
1622 	int error;
1623 	lwkt_tokref vvlock;
1624 
1625 	/*
1626 	 * We need the interlock to safely modify the v_ fields.  ZZZ it is
1627 	 * only legal to pass (1) the vnode's interlock and (2) only pass
1628 	 * NULL w/o LK_INTERLOCK if the vnode is *ALREADY* referenced or
1629 	 * held.
1630 	 */
1631 	if ((flags & LK_INTERLOCK) == 0) {
1632 		lwkt_gettoken(&vvlock, vp->v_interlock);
1633 		vlock = &vvlock;
1634 	}
1635 
1636 	/*
1637 	 * If the vnode is in the process of being cleaned out for
1638 	 * another use, we wait for the cleaning to finish and then
1639 	 * return failure. Cleaning is determined by checking that
1640 	 * the VXLOCK flag is set.  It is possible for the vnode to be
1641 	 * self-referenced during the cleaning operation.
1642 	 */
1643 	if (vp->v_flag & VXLOCK) {
1644 		if (vp->v_vxthread == curthread) {
1645 #if 0
1646 			/* this can now occur in normal operation */
1647 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1648 #endif
1649 		} else {
1650 			vp->v_flag |= VXWANT;
1651 			lwkt_reltoken(vlock);
1652 			tsleep((caddr_t)vp, 0, "vget", 0);
1653 			return (ENOENT);
1654 		}
1655 	}
1656 
1657 	/*
1658 	 * Bump v_usecount to prevent the vnode from being recycled.  The
1659 	 * usecount needs to be bumped before we successfully get our lock.
1660 	 */
1661 	vp->v_usecount++;
1662 	if (flags & LK_TYPE_MASK) {
1663 		if ((error = vn_lock(vp, vlock, flags | LK_INTERLOCK, td)) != 0) {
1664 			/*
1665 			 * must expand vrele here because we do not want
1666 			 * to call VOP_INACTIVE if the reference count
1667 			 * drops back to zero since it was never really
1668 			 * active. We must remove it from the free list
1669 			 * before sleeping so that multiple processes do
1670 			 * not try to recycle it.
1671 			 */
1672 			lwkt_gettokref(vlock);
1673 			vp->v_usecount--;
1674 			vmaybefree(vp);
1675 			lwkt_reltoken(vlock);
1676 		}
1677 		return (error);
1678 	}
1679 	if (VSHOULDBUSY(vp))
1680 		vbusy(vp);	/* interlock must be held on call */
1681 	lwkt_reltoken(vlock);
1682 	return (0);
1683 }
1684 
1685 void
1686 vref(struct vnode *vp)
1687 {
1688 	crit_enter();	/* YYY use crit section for moment / BGL protected */
1689 	vp->v_usecount++;
1690 	crit_exit();
1691 }
1692 
1693 /*
1694  * Vnode put/release.
1695  * If count drops to zero, call inactive routine and return to freelist.
1696  */
1697 void
1698 vrele(struct vnode *vp)
1699 {
1700 	struct thread *td = curthread;	/* XXX */
1701 	lwkt_tokref vlock;
1702 
1703 	KASSERT(vp != NULL && vp->v_usecount >= 0,
1704 	    ("vrele: null vp or <=0 v_usecount"));
1705 
1706 	lwkt_gettoken(&vlock, vp->v_interlock);
1707 
1708 	if (vp->v_usecount > 1) {
1709 		vp->v_usecount--;
1710 		lwkt_reltoken(&vlock);
1711 		return;
1712 	}
1713 
1714 	if (vp->v_usecount == 1) {
1715 		vp->v_usecount--;
1716 		/*
1717 		 * We must call VOP_INACTIVE with the node locked and the
1718 		 * usecount 0.  If we are doing a vpu, the node is already
1719 		 * locked, but, in the case of vrele, we must explicitly lock
1720 		 * the vnode before calling VOP_INACTIVE.
1721 		 */
1722 
1723 		if (vn_lock(vp, NULL, LK_EXCLUSIVE, td) == 0)
1724 			VOP_INACTIVE(vp, td);
1725 		vmaybefree(vp);
1726 		lwkt_reltoken(&vlock);
1727 	} else {
1728 #ifdef DIAGNOSTIC
1729 		vprint("vrele: negative ref count", vp);
1730 #endif
1731 		lwkt_reltoken(&vlock);
1732 		panic("vrele: negative ref cnt");
1733 	}
1734 }
1735 
1736 void
1737 vput(struct vnode *vp)
1738 {
1739 	struct thread *td = curthread;	/* XXX */
1740 	lwkt_tokref vlock;
1741 
1742 	KASSERT(vp != NULL, ("vput: null vp"));
1743 
1744 	lwkt_gettoken(&vlock, vp->v_interlock);
1745 
1746 	if (vp->v_usecount > 1) {
1747 		vp->v_usecount--;
1748 		VOP_UNLOCK(vp, &vlock, LK_INTERLOCK, td);
1749 		return;
1750 	}
1751 
1752 	if (vp->v_usecount == 1) {
1753 		vp->v_usecount--;
1754 		/*
1755 		 * We must call VOP_INACTIVE with the node locked.
1756 		 * If we are doing a vpu, the node is already locked,
1757 		 * so we just need to release the vnode mutex.
1758 		 */
1759 		VOP_INACTIVE(vp, td);
1760 		vmaybefree(vp);
1761 		lwkt_reltoken(&vlock);
1762 	} else {
1763 #ifdef DIAGNOSTIC
1764 		vprint("vput: negative ref count", vp);
1765 #endif
1766 		lwkt_reltoken(&vlock);
1767 		panic("vput: negative ref cnt");
1768 	}
1769 }
1770 
1771 /*
1772  * Somebody doesn't want the vnode recycled. ZZZ vnode interlock should
1773  * be held but isn't.
1774  */
1775 void
1776 vhold(vp)
1777 	struct vnode *vp;
1778 {
1779 	int s;
1780 
1781   	s = splbio();
1782 	vp->v_holdcnt++;
1783 	if (VSHOULDBUSY(vp))
1784 		vbusy(vp);	/* interlock must be held on call */
1785 	splx(s);
1786 }
1787 
1788 /*
1789  * One less who cares about this vnode.
1790  */
1791 void
1792 vdrop(vp)
1793 	struct vnode *vp;
1794 {
1795 	lwkt_tokref vlock;
1796 
1797 	lwkt_gettoken(&vlock, vp->v_interlock);
1798 	if (vp->v_holdcnt <= 0)
1799 		panic("vdrop: holdcnt");
1800 	vp->v_holdcnt--;
1801 	vmaybefree(vp);
1802 	lwkt_reltoken(&vlock);
1803 }
1804 
1805 int
1806 vmntvnodescan(
1807     struct mount *mp,
1808     int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1809     int (*slowfunc)(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data),
1810     void *data
1811 ) {
1812 	lwkt_tokref ilock;
1813 	lwkt_tokref vlock;
1814 	struct vnode *pvp;
1815 	struct vnode *vp;
1816 	int r = 0;
1817 
1818 	/*
1819 	 * Scan the vnodes on the mount's vnode list.  Use a placemarker
1820 	 */
1821 	pvp = zalloc(vnode_zone);
1822 	pvp->v_flag |= VPLACEMARKER;
1823 
1824 	lwkt_gettoken(&ilock, &mntvnode_token);
1825 	TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1826 
1827 	while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
1828 		/*
1829 		 * Move the placemarker and skip other placemarkers we
1830 		 * encounter.  The nothing can get in our way so the
1831 		 * mount point on the vp must be valid.
1832 		 */
1833 		TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1834 		TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
1835 		if (vp->v_flag & VPLACEMARKER)
1836 			continue;
1837 		KKASSERT(vp->v_mount == mp);
1838 
1839 		/*
1840 		 * Quick test
1841 		 */
1842 		if (fastfunc) {
1843 			if ((r = fastfunc(mp, vp, data)) < 0)
1844 				continue;
1845 			if (r)
1846 				break;
1847 		}
1848 
1849 		/*
1850 		 * Get the vnodes interlock and make sure it is still on the
1851 		 * mount list.  Skip it if it has moved (we may encounter it
1852 		 * later).  Then do the with-interlock test.  The callback
1853 		 * is responsible for releasing the vnode interlock.
1854 		 *
1855 		 * The interlock is type-stable.
1856 		 */
1857 		if (slowfunc) {
1858 			lwkt_gettoken(&vlock, vp->v_interlock);
1859 			if (vp != TAILQ_PREV(pvp, vnodelst, v_nmntvnodes)) {
1860 				printf("vmntvnodescan (debug info only): f=%p vp=%p vnode ripped out from under us\n", slowfunc, vp);
1861 				lwkt_reltoken(&vlock);
1862 				continue;
1863 			}
1864 			if ((r = slowfunc(mp, vp, &vlock, data)) != 0) {
1865 				KKASSERT(lwkt_havetokref(&vlock) == 0);
1866 				break;
1867 			}
1868 			KKASSERT(lwkt_havetokref(&vlock) == 0);
1869 		}
1870 	}
1871 	TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1872 	zfree(vnode_zone, pvp);
1873 	lwkt_reltoken(&ilock);
1874 	return(r);
1875 }
1876 
1877 /*
1878  * Remove any vnodes in the vnode table belonging to mount point mp.
1879  *
1880  * If FORCECLOSE is not specified, there should not be any active ones,
1881  * return error if any are found (nb: this is a user error, not a
1882  * system error). If FORCECLOSE is specified, detach any active vnodes
1883  * that are found.
1884  *
1885  * If WRITECLOSE is set, only flush out regular file vnodes open for
1886  * writing.
1887  *
1888  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1889  *
1890  * `rootrefs' specifies the base reference count for the root vnode
1891  * of this filesystem. The root vnode is considered busy if its
1892  * v_usecount exceeds this value. On a successful return, vflush()
1893  * will call vrele() on the root vnode exactly rootrefs times.
1894  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1895  * be zero.
1896  */
1897 #ifdef DIAGNOSTIC
1898 static int busyprt = 0;		/* print out busy vnodes */
1899 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1900 #endif
1901 
1902 static int vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data);
1903 
1904 struct vflush_info {
1905 	int flags;
1906 	int busy;
1907 	thread_t td;
1908 };
1909 
1910 int
1911 vflush(mp, rootrefs, flags)
1912 	struct mount *mp;
1913 	int rootrefs;
1914 	int flags;
1915 {
1916 	struct thread *td = curthread;	/* XXX */
1917 	struct vnode *rootvp = NULL;
1918 	int error;
1919 	lwkt_tokref vlock;
1920 	struct vflush_info vflush_info;
1921 
1922 	if (rootrefs > 0) {
1923 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1924 		    ("vflush: bad args"));
1925 		/*
1926 		 * Get the filesystem root vnode. We can vput() it
1927 		 * immediately, since with rootrefs > 0, it won't go away.
1928 		 */
1929 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1930 			return (error);
1931 		vput(rootvp);
1932 	}
1933 
1934 	vflush_info.busy = 0;
1935 	vflush_info.flags = flags;
1936 	vflush_info.td = td;
1937 	vmntvnodescan(mp, NULL, vflush_scan, &vflush_info);
1938 
1939 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1940 		/*
1941 		 * If just the root vnode is busy, and if its refcount
1942 		 * is equal to `rootrefs', then go ahead and kill it.
1943 		 */
1944 		lwkt_gettoken(&vlock, rootvp->v_interlock);
1945 		KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1946 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1947 		if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
1948 			vgonel(rootvp, &vlock, td);
1949 			vflush_info.busy = 0;
1950 		} else {
1951 			lwkt_reltoken(&vlock);
1952 		}
1953 	}
1954 	if (vflush_info.busy)
1955 		return (EBUSY);
1956 	for (; rootrefs > 0; rootrefs--)
1957 		vrele(rootvp);
1958 	return (0);
1959 }
1960 
1961 /*
1962  * The scan callback is made with an interlocked vnode.
1963  */
1964 static int
1965 vflush_scan(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
1966 {
1967 	struct vflush_info *info = data;
1968 	struct vattr vattr;
1969 
1970 	/*
1971 	 * Skip over a vnodes marked VSYSTEM.
1972 	 */
1973 	if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1974 		lwkt_reltoken(vlock);
1975 		return(0);
1976 	}
1977 
1978 	/*
1979 	 * If WRITECLOSE is set, flush out unlinked but still open
1980 	 * files (even if open only for reading) and regular file
1981 	 * vnodes open for writing.
1982 	 */
1983 	if ((info->flags & WRITECLOSE) &&
1984 	    (vp->v_type == VNON ||
1985 	    (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
1986 	    vattr.va_nlink > 0)) &&
1987 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1988 		lwkt_reltoken(vlock);
1989 		return(0);
1990 	}
1991 
1992 	/*
1993 	 * With v_usecount == 0, all we need to do is clear out the
1994 	 * vnode data structures and we are done.
1995 	 */
1996 	if (vp->v_usecount == 0) {
1997 		vgonel(vp, vlock, info->td);
1998 		return(0);
1999 	}
2000 
2001 	/*
2002 	 * If FORCECLOSE is set, forcibly close the vnode. For block
2003 	 * or character devices, revert to an anonymous device. For
2004 	 * all other files, just kill them.
2005 	 */
2006 	if (info->flags & FORCECLOSE) {
2007 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
2008 			vgonel(vp, vlock, info->td);
2009 		} else {
2010 			vclean(vp, vlock, 0, info->td);
2011 			vp->v_op = spec_vnodeop_p;
2012 			insmntque(vp, (struct mount *) 0);
2013 		}
2014 		return(0);
2015 	}
2016 #ifdef DIAGNOSTIC
2017 	if (busyprt)
2018 		vprint("vflush: busy vnode", vp);
2019 #endif
2020 	lwkt_reltoken(vlock);
2021 	++info->busy;
2022 	return(0);
2023 }
2024 
2025 /*
2026  * Disassociate the underlying file system from a vnode.
2027  */
2028 static void
2029 vclean(struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td)
2030 {
2031 	int active;
2032 
2033 	/*
2034 	 * Check to see if the vnode is in use. If so we have to reference it
2035 	 * before we clean it out so that its count cannot fall to zero and
2036 	 * generate a race against ourselves to recycle it.
2037 	 */
2038 	if ((active = vp->v_usecount))
2039 		vp->v_usecount++;
2040 
2041 	/*
2042 	 * Prevent the vnode from being recycled or brought into use while we
2043 	 * clean it out.
2044 	 */
2045 	if (vp->v_flag & VXLOCK)
2046 		panic("vclean: deadlock");
2047 	vp->v_flag |= VXLOCK;
2048 	vp->v_vxthread = curthread;
2049 
2050 	/*
2051 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2052 	 * have the object locked while it cleans it out. The VOP_LOCK
2053 	 * ensures that the VOP_INACTIVE routine is done with its work.
2054 	 * For active vnodes, it ensures that no other activity can
2055 	 * occur while the underlying object is being cleaned out.
2056 	 *
2057 	 * NOTE: we continue to hold the vnode interlock through to the
2058 	 * end of vclean().
2059 	 */
2060 	VOP_LOCK(vp, NULL, LK_DRAIN, td);
2061 
2062 	/*
2063 	 * Clean out any buffers associated with the vnode.
2064 	 */
2065 	vinvalbuf(vp, V_SAVE, td, 0, 0);
2066 	VOP_DESTROYVOBJECT(vp);
2067 
2068 	/*
2069 	 * If purging an active vnode, it must be closed and
2070 	 * deactivated before being reclaimed. Note that the
2071 	 * VOP_INACTIVE will unlock the vnode.
2072 	 */
2073 	if (active) {
2074 		if (flags & DOCLOSE)
2075 			VOP_CLOSE(vp, FNONBLOCK, td);
2076 		VOP_INACTIVE(vp, td);
2077 	} else {
2078 		/*
2079 		 * Any other processes trying to obtain this lock must first
2080 		 * wait for VXLOCK to clear, then call the new lock operation.
2081 		 */
2082 		VOP_UNLOCK(vp, NULL, 0, td);
2083 	}
2084 	/*
2085 	 * Reclaim the vnode.
2086 	 */
2087 	if (VOP_RECLAIM(vp, td))
2088 		panic("vclean: cannot reclaim");
2089 
2090 	if (active) {
2091 		/*
2092 		 * Inline copy of vrele() since VOP_INACTIVE
2093 		 * has already been called.
2094 		 */
2095 		if (--vp->v_usecount <= 0) {
2096 #ifdef DIAGNOSTIC
2097 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2098 				vprint("vclean: bad ref count", vp);
2099 				panic("vclean: ref cnt");
2100 			}
2101 #endif
2102 			vfree(vp);
2103 		}
2104 	}
2105 
2106 	cache_purge(vp);
2107 	vp->v_vnlock = NULL;
2108 	vmaybefree(vp);
2109 
2110 	/*
2111 	 * Done with purge, notify sleepers of the grim news.
2112 	 */
2113 	vp->v_op = dead_vnodeop_p;
2114 	vn_pollgone(vp);
2115 	vp->v_tag = VT_NON;
2116 	vp->v_flag &= ~VXLOCK;
2117 	vp->v_vxthread = NULL;
2118 	if (vp->v_flag & VXWANT) {
2119 		vp->v_flag &= ~VXWANT;
2120 		wakeup((caddr_t) vp);
2121 	}
2122 	lwkt_reltoken(vlock);
2123 }
2124 
2125 /*
2126  * Eliminate all activity associated with the requested vnode
2127  * and with all vnodes aliased to the requested vnode.
2128  */
2129 int
2130 vop_revoke(ap)
2131 	struct vop_revoke_args /* {
2132 		struct vnode *a_vp;
2133 		int a_flags;
2134 	} */ *ap;
2135 {
2136 	struct vnode *vp, *vq;
2137 	lwkt_tokref ilock;
2138 	dev_t dev;
2139 
2140 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2141 
2142 	vp = ap->a_vp;
2143 	/*
2144 	 * If a vgone (or vclean) is already in progress,
2145 	 * wait until it is done and return.
2146 	 */
2147 	if (vp->v_flag & VXLOCK) {
2148 		vp->v_flag |= VXWANT;
2149 		/*lwkt_reltoken(vlock); ZZZ */
2150 		tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2151 		return (0);
2152 	}
2153 	dev = vp->v_rdev;
2154 	for (;;) {
2155 		lwkt_gettoken(&ilock, &spechash_token);
2156 		vq = SLIST_FIRST(&dev->si_hlist);
2157 		lwkt_reltoken(&ilock);
2158 		if (!vq)
2159 			break;
2160 		vgone(vq);
2161 	}
2162 	return (0);
2163 }
2164 
2165 /*
2166  * Recycle an unused vnode to the front of the free list.
2167  * Release the passed interlock if the vnode will be recycled.
2168  */
2169 int
2170 vrecycle(struct vnode *vp, lwkt_tokref_t inter_lkp, struct thread *td)
2171 {
2172 	lwkt_tokref vlock;
2173 
2174 	lwkt_gettoken(&vlock, vp->v_interlock);
2175 	if (vp->v_usecount == 0) {
2176 		if (inter_lkp)
2177 			lwkt_reltoken(inter_lkp);
2178 		vgonel(vp, &vlock, td);
2179 		return (1);
2180 	}
2181 	lwkt_reltoken(&vlock);
2182 	return (0);
2183 }
2184 
2185 /*
2186  * Eliminate all activity associated with a vnode
2187  * in preparation for reuse.
2188  */
2189 void
2190 vgone(struct vnode *vp)
2191 {
2192 	struct thread *td = curthread;	/* XXX */
2193 	lwkt_tokref vlock;
2194 
2195 	lwkt_gettoken(&vlock, vp->v_interlock);
2196 	vgonel(vp, &vlock, td);
2197 }
2198 
2199 /*
2200  * vgone, with the vp interlock held.
2201  */
2202 void
2203 vgonel(struct vnode *vp, lwkt_tokref_t vlock, struct thread *td)
2204 {
2205 	lwkt_tokref ilock;
2206 	int s;
2207 
2208 	/*
2209 	 * If a vgone (or vclean) is already in progress,
2210 	 * wait until it is done and return.
2211 	 */
2212 	if (vp->v_flag & VXLOCK) {
2213 		vp->v_flag |= VXWANT;
2214 		lwkt_reltoken(vlock);
2215 		tsleep((caddr_t)vp, 0, "vgone", 0);
2216 		return;
2217 	}
2218 
2219 	/*
2220 	 * Clean out the filesystem specific data.
2221 	 */
2222 	vclean(vp, vlock, DOCLOSE, td);
2223 	lwkt_gettokref(vlock);
2224 
2225 	/*
2226 	 * Delete from old mount point vnode list, if on one.
2227 	 */
2228 	if (vp->v_mount != NULL)
2229 		insmntque(vp, (struct mount *)0);
2230 	/*
2231 	 * If special device, remove it from special device alias list
2232 	 * if it is on one.
2233 	 */
2234 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2235 		lwkt_gettoken(&ilock, &spechash_token);
2236 		SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext);
2237 		freedev(vp->v_rdev);
2238 		lwkt_reltoken(&ilock);
2239 		vp->v_rdev = NULL;
2240 	}
2241 
2242 	/*
2243 	 * If it is on the freelist and not already at the head,
2244 	 * move it to the head of the list. The test of the
2245 	 * VDOOMED flag and the reference count of zero is because
2246 	 * it will be removed from the free list by getnewvnode,
2247 	 * but will not have its reference count incremented until
2248 	 * after calling vgone. If the reference count were
2249 	 * incremented first, vgone would (incorrectly) try to
2250 	 * close the previous instance of the underlying object.
2251 	 */
2252 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2253 		s = splbio();
2254 		lwkt_gettoken(&ilock, &vnode_free_list_token);
2255 		if (vp->v_flag & VFREE)
2256 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2257 		else
2258 			freevnodes++;
2259 		vp->v_flag |= VFREE;
2260 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2261 		lwkt_reltoken(&ilock);
2262 		splx(s);
2263 	}
2264 	vp->v_type = VBAD;
2265 	lwkt_reltoken(vlock);
2266 }
2267 
2268 /*
2269  * Lookup a vnode by device number.
2270  */
2271 int
2272 vfinddev(dev, type, vpp)
2273 	dev_t dev;
2274 	enum vtype type;
2275 	struct vnode **vpp;
2276 {
2277 	lwkt_tokref ilock;
2278 	struct vnode *vp;
2279 
2280 	lwkt_gettoken(&ilock, &spechash_token);
2281 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2282 		if (type == vp->v_type) {
2283 			*vpp = vp;
2284 			lwkt_reltoken(&ilock);
2285 			return (1);
2286 		}
2287 	}
2288 	lwkt_reltoken(&ilock);
2289 	return (0);
2290 }
2291 
2292 /*
2293  * Calculate the total number of references to a special device.
2294  */
2295 int
2296 vcount(vp)
2297 	struct vnode *vp;
2298 {
2299 	lwkt_tokref ilock;
2300 	struct vnode *vq;
2301 	int count;
2302 
2303 	count = 0;
2304 	lwkt_gettoken(&ilock, &spechash_token);
2305 	SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext)
2306 		count += vq->v_usecount;
2307 	lwkt_reltoken(&ilock);
2308 	return (count);
2309 }
2310 
2311 /*
2312  * Same as above, but using the dev_t as argument
2313  */
2314 
2315 int
2316 count_dev(dev)
2317 	dev_t dev;
2318 {
2319 	struct vnode *vp;
2320 
2321 	vp = SLIST_FIRST(&dev->si_hlist);
2322 	if (vp == NULL)
2323 		return (0);
2324 	return(vcount(vp));
2325 }
2326 
2327 /*
2328  * Print out a description of a vnode.
2329  */
2330 static char *typename[] =
2331 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2332 
2333 void
2334 vprint(label, vp)
2335 	char *label;
2336 	struct vnode *vp;
2337 {
2338 	char buf[96];
2339 
2340 	if (label != NULL)
2341 		printf("%s: %p: ", label, (void *)vp);
2342 	else
2343 		printf("%p: ", (void *)vp);
2344 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2345 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2346 	    vp->v_holdcnt);
2347 	buf[0] = '\0';
2348 	if (vp->v_flag & VROOT)
2349 		strcat(buf, "|VROOT");
2350 	if (vp->v_flag & VTEXT)
2351 		strcat(buf, "|VTEXT");
2352 	if (vp->v_flag & VSYSTEM)
2353 		strcat(buf, "|VSYSTEM");
2354 	if (vp->v_flag & VXLOCK)
2355 		strcat(buf, "|VXLOCK");
2356 	if (vp->v_flag & VXWANT)
2357 		strcat(buf, "|VXWANT");
2358 	if (vp->v_flag & VBWAIT)
2359 		strcat(buf, "|VBWAIT");
2360 	if (vp->v_flag & VDOOMED)
2361 		strcat(buf, "|VDOOMED");
2362 	if (vp->v_flag & VFREE)
2363 		strcat(buf, "|VFREE");
2364 	if (vp->v_flag & VOBJBUF)
2365 		strcat(buf, "|VOBJBUF");
2366 	if (buf[0] != '\0')
2367 		printf(" flags (%s)", &buf[1]);
2368 	if (vp->v_data == NULL) {
2369 		printf("\n");
2370 	} else {
2371 		printf("\n\t");
2372 		VOP_PRINT(vp);
2373 	}
2374 }
2375 
2376 #ifdef DDB
2377 #include <ddb/ddb.h>
2378 /*
2379  * List all of the locked vnodes in the system.
2380  * Called when debugging the kernel.
2381  */
2382 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2383 {
2384 	struct thread *td = curthread;	/* XXX */
2385 	lwkt_tokref ilock;
2386 	struct mount *mp, *nmp;
2387 	struct vnode *vp;
2388 
2389 	printf("Locked vnodes\n");
2390 	lwkt_gettoken(&ilock, &mountlist_token);
2391 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2392 		if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
2393 			nmp = TAILQ_NEXT(mp, mnt_list);
2394 			continue;
2395 		}
2396 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2397 			if (VOP_ISLOCKED(vp, NULL))
2398 				vprint((char *)0, vp);
2399 		}
2400 		lwkt_gettokref(&ilock);
2401 		nmp = TAILQ_NEXT(mp, mnt_list);
2402 		vfs_unbusy(mp, td);
2403 	}
2404 	lwkt_reltoken(&ilock);
2405 }
2406 #endif
2407 
2408 /*
2409  * Top level filesystem related information gathering.
2410  */
2411 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2412 
2413 static int
2414 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2415 {
2416 	int *name = (int *)arg1 - 1;	/* XXX */
2417 	u_int namelen = arg2 + 1;	/* XXX */
2418 	struct vfsconf *vfsp;
2419 
2420 #if 1 || defined(COMPAT_PRELITE2)
2421 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2422 	if (namelen == 1)
2423 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2424 #endif
2425 
2426 #ifdef notyet
2427 	/* all sysctl names at this level are at least name and field */
2428 	if (namelen < 2)
2429 		return (ENOTDIR);		/* overloaded */
2430 	if (name[0] != VFS_GENERIC) {
2431 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2432 			if (vfsp->vfc_typenum == name[0])
2433 				break;
2434 		if (vfsp == NULL)
2435 			return (EOPNOTSUPP);
2436 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2437 		    oldp, oldlenp, newp, newlen, p));
2438 	}
2439 #endif
2440 	switch (name[1]) {
2441 	case VFS_MAXTYPENUM:
2442 		if (namelen != 2)
2443 			return (ENOTDIR);
2444 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2445 	case VFS_CONF:
2446 		if (namelen != 3)
2447 			return (ENOTDIR);	/* overloaded */
2448 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2449 			if (vfsp->vfc_typenum == name[2])
2450 				break;
2451 		if (vfsp == NULL)
2452 			return (EOPNOTSUPP);
2453 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2454 	}
2455 	return (EOPNOTSUPP);
2456 }
2457 
2458 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2459 	"Generic filesystem");
2460 
2461 #if 1 || defined(COMPAT_PRELITE2)
2462 
2463 static int
2464 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2465 {
2466 	int error;
2467 	struct vfsconf *vfsp;
2468 	struct ovfsconf ovfs;
2469 
2470 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2471 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2472 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2473 		ovfs.vfc_index = vfsp->vfc_typenum;
2474 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2475 		ovfs.vfc_flags = vfsp->vfc_flags;
2476 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2477 		if (error)
2478 			return error;
2479 	}
2480 	return 0;
2481 }
2482 
2483 #endif /* 1 || COMPAT_PRELITE2 */
2484 
2485 #if 0
2486 #define KINFO_VNODESLOP	10
2487 /*
2488  * Dump vnode list (via sysctl).
2489  * Copyout address of vnode followed by vnode.
2490  */
2491 /* ARGSUSED */
2492 static int
2493 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2494 {
2495 	struct proc *p = curproc;	/* XXX */
2496 	struct mount *mp, *nmp;
2497 	struct vnode *nvp, *vp;
2498 	lwkt_tokref ilock;
2499 	lwkt_tokref jlock;
2500 	int error;
2501 
2502 #define VPTRSZ	sizeof (struct vnode *)
2503 #define VNODESZ	sizeof (struct vnode)
2504 
2505 	req->lock = 0;
2506 	if (!req->oldptr) /* Make an estimate */
2507 		return (SYSCTL_OUT(req, 0,
2508 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2509 
2510 	lwkt_gettoken(&ilock, &mountlist_token);
2511 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2512 		if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
2513 			nmp = TAILQ_NEXT(mp, mnt_list);
2514 			continue;
2515 		}
2516 		lwkt_gettoken(&jlock, &mntvnode_token);
2517 again:
2518 		for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2519 		     vp != NULL;
2520 		     vp = nvp) {
2521 			/*
2522 			 * Check that the vp is still associated with
2523 			 * this filesystem.  RACE: could have been
2524 			 * recycled onto the same filesystem.
2525 			 */
2526 			if (vp->v_mount != mp)
2527 				goto again;
2528 			nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2529 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2530 			    (error = SYSCTL_OUT(req, vp, VNODESZ))) {
2531 				lwkt_reltoken(&jlock);
2532 				return (error);
2533 			}
2534 		}
2535 		lwkt_reltoken(&jlock);
2536 		lwkt_gettokref(&ilock);
2537 		nmp = TAILQ_NEXT(mp, mnt_list);	/* ZZZ */
2538 		vfs_unbusy(mp, p);
2539 	}
2540 	lwkt_reltoken(&ilock);
2541 
2542 	return (0);
2543 }
2544 #endif
2545 
2546 /*
2547  * XXX
2548  * Exporting the vnode list on large systems causes them to crash.
2549  * Exporting the vnode list on medium systems causes sysctl to coredump.
2550  */
2551 #if 0
2552 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2553 	0, 0, sysctl_vnode, "S,vnode", "");
2554 #endif
2555 
2556 /*
2557  * Check to see if a filesystem is mounted on a block device.
2558  */
2559 int
2560 vfs_mountedon(vp)
2561 	struct vnode *vp;
2562 {
2563 
2564 	if (vp->v_specmountpoint != NULL)
2565 		return (EBUSY);
2566 	return (0);
2567 }
2568 
2569 /*
2570  * Unmount all filesystems. The list is traversed in reverse order
2571  * of mounting to avoid dependencies.
2572  */
2573 void
2574 vfs_unmountall()
2575 {
2576 	struct mount *mp;
2577 	struct thread *td = curthread;
2578 	int error;
2579 
2580 	if (td->td_proc == NULL)
2581 		td = initproc->p_thread;	/* XXX XXX use proc0 instead? */
2582 
2583 	/*
2584 	 * Since this only runs when rebooting, it is not interlocked.
2585 	 */
2586 	while(!TAILQ_EMPTY(&mountlist)) {
2587 		mp = TAILQ_LAST(&mountlist, mntlist);
2588 		error = dounmount(mp, MNT_FORCE, td);
2589 		if (error) {
2590 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2591 			printf("unmount of %s failed (",
2592 			    mp->mnt_stat.f_mntonname);
2593 			if (error == EBUSY)
2594 				printf("BUSY)\n");
2595 			else
2596 				printf("%d)\n", error);
2597 		} else {
2598 			/* The unmount has removed mp from the mountlist */
2599 		}
2600 	}
2601 }
2602 
2603 /*
2604  * Build hash lists of net addresses and hang them off the mount point.
2605  * Called by ufs_mount() to set up the lists of export addresses.
2606  */
2607 static int
2608 vfs_hang_addrlist(mp, nep, argp)
2609 	struct mount *mp;
2610 	struct netexport *nep;
2611 	struct export_args *argp;
2612 {
2613 	struct netcred *np;
2614 	struct radix_node_head *rnh;
2615 	int i;
2616 	struct radix_node *rn;
2617 	struct sockaddr *saddr, *smask = 0;
2618 	struct domain *dom;
2619 	int error;
2620 
2621 	if (argp->ex_addrlen == 0) {
2622 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2623 			return (EPERM);
2624 		np = &nep->ne_defexported;
2625 		np->netc_exflags = argp->ex_flags;
2626 		np->netc_anon = argp->ex_anon;
2627 		np->netc_anon.cr_ref = 1;
2628 		mp->mnt_flag |= MNT_DEFEXPORTED;
2629 		return (0);
2630 	}
2631 
2632 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2633 		return (EINVAL);
2634 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2635 		return (EINVAL);
2636 
2637 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2638 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2639 	bzero((caddr_t) np, i);
2640 	saddr = (struct sockaddr *) (np + 1);
2641 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2642 		goto out;
2643 	if (saddr->sa_len > argp->ex_addrlen)
2644 		saddr->sa_len = argp->ex_addrlen;
2645 	if (argp->ex_masklen) {
2646 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2647 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2648 		if (error)
2649 			goto out;
2650 		if (smask->sa_len > argp->ex_masklen)
2651 			smask->sa_len = argp->ex_masklen;
2652 	}
2653 	i = saddr->sa_family;
2654 	if ((rnh = nep->ne_rtable[i]) == 0) {
2655 		/*
2656 		 * Seems silly to initialize every AF when most are not used,
2657 		 * do so on demand here
2658 		 */
2659 		for (dom = domains; dom; dom = dom->dom_next)
2660 			if (dom->dom_family == i && dom->dom_rtattach) {
2661 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2662 				    dom->dom_rtoffset);
2663 				break;
2664 			}
2665 		if ((rnh = nep->ne_rtable[i]) == 0) {
2666 			error = ENOBUFS;
2667 			goto out;
2668 		}
2669 	}
2670 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2671 	    np->netc_rnodes);
2672 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2673 		error = EPERM;
2674 		goto out;
2675 	}
2676 	np->netc_exflags = argp->ex_flags;
2677 	np->netc_anon = argp->ex_anon;
2678 	np->netc_anon.cr_ref = 1;
2679 	return (0);
2680 out:
2681 	free(np, M_NETADDR);
2682 	return (error);
2683 }
2684 
2685 /* ARGSUSED */
2686 static int
2687 vfs_free_netcred(rn, w)
2688 	struct radix_node *rn;
2689 	void *w;
2690 {
2691 	struct radix_node_head *rnh = (struct radix_node_head *) w;
2692 
2693 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2694 	free((caddr_t) rn, M_NETADDR);
2695 	return (0);
2696 }
2697 
2698 /*
2699  * Free the net address hash lists that are hanging off the mount points.
2700  */
2701 static void
2702 vfs_free_addrlist(nep)
2703 	struct netexport *nep;
2704 {
2705 	int i;
2706 	struct radix_node_head *rnh;
2707 
2708 	for (i = 0; i <= AF_MAX; i++)
2709 		if ((rnh = nep->ne_rtable[i])) {
2710 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2711 			    (caddr_t) rnh);
2712 			free((caddr_t) rnh, M_RTABLE);
2713 			nep->ne_rtable[i] = 0;
2714 		}
2715 }
2716 
2717 int
2718 vfs_export(mp, nep, argp)
2719 	struct mount *mp;
2720 	struct netexport *nep;
2721 	struct export_args *argp;
2722 {
2723 	int error;
2724 
2725 	if (argp->ex_flags & MNT_DELEXPORT) {
2726 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2727 			vfs_setpublicfs(NULL, NULL, NULL);
2728 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2729 		}
2730 		vfs_free_addrlist(nep);
2731 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2732 	}
2733 	if (argp->ex_flags & MNT_EXPORTED) {
2734 		if (argp->ex_flags & MNT_EXPUBLIC) {
2735 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2736 				return (error);
2737 			mp->mnt_flag |= MNT_EXPUBLIC;
2738 		}
2739 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2740 			return (error);
2741 		mp->mnt_flag |= MNT_EXPORTED;
2742 	}
2743 	return (0);
2744 }
2745 
2746 
2747 /*
2748  * Set the publicly exported filesystem (WebNFS). Currently, only
2749  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2750  */
2751 int
2752 vfs_setpublicfs(mp, nep, argp)
2753 	struct mount *mp;
2754 	struct netexport *nep;
2755 	struct export_args *argp;
2756 {
2757 	int error;
2758 	struct vnode *rvp;
2759 	char *cp;
2760 
2761 	/*
2762 	 * mp == NULL -> invalidate the current info, the FS is
2763 	 * no longer exported. May be called from either vfs_export
2764 	 * or unmount, so check if it hasn't already been done.
2765 	 */
2766 	if (mp == NULL) {
2767 		if (nfs_pub.np_valid) {
2768 			nfs_pub.np_valid = 0;
2769 			if (nfs_pub.np_index != NULL) {
2770 				FREE(nfs_pub.np_index, M_TEMP);
2771 				nfs_pub.np_index = NULL;
2772 			}
2773 		}
2774 		return (0);
2775 	}
2776 
2777 	/*
2778 	 * Only one allowed at a time.
2779 	 */
2780 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2781 		return (EBUSY);
2782 
2783 	/*
2784 	 * Get real filehandle for root of exported FS.
2785 	 */
2786 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2787 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2788 
2789 	if ((error = VFS_ROOT(mp, &rvp)))
2790 		return (error);
2791 
2792 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2793 		return (error);
2794 
2795 	vput(rvp);
2796 
2797 	/*
2798 	 * If an indexfile was specified, pull it in.
2799 	 */
2800 	if (argp->ex_indexfile != NULL) {
2801 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2802 		    M_WAITOK);
2803 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2804 		    MAXNAMLEN, (size_t *)0);
2805 		if (!error) {
2806 			/*
2807 			 * Check for illegal filenames.
2808 			 */
2809 			for (cp = nfs_pub.np_index; *cp; cp++) {
2810 				if (*cp == '/') {
2811 					error = EINVAL;
2812 					break;
2813 				}
2814 			}
2815 		}
2816 		if (error) {
2817 			FREE(nfs_pub.np_index, M_TEMP);
2818 			return (error);
2819 		}
2820 	}
2821 
2822 	nfs_pub.np_mount = mp;
2823 	nfs_pub.np_valid = 1;
2824 	return (0);
2825 }
2826 
2827 struct netcred *
2828 vfs_export_lookup(mp, nep, nam)
2829 	struct mount *mp;
2830 	struct netexport *nep;
2831 	struct sockaddr *nam;
2832 {
2833 	struct netcred *np;
2834 	struct radix_node_head *rnh;
2835 	struct sockaddr *saddr;
2836 
2837 	np = NULL;
2838 	if (mp->mnt_flag & MNT_EXPORTED) {
2839 		/*
2840 		 * Lookup in the export list first.
2841 		 */
2842 		if (nam != NULL) {
2843 			saddr = nam;
2844 			rnh = nep->ne_rtable[saddr->sa_family];
2845 			if (rnh != NULL) {
2846 				np = (struct netcred *)
2847 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2848 							      rnh);
2849 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2850 					np = NULL;
2851 			}
2852 		}
2853 		/*
2854 		 * If no address match, use the default if it exists.
2855 		 */
2856 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2857 			np = &nep->ne_defexported;
2858 	}
2859 	return (np);
2860 }
2861 
2862 /*
2863  * perform msync on all vnodes under a mount point.  The mount point must
2864  * be locked.  This code is also responsible for lazy-freeing unreferenced
2865  * vnodes whos VM objects no longer contain pages.
2866  *
2867  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2868  */
2869 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2870 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2871 				lwkt_tokref_t vlock, void *data);
2872 
2873 void
2874 vfs_msync(struct mount *mp, int flags)
2875 {
2876 	vmntvnodescan(mp, vfs_msync_scan1, vfs_msync_scan2, (void *)flags);
2877 }
2878 
2879 /*
2880  * scan1 is a fast pre-check.  There could be hundreds of thousands of
2881  * vnodes, we cannot afford to do anything heavy weight until we have a
2882  * fairly good indication that there is work to do.
2883  */
2884 static
2885 int
2886 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2887 {
2888 	int flags = (int)data;
2889 
2890 	if ((vp->v_flag & VXLOCK) == 0) {
2891 		if (VSHOULDFREE(vp))
2892 			return(0);
2893 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2894 		    (vp->v_flag & VOBJDIRTY) &&
2895 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2896 			return(0);
2897 		}
2898 	}
2899 	return(-1);
2900 }
2901 
2902 static
2903 int
2904 vfs_msync_scan2(struct mount *mp, struct vnode *vp, lwkt_tokref_t vlock, void *data)
2905 {
2906 	vm_object_t obj;
2907 	int error;
2908 	int flags = (int)data;
2909 
2910 	if (vp->v_flag & VXLOCK)
2911 		return(0);
2912 
2913 	if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2914 	    (vp->v_flag & VOBJDIRTY) &&
2915 	    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2916 		error = vget(vp, vlock, LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ | LK_INTERLOCK, curthread);
2917 		if (error == 0) {
2918 			if (VOP_GETVOBJECT(vp, &obj) == 0) {
2919 				vm_object_page_clean(obj, 0, 0,
2920 				 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2921 			}
2922 			vput(vp);
2923 		}
2924 		return(0);
2925 	}
2926 	vmaybefree(vp);
2927 	lwkt_reltoken(vlock);
2928 	return(0);
2929 }
2930 
2931 /*
2932  * Create the VM object needed for VMIO and mmap support.  This
2933  * is done for all VREG files in the system.  Some filesystems might
2934  * afford the additional metadata buffering capability of the
2935  * VMIO code by making the device node be VMIO mode also.
2936  *
2937  * vp must be locked when vfs_object_create is called.
2938  */
2939 int
2940 vfs_object_create(struct vnode *vp, struct thread *td)
2941 {
2942 	return (VOP_CREATEVOBJECT(vp, td));
2943 }
2944 
2945 /*
2946  * NOTE: the vnode interlock must be held during the call.  We have to recheck
2947  * the VFREE flag since the vnode may have been removed from the free list
2948  * while we were blocked on vnode_free_list_token.  The use or hold count
2949  * must have already been bumped by the caller.
2950  */
2951 static void
2952 vbusy(struct vnode *vp)
2953 {
2954 	lwkt_tokref ilock;
2955 
2956 	lwkt_gettoken(&ilock, &vnode_free_list_token);
2957 	if ((vp->v_flag & VFREE) != 0) {
2958 	    TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2959 	    freevnodes--;
2960 	    vp->v_flag &= ~(VFREE|VAGE);
2961 	}
2962 	lwkt_reltoken(&ilock);
2963 }
2964 
2965 /*
2966  * NOTE: the vnode interlock must be held during the call.  The use or hold
2967  * count must have already been bumped by the caller.  We use a VINFREE to
2968  * interlock against other calls to vfree() which might occur while we
2969  * are blocked.  The vnode cannot be reused until it has actually been
2970  * placed on the free list, so there are no other races even though the
2971  * use and hold counts are 0.
2972  */
2973 static void
2974 vfree(struct vnode *vp)
2975 {
2976 	lwkt_tokref ilock;
2977 
2978 	if ((vp->v_flag & VINFREE) == 0) {
2979 		vp->v_flag |= VINFREE;
2980 		lwkt_gettoken(&ilock, &vnode_free_list_token); /* can block */
2981 		KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2982 		if (vp->v_flag & VAGE) {
2983 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2984 		} else {
2985 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2986 		}
2987 		freevnodes++;
2988 		vp->v_flag &= ~(VAGE|VINFREE);
2989 		vp->v_flag |= VFREE;
2990 		lwkt_reltoken(&ilock);	/* can block */
2991 	}
2992 }
2993 
2994 
2995 /*
2996  * Record a process's interest in events which might happen to
2997  * a vnode.  Because poll uses the historic select-style interface
2998  * internally, this routine serves as both the ``check for any
2999  * pending events'' and the ``record my interest in future events''
3000  * functions.  (These are done together, while the lock is held,
3001  * to avoid race conditions.)
3002  */
3003 int
3004 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3005 {
3006 	lwkt_tokref ilock;
3007 
3008 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3009 	if (vp->v_pollinfo.vpi_revents & events) {
3010 		/*
3011 		 * This leaves events we are not interested
3012 		 * in available for the other process which
3013 		 * which presumably had requested them
3014 		 * (otherwise they would never have been
3015 		 * recorded).
3016 		 */
3017 		events &= vp->v_pollinfo.vpi_revents;
3018 		vp->v_pollinfo.vpi_revents &= ~events;
3019 
3020 		lwkt_reltoken(&ilock);
3021 		return events;
3022 	}
3023 	vp->v_pollinfo.vpi_events |= events;
3024 	selrecord(td, &vp->v_pollinfo.vpi_selinfo);
3025 	lwkt_reltoken(&ilock);
3026 	return 0;
3027 }
3028 
3029 /*
3030  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3031  * it is possible for us to miss an event due to race conditions, but
3032  * that condition is expected to be rare, so for the moment it is the
3033  * preferred interface.
3034  */
3035 void
3036 vn_pollevent(vp, events)
3037 	struct vnode *vp;
3038 	short events;
3039 {
3040 	lwkt_tokref ilock;
3041 
3042 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3043 	if (vp->v_pollinfo.vpi_events & events) {
3044 		/*
3045 		 * We clear vpi_events so that we don't
3046 		 * call selwakeup() twice if two events are
3047 		 * posted before the polling process(es) is
3048 		 * awakened.  This also ensures that we take at
3049 		 * most one selwakeup() if the polling process
3050 		 * is no longer interested.  However, it does
3051 		 * mean that only one event can be noticed at
3052 		 * a time.  (Perhaps we should only clear those
3053 		 * event bits which we note?) XXX
3054 		 */
3055 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
3056 		vp->v_pollinfo.vpi_revents |= events;
3057 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
3058 	}
3059 	lwkt_reltoken(&ilock);
3060 }
3061 
3062 /*
3063  * Wake up anyone polling on vp because it is being revoked.
3064  * This depends on dead_poll() returning POLLHUP for correct
3065  * behavior.
3066  */
3067 void
3068 vn_pollgone(vp)
3069 	struct vnode *vp;
3070 {
3071 	lwkt_tokref ilock;
3072 
3073 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3074 	if (vp->v_pollinfo.vpi_events) {
3075 		vp->v_pollinfo.vpi_events = 0;
3076 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
3077 	}
3078 	lwkt_reltoken(&ilock);
3079 }
3080 
3081 
3082 
3083 /*
3084  * Routine to create and manage a filesystem syncer vnode.
3085  */
3086 #define sync_close ((int (*) (struct  vop_close_args *))nullop)
3087 static int	sync_fsync (struct  vop_fsync_args *);
3088 static int	sync_inactive (struct  vop_inactive_args *);
3089 static int	sync_reclaim  (struct  vop_reclaim_args *);
3090 #define sync_lock ((int (*) (struct  vop_lock_args *))vop_nolock)
3091 #define sync_unlock ((int (*) (struct  vop_unlock_args *))vop_nounlock)
3092 static int	sync_print (struct vop_print_args *);
3093 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
3094 
3095 static vop_t **sync_vnodeop_p;
3096 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3097 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3098 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3099 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3100 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3101 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3102 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
3103 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
3104 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
3105 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
3106 	{ NULL, NULL }
3107 };
3108 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3109 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3110 
3111 VNODEOP_SET(sync_vnodeop_opv_desc);
3112 
3113 /*
3114  * Create a new filesystem syncer vnode for the specified mount point.
3115  * This vnode is placed on the worklist and is responsible for sync'ing
3116  * the filesystem.
3117  *
3118  * NOTE: read-only mounts are also placed on the worklist.  The filesystem
3119  * sync code is also responsible for cleaning up vnodes.
3120  */
3121 int
3122 vfs_allocate_syncvnode(struct mount *mp)
3123 {
3124 	struct vnode *vp;
3125 	static long start, incr, next;
3126 	int error;
3127 
3128 	/* Allocate a new vnode */
3129 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3130 		mp->mnt_syncer = NULL;
3131 		return (error);
3132 	}
3133 	vp->v_type = VNON;
3134 	/*
3135 	 * Place the vnode onto the syncer worklist. We attempt to
3136 	 * scatter them about on the list so that they will go off
3137 	 * at evenly distributed times even if all the filesystems
3138 	 * are mounted at once.
3139 	 */
3140 	next += incr;
3141 	if (next == 0 || next > syncer_maxdelay) {
3142 		start /= 2;
3143 		incr /= 2;
3144 		if (start == 0) {
3145 			start = syncer_maxdelay / 2;
3146 			incr = syncer_maxdelay;
3147 		}
3148 		next = start;
3149 	}
3150 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3151 	mp->mnt_syncer = vp;
3152 	return (0);
3153 }
3154 
3155 /*
3156  * Do a lazy sync of the filesystem.
3157  */
3158 static int
3159 sync_fsync(ap)
3160 	struct vop_fsync_args /* {
3161 		struct vnode *a_vp;
3162 		struct ucred *a_cred;
3163 		int a_waitfor;
3164 		struct thread *a_td;
3165 	} */ *ap;
3166 {
3167 	struct vnode *syncvp = ap->a_vp;
3168 	struct mount *mp = syncvp->v_mount;
3169 	struct thread *td = ap->a_td;
3170 	lwkt_tokref ilock;
3171 	int asyncflag;
3172 
3173 	/*
3174 	 * We only need to do something if this is a lazy evaluation.
3175 	 */
3176 	if (ap->a_waitfor != MNT_LAZY)
3177 		return (0);
3178 
3179 	/*
3180 	 * Move ourselves to the back of the sync list.
3181 	 */
3182 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3183 
3184 	/*
3185 	 * Walk the list of vnodes pushing all that are dirty and
3186 	 * not already on the sync list, and freeing vnodes which have
3187 	 * no refs and whos VM objects are empty.  vfs_msync() handles
3188 	 * the VM issues and must be called whether the mount is readonly
3189 	 * or not.
3190 	 */
3191 	lwkt_gettoken(&ilock, &mountlist_token);
3192 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &ilock, td) != 0) {
3193 		lwkt_reltoken(&ilock);
3194 		return (0);
3195 	}
3196 	if (mp->mnt_flag & MNT_RDONLY) {
3197 		vfs_msync(mp, MNT_NOWAIT);
3198 	} else {
3199 		asyncflag = mp->mnt_flag & MNT_ASYNC;
3200 		mp->mnt_flag &= ~MNT_ASYNC;	/* ZZZ hack */
3201 		vfs_msync(mp, MNT_NOWAIT);
3202 		VFS_SYNC(mp, MNT_LAZY, td);
3203 		if (asyncflag)
3204 			mp->mnt_flag |= MNT_ASYNC;
3205 	}
3206 	vfs_unbusy(mp, td);
3207 	return (0);
3208 }
3209 
3210 /*
3211  * The syncer vnode is no referenced.
3212  */
3213 static int
3214 sync_inactive(ap)
3215 	struct vop_inactive_args /* {
3216 		struct vnode *a_vp;
3217 		struct proc *a_p;
3218 	} */ *ap;
3219 {
3220 
3221 	vgone(ap->a_vp);
3222 	return (0);
3223 }
3224 
3225 /*
3226  * The syncer vnode is no longer needed and is being decommissioned.
3227  *
3228  * Modifications to the worklist must be protected at splbio().
3229  */
3230 static int
3231 sync_reclaim(ap)
3232 	struct vop_reclaim_args /* {
3233 		struct vnode *a_vp;
3234 	} */ *ap;
3235 {
3236 	struct vnode *vp = ap->a_vp;
3237 	int s;
3238 
3239 	s = splbio();
3240 	vp->v_mount->mnt_syncer = NULL;
3241 	if (vp->v_flag & VONWORKLST) {
3242 		LIST_REMOVE(vp, v_synclist);
3243 		vp->v_flag &= ~VONWORKLST;
3244 	}
3245 	splx(s);
3246 
3247 	return (0);
3248 }
3249 
3250 /*
3251  * Print out a syncer vnode.
3252  */
3253 static int
3254 sync_print(ap)
3255 	struct vop_print_args /* {
3256 		struct vnode *a_vp;
3257 	} */ *ap;
3258 {
3259 	struct vnode *vp = ap->a_vp;
3260 
3261 	printf("syncer vnode");
3262 	if (vp->v_vnlock != NULL)
3263 		lockmgr_printinfo(vp->v_vnlock);
3264 	printf("\n");
3265 	return (0);
3266 }
3267 
3268 /*
3269  * extract the dev_t from a VBLK or VCHR
3270  */
3271 dev_t
3272 vn_todev(vp)
3273 	struct vnode *vp;
3274 {
3275 	if (vp->v_type != VBLK && vp->v_type != VCHR)
3276 		return (NODEV);
3277 	return (vp->v_rdev);
3278 }
3279 
3280 /*
3281  * Check if vnode represents a disk device
3282  */
3283 int
3284 vn_isdisk(vp, errp)
3285 	struct vnode *vp;
3286 	int *errp;
3287 {
3288 	if (vp->v_type != VBLK && vp->v_type != VCHR) {
3289 		if (errp != NULL)
3290 			*errp = ENOTBLK;
3291 		return (0);
3292 	}
3293 	if (vp->v_rdev == NULL) {
3294 		if (errp != NULL)
3295 			*errp = ENXIO;
3296 		return (0);
3297 	}
3298 	if (!dev_dport(vp->v_rdev)) {
3299 		if (errp != NULL)
3300 			*errp = ENXIO;
3301 		return (0);
3302 	}
3303 	if (!(dev_dflags(vp->v_rdev) & D_DISK)) {
3304 		if (errp != NULL)
3305 			*errp = ENOTBLK;
3306 		return (0);
3307 	}
3308 	if (errp != NULL)
3309 		*errp = 0;
3310 	return (1);
3311 }
3312 
3313 void
3314 NDFREE(ndp, flags)
3315      struct nameidata *ndp;
3316      const uint flags;
3317 {
3318 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3319 	    (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
3320 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3321 		ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
3322 	}
3323 	if (!(flags & NDF_NO_DNCP_RELE) &&
3324 	    (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3325 	    ndp->ni_dncp) {
3326 		cache_drop(ndp->ni_dncp);
3327 		ndp->ni_dncp = NULL;
3328 	}
3329 	if (!(flags & NDF_NO_NCP_RELE) &&
3330 	    (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3331 	    ndp->ni_ncp) {
3332 		cache_drop(ndp->ni_ncp);
3333 		ndp->ni_ncp = NULL;
3334 	}
3335 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3336 	    (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
3337 	    ndp->ni_dvp != ndp->ni_vp) {
3338 		VOP_UNLOCK(ndp->ni_dvp, NULL, 0, ndp->ni_cnd.cn_td);
3339 	}
3340 	if (!(flags & NDF_NO_DVP_RELE) &&
3341 	    (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
3342 		vrele(ndp->ni_dvp);
3343 		ndp->ni_dvp = NULL;
3344 	}
3345 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3346 	    (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
3347 		VOP_UNLOCK(ndp->ni_vp, NULL, 0, ndp->ni_cnd.cn_td);
3348 	}
3349 	if (!(flags & NDF_NO_VP_RELE) &&
3350 	    ndp->ni_vp) {
3351 		vrele(ndp->ni_vp);
3352 		ndp->ni_vp = NULL;
3353 	}
3354 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3355 	    (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
3356 		vrele(ndp->ni_startdir);
3357 		ndp->ni_startdir = NULL;
3358 	}
3359 }
3360 
3361 #ifdef DEBUG_VFS_LOCKS
3362 
3363 void
3364 assert_vop_locked(struct vnode *vp, const char *str)
3365 {
3366 
3367 	if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
3368 		panic("%s: %p is not locked shared but should be", str, vp);
3369 	}
3370 }
3371 
3372 void
3373 assert_vop_unlocked(struct vnode *vp, const char *str)
3374 {
3375 
3376 	if (vp && IS_LOCKING_VFS(vp)) {
3377 		if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
3378 			panic("%s: %p is locked but should not be", str, vp);
3379 		}
3380 	}
3381 }
3382 
3383 #endif
3384