xref: /dflybsd-src/sys/kern/vfs_subr.c (revision bc49aa1be5400e3bdd801519c6936e8947d5d432)
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 1.35 2004/07/10 16:29:45 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,
91 			int flags, struct thread *td);
92 
93 static unsigned long numvnodes;
94 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
95 
96 enum vtype iftovt_tab[16] = {
97 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99 };
100 int vttoif_tab[9] = {
101 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102 	S_IFSOCK, S_IFIFO, S_IFMT,
103 };
104 
105 static TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
106 
107 static u_long wantfreevnodes = 25;
108 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW,
109 		&wantfreevnodes, 0, "");
110 static u_long freevnodes = 0;
111 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD,
112 		&freevnodes, 0, "");
113 
114 static int reassignbufcalls;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
116 		&reassignbufcalls, 0, "");
117 static int reassignbufloops;
118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
119 		&reassignbufloops, 0, "");
120 static int reassignbufsortgood;
121 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
122 		&reassignbufsortgood, 0, "");
123 static int reassignbufsortbad;
124 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
125 		&reassignbufsortbad, 0, "");
126 static int reassignbufmethod = 1;
127 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
128 		&reassignbufmethod, 0, "");
129 
130 #ifdef ENABLE_VFS_IOOPT
131 int vfs_ioopt = 0;
132 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
133 #endif
134 
135 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
136 struct lwkt_token mountlist_token;
137 struct lwkt_token mntvnode_token;
138 int	nfs_mount_type = -1;
139 static struct lwkt_token mntid_token;
140 static struct lwkt_token vnode_free_list_token;
141 static struct lwkt_token spechash_token;
142 struct nfs_public nfs_pub;	/* publicly exported FS */
143 static vm_zone_t vnode_zone;
144 
145 /*
146  * The workitem queue.
147  */
148 #define SYNCER_MAXDELAY		32
149 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
150 time_t syncdelay = 30;		/* max time to delay syncing data */
151 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW,
152 		&syncdelay, 0, "VFS data synchronization delay");
153 time_t filedelay = 30;		/* time to delay syncing files */
154 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW,
155 		&filedelay, 0, "File synchronization delay");
156 time_t dirdelay = 29;		/* time to delay syncing directories */
157 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW,
158 		&dirdelay, 0, "Directory synchronization delay");
159 time_t metadelay = 28;		/* time to delay syncing metadata */
160 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW,
161 		&metadelay, 0, "VFS metadata synchronization delay");
162 static int rushjob;			/* number of slots to run ASAP */
163 static int stat_rush_requests;	/* number of times I/O speeded up */
164 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW,
165 		&stat_rush_requests, 0, "");
166 
167 static int syncer_delayno = 0;
168 static long syncer_mask;
169 LIST_HEAD(synclist, vnode);
170 static struct synclist *syncer_workitem_pending;
171 
172 int desiredvnodes;
173 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
174 		&desiredvnodes, 0, "Maximum number of vnodes");
175 static int minvnodes;
176 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
177 		&minvnodes, 0, "Minimum number of vnodes");
178 static int vnlru_nowhere = 0;
179 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
180 		&vnlru_nowhere, 0,
181 		"Number of times the vnlru process ran without success");
182 
183 static void	vfs_free_addrlist (struct netexport *nep);
184 static int	vfs_free_netcred (struct radix_node *rn, void *w);
185 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
186 				       struct export_args *argp);
187 
188 #define VSHOULDFREE(vp) \
189 	(!((vp)->v_flag & (VFREE|VDOOMED)) && \
190 	 !(vp)->v_holdcnt && !(vp)->v_usecount && \
191 	 (!(vp)->v_object || \
192 	  !((vp)->v_object->ref_count || (vp)->v_object->resident_page_count)))
193 
194 #define VMIGHTFREE(vp) \
195 	(((vp)->v_flag & (VFREE|VDOOMED|VXLOCK)) == 0 &&   \
196 	 cache_leaf_test(vp) == 0 && (vp)->v_usecount == 0)
197 
198 #define VSHOULDBUSY(vp) \
199 	(((vp)->v_flag & VFREE) && \
200 	 ((vp)->v_holdcnt || (vp)->v_usecount))
201 
202 static void vbusy(struct vnode *vp);
203 static void vfree(struct vnode *vp);
204 static void vmaybefree(struct vnode *vp);
205 
206 extern int dev_ref_debug;
207 
208 /*
209  * NOTE: the vnode interlock must be held on call.
210  */
211 static __inline void
212 vmaybefree(struct vnode *vp)
213 {
214 	if (VSHOULDFREE(vp))
215 		vfree(vp);
216 }
217 
218 /*
219  * Initialize the vnode management data structures.
220  */
221 void
222 vntblinit(void)
223 {
224 
225 	/*
226 	 * Desired vnodes is a result of the physical page count
227 	 * and the size of kernel's heap.  It scales in proportion
228 	 * to the amount of available physical memory.  This can
229 	 * cause trouble on 64-bit and large memory platforms.
230 	 */
231 	/* desiredvnodes = maxproc + vmstats.v_page_count / 4; */
232 	desiredvnodes =
233 		min(maxproc + vmstats.v_page_count /4,
234 		    2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
235 		    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
236 
237 	minvnodes = desiredvnodes / 4;
238 	lwkt_token_init(&mountlist_token);
239 	lwkt_token_init(&mntvnode_token);
240 	lwkt_token_init(&mntid_token);
241 	lwkt_token_init(&spechash_token);
242 	TAILQ_INIT(&vnode_free_list);
243 	lwkt_token_init(&vnode_free_list_token);
244 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
245 	/*
246 	 * Initialize the filesystem syncer.
247 	 */
248 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
249 		&syncer_mask);
250 	syncer_maxdelay = syncer_mask + 1;
251 }
252 
253 /*
254  * Mark a mount point as busy. Used to synchronize access and to delay
255  * unmounting. Interlock is not released on failure.
256  */
257 int
258 vfs_busy(struct mount *mp, int flags,
259 	lwkt_tokref_t interlkp, struct thread *td)
260 {
261 	int lkflags;
262 
263 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
264 		if (flags & LK_NOWAIT)
265 			return (ENOENT);
266 		mp->mnt_kern_flag |= MNTK_MWAIT;
267 		/*
268 		 * Since all busy locks are shared except the exclusive
269 		 * lock granted when unmounting, the only place that a
270 		 * wakeup needs to be done is at the release of the
271 		 * exclusive lock at the end of dounmount.
272 		 *
273 		 * note: interlkp is a serializer and thus can be safely
274 		 * held through any sleep
275 		 */
276 		tsleep((caddr_t)mp, 0, "vfs_busy", 0);
277 		return (ENOENT);
278 	}
279 	lkflags = LK_SHARED | LK_NOPAUSE;
280 	if (interlkp)
281 		lkflags |= LK_INTERLOCK;
282 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
283 		panic("vfs_busy: unexpected lock failure");
284 	return (0);
285 }
286 
287 /*
288  * Free a busy filesystem.
289  */
290 void
291 vfs_unbusy(struct mount *mp, struct thread *td)
292 {
293 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
294 }
295 
296 /*
297  * Lookup a filesystem type, and if found allocate and initialize
298  * a mount structure for it.
299  *
300  * Devname is usually updated by mount(8) after booting.
301  */
302 int
303 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
304 {
305 	struct thread *td = curthread;	/* XXX */
306 	struct vfsconf *vfsp;
307 	struct mount *mp;
308 
309 	if (fstypename == NULL)
310 		return (ENODEV);
311 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
312 		if (!strcmp(vfsp->vfc_name, fstypename))
313 			break;
314 	}
315 	if (vfsp == NULL)
316 		return (ENODEV);
317 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
318 	bzero((char *)mp, (u_long)sizeof(struct mount));
319 	lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
320 	vfs_busy(mp, LK_NOWAIT, NULL, td);
321 	TAILQ_INIT(&mp->mnt_nvnodelist);
322 	TAILQ_INIT(&mp->mnt_reservedvnlist);
323 	mp->mnt_nvnodelistsize = 0;
324 	mp->mnt_vfc = vfsp;
325 	mp->mnt_op = vfsp->vfc_vfsops;
326 	mp->mnt_flag = MNT_RDONLY;
327 	mp->mnt_vnodecovered = NULLVP;
328 	vfsp->vfc_refcount++;
329 	mp->mnt_iosize_max = DFLTPHYS;
330 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
331 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
332 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
333 	mp->mnt_stat.f_mntonname[0] = '/';
334 	mp->mnt_stat.f_mntonname[1] = 0;
335 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
336 	*mpp = mp;
337 	return (0);
338 }
339 
340 /*
341  * Lookup a mount point by filesystem identifier.
342  */
343 struct mount *
344 vfs_getvfs(fsid_t *fsid)
345 {
346 	struct mount *mp;
347 	lwkt_tokref ilock;
348 
349 	lwkt_gettoken(&ilock, &mountlist_token);
350 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
351 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
352 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
353 			break;
354 	    }
355 	}
356 	lwkt_reltoken(&ilock);
357 	return (mp);
358 }
359 
360 /*
361  * Get a new unique fsid.  Try to make its val[0] unique, since this value
362  * will be used to create fake device numbers for stat().  Also try (but
363  * not so hard) make its val[0] unique mod 2^16, since some emulators only
364  * support 16-bit device numbers.  We end up with unique val[0]'s for the
365  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
366  *
367  * Keep in mind that several mounts may be running in parallel.  Starting
368  * the search one past where the previous search terminated is both a
369  * micro-optimization and a defense against returning the same fsid to
370  * different mounts.
371  */
372 void
373 vfs_getnewfsid(struct mount *mp)
374 {
375 	static u_int16_t mntid_base;
376 	lwkt_tokref ilock;
377 	fsid_t tfsid;
378 	int mtype;
379 
380 	lwkt_gettoken(&ilock, &mntid_token);
381 	mtype = mp->mnt_vfc->vfc_typenum;
382 	tfsid.val[1] = mtype;
383 	mtype = (mtype & 0xFF) << 24;
384 	for (;;) {
385 		tfsid.val[0] = makeudev(255,
386 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
387 		mntid_base++;
388 		if (vfs_getvfs(&tfsid) == NULL)
389 			break;
390 	}
391 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
392 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
393 	lwkt_reltoken(&ilock);
394 }
395 
396 /*
397  * Knob to control the precision of file timestamps:
398  *
399  *   0 = seconds only; nanoseconds zeroed.
400  *   1 = seconds and nanoseconds, accurate within 1/HZ.
401  *   2 = seconds and nanoseconds, truncated to microseconds.
402  * >=3 = seconds and nanoseconds, maximum precision.
403  */
404 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
405 
406 static int timestamp_precision = TSP_SEC;
407 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
408 		&timestamp_precision, 0, "");
409 
410 /*
411  * Get a current timestamp.
412  */
413 void
414 vfs_timestamp(struct timespec *tsp)
415 {
416 	struct timeval tv;
417 
418 	switch (timestamp_precision) {
419 	case TSP_SEC:
420 		tsp->tv_sec = time_second;
421 		tsp->tv_nsec = 0;
422 		break;
423 	case TSP_HZ:
424 		getnanotime(tsp);
425 		break;
426 	case TSP_USEC:
427 		microtime(&tv);
428 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
429 		break;
430 	case TSP_NSEC:
431 	default:
432 		nanotime(tsp);
433 		break;
434 	}
435 }
436 
437 /*
438  * Set vnode attributes to VNOVAL
439  */
440 void
441 vattr_null(struct vattr *vap)
442 {
443 	vap->va_type = VNON;
444 	vap->va_size = VNOVAL;
445 	vap->va_bytes = VNOVAL;
446 	vap->va_mode = VNOVAL;
447 	vap->va_nlink = VNOVAL;
448 	vap->va_uid = VNOVAL;
449 	vap->va_gid = VNOVAL;
450 	vap->va_fsid = VNOVAL;
451 	vap->va_fileid = VNOVAL;
452 	vap->va_blocksize = VNOVAL;
453 	vap->va_rdev = VNOVAL;
454 	vap->va_atime.tv_sec = VNOVAL;
455 	vap->va_atime.tv_nsec = VNOVAL;
456 	vap->va_mtime.tv_sec = VNOVAL;
457 	vap->va_mtime.tv_nsec = VNOVAL;
458 	vap->va_ctime.tv_sec = VNOVAL;
459 	vap->va_ctime.tv_nsec = VNOVAL;
460 	vap->va_flags = VNOVAL;
461 	vap->va_gen = VNOVAL;
462 	vap->va_vaflags = 0;
463 }
464 
465 /*
466  * This routine is called when we have too many vnodes.  It attempts
467  * to free <count> vnodes and will potentially free vnodes that still
468  * have VM backing store (VM backing store is typically the cause
469  * of a vnode blowout so we want to do this).  Therefore, this operation
470  * is not considered cheap.
471  *
472  * A number of conditions may prevent a vnode from being reclaimed.
473  * the buffer cache may have references on the vnode, a directory
474  * vnode may still have references due to the namei cache representing
475  * underlying files, or the vnode may be in active use.   It is not
476  * desireable to reuse such vnodes.  These conditions may cause the
477  * number of vnodes to reach some minimum value regardless of what
478  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
479  */
480 static int
481 vlrureclaim(struct mount *mp)
482 {
483 	struct vnode *vp;
484 	lwkt_tokref ilock;
485 	lwkt_tokref vlock;
486 	int done;
487 	int trigger;
488 	int usevnodes;
489 	int count;
490 
491 	/*
492 	 * Calculate the trigger point, don't allow user
493 	 * screwups to blow us up.   This prevents us from
494 	 * recycling vnodes with lots of resident pages.  We
495 	 * aren't trying to free memory, we are trying to
496 	 * free vnodes.
497 	 */
498 	usevnodes = desiredvnodes;
499 	if (usevnodes <= 0)
500 		usevnodes = 1;
501 	trigger = vmstats.v_page_count * 2 / usevnodes;
502 
503 	done = 0;
504 	lwkt_gettoken(&ilock, &mntvnode_token);
505 	count = mp->mnt_nvnodelistsize / 10 + 1;
506 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
507 		/*
508 		 * __VNODESCAN__
509 		 *
510 		 * The VP will stick around while we hold mntvnode_token,
511 		 * at least until we block, so we can safely do an initial
512 		 * check.  But we have to check again after obtaining
513 		 * the vnode interlock.  vp->v_interlock points to stable
514 		 * storage so it's ok if the vp gets ripped out from
515 		 * under us while we are blocked.
516 		 */
517 		if (vp->v_type == VNON ||
518 		    vp->v_type == VBAD ||
519 		    !VMIGHTFREE(vp) ||		/* critical path opt */
520 		    (vp->v_object &&
521 		     vp->v_object->resident_page_count >= trigger)
522 		) {
523 			TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
524 			TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
525 			--count;
526 			continue;
527 		}
528 
529 		/*
530 		 * Get the interlock, delay moving the node to the tail so
531 		 * we don't race against new additions to the mountlist.
532 		 */
533 		lwkt_gettoken(&vlock, vp->v_interlock);
534 		if (TAILQ_FIRST(&mp->mnt_nvnodelist) != vp) {
535 			lwkt_reltoken(&vlock);
536 			continue;
537 		}
538 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
539 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
540 
541 		/*
542 		 * Must check again
543 		 */
544 		if (vp->v_type == VNON ||
545 		    vp->v_type == VBAD ||
546 		    !VMIGHTFREE(vp) ||		/* critical path opt */
547 		    (vp->v_object &&
548 		     vp->v_object->resident_page_count >= trigger)
549 		) {
550 			lwkt_reltoken(&vlock);
551 			--count;
552 			continue;
553 		}
554 		vgonel(vp, &vlock, curthread);
555 		++done;
556 		--count;
557 	}
558 	lwkt_reltoken(&ilock);
559 	return done;
560 }
561 
562 /*
563  * Attempt to recycle vnodes in a context that is always safe to block.
564  * Calling vlrurecycle() from the bowels of file system code has some
565  * interesting deadlock problems.
566  */
567 static struct thread *vnlruthread;
568 static int vnlruproc_sig;
569 
570 static void
571 vnlru_proc(void)
572 {
573 	struct mount *mp, *nmp;
574 	lwkt_tokref ilock;
575 	int s;
576 	int done;
577 	struct thread *td = curthread;
578 
579 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
580 	    SHUTDOWN_PRI_FIRST);
581 
582 	s = splbio();
583 	for (;;) {
584 		kproc_suspend_loop();
585 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
586 			vnlruproc_sig = 0;
587 			wakeup(&vnlruproc_sig);
588 			tsleep(td, 0, "vlruwt", hz);
589 			continue;
590 		}
591 		done = 0;
592 		lwkt_gettoken(&ilock, &mountlist_token);
593 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
594 			if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
595 				nmp = TAILQ_NEXT(mp, mnt_list);
596 				continue;
597 			}
598 			done += vlrureclaim(mp);
599 			lwkt_gettokref(&ilock);
600 			nmp = TAILQ_NEXT(mp, mnt_list);
601 			vfs_unbusy(mp, td);
602 		}
603 		lwkt_reltoken(&ilock);
604 		if (done == 0) {
605 			vnlru_nowhere++;
606 			tsleep(td, 0, "vlrup", hz * 3);
607 		}
608 	}
609 	splx(s);
610 }
611 
612 static struct kproc_desc vnlru_kp = {
613 	"vnlru",
614 	vnlru_proc,
615 	&vnlruthread
616 };
617 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
618 
619 /*
620  * Routines having to do with the management of the vnode table.
621  */
622 extern vop_t **dead_vnodeop_p;
623 
624 /*
625  * Return the next vnode from the free list.
626  */
627 int
628 getnewvnode(enum vtagtype tag, struct mount *mp,
629 	    vop_t **vops, struct vnode **vpp)
630 {
631 	int s;
632 	struct thread *td = curthread;	/* XXX */
633 	struct vnode *vp = NULL;
634 	struct vnode *xvp;
635 	vm_object_t object;
636 	lwkt_tokref ilock;
637 	lwkt_tokref vlock;
638 
639 	s = splbio();	/* YYY remove me */
640 
641 	/*
642 	 * Try to reuse vnodes if we hit the max.  This situation only
643 	 * occurs in certain large-memory (2G+) situations.  We cannot
644 	 * attempt to directly reclaim vnodes due to nasty recursion
645 	 * problems.
646 	 */
647 	while (numvnodes - freevnodes > desiredvnodes) {
648 		if (vnlruproc_sig == 0) {
649 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
650 			wakeup(vnlruthread);
651 		}
652 		tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
653 	}
654 
655 
656 	/*
657 	 * Attempt to reuse a vnode already on the free list, allocating
658 	 * a new vnode if we can't find one or if we have not reached a
659 	 * good minimum for good LRU performance.
660 	 */
661 	lwkt_gettoken(&ilock, &vnode_free_list_token);
662 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
663 		int count;
664 
665 		for (count = 0; count < freevnodes; count++) {
666 			/*
667 			 * __VNODESCAN__
668 			 *
669 			 * Pull the next vnode off the free list and do some
670 			 * sanity checks.  Note that regardless of how we
671 			 * block, if freevnodes is non-zero there had better
672 			 * be something on the list.
673 			 */
674 			vp = TAILQ_FIRST(&vnode_free_list);
675 			if (vp == NULL)
676 				panic("getnewvnode: free vnode isn't");
677 
678 			/*
679 			 * Move the vnode to the end of the list so other
680 			 * processes do not double-block trying to recycle
681 			 * the same vnode (as an optimization), then get
682 			 * the interlock.
683 			 */
684 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
685 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
686 
687 			/*
688 			 * Skip vnodes that are in the process of being
689 			 * held or referenced.  Since the act of adding or
690 			 * removing a vnode on the freelist requires a token
691 			 * and may block, the ref count may be adjusted
692 			 * prior to its addition or removal.
693 			 */
694 			if (VSHOULDBUSY(vp)) {
695 				vp = NULL;
696 				continue;
697 			}
698 
699 
700 			/*
701 			 * Obtain the vnode interlock and check that the
702 			 * vnode is still on the free list.
703 			 *
704 			 * This normally devolves into a degenerate case so
705 			 * it is optimal.   Loop up if it isn't.  Note that
706 			 * the vnode could be in the middle of being moved
707 			 * off the free list (the VSHOULDBUSY() check) and
708 			 * must be skipped if so.
709 			 */
710 			lwkt_gettoken(&vlock, vp->v_interlock);
711 			TAILQ_FOREACH_REVERSE(xvp, &vnode_free_list,
712 			    freelst, v_freelist) {
713 				if (vp == xvp)
714 					break;
715 			}
716 			if (vp != xvp || VSHOULDBUSY(vp)) {
717 				vp = NULL;
718 				continue;
719 			}
720 
721 			/*
722 			 * We now safely own the vnode.  If the vnode has
723 			 * an object do not recycle it if its VM object
724 			 * has resident pages or references.
725 			 */
726 			if ((VOP_GETVOBJECT(vp, &object) == 0 &&
727 			    (object->resident_page_count || object->ref_count))
728 			) {
729 				lwkt_reltoken(&vlock);
730 				vp = NULL;
731 				continue;
732 			}
733 
734 			/*
735 			 * We can almost reuse this vnode.  But we don't want
736 			 * to recycle it if the vnode has children in the
737 			 * namecache because that breaks the namecache's
738 			 * path element chain.  (YYY use nc_refs for the
739 			 * check?)
740 			 */
741 			KKASSERT(vp->v_flag & VFREE);
742 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
743 
744 			if (TAILQ_FIRST(&vp->v_namecache) == NULL ||
745 			    cache_leaf_test(vp) >= 0) {
746 				/* ok, we can reuse this vnode */
747 				break;
748 			}
749 			lwkt_reltoken(&vlock);
750 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
751 			vp = NULL;
752 		}
753 	}
754 
755 	/*
756 	 * If vp is non-NULL we hold it's interlock.
757 	 */
758 	if (vp) {
759 		vp->v_flag |= VDOOMED;
760 		vp->v_flag &= ~VFREE;
761 		freevnodes--;
762 		lwkt_reltoken(&ilock);
763 		cache_purge(vp);	/* YYY may block */
764 		vp->v_lease = NULL;
765 		if (vp->v_type != VBAD) {
766 			vgonel(vp, &vlock, td);
767 		} else {
768 			lwkt_reltoken(&vlock);
769 		}
770 
771 #ifdef INVARIANTS
772 		{
773 			int s;
774 
775 			if (vp->v_data)
776 				panic("cleaned vnode isn't");
777 			s = splbio();
778 			if (vp->v_numoutput)
779 				panic("Clean vnode has pending I/O's");
780 			splx(s);
781 		}
782 #endif
783 		vp->v_flag = 0;
784 		vp->v_lastw = 0;
785 		vp->v_lasta = 0;
786 		vp->v_cstart = 0;
787 		vp->v_clen = 0;
788 		vp->v_socket = 0;
789 		vp->v_writecount = 0;	/* XXX */
790 	} else {
791 		lwkt_reltoken(&ilock);
792 		vp = zalloc(vnode_zone);
793 		bzero(vp, sizeof(*vp));
794 		vp->v_interlock = lwkt_token_pool_get(vp);
795 		lwkt_token_init(&vp->v_pollinfo.vpi_token);
796 		cache_purge(vp);
797 		TAILQ_INIT(&vp->v_namecache);
798 		numvnodes++;
799 	}
800 
801 	TAILQ_INIT(&vp->v_cleanblkhd);
802 	TAILQ_INIT(&vp->v_dirtyblkhd);
803 	vp->v_type = VNON;
804 	vp->v_tag = tag;
805 	vp->v_op = vops;
806 	*vpp = vp;
807 	vp->v_usecount = 1;
808 	vp->v_data = NULL;
809 	splx(s);
810 
811 	/*
812 	 * Placing the vnode on the mount point's queue makes it visible.
813 	 * We had better already have a ref on it.
814 	 */
815 	insmntque(vp, mp);
816 
817 	vfs_object_create(vp, td);
818 	return (0);
819 }
820 
821 /*
822  * Move a vnode from one mount queue to another.
823  */
824 static void
825 insmntque(struct vnode *vp, struct mount *mp)
826 {
827 	lwkt_tokref ilock;
828 
829 	lwkt_gettoken(&ilock, &mntvnode_token);
830 	/*
831 	 * Delete from old mount point vnode list, if on one.
832 	 */
833 	if (vp->v_mount != NULL) {
834 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
835 			("bad mount point vnode list size"));
836 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
837 		vp->v_mount->mnt_nvnodelistsize--;
838 	}
839 	/*
840 	 * Insert into list of vnodes for the new mount point, if available.
841 	 */
842 	if ((vp->v_mount = mp) == NULL) {
843 		lwkt_reltoken(&ilock);
844 		return;
845 	}
846 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
847 	mp->mnt_nvnodelistsize++;
848 	lwkt_reltoken(&ilock);
849 }
850 
851 /*
852  * Update outstanding I/O count and do wakeup if requested.
853  */
854 void
855 vwakeup(struct buf *bp)
856 {
857 	struct vnode *vp;
858 
859 	bp->b_flags &= ~B_WRITEINPROG;
860 	if ((vp = bp->b_vp)) {
861 		vp->v_numoutput--;
862 		if (vp->v_numoutput < 0)
863 			panic("vwakeup: neg numoutput");
864 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
865 			vp->v_flag &= ~VBWAIT;
866 			wakeup((caddr_t) &vp->v_numoutput);
867 		}
868 	}
869 }
870 
871 /*
872  * Flush out and invalidate all buffers associated with a vnode.
873  * Called with the underlying object locked.
874  */
875 int
876 vinvalbuf(struct vnode *vp, int flags, struct thread *td,
877 	int slpflag, int slptimeo)
878 {
879 	struct buf *bp;
880 	struct buf *nbp, *blist;
881 	int s, error;
882 	vm_object_t object;
883 	lwkt_tokref vlock;
884 
885 	if (flags & V_SAVE) {
886 		s = splbio();
887 		while (vp->v_numoutput) {
888 			vp->v_flag |= VBWAIT;
889 			error = tsleep((caddr_t)&vp->v_numoutput,
890 			    slpflag, "vinvlbuf", slptimeo);
891 			if (error) {
892 				splx(s);
893 				return (error);
894 			}
895 		}
896 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
897 			splx(s);
898 			if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0)
899 				return (error);
900 			s = splbio();
901 			if (vp->v_numoutput > 0 ||
902 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
903 				panic("vinvalbuf: dirty bufs");
904 		}
905 		splx(s);
906   	}
907 	s = splbio();
908 	for (;;) {
909 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
910 		if (!blist)
911 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
912 		if (!blist)
913 			break;
914 
915 		for (bp = blist; bp; bp = nbp) {
916 			nbp = TAILQ_NEXT(bp, b_vnbufs);
917 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
918 				error = BUF_TIMELOCK(bp,
919 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
920 				    "vinvalbuf", slpflag, slptimeo);
921 				if (error == ENOLCK)
922 					break;
923 				splx(s);
924 				return (error);
925 			}
926 			/*
927 			 * XXX Since there are no node locks for NFS, I
928 			 * believe there is a slight chance that a delayed
929 			 * write will occur while sleeping just above, so
930 			 * check for it.  Note that vfs_bio_awrite expects
931 			 * buffers to reside on a queue, while VOP_BWRITE and
932 			 * brelse do not.
933 			 */
934 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
935 				(flags & V_SAVE)) {
936 
937 				if (bp->b_vp == vp) {
938 					if (bp->b_flags & B_CLUSTEROK) {
939 						BUF_UNLOCK(bp);
940 						vfs_bio_awrite(bp);
941 					} else {
942 						bremfree(bp);
943 						bp->b_flags |= B_ASYNC;
944 						VOP_BWRITE(bp->b_vp, bp);
945 					}
946 				} else {
947 					bremfree(bp);
948 					(void) VOP_BWRITE(bp->b_vp, bp);
949 				}
950 				break;
951 			}
952 			bremfree(bp);
953 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
954 			bp->b_flags &= ~B_ASYNC;
955 			brelse(bp);
956 		}
957 	}
958 
959 	/*
960 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
961 	 * have write I/O in-progress but if there is a VM object then the
962 	 * VM object can also have read-I/O in-progress.
963 	 */
964 	do {
965 		while (vp->v_numoutput > 0) {
966 			vp->v_flag |= VBWAIT;
967 			tsleep(&vp->v_numoutput, 0, "vnvlbv", 0);
968 		}
969 		if (VOP_GETVOBJECT(vp, &object) == 0) {
970 			while (object->paging_in_progress)
971 				vm_object_pip_sleep(object, "vnvlbx");
972 		}
973 	} while (vp->v_numoutput > 0);
974 
975 	splx(s);
976 
977 	/*
978 	 * Destroy the copy in the VM cache, too.
979 	 */
980 	lwkt_gettoken(&vlock, vp->v_interlock);
981 	if (VOP_GETVOBJECT(vp, &object) == 0) {
982 		vm_object_page_remove(object, 0, 0,
983 			(flags & V_SAVE) ? TRUE : FALSE);
984 	}
985 	lwkt_reltoken(&vlock);
986 
987 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
988 		panic("vinvalbuf: flush failed");
989 	return (0);
990 }
991 
992 /*
993  * Truncate a file's buffer and pages to a specified length.  This
994  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
995  * sync activity.
996  */
997 int
998 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize)
999 {
1000 	struct buf *bp;
1001 	struct buf *nbp;
1002 	int s, anyfreed;
1003 	int trunclbn;
1004 
1005 	/*
1006 	 * Round up to the *next* lbn.
1007 	 */
1008 	trunclbn = (length + blksize - 1) / blksize;
1009 
1010 	s = splbio();
1011 restart:
1012 	anyfreed = 1;
1013 	for (;anyfreed;) {
1014 		anyfreed = 0;
1015 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1016 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1017 			if (bp->b_lblkno >= trunclbn) {
1018 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1019 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1020 					goto restart;
1021 				} else {
1022 					bremfree(bp);
1023 					bp->b_flags |= (B_INVAL | B_RELBUF);
1024 					bp->b_flags &= ~B_ASYNC;
1025 					brelse(bp);
1026 					anyfreed = 1;
1027 				}
1028 				if (nbp &&
1029 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1030 				    (nbp->b_vp != vp) ||
1031 				    (nbp->b_flags & B_DELWRI))) {
1032 					goto restart;
1033 				}
1034 			}
1035 		}
1036 
1037 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1038 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1039 			if (bp->b_lblkno >= trunclbn) {
1040 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1041 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1042 					goto restart;
1043 				} else {
1044 					bremfree(bp);
1045 					bp->b_flags |= (B_INVAL | B_RELBUF);
1046 					bp->b_flags &= ~B_ASYNC;
1047 					brelse(bp);
1048 					anyfreed = 1;
1049 				}
1050 				if (nbp &&
1051 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1052 				    (nbp->b_vp != vp) ||
1053 				    (nbp->b_flags & B_DELWRI) == 0)) {
1054 					goto restart;
1055 				}
1056 			}
1057 		}
1058 	}
1059 
1060 	if (length > 0) {
1061 restartsync:
1062 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1063 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1064 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1065 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1066 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1067 					goto restart;
1068 				} else {
1069 					bremfree(bp);
1070 					if (bp->b_vp == vp) {
1071 						bp->b_flags |= B_ASYNC;
1072 					} else {
1073 						bp->b_flags &= ~B_ASYNC;
1074 					}
1075 					VOP_BWRITE(bp->b_vp, bp);
1076 				}
1077 				goto restartsync;
1078 			}
1079 
1080 		}
1081 	}
1082 
1083 	while (vp->v_numoutput > 0) {
1084 		vp->v_flag |= VBWAIT;
1085 		tsleep(&vp->v_numoutput, 0, "vbtrunc", 0);
1086 	}
1087 
1088 	splx(s);
1089 
1090 	vnode_pager_setsize(vp, length);
1091 
1092 	return (0);
1093 }
1094 
1095 /*
1096  * Associate a buffer with a vnode.
1097  */
1098 void
1099 bgetvp(struct vnode *vp, struct buf *bp)
1100 {
1101 	int s;
1102 
1103 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1104 
1105 	vhold(vp);
1106 	bp->b_vp = vp;
1107 	bp->b_dev = vn_todev(vp);
1108 	/*
1109 	 * Insert onto list for new vnode.
1110 	 */
1111 	s = splbio();
1112 	bp->b_xflags |= BX_VNCLEAN;
1113 	bp->b_xflags &= ~BX_VNDIRTY;
1114 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1115 	splx(s);
1116 }
1117 
1118 /*
1119  * Disassociate a buffer from a vnode.
1120  */
1121 void
1122 brelvp(struct buf *bp)
1123 {
1124 	struct vnode *vp;
1125 	struct buflists *listheadp;
1126 	int s;
1127 
1128 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1129 
1130 	/*
1131 	 * Delete from old vnode list, if on one.
1132 	 */
1133 	vp = bp->b_vp;
1134 	s = splbio();
1135 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1136 		if (bp->b_xflags & BX_VNDIRTY)
1137 			listheadp = &vp->v_dirtyblkhd;
1138 		else
1139 			listheadp = &vp->v_cleanblkhd;
1140 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1141 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1142 	}
1143 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1144 		vp->v_flag &= ~VONWORKLST;
1145 		LIST_REMOVE(vp, v_synclist);
1146 	}
1147 	splx(s);
1148 	bp->b_vp = (struct vnode *) 0;
1149 	vdrop(vp);
1150 }
1151 
1152 /*
1153  * The workitem queue.
1154  *
1155  * It is useful to delay writes of file data and filesystem metadata
1156  * for tens of seconds so that quickly created and deleted files need
1157  * not waste disk bandwidth being created and removed. To realize this,
1158  * we append vnodes to a "workitem" queue. When running with a soft
1159  * updates implementation, most pending metadata dependencies should
1160  * not wait for more than a few seconds. Thus, mounted on block devices
1161  * are delayed only about a half the time that file data is delayed.
1162  * Similarly, directory updates are more critical, so are only delayed
1163  * about a third the time that file data is delayed. Thus, there are
1164  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
1165  * one each second (driven off the filesystem syncer process). The
1166  * syncer_delayno variable indicates the next queue that is to be processed.
1167  * Items that need to be processed soon are placed in this queue:
1168  *
1169  *	syncer_workitem_pending[syncer_delayno]
1170  *
1171  * A delay of fifteen seconds is done by placing the request fifteen
1172  * entries later in the queue:
1173  *
1174  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
1175  *
1176  */
1177 
1178 /*
1179  * Add an item to the syncer work queue.
1180  */
1181 static void
1182 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1183 {
1184 	int s, slot;
1185 
1186 	s = splbio();
1187 
1188 	if (vp->v_flag & VONWORKLST) {
1189 		LIST_REMOVE(vp, v_synclist);
1190 	}
1191 
1192 	if (delay > syncer_maxdelay - 2)
1193 		delay = syncer_maxdelay - 2;
1194 	slot = (syncer_delayno + delay) & syncer_mask;
1195 
1196 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1197 	vp->v_flag |= VONWORKLST;
1198 	splx(s);
1199 }
1200 
1201 struct  thread *updatethread;
1202 static void sched_sync (void);
1203 static struct kproc_desc up_kp = {
1204 	"syncer",
1205 	sched_sync,
1206 	&updatethread
1207 };
1208 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1209 
1210 /*
1211  * System filesystem synchronizer daemon.
1212  */
1213 void
1214 sched_sync(void)
1215 {
1216 	struct synclist *slp;
1217 	struct vnode *vp;
1218 	long starttime;
1219 	int s;
1220 	struct thread *td = curthread;
1221 
1222 	EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
1223 	    SHUTDOWN_PRI_LAST);
1224 
1225 	for (;;) {
1226 		kproc_suspend_loop();
1227 
1228 		starttime = time_second;
1229 
1230 		/*
1231 		 * Push files whose dirty time has expired.  Be careful
1232 		 * of interrupt race on slp queue.
1233 		 */
1234 		s = splbio();
1235 		slp = &syncer_workitem_pending[syncer_delayno];
1236 		syncer_delayno += 1;
1237 		if (syncer_delayno == syncer_maxdelay)
1238 			syncer_delayno = 0;
1239 		splx(s);
1240 
1241 		while ((vp = LIST_FIRST(slp)) != NULL) {
1242 			if (VOP_ISLOCKED(vp, NULL) == 0) {
1243 				vn_lock(vp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
1244 				(void) VOP_FSYNC(vp, MNT_LAZY, td);
1245 				VOP_UNLOCK(vp, NULL, 0, td);
1246 			}
1247 			s = splbio();
1248 			if (LIST_FIRST(slp) == vp) {
1249 				/*
1250 				 * Note: v_tag VT_VFS vps can remain on the
1251 				 * worklist too with no dirty blocks, but
1252 				 * since sync_fsync() moves it to a different
1253 				 * slot we are safe.
1254 				 */
1255 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1256 				    !vn_isdisk(vp, NULL))
1257 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1258 				/*
1259 				 * Put us back on the worklist.  The worklist
1260 				 * routine will remove us from our current
1261 				 * position and then add us back in at a later
1262 				 * position.
1263 				 */
1264 				vn_syncer_add_to_worklist(vp, syncdelay);
1265 			}
1266 			splx(s);
1267 		}
1268 
1269 		/*
1270 		 * Do soft update processing.
1271 		 */
1272 		if (bioops.io_sync)
1273 			(*bioops.io_sync)(NULL);
1274 
1275 		/*
1276 		 * The variable rushjob allows the kernel to speed up the
1277 		 * processing of the filesystem syncer process. A rushjob
1278 		 * value of N tells the filesystem syncer to process the next
1279 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1280 		 * is used by the soft update code to speed up the filesystem
1281 		 * syncer process when the incore state is getting so far
1282 		 * ahead of the disk that the kernel memory pool is being
1283 		 * threatened with exhaustion.
1284 		 */
1285 		if (rushjob > 0) {
1286 			rushjob -= 1;
1287 			continue;
1288 		}
1289 		/*
1290 		 * If it has taken us less than a second to process the
1291 		 * current work, then wait. Otherwise start right over
1292 		 * again. We can still lose time if any single round
1293 		 * takes more than two seconds, but it does not really
1294 		 * matter as we are just trying to generally pace the
1295 		 * filesystem activity.
1296 		 */
1297 		if (time_second == starttime)
1298 			tsleep(&lbolt, 0, "syncer", 0);
1299 	}
1300 }
1301 
1302 /*
1303  * Request the syncer daemon to speed up its work.
1304  * We never push it to speed up more than half of its
1305  * normal turn time, otherwise it could take over the cpu.
1306  *
1307  * YYY wchan field protected by the BGL.
1308  */
1309 int
1310 speedup_syncer(void)
1311 {
1312 	crit_enter();
1313 	if (updatethread->td_wchan == &lbolt) { /* YYY */
1314 		unsleep(updatethread);
1315 		lwkt_schedule(updatethread);
1316 	}
1317 	crit_exit();
1318 	if (rushjob < syncdelay / 2) {
1319 		rushjob += 1;
1320 		stat_rush_requests += 1;
1321 		return (1);
1322 	}
1323 	return(0);
1324 }
1325 
1326 /*
1327  * Associate a p-buffer with a vnode.
1328  *
1329  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1330  * with the buffer.  i.e. the bp has not been linked into the vnode or
1331  * ref-counted.
1332  */
1333 void
1334 pbgetvp(struct vnode *vp, struct buf *bp)
1335 {
1336 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1337 
1338 	bp->b_vp = vp;
1339 	bp->b_flags |= B_PAGING;
1340 	bp->b_dev = vn_todev(vp);
1341 }
1342 
1343 /*
1344  * Disassociate a p-buffer from a vnode.
1345  */
1346 void
1347 pbrelvp(struct buf *bp)
1348 {
1349 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1350 
1351 	/* XXX REMOVE ME */
1352 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1353 		panic(
1354 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1355 		    bp,
1356 		    (int)bp->b_flags
1357 		);
1358 	}
1359 	bp->b_vp = (struct vnode *) 0;
1360 	bp->b_flags &= ~B_PAGING;
1361 }
1362 
1363 void
1364 pbreassignbuf(struct buf *bp, struct vnode *newvp)
1365 {
1366 	if ((bp->b_flags & B_PAGING) == 0) {
1367 		panic(
1368 		    "pbreassignbuf() on non phys bp %p",
1369 		    bp
1370 		);
1371 	}
1372 	bp->b_vp = newvp;
1373 }
1374 
1375 /*
1376  * Reassign a buffer from one vnode to another.
1377  * Used to assign file specific control information
1378  * (indirect blocks) to the vnode to which they belong.
1379  */
1380 void
1381 reassignbuf(struct buf *bp, struct vnode *newvp)
1382 {
1383 	struct buflists *listheadp;
1384 	int delay;
1385 	int s;
1386 
1387 	if (newvp == NULL) {
1388 		printf("reassignbuf: NULL");
1389 		return;
1390 	}
1391 	++reassignbufcalls;
1392 
1393 	/*
1394 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1395 	 * is not fully linked in.
1396 	 */
1397 	if (bp->b_flags & B_PAGING)
1398 		panic("cannot reassign paging buffer");
1399 
1400 	s = splbio();
1401 	/*
1402 	 * Delete from old vnode list, if on one.
1403 	 */
1404 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1405 		if (bp->b_xflags & BX_VNDIRTY)
1406 			listheadp = &bp->b_vp->v_dirtyblkhd;
1407 		else
1408 			listheadp = &bp->b_vp->v_cleanblkhd;
1409 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1410 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1411 		if (bp->b_vp != newvp) {
1412 			vdrop(bp->b_vp);
1413 			bp->b_vp = NULL;	/* for clarification */
1414 		}
1415 	}
1416 	/*
1417 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1418 	 * of clean buffers.
1419 	 */
1420 	if (bp->b_flags & B_DELWRI) {
1421 		struct buf *tbp;
1422 
1423 		listheadp = &newvp->v_dirtyblkhd;
1424 		if ((newvp->v_flag & VONWORKLST) == 0) {
1425 			switch (newvp->v_type) {
1426 			case VDIR:
1427 				delay = dirdelay;
1428 				break;
1429 			case VCHR:
1430 			case VBLK:
1431 				if (newvp->v_rdev &&
1432 				    newvp->v_rdev->si_mountpoint != NULL) {
1433 					delay = metadelay;
1434 					break;
1435 				}
1436 				/* fall through */
1437 			default:
1438 				delay = filedelay;
1439 			}
1440 			vn_syncer_add_to_worklist(newvp, delay);
1441 		}
1442 		bp->b_xflags |= BX_VNDIRTY;
1443 		tbp = TAILQ_FIRST(listheadp);
1444 		if (tbp == NULL ||
1445 		    bp->b_lblkno == 0 ||
1446 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1447 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1448 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1449 			++reassignbufsortgood;
1450 		} else if (bp->b_lblkno < 0) {
1451 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1452 			++reassignbufsortgood;
1453 		} else if (reassignbufmethod == 1) {
1454 			/*
1455 			 * New sorting algorithm, only handle sequential case,
1456 			 * otherwise append to end (but before metadata)
1457 			 */
1458 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1459 			    (tbp->b_xflags & BX_VNDIRTY)) {
1460 				/*
1461 				 * Found the best place to insert the buffer
1462 				 */
1463 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1464 				++reassignbufsortgood;
1465 			} else {
1466 				/*
1467 				 * Missed, append to end, but before meta-data.
1468 				 * We know that the head buffer in the list is
1469 				 * not meta-data due to prior conditionals.
1470 				 *
1471 				 * Indirect effects:  NFS second stage write
1472 				 * tends to wind up here, giving maximum
1473 				 * distance between the unstable write and the
1474 				 * commit rpc.
1475 				 */
1476 				tbp = TAILQ_LAST(listheadp, buflists);
1477 				while (tbp && tbp->b_lblkno < 0)
1478 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1479 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1480 				++reassignbufsortbad;
1481 			}
1482 		} else {
1483 			/*
1484 			 * Old sorting algorithm, scan queue and insert
1485 			 */
1486 			struct buf *ttbp;
1487 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1488 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1489 				++reassignbufloops;
1490 				tbp = ttbp;
1491 			}
1492 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1493 		}
1494 	} else {
1495 		bp->b_xflags |= BX_VNCLEAN;
1496 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1497 		if ((newvp->v_flag & VONWORKLST) &&
1498 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1499 			newvp->v_flag &= ~VONWORKLST;
1500 			LIST_REMOVE(newvp, v_synclist);
1501 		}
1502 	}
1503 	if (bp->b_vp != newvp) {
1504 		bp->b_vp = newvp;
1505 		vhold(bp->b_vp);
1506 	}
1507 	splx(s);
1508 }
1509 
1510 /*
1511  * Create a vnode for a block device.
1512  * Used for mounting the root file system.
1513  */
1514 int
1515 bdevvp(dev_t dev, struct vnode **vpp)
1516 {
1517 	struct vnode *vp;
1518 	struct vnode *nvp;
1519 	int error;
1520 
1521 	if (dev == NODEV) {
1522 		*vpp = NULLVP;
1523 		return (ENXIO);
1524 	}
1525 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1526 	if (error) {
1527 		*vpp = NULLVP;
1528 		return (error);
1529 	}
1530 	vp = nvp;
1531 	vp->v_type = VCHR;
1532 	vp->v_udev = dev->si_udev;
1533 	*vpp = vp;
1534 	return (0);
1535 }
1536 
1537 int
1538 v_associate_rdev(struct vnode *vp, dev_t dev)
1539 {
1540 	lwkt_tokref ilock;
1541 
1542 	if (dev == NULL || dev == NODEV)
1543 		return(ENXIO);
1544 	if (dev_is_good(dev) == 0)
1545 		return(ENXIO);
1546 	KKASSERT(vp->v_rdev == NULL);
1547 	if (dev_ref_debug)
1548 		printf("Z1");
1549 	vp->v_rdev = reference_dev(dev);
1550 	lwkt_gettoken(&ilock, &spechash_token);
1551 	SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext);
1552 	lwkt_reltoken(&ilock);
1553 	return(0);
1554 }
1555 
1556 void
1557 v_release_rdev(struct vnode *vp)
1558 {
1559 	lwkt_tokref ilock;
1560 	dev_t dev;
1561 
1562 	if ((dev = vp->v_rdev) != NULL) {
1563 		lwkt_gettoken(&ilock, &spechash_token);
1564 		SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext);
1565 		if (dev_ref_debug && vp->v_opencount != 0) {
1566 			printf("releasing rdev with non-0 "
1567 				"v_opencount(%d) (revoked?)\n",
1568 				vp->v_opencount);
1569 		}
1570 		vp->v_rdev = NULL;
1571 		vp->v_opencount = 0;
1572 		release_dev(dev);
1573 		lwkt_reltoken(&ilock);
1574 	}
1575 }
1576 
1577 /*
1578  * Add a vnode to the alias list hung off the dev_t.  We only associate
1579  * the device number with the vnode.  The actual device is not associated
1580  * until the vnode is opened (usually in spec_open()), and will be
1581  * disassociated on last close.
1582  */
1583 void
1584 addaliasu(struct vnode *nvp, udev_t nvp_udev)
1585 {
1586 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1587 		panic("addaliasu on non-special vnode");
1588 	nvp->v_udev = nvp_udev;
1589 }
1590 
1591 /*
1592  * Grab a particular vnode from the free list, increment its
1593  * reference count and lock it. The vnode lock bit is set if the
1594  * vnode is being eliminated in vgone. The process is awakened
1595  * when the transition is completed, and an error returned to
1596  * indicate that the vnode is no longer usable (possibly having
1597  * been changed to a new file system type).
1598  *
1599  * This code is very sensitive.  We are depending on the vnode interlock
1600  * to be maintained through to the vn_lock() call, which means that we
1601  * cannot block which means that we cannot call vbusy() until after vn_lock().
1602  * If the interlock is not maintained, the VXLOCK check will not properly
1603  * interlock against a vclean()'s LK_DRAIN operation on the lock.
1604  */
1605 int
1606 vget(struct vnode *vp, lwkt_tokref_t vlock, int flags, thread_t td)
1607 {
1608 	int error;
1609 	lwkt_tokref vvlock;
1610 
1611 	/*
1612 	 * We need the interlock to safely modify the v_ fields.  ZZZ it is
1613 	 * only legal to pass (1) the vnode's interlock and (2) only pass
1614 	 * NULL w/o LK_INTERLOCK if the vnode is *ALREADY* referenced or
1615 	 * held.
1616 	 */
1617 	if ((flags & LK_INTERLOCK) == 0) {
1618 		lwkt_gettoken(&vvlock, vp->v_interlock);
1619 		vlock = &vvlock;
1620 	}
1621 
1622 	/*
1623 	 * If the vnode is in the process of being cleaned out for
1624 	 * another use, we wait for the cleaning to finish and then
1625 	 * return failure. Cleaning is determined by checking that
1626 	 * the VXLOCK flag is set.  It is possible for the vnode to be
1627 	 * self-referenced during the cleaning operation.
1628 	 */
1629 	if (vp->v_flag & VXLOCK) {
1630 		if (vp->v_vxthread == curthread) {
1631 #if 0
1632 			/* this can now occur in normal operation */
1633 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1634 #endif
1635 		} else {
1636 			vp->v_flag |= VXWANT;
1637 			lwkt_reltoken(vlock);
1638 			tsleep((caddr_t)vp, 0, "vget", 0);
1639 			return (ENOENT);
1640 		}
1641 	}
1642 
1643 	/*
1644 	 * Bump v_usecount to prevent the vnode from being recycled.  The
1645 	 * usecount needs to be bumped before we successfully get our lock.
1646 	 */
1647 	vp->v_usecount++;
1648 	if (flags & LK_TYPE_MASK) {
1649 		if ((error = vn_lock(vp, vlock, flags | LK_INTERLOCK, td)) != 0) {
1650 			/*
1651 			 * must expand vrele here because we do not want
1652 			 * to call VOP_INACTIVE if the reference count
1653 			 * drops back to zero since it was never really
1654 			 * active. We must remove it from the free list
1655 			 * before sleeping so that multiple processes do
1656 			 * not try to recycle it.
1657 			 */
1658 			lwkt_gettokref(vlock);
1659 			vp->v_usecount--;
1660 			vmaybefree(vp);
1661 			lwkt_reltoken(vlock);
1662 		}
1663 		return (error);
1664 	}
1665 	if (VSHOULDBUSY(vp))
1666 		vbusy(vp);	/* interlock must be held on call */
1667 	lwkt_reltoken(vlock);
1668 	return (0);
1669 }
1670 
1671 void
1672 vref(struct vnode *vp)
1673 {
1674 	crit_enter();	/* YYY use crit section for moment / BGL protected */
1675 	vp->v_usecount++;
1676 	crit_exit();
1677 }
1678 
1679 /*
1680  * Vnode put/release.
1681  * If count drops to zero, call inactive routine and return to freelist.
1682  */
1683 void
1684 vrele(struct vnode *vp)
1685 {
1686 	struct thread *td = curthread;	/* XXX */
1687 	lwkt_tokref vlock;
1688 
1689 	KASSERT(vp != NULL && vp->v_usecount >= 0,
1690 	    ("vrele: null vp or <=0 v_usecount"));
1691 
1692 	lwkt_gettoken(&vlock, vp->v_interlock);
1693 
1694 	if (vp->v_usecount > 1) {
1695 		vp->v_usecount--;
1696 		lwkt_reltoken(&vlock);
1697 		return;
1698 	}
1699 
1700 	if (vp->v_usecount == 1) {
1701 		vp->v_usecount--;
1702 		/*
1703 		 * We must call VOP_INACTIVE with the node locked and the
1704 		 * usecount 0.  If we are doing a vpu, the node is already
1705 		 * locked, but, in the case of vrele, we must explicitly lock
1706 		 * the vnode before calling VOP_INACTIVE.
1707 		 */
1708 
1709 		if (vn_lock(vp, NULL, LK_EXCLUSIVE, td) == 0)
1710 			VOP_INACTIVE(vp, td);
1711 		vmaybefree(vp);
1712 		lwkt_reltoken(&vlock);
1713 	} else {
1714 #ifdef DIAGNOSTIC
1715 		vprint("vrele: negative ref count", vp);
1716 #endif
1717 		lwkt_reltoken(&vlock);
1718 		panic("vrele: negative ref cnt");
1719 	}
1720 }
1721 
1722 void
1723 vput(struct vnode *vp)
1724 {
1725 	struct thread *td = curthread;	/* XXX */
1726 	lwkt_tokref vlock;
1727 
1728 	KASSERT(vp != NULL, ("vput: null vp"));
1729 
1730 	lwkt_gettoken(&vlock, vp->v_interlock);
1731 
1732 	if (vp->v_usecount > 1) {
1733 		vp->v_usecount--;
1734 		VOP_UNLOCK(vp, &vlock, LK_INTERLOCK, td);
1735 		return;
1736 	}
1737 
1738 	if (vp->v_usecount == 1) {
1739 		vp->v_usecount--;
1740 		/*
1741 		 * We must call VOP_INACTIVE with the node locked.
1742 		 * If we are doing a vpu, the node is already locked,
1743 		 * so we just need to release the vnode mutex.
1744 		 */
1745 		VOP_INACTIVE(vp, td);
1746 		vmaybefree(vp);
1747 		lwkt_reltoken(&vlock);
1748 	} else {
1749 #ifdef DIAGNOSTIC
1750 		vprint("vput: negative ref count", vp);
1751 #endif
1752 		lwkt_reltoken(&vlock);
1753 		panic("vput: negative ref cnt");
1754 	}
1755 }
1756 
1757 /*
1758  * Somebody doesn't want the vnode recycled. ZZZ vnode interlock should
1759  * be held but isn't.
1760  */
1761 void
1762 vhold(struct vnode *vp)
1763 {
1764 	int s;
1765 
1766   	s = splbio();
1767 	vp->v_holdcnt++;
1768 	if (VSHOULDBUSY(vp))
1769 		vbusy(vp);	/* interlock must be held on call */
1770 	splx(s);
1771 }
1772 
1773 /*
1774  * One less who cares about this vnode.
1775  */
1776 void
1777 vdrop(struct vnode *vp)
1778 {
1779 	lwkt_tokref vlock;
1780 
1781 	lwkt_gettoken(&vlock, vp->v_interlock);
1782 	if (vp->v_holdcnt <= 0)
1783 		panic("vdrop: holdcnt");
1784 	vp->v_holdcnt--;
1785 	vmaybefree(vp);
1786 	lwkt_reltoken(&vlock);
1787 }
1788 
1789 int
1790 vmntvnodescan(
1791     struct mount *mp,
1792     int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1793     int (*slowfunc)(struct mount *mp, struct vnode *vp,
1794 		    lwkt_tokref_t vlock, void *data),
1795     void *data
1796 ) {
1797 	lwkt_tokref ilock;
1798 	lwkt_tokref vlock;
1799 	struct vnode *pvp;
1800 	struct vnode *vp;
1801 	int r = 0;
1802 
1803 	/*
1804 	 * Scan the vnodes on the mount's vnode list.  Use a placemarker
1805 	 */
1806 	pvp = zalloc(vnode_zone);
1807 	pvp->v_flag |= VPLACEMARKER;
1808 
1809 	lwkt_gettoken(&ilock, &mntvnode_token);
1810 	TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1811 
1812 	while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
1813 		/*
1814 		 * Move the placemarker and skip other placemarkers we
1815 		 * encounter.  The nothing can get in our way so the
1816 		 * mount point on the vp must be valid.
1817 		 */
1818 		TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1819 		TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
1820 		if (vp->v_flag & VPLACEMARKER)
1821 			continue;
1822 		KKASSERT(vp->v_mount == mp);
1823 
1824 		/*
1825 		 * Quick test
1826 		 */
1827 		if (fastfunc) {
1828 			if ((r = fastfunc(mp, vp, data)) < 0)
1829 				continue;
1830 			if (r)
1831 				break;
1832 		}
1833 
1834 		/*
1835 		 * Get the vnodes interlock and make sure it is still on the
1836 		 * mount list.  Skip it if it has moved (we may encounter it
1837 		 * later).  Then do the with-interlock test.  The callback
1838 		 * is responsible for releasing the vnode interlock.
1839 		 *
1840 		 * The interlock is type-stable.
1841 		 */
1842 		if (slowfunc) {
1843 			lwkt_gettoken(&vlock, vp->v_interlock);
1844 			if (vp != TAILQ_PREV(pvp, vnodelst, v_nmntvnodes)) {
1845 				printf("vmntvnodescan (debug info only): f=%p vp=%p vnode ripped out from under us\n", slowfunc, vp);
1846 				lwkt_reltoken(&vlock);
1847 				continue;
1848 			}
1849 			if ((r = slowfunc(mp, vp, &vlock, data)) != 0) {
1850 				KKASSERT(lwkt_havetokref(&vlock) == 0);
1851 				break;
1852 			}
1853 			KKASSERT(lwkt_havetokref(&vlock) == 0);
1854 		}
1855 	}
1856 	TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
1857 	zfree(vnode_zone, pvp);
1858 	lwkt_reltoken(&ilock);
1859 	return(r);
1860 }
1861 
1862 /*
1863  * Remove any vnodes in the vnode table belonging to mount point mp.
1864  *
1865  * If FORCECLOSE is not specified, there should not be any active ones,
1866  * return error if any are found (nb: this is a user error, not a
1867  * system error). If FORCECLOSE is specified, detach any active vnodes
1868  * that are found.
1869  *
1870  * If WRITECLOSE is set, only flush out regular file vnodes open for
1871  * writing.
1872  *
1873  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1874  *
1875  * `rootrefs' specifies the base reference count for the root vnode
1876  * of this filesystem. The root vnode is considered busy if its
1877  * v_usecount exceeds this value. On a successful return, vflush()
1878  * will call vrele() on the root vnode exactly rootrefs times.
1879  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1880  * be zero.
1881  */
1882 #ifdef DIAGNOSTIC
1883 static int busyprt = 0;		/* print out busy vnodes */
1884 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1885 #endif
1886 
1887 static int vflush_scan(struct mount *mp, struct vnode *vp,
1888 			lwkt_tokref_t vlock, void *data);
1889 
1890 struct vflush_info {
1891 	int flags;
1892 	int busy;
1893 	thread_t td;
1894 };
1895 
1896 int
1897 vflush(struct mount *mp, int rootrefs, int flags)
1898 {
1899 	struct thread *td = curthread;	/* XXX */
1900 	struct vnode *rootvp = NULL;
1901 	int error;
1902 	lwkt_tokref vlock;
1903 	struct vflush_info vflush_info;
1904 
1905 	if (rootrefs > 0) {
1906 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1907 		    ("vflush: bad args"));
1908 		/*
1909 		 * Get the filesystem root vnode. We can vput() it
1910 		 * immediately, since with rootrefs > 0, it won't go away.
1911 		 */
1912 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1913 			return (error);
1914 		vput(rootvp);
1915 	}
1916 
1917 	vflush_info.busy = 0;
1918 	vflush_info.flags = flags;
1919 	vflush_info.td = td;
1920 	vmntvnodescan(mp, NULL, vflush_scan, &vflush_info);
1921 
1922 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1923 		/*
1924 		 * If just the root vnode is busy, and if its refcount
1925 		 * is equal to `rootrefs', then go ahead and kill it.
1926 		 */
1927 		lwkt_gettoken(&vlock, rootvp->v_interlock);
1928 		KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1929 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1930 		if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
1931 			vgonel(rootvp, &vlock, td);
1932 			vflush_info.busy = 0;
1933 		} else {
1934 			lwkt_reltoken(&vlock);
1935 		}
1936 	}
1937 	if (vflush_info.busy)
1938 		return (EBUSY);
1939 	for (; rootrefs > 0; rootrefs--)
1940 		vrele(rootvp);
1941 	return (0);
1942 }
1943 
1944 /*
1945  * The scan callback is made with an interlocked vnode.
1946  */
1947 static int
1948 vflush_scan(struct mount *mp, struct vnode *vp,
1949 	    lwkt_tokref_t vlock, void *data)
1950 {
1951 	struct vflush_info *info = data;
1952 	struct vattr vattr;
1953 
1954 	/*
1955 	 * Skip over a vnodes marked VSYSTEM.
1956 	 */
1957 	if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1958 		lwkt_reltoken(vlock);
1959 		return(0);
1960 	}
1961 
1962 	/*
1963 	 * If WRITECLOSE is set, flush out unlinked but still open
1964 	 * files (even if open only for reading) and regular file
1965 	 * vnodes open for writing.
1966 	 */
1967 	if ((info->flags & WRITECLOSE) &&
1968 	    (vp->v_type == VNON ||
1969 	    (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
1970 	    vattr.va_nlink > 0)) &&
1971 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1972 		lwkt_reltoken(vlock);
1973 		return(0);
1974 	}
1975 
1976 	/*
1977 	 * With v_usecount == 0, all we need to do is clear out the
1978 	 * vnode data structures and we are done.
1979 	 */
1980 	if (vp->v_usecount == 0) {
1981 		vgonel(vp, vlock, info->td);
1982 		return(0);
1983 	}
1984 
1985 	/*
1986 	 * If FORCECLOSE is set, forcibly close the vnode. For block
1987 	 * or character devices, revert to an anonymous device. For
1988 	 * all other files, just kill them.
1989 	 */
1990 	if (info->flags & FORCECLOSE) {
1991 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
1992 			vgonel(vp, vlock, info->td);
1993 		} else {
1994 			vclean(vp, vlock, 0, info->td);
1995 			vp->v_op = spec_vnodeop_p;
1996 			insmntque(vp, (struct mount *) 0);
1997 		}
1998 		return(0);
1999 	}
2000 #ifdef DIAGNOSTIC
2001 	if (busyprt)
2002 		vprint("vflush: busy vnode", vp);
2003 #endif
2004 	lwkt_reltoken(vlock);
2005 	++info->busy;
2006 	return(0);
2007 }
2008 
2009 /*
2010  * Disassociate the underlying file system from a vnode.
2011  */
2012 static void
2013 vclean(struct vnode *vp, lwkt_tokref_t vlock, int flags, struct thread *td)
2014 {
2015 	int active;
2016 
2017 	/*
2018 	 * Check to see if the vnode is in use. If so we have to reference it
2019 	 * before we clean it out so that its count cannot fall to zero and
2020 	 * generate a race against ourselves to recycle it.
2021 	 */
2022 	if ((active = vp->v_usecount))
2023 		vp->v_usecount++;
2024 
2025 	/*
2026 	 * Prevent the vnode from being recycled or brought into use while we
2027 	 * clean it out.
2028 	 */
2029 	if (vp->v_flag & VXLOCK)
2030 		panic("vclean: deadlock");
2031 	vp->v_flag |= VXLOCK;
2032 	vp->v_vxthread = curthread;
2033 
2034 	/*
2035 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2036 	 * have the object locked while it cleans it out. The VOP_LOCK
2037 	 * ensures that the VOP_INACTIVE routine is done with its work.
2038 	 * For active vnodes, it ensures that no other activity can
2039 	 * occur while the underlying object is being cleaned out.
2040 	 *
2041 	 * NOTE: we continue to hold the vnode interlock through to the
2042 	 * end of vclean().
2043 	 */
2044 	VOP_LOCK(vp, NULL, LK_DRAIN, td);
2045 
2046 	/*
2047 	 * Clean out any buffers associated with the vnode.
2048 	 */
2049 	vinvalbuf(vp, V_SAVE, td, 0, 0);
2050 	VOP_DESTROYVOBJECT(vp);
2051 
2052 	/*
2053 	 * If purging an active vnode, it must be closed and
2054 	 * deactivated before being reclaimed. Note that the
2055 	 * VOP_INACTIVE will unlock the vnode.
2056 	 */
2057 	if (active) {
2058 		if (flags & DOCLOSE)
2059 			VOP_CLOSE(vp, FNONBLOCK, td);
2060 		VOP_INACTIVE(vp, td);
2061 	} else {
2062 		/*
2063 		 * Any other processes trying to obtain this lock must first
2064 		 * wait for VXLOCK to clear, then call the new lock operation.
2065 		 */
2066 		VOP_UNLOCK(vp, NULL, 0, td);
2067 	}
2068 	/*
2069 	 * Reclaim the vnode.
2070 	 */
2071 	if (VOP_RECLAIM(vp, td))
2072 		panic("vclean: cannot reclaim");
2073 
2074 	if (active) {
2075 		/*
2076 		 * Inline copy of vrele() since VOP_INACTIVE
2077 		 * has already been called.
2078 		 */
2079 		if (--vp->v_usecount <= 0) {
2080 #ifdef DIAGNOSTIC
2081 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2082 				vprint("vclean: bad ref count", vp);
2083 				panic("vclean: ref cnt");
2084 			}
2085 #endif
2086 			vfree(vp);
2087 		}
2088 	}
2089 
2090 	cache_purge(vp);
2091 	vp->v_vnlock = NULL;
2092 	vmaybefree(vp);
2093 
2094 	/*
2095 	 * Done with purge, notify sleepers of the grim news.
2096 	 */
2097 	vp->v_op = dead_vnodeop_p;
2098 	vn_pollgone(vp);
2099 	vp->v_tag = VT_NON;
2100 	vp->v_flag &= ~VXLOCK;
2101 	vp->v_vxthread = NULL;
2102 	if (vp->v_flag & VXWANT) {
2103 		vp->v_flag &= ~VXWANT;
2104 		wakeup((caddr_t) vp);
2105 	}
2106 	lwkt_reltoken(vlock);
2107 }
2108 
2109 /*
2110  * Eliminate all activity associated with the requested vnode
2111  * and with all vnodes aliased to the requested vnode.
2112  *
2113  * revoke { struct vnode *a_vp, int a_flags }
2114  */
2115 int
2116 vop_revoke(struct vop_revoke_args *ap)
2117 {
2118 	struct vnode *vp, *vq;
2119 	lwkt_tokref ilock;
2120 	dev_t dev;
2121 
2122 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2123 
2124 	vp = ap->a_vp;
2125 	/*
2126 	 * If a vgone (or vclean) is already in progress,
2127 	 * wait until it is done and return.
2128 	 */
2129 	if (vp->v_flag & VXLOCK) {
2130 		vp->v_flag |= VXWANT;
2131 		/*lwkt_reltoken(vlock); ZZZ */
2132 		tsleep((caddr_t)vp, 0, "vop_revokeall", 0);
2133 		return (0);
2134 	}
2135 
2136 	/*
2137 	 * If the vnode has a device association, scrap all vnodes associated
2138 	 * with the device.  Don't let the device disappear on us while we
2139 	 * are scrapping the vnodes.
2140 	 */
2141 	if (vp->v_type != VCHR && vp->v_type != VBLK)
2142 		return(0);
2143 	if ((dev = vp->v_rdev) == NULL) {
2144 		if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV)
2145 			return(0);
2146 	}
2147 	reference_dev(dev);
2148 	for (;;) {
2149 		lwkt_gettoken(&ilock, &spechash_token);
2150 		vq = SLIST_FIRST(&dev->si_hlist);
2151 		lwkt_reltoken(&ilock);
2152 		if (vq == NULL)
2153 			break;
2154 		vgone(vq);
2155 	}
2156 	release_dev(dev);
2157 	return (0);
2158 }
2159 
2160 /*
2161  * Recycle an unused vnode to the front of the free list.
2162  * Release the passed interlock if the vnode will be recycled.
2163  */
2164 int
2165 vrecycle(struct vnode *vp, lwkt_tokref_t inter_lkp, struct thread *td)
2166 {
2167 	lwkt_tokref vlock;
2168 
2169 	lwkt_gettoken(&vlock, vp->v_interlock);
2170 	if (vp->v_usecount == 0) {
2171 		if (inter_lkp)
2172 			lwkt_reltoken(inter_lkp);
2173 		vgonel(vp, &vlock, td);
2174 		return (1);
2175 	}
2176 	lwkt_reltoken(&vlock);
2177 	return (0);
2178 }
2179 
2180 /*
2181  * Eliminate all activity associated with a vnode
2182  * in preparation for reuse.
2183  */
2184 void
2185 vgone(struct vnode *vp)
2186 {
2187 	struct thread *td = curthread;	/* XXX */
2188 	lwkt_tokref vlock;
2189 
2190 	lwkt_gettoken(&vlock, vp->v_interlock);
2191 	vgonel(vp, &vlock, td);
2192 }
2193 
2194 /*
2195  * vgone, with the vp interlock held.
2196  */
2197 void
2198 vgonel(struct vnode *vp, lwkt_tokref_t vlock, struct thread *td)
2199 {
2200 	lwkt_tokref ilock;
2201 	int s;
2202 
2203 	/*
2204 	 * If a vgone (or vclean) is already in progress,
2205 	 * wait until it is done and return.
2206 	 */
2207 	if (vp->v_flag & VXLOCK) {
2208 		vp->v_flag |= VXWANT;
2209 		lwkt_reltoken(vlock);
2210 		tsleep((caddr_t)vp, 0, "vgone", 0);
2211 		return;
2212 	}
2213 
2214 	/*
2215 	 * Clean out the filesystem specific data.
2216 	 */
2217 	vclean(vp, vlock, DOCLOSE, td);
2218 	lwkt_gettokref(vlock);
2219 
2220 	/*
2221 	 * Delete from old mount point vnode list, if on one.
2222 	 */
2223 	if (vp->v_mount != NULL)
2224 		insmntque(vp, (struct mount *)0);
2225 
2226 	/*
2227 	 * If special device, remove it from special device alias list
2228 	 * if it is on one.  This should normally only occur if a vnode is
2229 	 * being revoked as the device should otherwise have been released
2230 	 * naturally.
2231 	 */
2232 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
2233 		v_release_rdev(vp);
2234 	}
2235 
2236 	/*
2237 	 * If it is on the freelist and not already at the head,
2238 	 * move it to the head of the list. The test of the
2239 	 * VDOOMED flag and the reference count of zero is because
2240 	 * it will be removed from the free list by getnewvnode,
2241 	 * but will not have its reference count incremented until
2242 	 * after calling vgone. If the reference count were
2243 	 * incremented first, vgone would (incorrectly) try to
2244 	 * close the previous instance of the underlying object.
2245 	 */
2246 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2247 		s = splbio();
2248 		lwkt_gettoken(&ilock, &vnode_free_list_token);
2249 		if (vp->v_flag & VFREE)
2250 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2251 		else
2252 			freevnodes++;
2253 		vp->v_flag |= VFREE;
2254 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2255 		lwkt_reltoken(&ilock);
2256 		splx(s);
2257 	}
2258 	vp->v_type = VBAD;
2259 	lwkt_reltoken(vlock);
2260 }
2261 
2262 /*
2263  * Lookup a vnode by device number.
2264  */
2265 int
2266 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
2267 {
2268 	lwkt_tokref ilock;
2269 	struct vnode *vp;
2270 
2271 	lwkt_gettoken(&ilock, &spechash_token);
2272 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2273 		if (type == vp->v_type) {
2274 			*vpp = vp;
2275 			lwkt_reltoken(&ilock);
2276 			return (1);
2277 		}
2278 	}
2279 	lwkt_reltoken(&ilock);
2280 	return (0);
2281 }
2282 
2283 /*
2284  * Calculate the total number of references to a special device.  This
2285  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
2286  * an overloaded field.  Since udev2dev can now return NODEV, we have
2287  * to check for a NULL v_rdev.
2288  */
2289 int
2290 count_dev(dev_t dev)
2291 {
2292 	lwkt_tokref ilock;
2293 	struct vnode *vp;
2294 	int count = 0;
2295 
2296 	if (SLIST_FIRST(&dev->si_hlist)) {
2297 		lwkt_gettoken(&ilock, &spechash_token);
2298 		SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2299 			count += vp->v_usecount;
2300 		}
2301 		lwkt_reltoken(&ilock);
2302 	}
2303 	return(count);
2304 }
2305 
2306 int
2307 count_udev(udev_t udev)
2308 {
2309 	dev_t dev;
2310 
2311 	if ((dev = udev2dev(udev, 0)) == NODEV)
2312 		return(0);
2313 	return(count_dev(dev));
2314 }
2315 
2316 int
2317 vcount(struct vnode *vp)
2318 {
2319 	if (vp->v_rdev == NULL)
2320 		return(0);
2321 	return(count_dev(vp->v_rdev));
2322 }
2323 
2324 /*
2325  * Print out a description of a vnode.
2326  */
2327 static char *typename[] =
2328 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2329 
2330 void
2331 vprint(char *label, struct vnode *vp)
2332 {
2333 	char buf[96];
2334 
2335 	if (label != NULL)
2336 		printf("%s: %p: ", label, (void *)vp);
2337 	else
2338 		printf("%p: ", (void *)vp);
2339 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2340 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2341 	    vp->v_holdcnt);
2342 	buf[0] = '\0';
2343 	if (vp->v_flag & VROOT)
2344 		strcat(buf, "|VROOT");
2345 	if (vp->v_flag & VTEXT)
2346 		strcat(buf, "|VTEXT");
2347 	if (vp->v_flag & VSYSTEM)
2348 		strcat(buf, "|VSYSTEM");
2349 	if (vp->v_flag & VXLOCK)
2350 		strcat(buf, "|VXLOCK");
2351 	if (vp->v_flag & VXWANT)
2352 		strcat(buf, "|VXWANT");
2353 	if (vp->v_flag & VBWAIT)
2354 		strcat(buf, "|VBWAIT");
2355 	if (vp->v_flag & VDOOMED)
2356 		strcat(buf, "|VDOOMED");
2357 	if (vp->v_flag & VFREE)
2358 		strcat(buf, "|VFREE");
2359 	if (vp->v_flag & VOBJBUF)
2360 		strcat(buf, "|VOBJBUF");
2361 	if (buf[0] != '\0')
2362 		printf(" flags (%s)", &buf[1]);
2363 	if (vp->v_data == NULL) {
2364 		printf("\n");
2365 	} else {
2366 		printf("\n\t");
2367 		VOP_PRINT(vp);
2368 	}
2369 }
2370 
2371 #ifdef DDB
2372 #include <ddb/ddb.h>
2373 /*
2374  * List all of the locked vnodes in the system.
2375  * Called when debugging the kernel.
2376  */
2377 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2378 {
2379 	struct thread *td = curthread;	/* XXX */
2380 	lwkt_tokref ilock;
2381 	struct mount *mp, *nmp;
2382 	struct vnode *vp;
2383 
2384 	printf("Locked vnodes\n");
2385 	lwkt_gettoken(&ilock, &mountlist_token);
2386 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2387 		if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
2388 			nmp = TAILQ_NEXT(mp, mnt_list);
2389 			continue;
2390 		}
2391 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2392 			if (VOP_ISLOCKED(vp, NULL))
2393 				vprint((char *)0, vp);
2394 		}
2395 		lwkt_gettokref(&ilock);
2396 		nmp = TAILQ_NEXT(mp, mnt_list);
2397 		vfs_unbusy(mp, td);
2398 	}
2399 	lwkt_reltoken(&ilock);
2400 }
2401 #endif
2402 
2403 /*
2404  * Top level filesystem related information gathering.
2405  */
2406 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
2407 
2408 static int
2409 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2410 {
2411 	int *name = (int *)arg1 - 1;	/* XXX */
2412 	u_int namelen = arg2 + 1;	/* XXX */
2413 	struct vfsconf *vfsp;
2414 
2415 #if 1 || defined(COMPAT_PRELITE2)
2416 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2417 	if (namelen == 1)
2418 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2419 #endif
2420 
2421 #ifdef notyet
2422 	/* all sysctl names at this level are at least name and field */
2423 	if (namelen < 2)
2424 		return (ENOTDIR);		/* overloaded */
2425 	if (name[0] != VFS_GENERIC) {
2426 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2427 			if (vfsp->vfc_typenum == name[0])
2428 				break;
2429 		if (vfsp == NULL)
2430 			return (EOPNOTSUPP);
2431 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2432 		    oldp, oldlenp, newp, newlen, p));
2433 	}
2434 #endif
2435 	switch (name[1]) {
2436 	case VFS_MAXTYPENUM:
2437 		if (namelen != 2)
2438 			return (ENOTDIR);
2439 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2440 	case VFS_CONF:
2441 		if (namelen != 3)
2442 			return (ENOTDIR);	/* overloaded */
2443 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2444 			if (vfsp->vfc_typenum == name[2])
2445 				break;
2446 		if (vfsp == NULL)
2447 			return (EOPNOTSUPP);
2448 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2449 	}
2450 	return (EOPNOTSUPP);
2451 }
2452 
2453 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2454 	"Generic filesystem");
2455 
2456 #if 1 || defined(COMPAT_PRELITE2)
2457 
2458 static int
2459 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2460 {
2461 	int error;
2462 	struct vfsconf *vfsp;
2463 	struct ovfsconf ovfs;
2464 
2465 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2466 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2467 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2468 		ovfs.vfc_index = vfsp->vfc_typenum;
2469 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2470 		ovfs.vfc_flags = vfsp->vfc_flags;
2471 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2472 		if (error)
2473 			return error;
2474 	}
2475 	return 0;
2476 }
2477 
2478 #endif /* 1 || COMPAT_PRELITE2 */
2479 
2480 #if 0
2481 #define KINFO_VNODESLOP	10
2482 /*
2483  * Dump vnode list (via sysctl).
2484  * Copyout address of vnode followed by vnode.
2485  */
2486 /* ARGSUSED */
2487 static int
2488 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2489 {
2490 	struct proc *p = curproc;	/* XXX */
2491 	struct mount *mp, *nmp;
2492 	struct vnode *nvp, *vp;
2493 	lwkt_tokref ilock;
2494 	lwkt_tokref jlock;
2495 	int error;
2496 
2497 #define VPTRSZ	sizeof (struct vnode *)
2498 #define VNODESZ	sizeof (struct vnode)
2499 
2500 	req->lock = 0;
2501 	if (!req->oldptr) /* Make an estimate */
2502 		return (SYSCTL_OUT(req, 0,
2503 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2504 
2505 	lwkt_gettoken(&ilock, &mountlist_token);
2506 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2507 		if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) {
2508 			nmp = TAILQ_NEXT(mp, mnt_list);
2509 			continue;
2510 		}
2511 		lwkt_gettoken(&jlock, &mntvnode_token);
2512 again:
2513 		for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2514 		     vp != NULL;
2515 		     vp = nvp) {
2516 			/*
2517 			 * Check that the vp is still associated with
2518 			 * this filesystem.  RACE: could have been
2519 			 * recycled onto the same filesystem.
2520 			 */
2521 			if (vp->v_mount != mp)
2522 				goto again;
2523 			nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2524 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2525 			    (error = SYSCTL_OUT(req, vp, VNODESZ))) {
2526 				lwkt_reltoken(&jlock);
2527 				return (error);
2528 			}
2529 		}
2530 		lwkt_reltoken(&jlock);
2531 		lwkt_gettokref(&ilock);
2532 		nmp = TAILQ_NEXT(mp, mnt_list);	/* ZZZ */
2533 		vfs_unbusy(mp, p);
2534 	}
2535 	lwkt_reltoken(&ilock);
2536 
2537 	return (0);
2538 }
2539 #endif
2540 
2541 /*
2542  * XXX
2543  * Exporting the vnode list on large systems causes them to crash.
2544  * Exporting the vnode list on medium systems causes sysctl to coredump.
2545  */
2546 #if 0
2547 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2548 	0, 0, sysctl_vnode, "S,vnode", "");
2549 #endif
2550 
2551 /*
2552  * Check to see if a filesystem is mounted on a block device.
2553  */
2554 int
2555 vfs_mountedon(struct vnode *vp)
2556 {
2557 	dev_t dev;
2558 
2559 	if ((dev = vp->v_rdev) == NULL)
2560 		dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
2561 	if (dev != NODEV && dev->si_mountpoint)
2562 		return (EBUSY);
2563 	return (0);
2564 }
2565 
2566 /*
2567  * Unmount all filesystems. The list is traversed in reverse order
2568  * of mounting to avoid dependencies.
2569  */
2570 void
2571 vfs_unmountall(void)
2572 {
2573 	struct mount *mp;
2574 	struct thread *td = curthread;
2575 	int error;
2576 
2577 	if (td->td_proc == NULL)
2578 		td = initproc->p_thread;	/* XXX XXX use proc0 instead? */
2579 
2580 	/*
2581 	 * Since this only runs when rebooting, it is not interlocked.
2582 	 */
2583 	while(!TAILQ_EMPTY(&mountlist)) {
2584 		mp = TAILQ_LAST(&mountlist, mntlist);
2585 		error = dounmount(mp, MNT_FORCE, td);
2586 		if (error) {
2587 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2588 			printf("unmount of %s failed (",
2589 			    mp->mnt_stat.f_mntonname);
2590 			if (error == EBUSY)
2591 				printf("BUSY)\n");
2592 			else
2593 				printf("%d)\n", error);
2594 		} else {
2595 			/* The unmount has removed mp from the mountlist */
2596 		}
2597 	}
2598 }
2599 
2600 /*
2601  * Build hash lists of net addresses and hang them off the mount point.
2602  * Called by ufs_mount() to set up the lists of export addresses.
2603  */
2604 static int
2605 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
2606 		struct export_args *argp)
2607 {
2608 	struct netcred *np;
2609 	struct radix_node_head *rnh;
2610 	int i;
2611 	struct radix_node *rn;
2612 	struct sockaddr *saddr, *smask = 0;
2613 	struct domain *dom;
2614 	int error;
2615 
2616 	if (argp->ex_addrlen == 0) {
2617 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2618 			return (EPERM);
2619 		np = &nep->ne_defexported;
2620 		np->netc_exflags = argp->ex_flags;
2621 		np->netc_anon = argp->ex_anon;
2622 		np->netc_anon.cr_ref = 1;
2623 		mp->mnt_flag |= MNT_DEFEXPORTED;
2624 		return (0);
2625 	}
2626 
2627 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
2628 		return (EINVAL);
2629 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
2630 		return (EINVAL);
2631 
2632 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2633 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
2634 	bzero((caddr_t) np, i);
2635 	saddr = (struct sockaddr *) (np + 1);
2636 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2637 		goto out;
2638 	if (saddr->sa_len > argp->ex_addrlen)
2639 		saddr->sa_len = argp->ex_addrlen;
2640 	if (argp->ex_masklen) {
2641 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2642 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2643 		if (error)
2644 			goto out;
2645 		if (smask->sa_len > argp->ex_masklen)
2646 			smask->sa_len = argp->ex_masklen;
2647 	}
2648 	i = saddr->sa_family;
2649 	if ((rnh = nep->ne_rtable[i]) == 0) {
2650 		/*
2651 		 * Seems silly to initialize every AF when most are not used,
2652 		 * do so on demand here
2653 		 */
2654 		for (dom = domains; dom; dom = dom->dom_next)
2655 			if (dom->dom_family == i && dom->dom_rtattach) {
2656 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2657 				    dom->dom_rtoffset);
2658 				break;
2659 			}
2660 		if ((rnh = nep->ne_rtable[i]) == 0) {
2661 			error = ENOBUFS;
2662 			goto out;
2663 		}
2664 	}
2665 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2666 	    np->netc_rnodes);
2667 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2668 		error = EPERM;
2669 		goto out;
2670 	}
2671 	np->netc_exflags = argp->ex_flags;
2672 	np->netc_anon = argp->ex_anon;
2673 	np->netc_anon.cr_ref = 1;
2674 	return (0);
2675 out:
2676 	free(np, M_NETADDR);
2677 	return (error);
2678 }
2679 
2680 /* ARGSUSED */
2681 static int
2682 vfs_free_netcred(struct radix_node *rn, void *w)
2683 {
2684 	struct radix_node_head *rnh = (struct radix_node_head *) w;
2685 
2686 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2687 	free((caddr_t) rn, M_NETADDR);
2688 	return (0);
2689 }
2690 
2691 /*
2692  * Free the net address hash lists that are hanging off the mount points.
2693  */
2694 static void
2695 vfs_free_addrlist(struct netexport *nep)
2696 {
2697 	int i;
2698 	struct radix_node_head *rnh;
2699 
2700 	for (i = 0; i <= AF_MAX; i++)
2701 		if ((rnh = nep->ne_rtable[i])) {
2702 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2703 			    (caddr_t) rnh);
2704 			free((caddr_t) rnh, M_RTABLE);
2705 			nep->ne_rtable[i] = 0;
2706 		}
2707 }
2708 
2709 int
2710 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
2711 {
2712 	int error;
2713 
2714 	if (argp->ex_flags & MNT_DELEXPORT) {
2715 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2716 			vfs_setpublicfs(NULL, NULL, NULL);
2717 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2718 		}
2719 		vfs_free_addrlist(nep);
2720 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2721 	}
2722 	if (argp->ex_flags & MNT_EXPORTED) {
2723 		if (argp->ex_flags & MNT_EXPUBLIC) {
2724 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2725 				return (error);
2726 			mp->mnt_flag |= MNT_EXPUBLIC;
2727 		}
2728 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2729 			return (error);
2730 		mp->mnt_flag |= MNT_EXPORTED;
2731 	}
2732 	return (0);
2733 }
2734 
2735 
2736 /*
2737  * Set the publicly exported filesystem (WebNFS). Currently, only
2738  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2739  */
2740 int
2741 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
2742 		struct export_args *argp)
2743 {
2744 	int error;
2745 	struct vnode *rvp;
2746 	char *cp;
2747 
2748 	/*
2749 	 * mp == NULL -> invalidate the current info, the FS is
2750 	 * no longer exported. May be called from either vfs_export
2751 	 * or unmount, so check if it hasn't already been done.
2752 	 */
2753 	if (mp == NULL) {
2754 		if (nfs_pub.np_valid) {
2755 			nfs_pub.np_valid = 0;
2756 			if (nfs_pub.np_index != NULL) {
2757 				FREE(nfs_pub.np_index, M_TEMP);
2758 				nfs_pub.np_index = NULL;
2759 			}
2760 		}
2761 		return (0);
2762 	}
2763 
2764 	/*
2765 	 * Only one allowed at a time.
2766 	 */
2767 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2768 		return (EBUSY);
2769 
2770 	/*
2771 	 * Get real filehandle for root of exported FS.
2772 	 */
2773 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2774 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2775 
2776 	if ((error = VFS_ROOT(mp, &rvp)))
2777 		return (error);
2778 
2779 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2780 		return (error);
2781 
2782 	vput(rvp);
2783 
2784 	/*
2785 	 * If an indexfile was specified, pull it in.
2786 	 */
2787 	if (argp->ex_indexfile != NULL) {
2788 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2789 		    M_WAITOK);
2790 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2791 		    MAXNAMLEN, (size_t *)0);
2792 		if (!error) {
2793 			/*
2794 			 * Check for illegal filenames.
2795 			 */
2796 			for (cp = nfs_pub.np_index; *cp; cp++) {
2797 				if (*cp == '/') {
2798 					error = EINVAL;
2799 					break;
2800 				}
2801 			}
2802 		}
2803 		if (error) {
2804 			FREE(nfs_pub.np_index, M_TEMP);
2805 			return (error);
2806 		}
2807 	}
2808 
2809 	nfs_pub.np_mount = mp;
2810 	nfs_pub.np_valid = 1;
2811 	return (0);
2812 }
2813 
2814 struct netcred *
2815 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2816 		struct sockaddr *nam)
2817 {
2818 	struct netcred *np;
2819 	struct radix_node_head *rnh;
2820 	struct sockaddr *saddr;
2821 
2822 	np = NULL;
2823 	if (mp->mnt_flag & MNT_EXPORTED) {
2824 		/*
2825 		 * Lookup in the export list first.
2826 		 */
2827 		if (nam != NULL) {
2828 			saddr = nam;
2829 			rnh = nep->ne_rtable[saddr->sa_family];
2830 			if (rnh != NULL) {
2831 				np = (struct netcred *)
2832 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2833 							      rnh);
2834 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2835 					np = NULL;
2836 			}
2837 		}
2838 		/*
2839 		 * If no address match, use the default if it exists.
2840 		 */
2841 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2842 			np = &nep->ne_defexported;
2843 	}
2844 	return (np);
2845 }
2846 
2847 /*
2848  * perform msync on all vnodes under a mount point.  The mount point must
2849  * be locked.  This code is also responsible for lazy-freeing unreferenced
2850  * vnodes whos VM objects no longer contain pages.
2851  *
2852  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2853  */
2854 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2855 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2856 			    lwkt_tokref_t vlock, void *data);
2857 
2858 void
2859 vfs_msync(struct mount *mp, int flags)
2860 {
2861 	vmntvnodescan(mp, vfs_msync_scan1, vfs_msync_scan2, (void *)flags);
2862 }
2863 
2864 /*
2865  * scan1 is a fast pre-check.  There could be hundreds of thousands of
2866  * vnodes, we cannot afford to do anything heavy weight until we have a
2867  * fairly good indication that there is work to do.
2868  */
2869 static
2870 int
2871 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2872 {
2873 	int flags = (int)data;
2874 
2875 	if ((vp->v_flag & VXLOCK) == 0) {
2876 		if (VSHOULDFREE(vp))
2877 			return(0);
2878 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2879 		    (vp->v_flag & VOBJDIRTY) &&
2880 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2881 			return(0);
2882 		}
2883 	}
2884 	return(-1);
2885 }
2886 
2887 static
2888 int
2889 vfs_msync_scan2(struct mount *mp, struct vnode *vp,
2890 		lwkt_tokref_t vlock, void *data)
2891 {
2892 	vm_object_t obj;
2893 	int error;
2894 	int flags = (int)data;
2895 
2896 	if (vp->v_flag & VXLOCK)
2897 		return(0);
2898 
2899 	if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2900 	    (vp->v_flag & VOBJDIRTY) &&
2901 	    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2902 		error = vget(vp, vlock, LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ | LK_INTERLOCK, curthread);
2903 		if (error == 0) {
2904 			if (VOP_GETVOBJECT(vp, &obj) == 0) {
2905 				vm_object_page_clean(obj, 0, 0,
2906 				 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2907 			}
2908 			vput(vp);
2909 		}
2910 		return(0);
2911 	}
2912 	vmaybefree(vp);
2913 	lwkt_reltoken(vlock);
2914 	return(0);
2915 }
2916 
2917 /*
2918  * Create the VM object needed for VMIO and mmap support.  This
2919  * is done for all VREG files in the system.  Some filesystems might
2920  * afford the additional metadata buffering capability of the
2921  * VMIO code by making the device node be VMIO mode also.
2922  *
2923  * vp must be locked when vfs_object_create is called.
2924  */
2925 int
2926 vfs_object_create(struct vnode *vp, struct thread *td)
2927 {
2928 	return (VOP_CREATEVOBJECT(vp, td));
2929 }
2930 
2931 /*
2932  * NOTE: the vnode interlock must be held during the call.  We have to recheck
2933  * the VFREE flag since the vnode may have been removed from the free list
2934  * while we were blocked on vnode_free_list_token.  The use or hold count
2935  * must have already been bumped by the caller.
2936  */
2937 static void
2938 vbusy(struct vnode *vp)
2939 {
2940 	lwkt_tokref ilock;
2941 
2942 	lwkt_gettoken(&ilock, &vnode_free_list_token);
2943 	if ((vp->v_flag & VFREE) != 0) {
2944 	    TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2945 	    freevnodes--;
2946 	    vp->v_flag &= ~(VFREE|VAGE);
2947 	}
2948 	lwkt_reltoken(&ilock);
2949 }
2950 
2951 /*
2952  * NOTE: the vnode interlock must be held during the call.  The use or hold
2953  * count must have already been bumped by the caller.  We use a VINFREE to
2954  * interlock against other calls to vfree() which might occur while we
2955  * are blocked.  The vnode cannot be reused until it has actually been
2956  * placed on the free list, so there are no other races even though the
2957  * use and hold counts are 0.
2958  */
2959 static void
2960 vfree(struct vnode *vp)
2961 {
2962 	lwkt_tokref ilock;
2963 
2964 	if ((vp->v_flag & VINFREE) == 0) {
2965 		vp->v_flag |= VINFREE;
2966 		lwkt_gettoken(&ilock, &vnode_free_list_token); /* can block */
2967 		KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2968 		if (vp->v_flag & VAGE) {
2969 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2970 		} else {
2971 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2972 		}
2973 		freevnodes++;
2974 		vp->v_flag &= ~(VAGE|VINFREE);
2975 		vp->v_flag |= VFREE;
2976 		lwkt_reltoken(&ilock);	/* can block */
2977 	}
2978 }
2979 
2980 
2981 /*
2982  * Record a process's interest in events which might happen to
2983  * a vnode.  Because poll uses the historic select-style interface
2984  * internally, this routine serves as both the ``check for any
2985  * pending events'' and the ``record my interest in future events''
2986  * functions.  (These are done together, while the lock is held,
2987  * to avoid race conditions.)
2988  */
2989 int
2990 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
2991 {
2992 	lwkt_tokref ilock;
2993 
2994 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2995 	if (vp->v_pollinfo.vpi_revents & events) {
2996 		/*
2997 		 * This leaves events we are not interested
2998 		 * in available for the other process which
2999 		 * which presumably had requested them
3000 		 * (otherwise they would never have been
3001 		 * recorded).
3002 		 */
3003 		events &= vp->v_pollinfo.vpi_revents;
3004 		vp->v_pollinfo.vpi_revents &= ~events;
3005 
3006 		lwkt_reltoken(&ilock);
3007 		return events;
3008 	}
3009 	vp->v_pollinfo.vpi_events |= events;
3010 	selrecord(td, &vp->v_pollinfo.vpi_selinfo);
3011 	lwkt_reltoken(&ilock);
3012 	return 0;
3013 }
3014 
3015 /*
3016  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3017  * it is possible for us to miss an event due to race conditions, but
3018  * that condition is expected to be rare, so for the moment it is the
3019  * preferred interface.
3020  */
3021 void
3022 vn_pollevent(struct vnode *vp, int events)
3023 {
3024 	lwkt_tokref ilock;
3025 
3026 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3027 	if (vp->v_pollinfo.vpi_events & events) {
3028 		/*
3029 		 * We clear vpi_events so that we don't
3030 		 * call selwakeup() twice if two events are
3031 		 * posted before the polling process(es) is
3032 		 * awakened.  This also ensures that we take at
3033 		 * most one selwakeup() if the polling process
3034 		 * is no longer interested.  However, it does
3035 		 * mean that only one event can be noticed at
3036 		 * a time.  (Perhaps we should only clear those
3037 		 * event bits which we note?) XXX
3038 		 */
3039 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
3040 		vp->v_pollinfo.vpi_revents |= events;
3041 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
3042 	}
3043 	lwkt_reltoken(&ilock);
3044 }
3045 
3046 /*
3047  * Wake up anyone polling on vp because it is being revoked.
3048  * This depends on dead_poll() returning POLLHUP for correct
3049  * behavior.
3050  */
3051 void
3052 vn_pollgone(struct vnode *vp)
3053 {
3054 	lwkt_tokref ilock;
3055 
3056 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
3057 	if (vp->v_pollinfo.vpi_events) {
3058 		vp->v_pollinfo.vpi_events = 0;
3059 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
3060 	}
3061 	lwkt_reltoken(&ilock);
3062 }
3063 
3064 
3065 
3066 /*
3067  * Routine to create and manage a filesystem syncer vnode.
3068  */
3069 #define sync_close ((int (*) (struct  vop_close_args *))nullop)
3070 static int	sync_fsync (struct  vop_fsync_args *);
3071 static int	sync_inactive (struct  vop_inactive_args *);
3072 static int	sync_reclaim  (struct  vop_reclaim_args *);
3073 #define sync_lock ((int (*) (struct  vop_lock_args *))vop_nolock)
3074 #define sync_unlock ((int (*) (struct  vop_unlock_args *))vop_nounlock)
3075 static int	sync_print (struct vop_print_args *);
3076 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked)
3077 
3078 static vop_t **sync_vnodeop_p;
3079 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3080 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3081 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3082 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3083 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3084 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3085 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
3086 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
3087 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
3088 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
3089 	{ NULL, NULL }
3090 };
3091 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3092 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3093 
3094 VNODEOP_SET(sync_vnodeop_opv_desc);
3095 
3096 /*
3097  * Create a new filesystem syncer vnode for the specified mount point.
3098  * This vnode is placed on the worklist and is responsible for sync'ing
3099  * the filesystem.
3100  *
3101  * NOTE: read-only mounts are also placed on the worklist.  The filesystem
3102  * sync code is also responsible for cleaning up vnodes.
3103  */
3104 int
3105 vfs_allocate_syncvnode(struct mount *mp)
3106 {
3107 	struct vnode *vp;
3108 	static long start, incr, next;
3109 	int error;
3110 
3111 	/* Allocate a new vnode */
3112 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3113 		mp->mnt_syncer = NULL;
3114 		return (error);
3115 	}
3116 	vp->v_type = VNON;
3117 	/*
3118 	 * Place the vnode onto the syncer worklist. We attempt to
3119 	 * scatter them about on the list so that they will go off
3120 	 * at evenly distributed times even if all the filesystems
3121 	 * are mounted at once.
3122 	 */
3123 	next += incr;
3124 	if (next == 0 || next > syncer_maxdelay) {
3125 		start /= 2;
3126 		incr /= 2;
3127 		if (start == 0) {
3128 			start = syncer_maxdelay / 2;
3129 			incr = syncer_maxdelay;
3130 		}
3131 		next = start;
3132 	}
3133 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3134 	mp->mnt_syncer = vp;
3135 	return (0);
3136 }
3137 
3138 /*
3139  * Do a lazy sync of the filesystem.
3140  *
3141  * sync_fsync { struct vnode *a_vp, struct ucred *a_cred, int a_waitfor,
3142  *		struct thread *a_td }
3143  */
3144 static int
3145 sync_fsync(struct vop_fsync_args *ap)
3146 {
3147 	struct vnode *syncvp = ap->a_vp;
3148 	struct mount *mp = syncvp->v_mount;
3149 	struct thread *td = ap->a_td;
3150 	lwkt_tokref ilock;
3151 	int asyncflag;
3152 
3153 	/*
3154 	 * We only need to do something if this is a lazy evaluation.
3155 	 */
3156 	if (ap->a_waitfor != MNT_LAZY)
3157 		return (0);
3158 
3159 	/*
3160 	 * Move ourselves to the back of the sync list.
3161 	 */
3162 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3163 
3164 	/*
3165 	 * Walk the list of vnodes pushing all that are dirty and
3166 	 * not already on the sync list, and freeing vnodes which have
3167 	 * no refs and whos VM objects are empty.  vfs_msync() handles
3168 	 * the VM issues and must be called whether the mount is readonly
3169 	 * or not.
3170 	 */
3171 	lwkt_gettoken(&ilock, &mountlist_token);
3172 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &ilock, td) != 0) {
3173 		lwkt_reltoken(&ilock);
3174 		return (0);
3175 	}
3176 	if (mp->mnt_flag & MNT_RDONLY) {
3177 		vfs_msync(mp, MNT_NOWAIT);
3178 	} else {
3179 		asyncflag = mp->mnt_flag & MNT_ASYNC;
3180 		mp->mnt_flag &= ~MNT_ASYNC;	/* ZZZ hack */
3181 		vfs_msync(mp, MNT_NOWAIT);
3182 		VFS_SYNC(mp, MNT_LAZY, td);
3183 		if (asyncflag)
3184 			mp->mnt_flag |= MNT_ASYNC;
3185 	}
3186 	vfs_unbusy(mp, td);
3187 	return (0);
3188 }
3189 
3190 /*
3191  * The syncer vnode is no referenced.
3192  *
3193  * sync_inactive { struct vnode *a_vp, struct proc *a_p }
3194  */
3195 static int
3196 sync_inactive(struct vop_inactive_args *ap)
3197 {
3198 	vgone(ap->a_vp);
3199 	return (0);
3200 }
3201 
3202 /*
3203  * The syncer vnode is no longer needed and is being decommissioned.
3204  *
3205  * Modifications to the worklist must be protected at splbio().
3206  *
3207  *	sync_reclaim { struct vnode *a_vp }
3208  */
3209 static int
3210 sync_reclaim(struct vop_reclaim_args *ap)
3211 {
3212 	struct vnode *vp = ap->a_vp;
3213 	int s;
3214 
3215 	s = splbio();
3216 	vp->v_mount->mnt_syncer = NULL;
3217 	if (vp->v_flag & VONWORKLST) {
3218 		LIST_REMOVE(vp, v_synclist);
3219 		vp->v_flag &= ~VONWORKLST;
3220 	}
3221 	splx(s);
3222 
3223 	return (0);
3224 }
3225 
3226 /*
3227  * Print out a syncer vnode.
3228  *
3229  *	sync_print { struct vnode *a_vp }
3230  */
3231 static int
3232 sync_print(struct vop_print_args *ap)
3233 {
3234 	struct vnode *vp = ap->a_vp;
3235 
3236 	printf("syncer vnode");
3237 	if (vp->v_vnlock != NULL)
3238 		lockmgr_printinfo(vp->v_vnlock);
3239 	printf("\n");
3240 	return (0);
3241 }
3242 
3243 /*
3244  * extract the dev_t from a VBLK or VCHR.  The vnode must have been opened
3245  * (or v_rdev might be NULL).
3246  */
3247 dev_t
3248 vn_todev(struct vnode *vp)
3249 {
3250 	if (vp->v_type != VBLK && vp->v_type != VCHR)
3251 		return (NODEV);
3252 	KKASSERT(vp->v_rdev != NULL);
3253 	return (vp->v_rdev);
3254 }
3255 
3256 /*
3257  * Check if vnode represents a disk device.  The vnode does not need to be
3258  * opened.
3259  */
3260 int
3261 vn_isdisk(struct vnode *vp, int *errp)
3262 {
3263 	dev_t dev;
3264 
3265 	if (vp->v_type != VBLK && vp->v_type != VCHR) {
3266 		if (errp != NULL)
3267 			*errp = ENOTBLK;
3268 		return (0);
3269 	}
3270 
3271 	if ((dev = vp->v_rdev) == NULL)
3272 		dev = udev2dev(vp->v_udev, (vp->v_type == VBLK));
3273 	if (dev == NULL || dev == NODEV) {
3274 		if (errp != NULL)
3275 			*errp = ENXIO;
3276 		return (0);
3277 	}
3278 	if (dev_is_good(dev) == 0) {
3279 		if (errp != NULL)
3280 			*errp = ENXIO;
3281 		return (0);
3282 	}
3283 	if ((dev_dflags(dev) & D_DISK) == 0) {
3284 		if (errp != NULL)
3285 			*errp = ENOTBLK;
3286 		return (0);
3287 	}
3288 	if (errp != NULL)
3289 		*errp = 0;
3290 	return (1);
3291 }
3292 
3293 void
3294 NDFREE(struct nameidata *ndp, const uint flags)
3295 {
3296 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3297 	    (ndp->ni_cnd.cn_flags & CNP_HASBUF)) {
3298 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3299 		ndp->ni_cnd.cn_flags &= ~CNP_HASBUF;
3300 	}
3301 	if (!(flags & NDF_NO_DNCP_RELE) &&
3302 	    (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) &&
3303 	    ndp->ni_dncp) {
3304 		cache_drop(ndp->ni_dncp);
3305 		ndp->ni_dncp = NULL;
3306 	}
3307 	if (!(flags & NDF_NO_NCP_RELE) &&
3308 	    (ndp->ni_cnd.cn_flags & CNP_WANTNCP) &&
3309 	    ndp->ni_ncp) {
3310 		cache_drop(ndp->ni_ncp);
3311 		ndp->ni_ncp = NULL;
3312 	}
3313 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3314 	    (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) &&
3315 	    ndp->ni_dvp != ndp->ni_vp) {
3316 		VOP_UNLOCK(ndp->ni_dvp, NULL, 0, ndp->ni_cnd.cn_td);
3317 	}
3318 	if (!(flags & NDF_NO_DVP_RELE) &&
3319 	    (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) {
3320 		vrele(ndp->ni_dvp);
3321 		ndp->ni_dvp = NULL;
3322 	}
3323 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3324 	    (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) {
3325 		VOP_UNLOCK(ndp->ni_vp, NULL, 0, ndp->ni_cnd.cn_td);
3326 	}
3327 	if (!(flags & NDF_NO_VP_RELE) &&
3328 	    ndp->ni_vp) {
3329 		vrele(ndp->ni_vp);
3330 		ndp->ni_vp = NULL;
3331 	}
3332 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3333 	    (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) {
3334 		vrele(ndp->ni_startdir);
3335 		ndp->ni_startdir = NULL;
3336 	}
3337 }
3338 
3339 #ifdef DEBUG_VFS_LOCKS
3340 
3341 void
3342 assert_vop_locked(struct vnode *vp, const char *str)
3343 {
3344 	if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) {
3345 		panic("%s: %p is not locked shared but should be", str, vp);
3346 	}
3347 }
3348 
3349 void
3350 assert_vop_unlocked(struct vnode *vp, const char *str)
3351 {
3352 	if (vp && IS_LOCKING_VFS(vp)) {
3353 		if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) {
3354 			panic("%s: %p is locked but should not be", str, vp);
3355 		}
3356 	}
3357 }
3358 
3359 #endif
3360